Handle SIGTERM in display(1).

Display a final frame with a message explaining what is happening that is
displayed while the system powers off, reboots, halts, when the user logs
out, or just exits the compositor.

Makefile

@@ -20,13 +20,16 @@ display \
 dnsconfig \
 editor \
 ext \
+fat \
 games \
 hostname \
 ifconfig \
 init \
+iso9660 \
 kblayout \
 kblayout-compiler \
 login \
+nyan \
 ping \
 regress \
 rw \
@@ -68,10 +71,13 @@ else
   SORTIX_INCLUDE_SOURCE?=yes
 endif
 
+ISO_MOUNT?=no
+
 include build-aux/dirs.mak
 
 BUILD_NAME:=sortix-$(RELEASE)-$(MACHINE)
 
+CHAIN_INITRD:=$(SORTIX_BUILDS_DIR)/$(BUILD_NAME).chain.tar
 LIVE_INITRD:=$(SORTIX_BUILDS_DIR)/$(BUILD_NAME).live.tar
 OVERLAY_INITRD:=$(SORTIX_BUILDS_DIR)/$(BUILD_NAME).overlay.tar
 SRC_INITRD:=$(SORTIX_BUILDS_DIR)/$(BUILD_NAME).src.tar
@@ -226,11 +232,13 @@ sysroot-system: sysroot-fsh sysroot-base-headers
 	 echo 'ID=sortix' && \
 	 echo 'VERSION_ID="$(VERSION)"' && \
 	 echo 'PRETTY_NAME="Sortix $(VERSION)"' && \
-	 echo 'SORTIX_ABI=2.0' && \
+	 echo 'SORTIX_ABI=2.1' && \
 	 true) > "$(SYSROOT)/etc/sortix-release"
 	echo /etc/sortix-release >> "$(SYSROOT)/tix/manifest/system"
 	ln -sf sortix-release "$(SYSROOT)/etc/os-release"
 	echo /etc/os-release >> "$(SYSROOT)/tix/manifest/system"
+	find etc | sed -e 's,^,/,' >> "$(SYSROOT)/tix/manifest/system"
+	cp -RT etc "$(SYSROOT)/etc"
 	find share | sed -e 's,^,/,' >> "$(SYSROOT)/tix/manifest/system"
 	cp -RT share "$(SYSROOT)/share"
 	export SYSROOT="$(SYSROOT)" && \
@@ -269,6 +277,7 @@ else ifneq ($(SORTIX_INCLUDE_SOURCE),no)
 	cp Makefile -t "$(SYSROOT)/src"
 	cp README -t "$(SYSROOT)/src"
 	cp -RT build-aux "$(SYSROOT)/src/build-aux"
+	cp -RT etc "$(SYSROOT)/src/etc"
 	cp -RT share "$(SYSROOT)/src/share"
 	(for D in $(MODULES); do (cp -R $$D -t "$(SYSROOT)/src" && $(MAKE) -C "$(SYSROOT)/src/$$D" clean) || exit $$?; done)
 endif
@@ -441,6 +450,25 @@ release-all-archs:
 
 # Initial ramdisk
 
+$(CHAIN_INITRD).uuid:
+	mkdir -p `dirname $@`
+	uuidgen > $@
+
+$(CHAIN_INITRD): $(CHAIN_INITRD).uuid sysroot
+	mkdir -p `dirname $(CHAIN_INITRD)`
+	rm -rf $(CHAIN_INITRD).d
+	mkdir -p $(CHAIN_INITRD).d
+	mkdir -p $(CHAIN_INITRD).d/etc
+	echo "UUID=`cat $(CHAIN_INITRD).uuid` / iso9660 ro 0 1" > $(CHAIN_INITRD).d/etc/fstab
+	mkdir -p $(CHAIN_INITRD).d/etc/init
+	echo require chain exit-code > $(CHAIN_INITRD).d/etc/init/default
+	mkdir -p $(CHAIN_INITRD).d/sbin
+	cp "$(SYSROOT)/sbin/init" $(CHAIN_INITRD).d/sbin
+	cp "$(SYSROOT)/sbin/iso9660fs" $(CHAIN_INITRD).d/sbin
+	LC_ALL=C ls -A $(CHAIN_INITRD).d | \
+	tar -cf $(CHAIN_INITRD) -C $(CHAIN_INITRD).d --numeric-owner --owner=0 --group=0 -T -
+	rm -rf $(CHAIN_INITRD).d
+
 $(LIVE_INITRD): sysroot
 	mkdir -p `dirname $(LIVE_INITRD)`
 	rm -rf $(LIVE_INITRD).d
@@ -452,6 +480,9 @@ $(LIVE_INITRD): sysroot
 	echo "include /etc/default/passwd.d/*" >> $(LIVE_INITRD).d/etc/passwd
 	echo "root::0:root" > $(LIVE_INITRD).d/etc/group
 	echo "include /etc/default/group.d/*" >> $(LIVE_INITRD).d/etc/group
+	(echo 'channel = $(CHANNEL)' && \
+	 echo 'release_key = $(RELEASE_KEY)' && \
+	 echo 'release_sig_url = $(RELEASE_SIG_URL)') > $(LIVE_INITRD).d/etc/upgrade.conf
 	mkdir -p $(LIVE_INITRD).d/root -m 700
 	cp -RT "$(SYSROOT)/etc/skel" $(LIVE_INITRD).d/root
 	(echo "You can view the documentation for new users by typing:" && \
@@ -490,6 +521,42 @@ $(SORTIX_BUILDS_DIR):
 
 # Bootable images
 
+ifeq ($(ISO_MOUNT),yes)
+
+$(SORTIX_BUILDS_DIR)/$(BUILD_NAME).iso: sysroot $(CHAIN_INITRD) $(CHAIN_INITRD).uuid $(SORTIX_BUILDS_DIR)
+	rm -rf $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
+	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
+	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/boot
+	cp "$(SYSROOT)/boot/sortix.bin" $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/boot/sortix.bin
+	cp $(CHAIN_INITRD) $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/boot/sortix.initrd
+	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc
+	echo "UUID=`cat $(CHAIN_INITRD).uuid` / iso9660 ro 0 1" > $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/fstab
+	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/init
+	echo require single-user exit-code > $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/init/default
+	echo "root::0:0:root:/root:sh" > $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/passwd
+	echo "include /etc/default/passwd.d/*" >> $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/passwd
+	echo "root::0:root" > $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/group
+	echo "include /etc/default/group.d/*" >> $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/group
+	(echo 'channel = $(CHANNEL)' && \
+	 echo 'release_key = $(RELEASE_KEY)' && \
+	 echo 'release_sig_url = $(RELEASE_SIG_URL)') > $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/etc/upgrade.conf
+	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/root -m 700
+	cp -RT "$(SYSROOT)/etc/skel" $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/root
+	(echo "You can view the documentation for new users by typing:" && \
+	 echo && \
+	 echo "  man user-guide" && \
+	 echo && \
+	 echo "You can view the installation instructions by typing:" && \
+	 echo && \
+	 echo "  man installation") > $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/root/welcome
+	if [ -d "$(SYSROOT_OVERLAY)" ]; then cp -RT "$(SYSROOT_OVERLAY)" $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso; fi
+	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso/boot/grub
+	build-aux/iso-grub-cfg.sh --platform $(HOST) --version $(VERSION) --mount $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
+	grub-mkrescue -o $(SORTIX_BUILDS_DIR)/$(BUILD_NAME).iso sysroot $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso -- -volid Sortix -volset_id `cat $(CHAIN_INITRD).uuid`
+	rm -rf $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
+
+else
+
 $(SORTIX_BUILDS_DIR)/$(BUILD_NAME).iso: sysroot $(LIVE_INITRD) $(OVERLAY_INITRD) $(SRC_INITRD) $(SYSTEM_INITRD) $(SORTIX_BUILDS_DIR)
 	rm -rf $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
 	mkdir -p $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
@@ -530,6 +597,8 @@ else # none
 endif
 	rm -rf $(SORTIX_BUILDS_DIR)/$(BUILD_NAME)-iso
 
+endif
+
 .PHONY: iso
 iso: $(SORTIX_BUILDS_DIR)/$(BUILD_NAME).iso
 
@@ -618,6 +687,7 @@ release-repository: sysroot $(SYSTEM_INITRD) $(SORTIX_RELEASE_DIR)/$(RELEASE)/re
 	  cp $(SORTIX_REPOSITORY_DIR)/$(HOST)/$$port.tix.tar.xz $(SORTIX_RELEASE_DIR)/$(RELEASE)/repository/$(HOST) && \
 	  cp $(SORTIX_REPOSITORY_DIR)/$(HOST)/$$port.version $(SORTIX_RELEASE_DIR)/$(RELEASE)/repository/$(HOST); \
 	done
+	tix-repository --generation=3 $(SORTIX_RELEASE_DIR)/$(RELEASE)/repository/$(HOST)
 
 .PHONY: release-scripts
 release-scripts: \

build-aux/compiler.mak

@@ -124,9 +124,9 @@ endif
 DEFAULT_GENERIC_OPTLEVEL_BASE:=-Os -s
 DEFAULT_OPTLEVEL_FOR_BUILD:=$(DEFAULT_GENERIC_OPTLEVEL_BASE)
 ifeq ($(BUILD_IS_SORTIX),1)
-  DEFAULT_OPTLEVEL_FOR_BUILD+=
+  DEFAULT_OPTLEVEL_FOR_BUILD+=-fsanitize=undefined -fstack-protector-all
 endif
 DEFAULT_OPTLEVEL:=$(DEFAULT_GENERIC_OPTLEVEL_BASE)
 ifeq ($(HOST_IS_SORTIX),1)
-  DEFAULT_OPTLEVEL+=
+  DEFAULT_OPTLEVEL+=-fsanitize=undefined -fstack-protector-all
 endif

build-aux/iso-grub-cfg.sh

@@ -23,8 +23,9 @@ set -e
 this=$(which -- "$0")
 thisdir=$(dirname -- "$this")
 
-platform=
 directory=
+mount=false
+platform=
 version=
 
 dashdash=
@@ -44,8 +45,9 @@ for argument do
 
   case $dashdash$argument in
   --) dashdash=yes ;;
-  --platform=*) platform=$parameter ;;
   --platform) previous_option=platform ;;
+  --platform=*) platform=$parameter ;;
+  --mount) mount=true ;;
   --version=*) version=$parameter ;;
   --version) previous_option=version ;;
   -*) echo "$0: unrecognized option $argument" >&2
@@ -111,14 +113,24 @@ isinset() {
 cd "$directory"
 
 kernel=$(maybe_compressed boot/sortix.bin)
-live_initrd=$(maybe_compressed boot/live.tar)
-overlay_initrd=$(maybe_compressed boot/overlay.tar)
-src_initrd=$(maybe_compressed boot/src.tar)
-system_initrd=$(maybe_compressed repository/system.tix.tar)
-ports=$(ls repository |
-        grep -E '\.tix\.tar\.xz$' |
-        sed -E 's/\.tix\.tar\.xz$//' |
-        (grep -Ev '^system$' || true))
+if $mount; then
+  initrd=$(maybe_compressed boot/sortix.initrd)
+  initrds=$initrd
+else
+  live_initrd=$(maybe_compressed boot/live.tar)
+  overlay_initrd=$(maybe_compressed boot/overlay.tar)
+  src_initrd=$(maybe_compressed boot/src.tar)
+  system_initrd=$(maybe_compressed repository/system.tix.tar)
+  initrds="$system_initrd $src_initrd $live_initrd $overlay_initrd"
+fi
+if $mount; then
+  ports=
+else
+  ports=$(ls repository |
+          grep -E '\.tix\.tar\.xz$' |
+          sed -E 's/\.tix\.tar\.xz$//' |
+          (grep -Ev '^system$' || true))
+fi
 
 mkdir -p boot/grub
 mkdir -p boot/grub/init
@@ -184,6 +196,12 @@ fi
 
 set version="$version"
 set machine="$machine"
+set mount=$mount
+if \$mount; then
+  chain='-- /sbin/init --target=chain'
+else
+  chain=
+fi
 set base_menu_title="Sortix \$version for \$machine"
 set menu_title="\$base_menu_title"
 set title_single_user='live environment'
@@ -205,6 +223,8 @@ set enable_sshd=false
 
 export version
 export machine
+export mount
+export chain
 export base_menu_title
 export menu_title
 export title_single_user
@@ -305,9 +325,10 @@ esac
 cat << EOF
   hook_kernel_pre
   echo -n "Loading /$kernel ($(human_size $kernel)) ... "
-  multiboot /$kernel \$no_random_seed \$enable_network_drivers "\$@"
+  multiboot /$kernel --firmware=\$grub_platform \$no_random_seed \$enable_network_drivers \$chain "\$@"
   echo done
   hook_kernel_post
+  # TODO: Injecting configuration doesn't work for mounted cdroms.
   if ! \$enable_dhclient; then
     echo -n "Disabling dhclient ... "
     module /boot/grub/init/furthermore --create-to /etc/init/network
@@ -332,7 +353,7 @@ cat << EOF
     echo done
   fi
 EOF
-for initrd in $system_initrd $src_initrd $live_initrd $overlay_initrd; do
+for initrd in $initrds; do
   if [ "$initrd" = "$src_initrd" ]; then
     cat << EOF
   if \$enable_src; then
@@ -428,9 +449,11 @@ menuentry "\$title_sysupgrade" '-- /sbin/init --target=sysupgrade'
 
 cat << EOF
 
+if ! $mount; then
 menuentry "Select ports..." {
   configfile /boot/grub/ports.cfg
 }
+fi
 
 menuentry "Advanced..." {
   configfile /boot/grub/advanced.cfg
@@ -462,6 +485,7 @@ else
   }
 fi
 
+if ! $mount; then
 if "\$enable_src"; then
   menuentry "Disable loading source code" {
     enable_src=false
@@ -473,6 +497,7 @@ else
     configfile /boot/grub/advanced.cfg
   }
 fi
+fi
 
 if [ "\$enable_network_drivers" = --disable-network-drivers ]; then
   menuentry "Enable networking drivers" {
@@ -486,6 +511,7 @@ else
   }
 fi
 
+if ! $mount; then
 if \$enable_dhclient; then
   menuentry "Disable DHCP client" {
     enable_dhclient=false
@@ -525,6 +551,7 @@ fi
 menuentry "Select binary packages..." {
   configfile /boot/grub/tix.cfg
 }
+fi
 
 hook_advanced_menu_post
 EOF

build-aux/ports.conf

@@ -1,3 +1,3 @@
-set_minimal="cut dash e2fsprogs grep grub mdocml sed xargs"
-set_basic="$set_minimal binutils bison bzip2 diffutils ed flex gawk gcc git gzip irssi libcurl libcurses libssl libstdc++ m4 make nano ntpd patch perl pkg-config python ssh tar texinfo vim wget xz xorriso"
+set_minimal="cut dash dosfstools e2fsprogs grep grub grub-i386-pc grub-i386-efi grub-x86_64-efi libssl mdocml sed signify tar wget xargs xz"
+set_basic="$set_minimal binutils bison bzip2 diffutils ed flex gawk gcc git gzip irssi libcurl libcurses libstdc++ m4 make mtools nano ntpd patch perl pkg-config python ssh texinfo vim xorriso"
 sets="basic minimal"

build-aux/version.mak

@@ -1,2 +1,6 @@
 VERSION=1.1dev
+CHANNEL?=volatile
 RELEASE?=$(VERSION)
+RELEASE_MASTER?=https://sortix.org/release
+RELEASE_KEY=/etc/signify/sortix-$(VERSION).pub
+RELEASE_SIG_URL?=$(RELEASE_MASTER)/$(CHANNEL)/$(HOST_MACHINE).sh.sig

checksum/Makefile

@@ -1,4 +1,3 @@
-SOFTWARE_MEANT_FOR_SORTIX=1
 include ../build-aux/platform.mak
 include ../build-aux/compiler.mak
 include ../build-aux/version.mak

checksum/checksum.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2020 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2017, 2020, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -17,6 +17,8 @@
  * Compute and check cryptographic hashes.
  */
 
+#include <sys/stat.h>
+
 #include <err.h>
 #include <errno.h>
 #include <fcntl.h>
@@ -97,8 +99,20 @@ static struct hash* hashes[] =
 	NULL,
 };
 
+struct checklist
+{
+	const char* file;
+	uint8_t checksum[DIGEST_MAX_LENGTH];
+	bool initialized;
+};
+
+static struct checklist** cache = NULL;
+static size_t cache_used = 0;
+static size_t cache_length = 0;
+static struct timespec cache_time;
 static struct hash* hash = NULL;
 static const char* algorithm = NULL;
+static const char* cache_path = NULL;
 static const char* checklist = NULL;
 static bool check = false;
 static bool ignore_missing = false;
@@ -116,19 +130,182 @@ int debase(char c)
 	return -1;
 }
 
-static void printhex(const uint8_t* buffer, size_t size)
+static void fprinthex(FILE* fp, const uint8_t* buffer, size_t size)
 {
 	for ( size_t i = 0; i < size; i++ )
 	{
-		putchar(hexchars[buffer[i] >> 4]);
-		putchar(hexchars[buffer[i] & 0xF]);
+		fputc(hexchars[buffer[i] >> 4], fp);
+		fputc(hexchars[buffer[i] & 0xF], fp);
 	}
 }
 
+static int compare_checklist_file(const void* a_ptr, const void* b_ptr)
+{
+	struct checklist* a = *(struct checklist**) a_ptr;
+	struct checklist* b = *(struct checklist**) b_ptr;
+	return strcmp(a->file, b->file);
+}
+
+static int search_checklist_file(const void* file_ptr, const void* elem_ptr)
+{
+	const char* file = (const char*) file_ptr;
+	struct checklist* elem = *(struct checklist**) elem_ptr;
+	return strcmp(file, elem->file);
+}
+
+static struct checklist* checklist_lookup(struct checklist** sorted,
+                                          size_t used,
+                                          const char* file)
+{
+	struct checklist** entry_ptr =
+		bsearch(file, sorted, used,
+                sizeof(struct checksum*), search_checklist_file);
+	return entry_ptr ? *entry_ptr : NULL;
+}
+
+static void checklist_add(struct checklist*** checklist_ptr,
+                          size_t* used_ptr,
+                          size_t* length_ptr,
+                          uint8_t digest[DIGEST_MAX_LENGTH],
+                          const char* file)
+{
+	struct checklist* entry = calloc(1, sizeof(struct checklist));
+	if ( !entry || !(entry->file = strdup(file)) )
+		err(1, "malloc");
+	memcpy(entry->checksum, digest, hash->digest_size);
+	if ( *used_ptr == *length_ptr )
+	{
+		struct checklist** new_checklist =
+			reallocarray(*checklist_ptr, *length_ptr,
+			             2 * sizeof(struct checklist*));
+		if ( !new_checklist )
+			err(1, "malloc");
+		*checklist_ptr = new_checklist;
+		*length_ptr *= 2;
+	}
+	(*checklist_ptr)[(*used_ptr)++] = entry;
+}
+
+static void  checklist_parse(char* line,
+                             size_t line_length,
+                             struct checklist* entry,
+                             const char* path,
+                             off_t line_number)
+{
+	if ( line[line_length - 1] != '\n' )
+		errx(1, "%s:%ji: Line was not terminated with a newline",
+		     path, (intmax_t) line_number);
+	line[--line_length] = '\0';
+	if ( (size_t) line_length < 2 * hash->digest_size )
+		errx(1, "%s:%ji: Improperly formatted %s checksum line",
+		     path, (intmax_t) line_number, hash->name);
+	for ( size_t i = 0; i < hash->digest_size; i++ )
+	{
+		int higher = debase(line[i*2 + 0]);
+		int lower = debase(line[i*2 + 1]);
+		if ( higher == -1 || lower == -1 )
+			errx(1, "%s:%ji: Improperly formatted %s checksum line",
+			     path, (intmax_t) line_number, hash->name);
+		entry->checksum[i] = higher << 4 | lower;
+	}
+	if ( line[2 * hash->digest_size + 0] != ' ' ||
+	     line[2 * hash->digest_size + 1] != ' ' ||
+	     line[2 * hash->digest_size + 2] == '\0' )
+		errx(1, "%s:%ji: Improperly formatted %s checksum line",
+		     path, (intmax_t) line_number, hash->name);
+	entry->file = line + 2 * hash->digest_size + 2;
+	if ( !strcmp(path, "-") && !strcmp(entry->file, "-") )
+		errx(1, "%s:%ji: Improperly formatted %s checksum line",
+		     path, (intmax_t) line_number, hash->name);
+}
+
+static void checklist_read(struct checklist*** ptr,
+                           size_t* used_ptr,
+                           size_t* length_ptr,
+                           struct timespec* time_ptr,
+                           const char* path,
+                           bool allow_missing)
+{
+	if ( !(*ptr = malloc(sizeof(struct checklist*) * 4)) )
+		err(1, "malloc");
+	*used_ptr = 0;
+	*length_ptr = 4;
+	FILE* fp = fopen(path, "r");
+	if ( !fp )
+	{
+		if ( allow_missing )
+			return;
+		err(1, "malloc");
+	}
+	struct stat st;
+	fstat(fileno(fp), &st);
+	*time_ptr = st.st_mtim;
+	char* line = NULL;
+	size_t line_size = 0;
+	ssize_t line_length;
+	off_t line_number = 0;
+	struct checklist input;
+	while ( 0 < (line_length = getline(&line, &line_size, fp)) )
+	{
+		line_number++;
+		checklist_parse(line, (size_t) line_length, &input, path, line_number);
+		checklist_add(ptr, used_ptr, length_ptr, input.checksum, input.file);
+	}
+	free(line);
+	if ( ferror(fp) )
+		err(1, "%s", path);
+	fclose(fp);
+	qsort(*ptr, *used_ptr, sizeof(struct checklist*), compare_checklist_file);
+}
+
+static void checklist_write(struct checklist** checklist,
+                            size_t used,
+                            const char* path)
+{
+	char* out_path;
+	if ( asprintf(&out_path, "%s.XXXXXX", path) < 0 )
+		err(1, "malloc");
+	int out_fd = mkstemp(out_path);
+	if ( out_fd < 0 )
+		err(1, "mkstemp: %s.XXXXXX", path);
+	FILE* out = fdopen(out_fd, "w");
+	if ( !out )
+	{
+		unlink(out_path);
+		err(1, "fdopen");
+	}
+	for ( size_t i = 0; i < used; i++ )
+	{
+		fprinthex(out, checklist[i]->checksum, hash->digest_size);
+		fprintf(out, "  %s\n", checklist[i]->file);
+	}
+	if ( ferror(out) || fclose(out) == EOF )
+	{
+		unlink(out_path);
+		err(1, "%s", out_path);
+	}
+	if ( rename(out_path, path) < 0 )
+		err(1, "rename: %s -> %s", out_path, path);
+	free(out_path);
+}
+
 static int digest_fd(uint8_t digest[DIGEST_MAX_LENGTH],
                      int fd,
                      const char* path)
 {
+	struct checklist* entry = NULL;
+	if ( cache && (entry = checklist_lookup(cache, cache_used, path)) )
+	{
+		struct stat st;
+		fstat(fd, &st);
+		if ( st.st_mtim.tv_sec < cache_time.tv_sec ||
+		     (st.st_mtim.tv_sec == cache_time.tv_sec &&
+		      st.st_mtim.tv_nsec <= cache_time.tv_nsec) )
+		{
+			memcpy(digest, entry->checksum, hash->digest_size);
+			return 0;
+		}
+	}
 	union ctx ctx;
 	hash->init(&ctx);
 	ssize_t amount;
@@ -140,6 +317,22 @@ static int digest_fd(uint8_t digest[DIGEST_MAX_LENGTH],
 		return 1;
 	}
 	hash->final(digest, &ctx);
+	if ( cache )
+	{
+		if ( entry )
+			memcpy(entry->checksum, digest, hash->digest_size);
+		else
+		{
+			checklist_add(&cache, &cache_used, &cache_length, digest, path);
+			size_t i = cache_used - 1;
+			while ( i && 0 < strcmp(cache[i - 1]->file, cache[i]->file) )
+			{
+				struct checklist* t = cache[i - 1];
+				cache[i - 1] = cache[i];
+				cache[i--] = t;
+			}
+		}
+	}
 	return 0;
 }
 
@@ -175,27 +368,6 @@ static int verify_path(uint8_t checksum[], const char* path)
 	return status;
 }
 
-struct checklist
-{
-	const char* file;
-	uint8_t checksum[DIGEST_MAX_LENGTH];
-	bool initialized;
-};
-
-static int compare_checklist_file(const void* a_ptr, const void* b_ptr)
-{
-	struct checklist* a = *(struct checklist**) a_ptr;
-	struct checklist* b = *(struct checklist**) b_ptr;
-	return strcmp(a->file, b->file);
-}
-
-static int search_checklist_file(const void* file_ptr, const void* elem_ptr)
-{
-	const char* file = (const char*) file_ptr;
-	struct checklist* elem = *(struct checklist**) elem_ptr;
-	return strcmp(file, elem->file);
-}
-
 static int checklist_fp(FILE* fp,
                         const char* path,
                         size_t files_count,
@@ -217,7 +389,6 @@ static int checklist_fp(FILE* fp,
 		qsort(checklist_sorted, files_count, sizeof(struct checklist*),
 		      compare_checklist_file);
 	}
-	uint8_t checksum[DIGEST_MAX_LENGTH];
 	bool any = false;
 	char* line = NULL;
 	size_t line_size = 0;
@@ -225,52 +396,29 @@ static int checklist_fp(FILE* fp,
 	off_t line_number = 0;
 	size_t read_failures = 0;
 	size_t check_failures = 0;
+	struct checklist input;
 	while ( 0 < (line_length = getline(&line, &line_size, fp)) )
 	{
 		line_number++;
-		if ( line[line_length - 1] != '\n' )
-			errx(1, "%s:%ji: Line was not terminated with a newline",
-			     path, (intmax_t) line_number);
-		line[--line_length] = '\0';
-		if ( (size_t) line_length < 2 * hash->digest_size )
-			errx(1, "%s:%ji: Improperly formatted %s checksum line",
-			     path, (intmax_t) line_number, hash->name);
-		for ( size_t i = 0; i < hash->digest_size; i++ )
-		{
-			int higher = debase(line[i*2 + 0]);
-			int lower = debase(line[i*2 + 1]);
-			if ( higher == -1 || lower == -1 )
-				errx(1, "%s:%ji: Improperly formatted %s checksum line",
-				     path, (intmax_t) line_number, hash->name);
-			checksum[i] = higher << 4 | lower;
-		}
-		if ( line[2 * hash->digest_size + 0] != ' ' ||
-		     line[2 * hash->digest_size + 1] != ' ' ||
-		     line[2 * hash->digest_size + 2] == '\0' )
-			errx(1, "%s:%ji: Improperly formatted %s checksum line",
-			     path, (intmax_t) line_number, hash->name);
-		const char* file = line + 2 * hash->digest_size + 2;
-		if ( !strcmp(path, "-") && !strcmp(file, "-") )
-			errx(1, "%s:%ji: Improperly formatted %s checksum line",
-			     path, (intmax_t) line_number, hash->name);
+		checklist_parse(line, (size_t) line_length, &input, path, line_number);
 		if ( files )
 		{
 			struct checklist** entry_ptr =
-				bsearch(file, checklist_sorted, files_count,
+				bsearch(input.file, checklist_sorted, files_count,
                         sizeof(struct checksum*), search_checklist_file);
 			if ( entry_ptr )
 			{
 				struct checklist* entry = *entry_ptr;
 				if ( entry->initialized )
 					errx(1, "%s:%ji: Duplicate hash found for: %s", path,
-					     (intmax_t) line_number, file);
-				memcpy(entry->checksum, checksum, DIGEST_MAX_LENGTH);
+					     (intmax_t) line_number, input.file);
+				memcpy(entry->checksum, input.checksum, DIGEST_MAX_LENGTH);
 				entry->initialized = true;
 			}
 		}
 		else
 		{
-			int status = verify_path(checksum, file);
+			int status = verify_path(input.checksum, input.file);
 			if ( status == 1 )
 				read_failures++;
 			else if ( status == 2 )
@@ -296,7 +444,7 @@ static int checklist_fp(FILE* fp,
 		else if ( status == 2 )
 			check_failures++;
 	}
-	explicit_bzero(checksum, sizeof(checksum));
+	explicit_bzero(input.checksum, sizeof(input.checksum));
 	free(checklist);
 	free(checklist_sorted);
 	if ( read_failures )
@@ -391,6 +539,15 @@ int main(int argc, char* argv[])
 		}
 		else if ( !strncmp(arg, "--algorithm=", strlen("--algorithm=")) )
 			algorithm = arg + strlen("--algorithm=");
+		else if ( !strcmp(arg, "--cache") )
+		{
+			if ( i + 1 == argc )
+				errx(1, "option '--cache' requires an argument");
+			cache_path = argv[i+1];
+			argv[++i] = NULL;
+		}
+		else if ( !strncmp(arg, "--cache=", strlen("--cache=")) )
+			cache_path = arg + strlen("--cache=");
 		else if ( !strcmp(arg, "--check") )
 			check = true;
 		else if ( !strcmp(arg, "--checklist") )
@@ -439,6 +596,14 @@ int main(int argc, char* argv[])
 	else
 		errx(1, "No hash algorithm was specified with -a");
 
+	if ( cache_path )
+	{
+		if ( !strcmp(cache_path, "-") )
+			errx(1, "cache cannot be the standard input");
+		checklist_read(&cache, &cache_used, &cache_length, &cache_time,
+		               cache_path, true);
+	}
+
 	bool read_failures = false;
 	bool check_failures = false;
 
@@ -464,10 +629,10 @@ int main(int argc, char* argv[])
 		else
 		{
 			uint8_t digest[DIGEST_MAX_LENGTH];
-			int result = digest_fd(digest, 0, "-");
+			int result = digest_path(digest, "-");
 			if ( result == 0 )
 			{
-				printhex(digest, hash->digest_size);
+				fprinthex(stdout, digest, hash->digest_size);
 				puts("  -");
 				explicit_bzero(digest, sizeof(digest));
 			}
@@ -491,7 +656,7 @@ int main(int argc, char* argv[])
 			int result = digest_path(digest, argv[i]);
 			if ( result == 0 )
 			{
-				printhex(digest, hash->digest_size);
+				fprinthex(stdout, digest, hash->digest_size);
 				printf("  %s\n", argv[i]);
 				explicit_bzero(digest, sizeof(digest));
 			}
@@ -502,5 +667,9 @@ int main(int argc, char* argv[])
 
 	if ( ferror(stdout) || fflush(stdout) == EOF )
 		return 1;
+
+	if ( cache_path)
+	     checklist_write(cache, cache_used, cache_path);
+
 	return read_failures ? 1 : check_failures ? 2 : 0;
 }

checksum/compat/sha2.h

@@ -0,0 +1,14 @@
+#ifndef SHA2_H
+#define SHA2_H
+#if __has_include_next(<sha2.h>)
+#include_next <sha2.h>
+#else
+#include "../../libc/include/sha2.h"
+#include "../../libc/sha2/sha224.c"
+#include "../../libc/sha2/sha256.c"
+#include "../../libc/sha2/sha384.c"
+#include "../../libc/sha2/sha512_256.c"
+#include "../../libc/sha2/sha512.c"
+#define timingsafe_memcmp memcmp
+#endif
+#endif

disked/disked.c

@@ -49,6 +49,7 @@
 #include <mount/blockdevice.h>
 #include <mount/devices.h>
 #include <mount/ext2.h>
+#include <mount/fat.h>
 #include <mount/filesystem.h>
 #include <mount/gpt.h>
 #include <mount/harddisk.h>
@@ -1496,7 +1497,12 @@ static char* display_area_format(void* ctx, size_t row, size_t column)
 		{
 		case FILESYSTEM_ERROR_NONE: return strdup("(no error)");
 		case FILESYSTEM_ERROR_ABSENT: return strdup("none");
-		case FILESYSTEM_ERROR_UNRECOGNIZED: return strdup("unrecognized");
+		case FILESYSTEM_ERROR_UNRECOGNIZED:
+		{
+			char uuid[UUID_STRING_LENGTH + 1];
+			uuid_to_string(area->p->gpt_type_guid, uuid);
+			return strdup(uuid);
+		}
 		case FILESYSTEM_ERROR_ERRNO: return strdup("(error)");
 		}
 		return strdup("(unknown error)");
@@ -1771,9 +1777,9 @@ static void on_mkpart(size_t argc, char** argv)
 		bool is_gpt = current_pt_type == PARTITION_TABLE_TYPE_GPT;
 		const char* question = "Format a filesystem? (no/ext2)";
 		if ( is_mbr )
-			question = "Format a filesystem? (no/ext2/extended)";
+			question = "Format a filesystem? (no/ext2/extended/fat)";
 		else if ( is_gpt )
-			question = "Format a filesystem? (no/ext2/biosboot)";
+			question = "Format a filesystem? (no/ext2/biosboot/efi/fat)";
 		if ( 5 <= argc )
 			strlcpy(fstype, argv[4], sizeof(fstype));
 		else
@@ -1781,7 +1787,9 @@ static void on_mkpart(size_t argc, char** argv)
 		if ( strcmp(fstype, "no") != 0 &&
 		     strcmp(fstype, "ext2") != 0 &&
 		     (!is_mbr || strcmp(fstype, "extended") != 0) &&
-		     (!is_gpt || strcmp(fstype, "biosboot") != 0) )
+		     (!is_gpt || strcmp(fstype, "biosboot") != 0) &&
+		     (!is_gpt || strcmp(fstype, "efi") != 0) &&
+		     strcmp(fstype, "fat") != 0 )
 		{
 			command_errorx("%s: %s: Invalid filesystem choice: %s",
 			               argv[0], device_name(current_hd->path), fstype);
@@ -1801,14 +1809,18 @@ static void on_mkpart(size_t argc, char** argv)
 		break;
 	}
 	char mountpoint[256] = "";
-	bool mountable = !strcmp(fstype, "ext2");
+	// TODO: Get this information from libmount.
+	bool mountable = !strcmp(fstype, "ext2") ||
+	                 !strcmp(fstype, "fat") ||
+	                 !strcmp(fstype, "efi");
 	while ( mountable )
 	{
+		const char* def = !strcmp(fstype, "efi") ? "/boot/efi" : "no";
 		if ( 6 <= argc )
 			strlcpy(mountpoint, argv[5], sizeof(mountpoint));
 		else
 			prompt(mountpoint, sizeof(mountpoint),
-			       "Where to mount partition? (mountpoint or 'no')", "no");
+			       "Where to mount partition? (mountpoint or 'no')", def);
 		if ( !strcmp(mountpoint, "no") )
 		{
 			mountpoint[0] = '\0';
@@ -2003,6 +2015,10 @@ static void on_mkpart(size_t argc, char** argv)
 		const char* type_uuid_str = "0FC63DAF-8483-4772-8E79-3D69D8477DE4";
 		if ( !strcmp(fstype, "biosboot") )
 			type_uuid_str = BIOSBOOT_GPT_TYPE_UUID;
+		else if ( !strcmp(fstype, "efi") )
+			type_uuid_str = ESP_GPT_TYPE_UUID;
+		else if ( !strcmp(fstype, "fat") )
+			type_uuid_str = BDP_GPT_TYPE_UUID;
 		uuid_from_string(p.partition_type_guid, type_uuid_str);
 		arc4random_buf(p.unique_partition_guid, sizeof(p.unique_partition_guid));
 		off_t pstart = hole->start + start;
@@ -2162,6 +2178,68 @@ static void on_mkpart(size_t argc, char** argv)
 			}
 		}
 	}
+	if ( !strcmp(fstype, "efi") || !strcmp(fstype, "fat") )
+	{
+		printf("(Formatting %s as %s...)\n", device_name(p->path), fstype);
+		// TODO: Zero superblock?
+		// TODO: Run this in its own foreground process group so ^C works.
+		pid_t child_pid = fork();
+		if ( child_pid < 0 )
+		{
+			command_error("%s: fork", argv[0]);
+			return;
+		}
+		const char* mkfs_argv[] =
+		{
+			"mkfs.fat",
+			"-F", // TODO: Force FAT32 until FAT12/16 root dir writing is added.
+			"32",
+			p->path,
+			NULL
+		};
+		if ( child_pid == 0 )
+		{
+			execvp(mkfs_argv[0], (char* const*) mkfs_argv);
+			warn("%s", mkfs_argv[0]);
+			_exit(127);
+		}
+		int status;
+		waitpid(child_pid, &status, 0);
+		if ( WIFEXITED(status) && WEXITSTATUS(status) == 127 )
+		{
+			command_errorx("%s: Failed to format filesystem (%s is not installed)",
+			               argv[0], mkfs_argv[0]);
+			return;
+		}
+		else if ( WIFEXITED(status) && WEXITSTATUS(status) != 0 )
+		{
+			command_errorx("%s: Failed to format filesystem", argv[0]);
+			return;
+		}
+		else if ( WIFSIGNALED(status) )
+		{
+			command_errorx("%s: Failed to format filesystem (%s)",
+			               argv[0], strsignal(WTERMSIG(status)));
+			return;
+		}
+		printf("(Formatted %s as %s)\n", device_name(p->path), fstype);
+		scan_partition(p);
+		if ( !p->bdev.fs /* TODO: || !(p->bdev.fs->flags & FILESYSTEM_FLAG_UUID) */ )
+		{
+			command_errorx("%s: %s: Failed to scan expected %s filesystem",
+			               argv[0], device_name(p->path), fstype);
+			return;
+		}
+		if ( mountpoint[0] )
+		{
+			if ( !add_blockdevice_to_fstab(&p->bdev, mountpoint) )
+			{
+				command_error("%s: %s: Failed to add partition", argv[0], fstab_path);
+				return;
+			}
+		}
+
+	}
 }
 
 static void on_mktable(size_t argc, char** argv)

display/connection.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2015, 2016, 2023 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2014, 2015, 2016, 2023, 2024 Jonas 'Sortie' Termansen.
  * Copyright (c) 2023 Juhani 'nortti' Krekelä.
  *
  * Permission to use, copy, modify, and distribute this software for any
@@ -149,15 +149,15 @@ CONNECTION_MESSAGE_HANDLER_NO_AUX(shutdown)
 {
 	(void) connection;
 	if ( msg->code == 0 )
-		exit(0);
+		display_exit(server->display, 0);
 	else if ( msg->code == 1 )
-		exit(1);
+		display_exit(server->display, 1);
 	else if ( msg->code == 2 )
-		exit(2);
+		display_exit(server->display, 2);
 	else if ( msg->code == 3 )
-		exit(3);
+		display_exit(server->display, 3);
 	else
-		exit(0);
+		display_exit(server->display, 0);
 }
 
 CONNECTION_MESSAGE_HANDLER_NO_AUX(create_window)

display/connection.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2015, 2016 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2014, 2015, 2016, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -25,6 +25,7 @@
 
 #include <display-protocol.h>
 
+#include "display.h"
 #include "server.h"
 #include "window.h"
 

display/display-code.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2015, 2016, 2018, 2022, 2023 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2014-2016, 2018, 2022-2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -23,6 +23,8 @@
 
 #include <assert.h>
 #include <err.h>
+#include <inttypes.h>
+#include <signal.h>
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
@@ -47,6 +49,43 @@ void display_initialize(struct display* display)
 {
 	memset(display, 0, sizeof(*display));
 	display->redraw = true;
+	display->running = true;
+}
+
+static int get_init_exit_plan(void)
+{
+	FILE* fp = popen("/sbin/service default exit-code", "r");
+	if ( !fp )
+		return -1;
+	int result = -1;
+	char buffer[sizeof(int) * 3];
+	if ( fgets(buffer, sizeof(buffer), fp) && buffer[0] )
+		result = atoi(buffer);
+	pclose(fp);
+	return result;
+}
+
+void display_exit(struct display* display, int exit_code)
+{
+	display->running = false;
+	display->exit_code = exit_code;
+	int plan = exit_code == -1 ? get_init_exit_plan() : exit_code;
+	display->announcement = "Exiting...";
+	if ( plan == 0 )
+		display->announcement = "Powering off...";
+	else if ( plan == 1 )
+		display->announcement = "Rebooting...";
+	else if ( plan == 2 )
+		display->announcement = "Halting...";
+	else if ( plan == 3 )
+		display->announcement = "Reinitializing operating system...";
+	else if ( getenv("LOGIN_PID") ) // TODO: display -l?
+	{
+		intmax_t login_pid = strtoimax(getenv("LOGIN_PID"), NULL, 10);
+		if ( login_pid == getppid() )
+			display->announcement = "Logging out...";
+	}
+	display->redraw = true;
 }
 
 void assert_is_well_formed_display_list(struct display* display)
@@ -281,6 +320,31 @@ static void wallpaper(struct framebuffer fb)
 	}
 }
 
+static void display_render_exit(struct display* display, struct framebuffer fb)
+{
+	for ( int yoff = -1; yoff <= 1; yoff++ )
+	{
+		for ( int xoff = -1; xoff <= 1; xoff++ )
+		{
+			struct framebuffer msgfb = fb;
+			int y = (fb.yres - FONT_HEIGHT) / 2 + yoff;
+			msgfb = framebuffer_cut_top_y(msgfb, y);
+			int w = strlen(display->announcement) * (FONT_WIDTH+1);
+			int x = (fb.xres - w) / 2 + xoff;
+			msgfb = framebuffer_cut_left_x(msgfb, x);
+			render_text(msgfb, display->announcement, make_color_a(0, 0, 0, 64));
+		}
+	}
+
+	struct framebuffer msgfb = fb;
+	int y = (fb.yres - FONT_HEIGHT) / 2;
+	msgfb = framebuffer_cut_top_y(msgfb, y);
+	int w = strlen(display->announcement) * (FONT_WIDTH+1);
+	int x = (fb.xres - w) / 2;
+	msgfb = framebuffer_cut_left_x(msgfb, x);
+	render_text(msgfb, display->announcement, make_color(255, 255, 255));
+}
+
 void display_composit(struct display* display, struct framebuffer fb)
 {
 	struct damage_rect damage_rect = display->damage_rect;
@@ -301,6 +365,12 @@ void display_composit(struct display* display, struct framebuffer fb)
 
 	framebuffer_copy_to_framebuffer(fb, display->wallpaper);
 
+	if ( display->announcement )
+	{
+		display_render_exit(display, fb);
+		return;
+	}
+
 	for ( struct window* window = display->bottom_window;
 	      window;
 	      window = window->above_window )
@@ -482,11 +552,15 @@ void display_keyboard_event(struct display* display, uint32_t codepoint)
 		case KBKEY_RSUPER: display->key_rsuper = kbkey > 0; break;
 		}
 		if ( display->key_lctrl && display->key_lalt && kbkey == KBKEY_DELETE )
-			exit(0);
+			display_exit(display, -1);
 		if ( display->key_lctrl && display->key_lalt && kbkey == KBKEY_T )
 		{
 			if ( !fork() )
 			{
+				sigset_t sigterm;
+				sigemptyset(&sigterm);
+				sigaddset(&sigterm, SIGTERM);
+				sigprocmask(SIG_UNBLOCK, &sigterm, NULL);
 				execlp("terminal", "terminal", (char*) NULL);
 				_exit(127);
 			}

display/display.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2015, 2016, 2017, 2022, 2023 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2014-2017, 2022-2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -119,5 +119,5 @@ int main(int argc, char* argv[])
 
 	server_mainloop(&server);
 
-	return 0;
+	return display.exit_code;
 }

display/display.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2015, 2016, 2022, 2023 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2014, 2015, 2016, 2022, 2023, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -68,14 +68,19 @@ struct display
 	bool key_rsuper;
 	bool codepoint_discard;
 	bool redraw;
+	bool running;
 	int pointer_x;
 	int pointer_y;
 	enum mouse_state mouse_state;
 	size_t mouse_byte_count;
 	uint8_t mouse_bytes[MOUSE_PACKET_SIZE];
+	const char* announcement;
+	int exit_code;
+
 };
 
 void display_initialize(struct display* display);
+void display_exit(struct display* display, int exit_code);
 void assert_is_well_formed_display_list(struct display* display);
 void assert_is_well_formed_display(struct display* display);
 void display_link_window_at_top(struct display* display, struct window* window);

display/server.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2015, 2016, 2023 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2014, 2015, 2016, 2023, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -28,6 +28,7 @@
 #include <errno.h>
 #include <fcntl.h>
 #include <poll.h>
+#include <signal.h>
 #include <stddef.h>
 #include <stdlib.h>
 #include <string.h>
@@ -40,6 +41,14 @@
 #include "server.h"
 #include "vgafont.h"
 
+static volatile sig_atomic_t got_sigterm;
+
+static void on_sigterm(int signum)
+{
+	(void) signum;
+	got_sigterm = 1;
+}
+
 static int open_local_server_socket(const char* path, int flags)
 {
 	size_t path_length = strlen(path);
@@ -196,9 +205,24 @@ void server_poll(struct server* server)
 	}
 	size_t pfds_used = cpfd_off + connections_polled;
 
-	int num_events = ppoll(pfds, pfds_used, NULL, NULL);
+	sigset_t mask;
+	sigprocmask(SIG_SETMASK, NULL, &mask);
+	sigdelset(&mask, SIGTERM);
+
+	int num_events = ppoll(pfds, pfds_used, NULL, &mask);
+
+	if ( got_sigterm )
+	{
+		display_exit(server->display, -1);
+		got_sigterm = 0;
+	}
+
 	if ( num_events < 0 )
+	{
+		if ( errno == EINTR )
+			return;
 		err(1, "poll");
+	}
 
 	if ( pfds[0].revents )
 	{
@@ -273,9 +297,20 @@ void server_poll(struct server* server)
 
 void server_mainloop(struct server* server)
 {
-	while ( true )
+	sigset_t sigterm, oldset;
+	sigemptyset(&sigterm);
+	sigaddset(&sigterm, SIGTERM);
+	sigprocmask(SIG_BLOCK, &sigterm, &oldset);
+	struct sigaction sa = { .sa_handler = on_sigterm }, old_sa;
+	sigaction(SIGTERM, &sa, &old_sa);
+
+	while ( server->display->running )
 	{
 		display_render(server->display);
 		server_poll(server);
 	}
+	display_render(server->display);
+
+	sigaction(SIGTERM, &old_sa, NULL);
+	sigprocmask(SIG_SETMASK, &oldset, NULL);
 }

editor/editor.c

@@ -25,6 +25,7 @@
 #include <limits.h>
 #include <locale.h>
 #include <pwd.h>
+#include <signal.h>
 #include <stdbool.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -270,6 +271,14 @@ bool editor_save_file(struct editor* editor, const char* path)
 	return fclose(fp) != EOF;
 }
 
+volatile sig_atomic_t had_sigwinch = 0;
+
+static void sigwinch(int signum)
+{
+	(void) signum;
+	had_sigwinch = 1;
+}
+
 int main(int argc, char* argv[])
 {
 	setlocale(LC_ALL, "");
@@ -279,6 +288,13 @@ int main(int argc, char* argv[])
 	if ( !isatty(1) )
 		err(1, "standard output");
 
+	sigset_t sigwinch_set;
+	sigemptyset(&sigwinch_set);
+	sigaddset(&sigwinch_set, SIGWINCH);
+	sigprocmask(SIG_BLOCK, &sigwinch_set, NULL);
+	struct sigaction sa = { .sa_handler = sigwinch };
+	sigaction(SIGWINCH, &sa, NULL);
+
 	struct editor editor;
 	initialize_editor(&editor);
 

editor/input.c

@@ -353,12 +353,22 @@ void editor_input_process_byte(struct editor_input* editor_input,
 	}
 }
 
+extern volatile sig_atomic_t had_sigwinch;
+
 void editor_input_process(struct editor_input* editor_input,
                           struct editor* editor)
 {
+	sigset_t sigset;
+	sigemptyset(&sigset);
 	struct pollfd pfd = { .fd = 0, .events = POLLIN };
-	do editor_input_process_byte(editor_input, editor);
-	while ( poll(&pfd, 1, 0) == 1 );
+	struct timespec timeout = { .tv_sec = 0 };
+	struct timespec* timeout_ptr = NULL;
+	while ( !had_sigwinch && ppoll(&pfd, 1, timeout_ptr, &sigset) == 1 )
+	{
+		editor_input_process_byte(editor_input, editor);
+		timeout_ptr = &timeout;
+	}
+	had_sigwinch = 0;
 }
 
 void editor_input_end(struct editor_input* editor_input)

etc/signify/sortix-1.1.pub

@@ -0,0 +1,2 @@
+untrusted comment: signify public key
+RWQiTQbFzyZJVobf/pn53Jp3njhRB9DgwkMaNakCpDE9RaTABMjlbz9W

etc/signify/sortix-1.1dev.pub

@@ -0,0 +1,2 @@
+untrusted comment: signify public key
+RWQnkSm9lj1YIZYpt1Y3mHYzFsaky82gQF6CrW4lme9OoEYzSIl2ZsIC

etc/signify/sortix-1.2.pub

@@ -0,0 +1,2 @@
+untrusted comment: signify public key
+RWTGrBXmGvl2zUpCa47ui5EyPsnitKLjsCZ2YZphNY8F3b33t6QWYDs1

etc/signify/sortix-1.2dev.pub

@@ -0,0 +1,2 @@
+untrusted comment: signify public key
+RWRTbLQ+3+a9I5yche2BEVP03TRtumGO4Vgq1AQ/5bRj8JAJ1R0+vpxE

fat/.gitignore

@@ -0,0 +1,2 @@
+fatfs
+*.o

fat/Makefile

@@ -0,0 +1,34 @@
+include ../build-aux/platform.mak
+include ../build-aux/compiler.mak
+include ../build-aux/version.mak
+include ../build-aux/dirs.mak
+
+OPTLEVEL?=$(DEFAULT_OPTLEVEL)
+CXXFLAGS?=$(OPTLEVEL)
+
+CPPFLAGS:=$(CPPFLAGS) -DVERSIONSTR=\"$(VERSION)\"
+CXXFLAGS:=$(CXXFLAGS) -Wall -Wextra -fno-exceptions -fno-rtti -fcheck-new
+
+LIBS:=$(LIBS)
+
+ifeq ($(HOST_IS_SORTIX),0)
+  PTHREAD_OPTION:=-pthread
+  LIBS:=$(LIBS) -lfuse
+  CPPFLAGS:=$(CPPFLAGS) -D_FILE_OFFSET_BITS=64
+endif
+
+BINARIES:=fatfs
+
+all: $(BINARIES)
+
+.PHONY: all install clean
+
+install: all
+	mkdir -p $(DESTDIR)$(SBINDIR)
+	install $(BINARIES) $(DESTDIR)$(SBINDIR)
+
+fatfs: *.cpp *.h
+	$(CXX) $(PTHREAD_OPTION) -std=gnu++11 $(CPPFLAGS) $(CXXFLAGS) *.cpp -o $@ $(LIBS)
+
+clean:
+	rm -f $(BINARIES) *.o

fat/block.cpp

@@ -0,0 +1,165 @@
+/*
+ * Copyright (c) 2013, 2014, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * block.cpp
+ * Blocks in the filesystem.
+ */
+
+#include <sys/types.h>
+
+#include <assert.h>
+#include <pthread.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#include "block.h"
+#include "device.h"
+#include "ioleast.h"
+
+Block::Block()
+{
+	this->block_data = NULL;
+}
+
+Block::Block(Device* device, uint32_t block_id)
+{
+	Construct(device, block_id);
+}
+
+void Block::Construct(Device* device, uint32_t block_id)
+{
+	this->modify_lock = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
+	this->transit_done_cond = PTHREAD_COND_INITIALIZER;
+	this->prev_block = NULL;
+	this->next_block = NULL;
+	this->prev_hashed = NULL;
+	this->next_hashed = NULL;
+	this->prev_dirty = NULL;
+	this->next_dirty = NULL;
+	this->device = device;
+	this->reference_count = 1;
+	this->block_id = block_id;
+	this->dirty = false;
+	this->is_in_transit = false;
+}
+
+Block::~Block()
+{
+	Destruct();
+	delete[] block_data;
+}
+
+void Block::Destruct()
+{
+	Sync();
+	Unlink();
+}
+
+void Block::Refer()
+{
+	reference_count++;
+}
+
+void Block::Unref()
+{
+	if ( !--reference_count )
+	{
+#if 0
+		device->block_count--;
+		delete this;
+#endif
+	}
+}
+
+void Block::Sync()
+{
+	if ( device->has_sync_thread )
+	{
+		pthread_mutex_lock(&device->sync_thread_lock);
+		while ( dirty || is_in_transit )
+			pthread_cond_wait(&transit_done_cond, &device->sync_thread_lock);
+		pthread_mutex_unlock(&device->sync_thread_lock);
+		return;
+	}
+
+	if ( !dirty )
+		return;
+	dirty = false;
+	(prev_dirty ? prev_dirty->next_dirty : device->dirty_block) = next_dirty;
+	if ( next_dirty )
+		next_dirty->prev_dirty = prev_dirty;
+	prev_dirty = NULL;
+	next_dirty = NULL;
+	if ( !device->write )
+		return;
+	off_t file_offset = (off_t) device->block_size * (off_t) block_id;
+	pwriteall(device->fd, block_data, device->block_size, file_offset);
+}
+
+void Block::BeginWrite()
+{
+	assert(device->write);
+	pthread_mutex_lock(&modify_lock);
+}
+
+void Block::FinishWrite()
+{
+	pthread_mutex_unlock(&modify_lock);
+	pthread_mutex_lock(&device->sync_thread_lock);
+	if ( !dirty )
+	{
+		dirty = true;
+		prev_dirty = NULL;
+		next_dirty = device->dirty_block;
+		if ( next_dirty )
+			next_dirty->prev_dirty = this;
+		device->dirty_block = this;
+		pthread_cond_signal(&device->sync_thread_cond);
+	}
+	pthread_mutex_unlock(&device->sync_thread_lock);
+	Use();
+}
+
+void Block::Use()
+{
+	Unlink();
+	Prelink();
+}
+
+void Block::Unlink()
+{
+	(prev_block ? prev_block->next_block : device->mru_block) = next_block;
+	(next_block ? next_block->prev_block : device->lru_block) = prev_block;
+	size_t bin = block_id % DEVICE_HASH_LENGTH;
+	(prev_hashed ? prev_hashed->next_hashed : device->hash_blocks[bin]) = next_hashed;
+	if ( next_hashed ) next_hashed->prev_hashed = prev_hashed;
+}
+
+void Block::Prelink()
+{
+	prev_block = NULL;
+	next_block = device->mru_block;
+	if ( device->mru_block )
+		device->mru_block->prev_block = this;
+	device->mru_block = this;
+	if ( !device->lru_block )
+		device->lru_block = this;
+	size_t bin = block_id % DEVICE_HASH_LENGTH;
+	prev_hashed = NULL;
+	next_hashed = device->hash_blocks[bin];
+	device->hash_blocks[bin] = this;
+	if ( next_hashed )
+		next_hashed->prev_hashed = this;
+}

fat/block.h

@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2013, 2014, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * block.h
+ * Blocks in the filesystem.
+ */
+
+#ifndef BLOCK_H
+#define BLOCK_H
+
+class Device;
+
+class Block
+{
+public:
+	Block();
+	Block(Device* device, uint32_t block_id);
+	~Block();
+	void Construct(Device* device, uint32_t block_id);
+	void Destruct();
+
+public:
+	pthread_mutex_t modify_lock;
+	pthread_cond_t transit_done_cond;
+	Block* prev_block;
+	Block* next_block;
+	Block* prev_hashed;
+	Block* next_hashed;
+	Block* prev_dirty;
+	Block* next_dirty;
+	Device* device;
+	size_t reference_count;
+	uint32_t block_id;
+	bool dirty;
+	bool is_in_transit;
+	uint8_t* block_data;
+
+public:
+	void Refer();
+	void Unref();
+	void Sync();
+	void BeginWrite();
+	void FinishWrite();
+	void Use();
+	void Unlink();
+	void Prelink();
+
+};
+
+#endif

fat/device.cpp

@@ -0,0 +1,220 @@
+/*
+ * Copyright (c) 2013, 2014, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * device.cpp
+ * Block device.
+ */
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <pthread.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "block.h"
+#include "device.h"
+#include "ioleast.h"
+
+void* Device__SyncThread(void* ctx)
+{
+	((Device*) ctx)->SyncThread();
+	return NULL;
+}
+
+Device::Device(int fd, const char* path, uint32_t block_size, bool write)
+{
+	// sync_thread unset.
+	this->sync_thread_cond = PTHREAD_COND_INITIALIZER;
+	this->sync_thread_idle_cond = PTHREAD_COND_INITIALIZER;
+	this->sync_thread_lock = PTHREAD_MUTEX_INITIALIZER;
+	this->mru_block = NULL;
+	this->lru_block = NULL;
+	this->dirty_block = NULL;
+	for ( size_t i = 0; i < DEVICE_HASH_LENGTH; i++ )
+		hash_blocks[i] = NULL;
+	struct stat st;
+	fstat(fd, &st);
+	this->device_size = st.st_size;
+	this->path = path;
+	this->block_size = block_size;
+	this->fd = fd;
+	this->write = write;
+	this->has_sync_thread = false;
+	this->sync_thread_should_exit = false;
+	this->sync_in_transit = false;
+	this->block_count = 0;
+#ifdef __sortix__
+	// TODO: This isn't scaleable if there's multiple filesystems mounted.
+	size_t memory;
+	memstat(NULL, &memory);
+	this->block_limit = (memory / 10) / block_size;
+#else
+	this->block_limit = 32768;
+#endif
+}
+
+Device::~Device()
+{
+	if ( has_sync_thread )
+	{
+		pthread_mutex_lock(&sync_thread_lock);
+		sync_thread_should_exit = true;
+		pthread_cond_signal(&sync_thread_cond);
+		pthread_mutex_unlock(&sync_thread_lock);
+		pthread_join(sync_thread, NULL);
+		has_sync_thread = false;
+	}
+	Sync();
+	while ( mru_block )
+		delete mru_block;
+	close(fd);
+}
+
+void Device::SpawnSyncThread()
+{
+	if ( this->has_sync_thread )
+		return;
+	this->has_sync_thread = write &&
+		pthread_create(&this->sync_thread, NULL, Device__SyncThread, this) == 0;
+}
+
+Block* Device::AllocateBlock()
+{
+	if ( block_limit <= block_count )
+	{
+		for ( Block* block = lru_block; block; block = block->prev_block )
+		{
+			if ( block->reference_count )
+				continue;
+			block->Destruct(); // Syncs.
+			return block;
+		}
+	}
+	uint8_t* data = new uint8_t[block_size];
+	if ( !data ) // TODO: Use operator new nothrow!
+		return NULL;
+	Block* block = new Block();
+	if ( !block ) // TODO: Use operator new nothrow!
+		return delete[] data, (Block*) NULL;
+	block->block_data = data;
+	block_count++;
+	return block;
+}
+
+Block* Device::GetBlock(uint32_t block_id)
+{
+	if ( Block* block = GetCachedBlock(block_id) )
+		return block;
+	Block* block = AllocateBlock();
+	if ( !block )
+		return NULL;
+	block->Construct(this, block_id);
+	off_t file_offset = (off_t) block_size * (off_t) block_id;
+	preadall(fd, block->block_data, block_size, file_offset);
+	block->Prelink();
+	return block;
+}
+
+Block* Device::GetBlockZeroed(uint32_t block_id)
+{
+	assert(write);
+	if ( Block* block = GetCachedBlock(block_id) )
+	{
+		block->BeginWrite();
+		memset(block->block_data, 0, block_size);
+		block->FinishWrite();
+		return block;
+	}
+	Block* block = AllocateBlock();
+	if ( !block )
+		return NULL;
+	block->Construct(this, block_id);
+	memset(block->block_data, 0, block_size);
+	block->Prelink();
+	block->BeginWrite();
+	block->FinishWrite();
+	return block;
+}
+
+Block* Device::GetCachedBlock(uint32_t block_id)
+{
+	size_t bin = block_id % DEVICE_HASH_LENGTH;
+	for ( Block* iter = hash_blocks[bin]; iter; iter = iter->next_hashed )
+		if ( iter->block_id == block_id )
+			return iter->Refer(), iter;
+	return NULL;
+}
+
+void Device::Sync()
+{
+	if ( has_sync_thread )
+	{
+		pthread_mutex_lock(&sync_thread_lock);
+		while ( dirty_block || sync_in_transit )
+			pthread_cond_wait(&sync_thread_cond, &sync_thread_lock);
+		pthread_mutex_unlock(&sync_thread_lock);
+		fsync(fd);
+		return;
+	}
+
+	while ( dirty_block )
+		dirty_block->Sync();
+	fsync(fd);
+}
+
+void Device::SyncThread()
+{
+	uint8_t transit_block_data[block_size];
+	pthread_mutex_lock(&sync_thread_lock);
+	while ( true )
+	{
+		while ( !(dirty_block || sync_thread_should_exit) )
+			pthread_cond_wait(&sync_thread_cond, &sync_thread_lock);
+		if ( sync_thread_should_exit )
+			break;
+
+		Block* block = dirty_block;
+
+		if ( block->next_dirty )
+			block->next_dirty->prev_dirty = NULL;
+		dirty_block = block->next_dirty;
+		block->next_dirty = NULL;
+
+		block->dirty = false;
+		block->is_in_transit = true;
+		sync_in_transit = true;
+
+		pthread_mutex_unlock(&sync_thread_lock);
+
+		pthread_mutex_lock(&block->modify_lock);
+		memcpy(transit_block_data, block->block_data, block_size);
+		pthread_mutex_unlock(&block->modify_lock);
+
+		off_t offset = (off_t) block_size * (off_t) block->block_id;
+		pwriteall(fd, transit_block_data, block_size, offset);
+
+		pthread_mutex_lock(&sync_thread_lock);
+		block->is_in_transit = false;
+		sync_in_transit = false;
+		pthread_cond_signal(&block->transit_done_cond);
+		if ( !dirty_block )
+			pthread_cond_signal(&sync_thread_idle_cond);
+	}
+	pthread_mutex_unlock(&sync_thread_lock);
+}

fat/device.h

@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2013, 2014, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * device.h
+ * Block device.
+ */
+
+#ifndef DEVICE_H
+#define DEVICE_H
+
+class Block;
+
+static const size_t DEVICE_HASH_LENGTH = 1 << 16;
+
+class Device
+{
+public:
+	Device(int fd, const char* path, uint32_t block_size, bool write);
+	~Device();
+
+public:
+	pthread_t sync_thread;
+	pthread_cond_t sync_thread_cond;
+	pthread_cond_t sync_thread_idle_cond;
+	pthread_mutex_t sync_thread_lock;
+	Block* mru_block;
+	Block* lru_block;
+	Block* dirty_block;
+	Block* hash_blocks[DEVICE_HASH_LENGTH];
+	off_t device_size;
+	const char* path;
+	uint32_t block_size;
+	int fd;
+	bool write;
+	bool has_sync_thread;
+	bool sync_thread_should_exit;
+	bool sync_in_transit;
+	size_t block_count;
+	size_t block_limit;
+
+public:
+	void SpawnSyncThread();
+	Block* AllocateBlock();
+	Block* GetBlock(uint32_t block_id);
+	Block* GetBlockZeroed(uint32_t block_id);
+	Block* GetCachedBlock(uint32_t block_id);
+	void Sync();
+	void SyncThread();
+
+};
+
+#endif

fat/fat.h

@@ -0,0 +1,124 @@
+/*
+ * Copyright (c) 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fat.h
+ * The File Allocation Table (FAT) filesystem.
+ */
+
+#ifndef FAT_H
+#define FAT_H
+
+#include <assert.h>
+#include <stdint.h>
+
+// TODO: Dammit. I forgot to swap endian all across the codebase.
+
+struct fat_bpb
+{
+	uint8_t jump[3];
+	char oem[8];
+	uint8_t bytes_per_sector_low;
+	uint8_t bytes_per_sector_high;
+	uint8_t sectors_per_cluster;
+	uint16_t reserved_sectors;
+	uint8_t fat_count;
+	uint8_t root_dirent_count_low;
+	uint8_t root_dirent_count_high;
+	uint8_t total_sectors_low;
+	uint8_t total_sectors_high;
+	uint8_t media_descriptor_type;
+	uint16_t sectors_per_fat;
+	uint16_t sectors_per_track;
+	uint16_t head_count;
+	uint16_t hidden_sectors;
+	uint32_t total_sectors_large;
+	union
+	{
+		struct
+		{
+			uint8_t fat12_drive_number;
+			uint8_t fat12_reserved;
+			uint8_t fat12_signature;
+			uint8_t fat12_volume_id[4];
+			uint8_t fat12_volume_label[11];
+			uint8_t fat12_system[8];
+		};
+		struct
+		{
+			uint32_t fat32_sectors_per_fat;
+			uint16_t fat32_flags;
+			uint16_t fat32_version;
+			uint32_t fat32_root_cluster;
+			uint16_t fat32_fsinfo;
+			uint16_t fat32_backup_boot;
+			uint32_t fat32_reserved1[3];
+			uint8_t fat32_drive_number;
+			uint8_t fat32_reserved2;
+			uint8_t fat32_signature;
+			uint8_t fat32_volume_id[4];
+			uint8_t fat32_volume_label[11];
+			uint8_t fat32_system[8];
+		};
+		struct
+		{
+			uint8_t bootloader[510 - 36];
+			uint8_t boot_signature[2];
+		};
+	};
+};
+
+static_assert(sizeof(struct fat_bpb) == 512, "sizeof(struct fat_bpb) == 512");
+
+struct fat_fsinfo
+{
+	uint32_t signature1;
+	uint32_t reserved1[120];
+	uint32_t signature2;
+	uint32_t free_count;
+	uint32_t next_free;
+	uint32_t reserved2[3];
+	uint32_t signature3;
+};
+
+static_assert(sizeof(struct fat_fsinfo) == 512, "sizeof(struct fat_fsinfo) == 512");
+
+struct fat_dirent
+{
+	char name[11];
+	uint8_t attributes;
+	uint8_t reserved;
+	uint8_t creation_tenths; // TODO: Misnamed semantically.
+	uint16_t creation_time;
+	uint16_t creation_date;
+	uint16_t access_date;
+	uint16_t cluster_high;
+	uint16_t modified_time;
+	uint16_t modified_date;
+	uint16_t cluster_low;
+	uint32_t size;
+};
+
+static_assert(sizeof(struct fat_dirent) == 32, "sizeof(struct fat_dirent) == 32");
+
+#define FAT_ATTRIBUTE_READ_ONLY (1 << 0)
+#define FAT_ATTRIBUTE_HIDDEN (1 << 1)
+#define FAT_ATTRIBUTE_SYSTEM (1 << 2)
+#define FAT_ATTRIBUTE_VOLUME_ID (1 << 3)
+#define FAT_ATTRIBUTE_DIRECTORY (1 << 4)
+#define FAT_ATTRIBUTE_ARCHIVE (1 << 5)
+
+#define FAT_ATTRIBUTE_LONG_NAME 0x0F
+
+#endif

fat/fatfs.cpp

@@ -0,0 +1,275 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2016, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fatfs.cpp
+ * The File Allocation Table (FAT) filesystem.
+ */
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <dirent.h>
+#include <err.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <unistd.h>
+
+#if defined(__sortix__)
+#include "fsmarshall.h"
+#else
+#include "fuse.h"
+#endif
+
+#include "block.h"
+#include "device.h"
+#include "fat.h"
+#include "fatfs.h"
+#include "filesystem.h"
+#include "inode.h"
+#include "ioleast.h"
+#include "util.h"
+
+uid_t request_uid;
+uid_t request_gid;
+
+// TODO: Encapsulate.
+void StatInode(Inode* inode, struct stat* st)
+{
+	memset(st, 0, sizeof(*st));
+	st->st_ino = inode->inode_id;
+	st->st_mode = inode->Mode();
+	st->st_nlink = 1; // TODO: Encapsulate.
+	st->st_uid = inode->UserId();
+	st->st_gid = inode->GroupId();
+	st->st_size = inode->Size();
+	// TODO: Encapsulate.
+	// TODO: For the root dir, mount time, or maybe volume label or first file?
+	//       Or maybe the time of the mount point?
+	time_t mtime = 0;
+	if ( inode->dirent )
+	{
+		struct tm tm;
+		memset(&tm, 0, sizeof(tm));
+		tm.tm_sec = ((inode->dirent->modified_time >> 0) & 0x1F) * 2;
+		tm.tm_min = (inode->dirent->modified_time >> 5) & 0x3F;
+		tm.tm_hour = (inode->dirent->modified_time >> 11) & 0x1F;
+		tm.tm_mday = (inode->dirent->modified_date >> 0) & 0x1F;
+		tm.tm_mon = ((inode->dirent->modified_date >> 5) & 0xF) - 1;
+		tm.tm_year = ((inode->dirent->modified_date >> 9) & 0x7F) + 80;
+		mtime = mktime(&tm);
+	}
+	st->st_atim.tv_sec = mtime; // TODO: The actual accessed time;
+	st->st_atim.tv_nsec = 0;
+	st->st_ctim.tv_sec = mtime; // TODO: Probably fine to keep as modified time.
+	st->st_ctim.tv_nsec = 0;
+	st->st_mtim.tv_sec = mtime;
+	st->st_mtim.tv_nsec = 0;
+	st->st_blksize = inode->filesystem->bytes_per_sector *
+	                 inode->filesystem->bpb->sectors_per_cluster;
+	// TODO: Encapsulate.
+	st->st_blocks = 0; // TODO inode->data->i_blocks;
+}
+
+static void compact_arguments(int* argc, char*** argv)
+{
+	for ( int i = 0; i < *argc; i++ )
+	{
+		while ( i < *argc && !(*argv)[i] )
+		{
+			for ( int n = i; n < *argc; n++ )
+				(*argv)[n] = (*argv)[n+1];
+			(*argc)--;
+		}
+	}
+}
+
+static void help(FILE* fp, const char* argv0)
+{
+	fprintf(fp, "Usage: %s [OPTION]... DEVICE [MOUNT-POINT]\n", argv0);
+}
+
+static void version(FILE* fp, const char* argv0)
+{
+	fprintf(fp, "%s (Sortix) %s\n", argv0, VERSIONSTR);
+}
+
+int main(int argc, char* argv[])
+{
+	const char* argv0 = argv[0];
+	const char* pretend_mount_path = NULL;
+	bool foreground = false;
+	bool read = false;
+	bool write = false;
+	for ( int i = 1; i < argc; i++ )
+	{
+		const char* arg = argv[i];
+		if ( arg[0] != '-' || !arg[1] )
+			continue;
+		argv[i] = NULL;
+		if ( !strcmp(arg, "--") )
+			break;
+		if ( arg[1] != '-' )
+		{
+			while ( char c = *++arg ) switch ( c )
+			{
+			case 'r': read = true; break;
+			case 'w': write = true; break;
+			default:
+				fprintf(stderr, "%s: unknown option -- '%c'\n", argv0, c);
+				help(stderr, argv0);
+				exit(1);
+			}
+		}
+		else if ( !strcmp(arg, "--help") )
+			help(stdout, argv0), exit(0);
+		else if ( !strcmp(arg, "--version") )
+			version(stdout, argv0), exit(0);
+		else if ( !strcmp(arg, "--background") )
+			foreground = false;
+		else if ( !strcmp(arg, "--foreground") )
+			foreground = true;
+		else if ( !strcmp(arg, "--read") )
+			read = true;
+		else if ( !strcmp(arg, "--write") )
+			write = true;
+		else if ( !strncmp(arg, "--pretend-mount-path=", strlen("--pretend-mount-path=")) )
+			pretend_mount_path = arg + strlen("--pretend-mount-path=");
+		else if ( !strcmp(arg, "--pretend-mount-path") )
+		{
+			if ( i+1 == argc )
+			{
+				fprintf(stderr, "%s: --pretend-mount-path: Missing operand\n", argv0);
+				exit(1);
+			}
+			pretend_mount_path = argv[++i];
+			argv[i] = NULL;
+		}
+		else
+		{
+			fprintf(stderr, "%s: unknown option: %s\n", argv0, arg);
+			help(stderr, argv0);
+			exit(1);
+		}
+	}
+
+	// It doesn't make sense to have a write-only filesystem.
+	read = read || write;
+
+	// Default to read and write filesystem access.
+	if ( !read && !write )
+		 read = write = true;
+
+	if ( argc == 1 )
+	{
+		help(stdout, argv0);
+		exit(0);
+	}
+
+	compact_arguments(&argc, &argv);
+
+	const char* device_path = 2 <= argc ? argv[1] : NULL;
+	const char* mount_path = 3 <= argc ? argv[2] : NULL;
+
+	if ( !device_path )
+	{
+		help(stderr, argv0);
+		exit(1);
+	}
+
+	if ( !pretend_mount_path )
+		pretend_mount_path = mount_path;
+
+	int fd = open(device_path, write ? O_RDWR : O_RDONLY);
+	if ( fd < 0 )
+		err(1, "%s", device_path);
+
+	// Read the bios parameter block from the filesystem so we can verify it.
+	struct fat_bpb bpb;
+	if ( preadall(fd, &bpb, sizeof(bpb), 0) != sizeof(bpb) )
+	{
+		if ( errno == EEOF )
+			errx(1, "%s: Isn't a valid FAT filesystem (too short)", device_path);
+		else
+			err(1, "read: %s", device_path);
+	}
+
+	// Verify the boot.
+	if ( !(bpb.boot_signature[0] == 0x55 && bpb.boot_signature[1] == 0xAA) )
+		errx(1, "%s: Isn't a valid FAT filesystem (no boot signature)", device_path);
+
+	// Verify the jump instruction at the start of the boot sector.
+	if ( !(bpb.jump[0] == 0xEB && bpb.jump[2] == 0x90) &&
+	     !(bpb.jump[0] == 0xE9) )
+		errx(1, "%s: Isn't a valid FAT filesystem (bad jump)", device_path);
+
+	// TODO: Validate all parameters make sense.
+
+	uint16_t bytes_per_sector =
+		bpb.bytes_per_sector_low | bpb.bytes_per_sector_high << 8;
+	uint16_t root_dirent_count =
+		bpb.root_dirent_count_low | bpb.root_dirent_count_high << 8;
+	uint32_t root_dir_sectors =
+		divup<uint32_t>(root_dirent_count * sizeof(fat_dirent), bytes_per_sector);
+	uint32_t sectors_per_fat =
+		bpb.sectors_per_fat ? bpb.sectors_per_fat : bpb.fat32_sectors_per_fat;
+	uint32_t total_sectors =
+		bpb.total_sectors_low | bpb.total_sectors_high << 8;
+	if ( !total_sectors )
+		total_sectors = bpb.total_sectors_large;
+
+	uint32_t data_offset =
+		bpb.reserved_sectors + bpb.fat_count * sectors_per_fat + root_dir_sectors;
+	uint32_t data_sectors = total_sectors - data_offset;
+	uint32_t cluster_count = data_sectors / bpb.sectors_per_cluster;
+
+	uint8_t fat_type =
+		cluster_count < 4085 ? 12 : cluster_count < 65525 ? 16 : 32;
+
+	// Verify the filesystem version.
+	if ( fat_type == 32 && bpb.fat32_version != 0x0000 )
+		errx(1, "%s: Unsupported filesystem version 0x%04x", device_path,
+		     bpb.fat32_version);
+
+	// TODO: On FAT16/32 check FAT entry 1 for the high bits to see if fsck is needed.
+	// Check whether the filesystem was unmounted cleanly.
+	//if ( !(fat[1] & FAT_UNMOUNTED) || !(fat[1] & FAT_NO_IO_ERR) )
+	//	warn("warning: %s: Filesystem wasn't unmounted cleanly\n", device_path);
+
+	// TODO: The FAT and clusters are not aligned to cluster size so
+	//       we can't use the cluster size here. Perhaps refactor the
+	//       device so we can deal with whole clusters.
+	Device* dev = new Device(fd, device_path, bytes_per_sector, write);
+	if ( !dev ) // TODO: Use operator new nothrow!
+		err(1, "malloc");
+	Filesystem* fs = new Filesystem(dev, pretend_mount_path);
+	if ( !fs ) // TODO: Use operator new nothrow!
+		err(1, "malloc");
+
+	if ( !mount_path )
+		return 0;
+
+#if defined(__sortix__)
+	return fsmarshall_main(argv0, mount_path, foreground, fs, dev);
+#else
+	return fat_fuse_main(argv0, mount_path, foreground, fs, dev);
+#endif
+}

fat/fatfs.h

@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fatfs.h
+ * Implementation of the extended filesystem.
+ */
+
+#ifndef EXTFS_H
+#define EXTFS_H
+
+extern uid_t request_uid;
+extern gid_t request_gid;
+
+class Inode;
+
+mode_t HostModeFromExtMode(uint32_t extmode);
+uint32_t ExtModeFromHostMode(mode_t hostmode);
+uint8_t HostDTFromExtDT(uint8_t extdt);
+void StatInode(Inode* inode, struct stat* st);
+
+#endif

fat/filesystem.cpp

@@ -0,0 +1,446 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * filesystem.cpp
+ * Filesystem.
+ */
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <endian.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <time.h>
+
+#include "fat.h"
+
+#include "block.h"
+#include "device.h"
+#include "filesystem.h"
+#include "inode.h"
+#include "util.h"
+
+// TODO: Be more precise.
+static bool is_8_3_char(char c)
+{
+		return ('A' <= c && c <= 'Z') || ('0' <= c && c <= '9');
+}
+
+// TODO: Is this fully precise? What about .FOO?
+bool is_8_3(const char* name)
+{
+	if ( !name[0] )
+		return false;
+	size_t b = 0;
+	while ( name[b] && is_8_3_char(name[b]) )
+		b++;
+	if ( 8 < b )
+		return false;
+	if ( !name[b] )
+		return true;
+	if ( name[b] != '.' )
+		return false;
+	size_t e = 0;
+	while ( name[b+1+e] && is_8_3_char(name[b+1+e]) )
+		e++;
+	if ( 3 < e )
+		return false;
+	if ( name[b+1+e] )
+		return false;
+	return true;
+}
+
+void encode_8_3(const char* decoded, char encoded[8 + 3])
+{
+	size_t i = 0;
+	for ( size_t o = 0; o < 8 + 3; o++ )
+	{
+		char c = ' ';
+		if ( decoded[i] == '.' && o == 8 )
+			i++;
+		if ( decoded[i] && decoded[i] != '.' )
+			c = decoded[i++];
+		if ( (unsigned char) c == 0xE5 )
+			c = 0x05;
+		encoded[o] = c;
+	}
+}
+
+void decode_8_3(const char encoded[8 + 3], char decoded[8 + 1 + 3 + 1])
+{
+	size_t o = 0;
+	for ( size_t i = 0; i < 8; i++ )
+	{
+		char c = encoded[i];
+		if ( !c || c == ' ' )
+			break;
+		if ( c == 0x05 )
+			c = (char) 0xE5;
+		decoded[o++] = c;
+	}
+	for ( size_t i = 8; i < 8 + 3; i++ )
+	{
+		char c = encoded[i];
+		if ( !c || c == ' ' )
+			break;
+		if ( i == 8 )
+			decoded[o++] = '.';
+		if ( c == 0x05 )
+			c = (char) 0xE5;
+		decoded[o++] = c;
+	}
+	decoded[o] = '\0';
+}
+
+uint8_t timespec_to_fat_tenths(struct timespec* ts)
+{
+	// TODO: Work with a struct tm instead.
+	uint16_t hundreds = ts->tv_nsec / 10000000;
+	struct tm tm;
+	gmtime_r(&ts->tv_sec, &tm);
+	if ( tm.tm_sec & 1 )
+		hundreds += 100;
+	return hundreds;
+}
+
+uint16_t timespec_to_fat_time(struct timespec* ts)
+{
+	// TODO: Work with a struct tm instead.
+	struct tm tm;
+	gmtime_r(&ts->tv_sec, &tm);
+	return (tm.tm_sec / 2) << 0 | tm.tm_min << 5 | tm.tm_hour << 11;
+}
+
+uint16_t timespec_to_fat_date(struct timespec* ts)
+{
+	// TODO: Work with a struct tm instead.
+	struct tm tm;
+	gmtime_r(&ts->tv_sec, &tm);
+	return tm.tm_mday << 0 | (tm.tm_mon + 1) << 5 | (tm.tm_year - 80) << 9;
+}
+
+// TODO: Rename tenths to a better name.
+void timespec_to_fat(const struct timespec* ts, uint16_t* date, uint16_t* time,
+                     uint8_t* tenths)
+{
+	struct tm tm;
+	gmtime_r(&ts->tv_sec, &tm);
+	// TODO: Endian.
+	*date = tm.tm_mday << 0 | (tm.tm_mon + 1) << 5 | (tm.tm_year - 80) << 9;
+	*time = (tm.tm_sec / 2) << 0 | tm.tm_min << 5 | tm.tm_hour << 11;
+	*tenths = ts->tv_nsec / 10000000 + (tm.tm_sec & 1 ? 100 : 0);
+}
+
+Filesystem::Filesystem(Device* device, const char* mount_path)
+{
+	this->bpb_block = device->GetBlock(0);
+	assert(bpb_block); // TODO: This can fail.
+	this->bpb = (struct fat_bpb*) bpb_block->block_data;
+	this->device = device;
+	this->mount_path = mount_path;
+	this->mode_reg = S_IFREG | 0644;
+	this->mode_dir = S_IFDIR | 0755;
+	this->block_size = device->block_size;
+	this->bytes_per_sector =
+		bpb->bytes_per_sector_low | bpb->bytes_per_sector_high << 8;
+	this->root_dirent_count =
+		bpb->root_dirent_count_low | bpb->root_dirent_count_high << 8;
+	uint32_t root_dir_sectors =
+		divup<uint32_t>(root_dirent_count * sizeof(fat_dirent), bytes_per_sector);
+	this->sectors_per_fat =
+		bpb->sectors_per_fat ? bpb->sectors_per_fat : bpb->fat32_sectors_per_fat;
+	this->total_sectors =
+		bpb->total_sectors_low | bpb->total_sectors_high << 8;
+	if ( !this->total_sectors )
+		this->total_sectors = bpb->total_sectors_large;
+	this->fat_sector = bpb->reserved_sectors;
+	this->root_sector = fat_sector + bpb->fat_count * sectors_per_fat;
+	this->data_sector = root_sector + root_dir_sectors;
+	uint32_t data_sectors = total_sectors - data_sector;
+	this->cluster_count = data_sectors / bpb->sectors_per_cluster;
+	this->cluster_size = bpb->sectors_per_cluster * bytes_per_sector;
+	this->fat_type = cluster_count < 4085 ? 12 : cluster_count < 65525 ? 16 : 32;
+	// Use cluster 1 as the root inode on FAT12/FAT16 since it's not a valid
+	// cluster for use in the FAT.
+	this->root_inode_id = fat_type == 32 ? bpb->fat32_root_cluster : 1;
+	// TODO: Okay we actually need to compare with their lower bounds.
+	this->eio_cluster =
+		fat_type == 12 ? 0xFF7 : fat_type == 16 ? 0xFFF7 : 0xFFFFFF7;
+	this->eof_cluster =
+		fat_type == 12 ? 0xFFF : fat_type == 16 ? 0xFFFF : 0xFFFFFFF;
+	// TODO: Obtain and verify this from the fsinfo.
+	this->free_search = 0;
+	this->mru_inode = NULL;
+	this->lru_inode = NULL;
+	this->dirty_inode = NULL;
+	for ( size_t i = 0; i < INODE_HASH_LENGTH; i++ )
+		this->hash_inodes[i] = NULL;
+	this->dirty = false;
+
+	if ( device->write )
+	{
+		BeginWrite();
+		// TODO: Mark as mounted in fat[1] if FAT16/32.
+		FinishWrite();
+		Sync();
+	}
+}
+
+Filesystem::~Filesystem()
+{
+	Sync();
+	while ( mru_inode )
+		delete mru_inode;
+	if ( device->write )
+	{
+		BeginWrite();
+		// TODO: Mark as unounted in fat[1] if FAT16/32.
+		FinishWrite();
+		Sync();
+	}
+	bpb_block->Unref();
+}
+
+void Filesystem::BeginWrite()
+{
+	bpb_block->BeginWrite();
+}
+
+void Filesystem::FinishWrite()
+{
+	dirty = true;
+	bpb_block->FinishWrite();
+}
+
+void Filesystem::Sync()
+{
+	// TODO: Replacement concept?
+	while ( dirty_inode )
+		dirty_inode->Sync();
+	if ( dirty )
+	{
+		bpb_block->Sync();
+		dirty = false;
+	}
+	device->Sync();
+}
+
+Inode* Filesystem::GetInode(uint32_t inode_id, Block* dirent_block,
+                            struct fat_dirent* dirent)
+{
+#if 0
+	if ( !inode_id || num_inodes <= inode_id )
+		return errno = EBADF, (Inode*) NULL;
+	if ( !inode_id )
+		return errno = EBADF, (Inode*) NULL;
+#endif
+
+	size_t bin = inode_id % INODE_HASH_LENGTH;
+	for ( Inode* iter = hash_inodes[bin]; iter; iter = iter->next_hashed )
+		if ( iter->inode_id == inode_id )
+			return iter->Refer(), iter;
+
+	if ( inode_id != root_inode_id && !dirent_block )
+		return errno = EBADF, (Inode*) NULL;
+
+	Inode* inode = new Inode(this, inode_id);
+	if ( !inode )
+		return (Inode*) NULL;
+	inode->first_cluster =
+		inode_id == root_inode_id && fat_type != 32 ? 0 : inode_id;
+	if ( (inode->data_block = dirent_block) )
+		inode->data_block->Refer();
+	inode->dirent = dirent;
+	inode->Prelink();
+
+	return inode;
+}
+
+uint32_t Filesystem::AllocateCluster()
+{
+	for ( size_t i = 0; i < cluster_count; i++ )
+	{
+		size_t n = 2 + (free_search + i) % cluster_count;
+		if ( !ReadFAT(n) )
+		{
+			free_search = (i + 1) % cluster_count;
+			if ( fat_type == 32 )
+			{
+				Block* block = device->GetBlock(bpb->fat32_fsinfo);
+				if ( block )
+				{
+					struct fat_fsinfo* fsinfo =
+						(struct fat_fsinfo*) block->block_data;
+					block->BeginWrite();
+					uint32_t free_count = le32toh(fsinfo->free_count);
+					if ( free_count )
+						free_count--;
+					fsinfo->free_count = htole32(free_count);
+					fsinfo->next_free = n;
+					block->FinishWrite();
+					block->Unref();
+				}
+			}
+			return n;
+		}
+	}
+	return errno = ENOSPC, 0;
+}
+
+void Filesystem::FreeCluster(uint32_t cluster)
+{
+	if ( fat_type != 32 )
+		return;
+	Block* block = device->GetBlock(bpb->fat32_fsinfo);
+	if ( !block )
+		return;
+	struct fat_fsinfo* fsinfo = (struct fat_fsinfo*) block->block_data;
+	block->BeginWrite();
+	uint32_t free_count = le32toh(fsinfo->free_count);
+	if ( free_count < cluster_count )
+		free_count++;
+	fsinfo->free_count = htole32(free_count);
+	if ( !fsinfo->free_count || le32toh(fsinfo->next_free) == cluster + 1 )
+	{
+		fsinfo->next_free = htole32(cluster);
+		free_search = cluster - 2;
+	}
+	block->FinishWrite();
+	block->Unref();
+}
+
+uint32_t Filesystem::ReadFAT(uint32_t cluster)
+{
+	// TODO: Bounds check.
+	if ( fat_type == 12 )
+	{
+		size_t position = cluster + (cluster / 2);
+		size_t lba = position / bytes_per_sector;
+		size_t offset = position % bytes_per_sector;
+		Block* block = device->GetBlock(fat_sector + lba);
+		if ( !block )
+			return eio_cluster;
+		uint8_t lower = block->block_data[offset];
+		if ( ++offset == bytes_per_sector )
+		{
+			block->Unref();
+			if ( !(block = device->GetBlock(fat_sector + lba)) )
+				return eio_cluster;
+			offset = 0;
+		}
+		uint8_t higher = block->block_data[offset];
+		block->Unref();
+		uint16_t value = lower | higher << 8;
+		if ( cluster & 1 )
+			return value >> 4;
+		else
+			return value & 0xFFF;
+	}
+	size_t fat_size = fat_type / 8;
+	size_t position = cluster * fat_size;
+	size_t lba = position / bytes_per_sector;
+	size_t entry = (position % bytes_per_sector) / fat_size;
+	Block* block = device->GetBlock(fat_sector + lba);
+	if ( !block )
+		return eio_cluster;
+	uint32_t result = 0;
+	if ( fat_type == 16 )
+		result = ((uint16_t*) block->block_data)[entry];
+	else if ( fat_type == 32 )
+		result = ((uint32_t*) block->block_data)[entry] & 0x0FFFFFFF;
+	block->Unref();
+	return result;
+}
+
+bool Filesystem::WriteFAT(uint32_t cluster, uint32_t value)
+{
+	assert(device->write);
+	// TODO: Bounds check.
+	if ( fat_type == 12 )
+	{
+		size_t position = cluster + (cluster / 2);
+		size_t lba = position / bytes_per_sector;
+		size_t offset = position % bytes_per_sector;
+		Block* block = device->GetBlock(fat_sector + lba);
+		if ( !block )
+			return false;
+		value = cluster & 1 ? value << 4 : value;
+		uint16_t mask = cluster & 1 ? 0xFFF0 : 0x0FFF;
+		block->BeginWrite();
+		block->block_data[offset] &= ~mask;
+		block->block_data[offset] |= value;
+		if ( ++offset == bytes_per_sector )
+		{
+			block->FinishWrite();
+			block->Unref();
+			if ( !(block = device->GetBlock(fat_sector + lba)) )
+				return false;
+			offset = 0;
+			block->BeginWrite();
+		}
+		block->block_data[offset] &= ~(mask >> 8);
+		block->block_data[offset] |= value >> 8;
+		block->FinishWrite();
+		block->Unref();
+		return true;
+	}
+
+	// TODO: Mirror to the other FATs.
+	size_t fat_size = fat_type / 8;
+	size_t position = cluster * fat_size;
+	size_t lba = position / bytes_per_sector;
+	size_t entry = (position % bytes_per_sector) / fat_size;
+	Block* block = device->GetBlock(fat_sector + lba);
+	if ( !block )
+		return false;
+	block->BeginWrite();
+	if ( fat_type == 16 )
+		((uint16_t*) block->block_data)[entry] = value;
+	else if ( fat_type == 32 )
+	{
+		uint32_t old = ((uint32_t*) block->block_data)[entry] & 0xF0000000;
+		((uint32_t*) block->block_data)[entry] = value | old;
+	}
+	block->FinishWrite();
+	block->Unref();
+	return true;
+}
+
+uint32_t Filesystem::CalculateFreeCount()
+{
+	if ( fat_type == 32 )
+	{
+		// TODO: Verify the fsinfo.
+		Block* block = device->GetBlock(bpb->fat32_fsinfo);
+		if ( !block )
+			return 0xFFFFFFFF;
+		const struct fat_fsinfo* fsinfo =
+			(const struct fat_fsinfo*) block->block_data;
+		uint32_t result = le32toh(fsinfo->free_count);
+		block->Unref();
+		if ( result != 0xFFFFFFFF )
+			return result;
+	}
+	// TODO: Cache these.
+	size_t count = 0;
+	for ( size_t i = 0; i < cluster_count; i++ )
+		if ( !ReadFAT(2 + i) )
+			count++;
+	return count;
+}

fat/filesystem.h

@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * filesystem.h
+ * Filesystem.
+ */
+
+#ifndef FILESYSTEM_H
+#define FILESYSTEM_H
+
+bool is_8_3(const char* name);
+void encode_8_3(const char* decoded, char encoded[8 + 3]);
+void decode_8_3(const char encoded[8 + 3], char decoded[8 + 1 + 3 + 1]);
+void timespec_to_fat(const struct timespec* ts, uint16_t* date, uint16_t* time,
+                     uint8_t* tenths);
+// TODO: Unify into the above.
+uint8_t timespec_to_fat_tenths(struct timespec* ts);
+uint16_t timespec_to_fat_time(struct timespec* ts);
+uint16_t timespec_to_fat_date(struct timespec* ts);
+
+class Device;
+class Inode;
+
+static const size_t INODE_HASH_LENGTH = 1 << 16;
+
+class Filesystem
+{
+public:
+	Filesystem(Device* device, const char* mount_path);
+	~Filesystem();
+
+public:
+	Block* bpb_block;
+	struct fat_bpb* bpb;
+	Device* device;
+	const char* mount_path;
+	mode_t mode_reg;
+	mode_t mode_dir;
+	uid_t uid;
+	gid_t gid;
+	uint32_t block_size;
+	uint16_t bytes_per_sector;
+	uint16_t root_dirent_count;
+	uint32_t sectors_per_fat;
+	uint32_t root_inode_id;
+	uint32_t total_sectors;
+	uint32_t fat_sector; // TODO: Rename to lba
+	uint32_t root_sector; // TODO: Rename to lba
+	uint32_t data_sector; // TODO: Rename to lba
+	uint32_t cluster_count;
+	uint32_t cluster_size;
+	uint8_t fat_type;
+	uint32_t eio_cluster;
+	uint32_t eof_cluster;
+	uint32_t free_search;
+	Inode* mru_inode;
+	Inode* lru_inode;
+	Inode* dirty_inode;
+	Inode* hash_inodes[INODE_HASH_LENGTH];
+	bool dirty;
+
+public:
+	Inode* GetInode(uint32_t inode_id, Block* dirent_block = NULL,
+	                struct fat_dirent* dirent = NULL);
+	uint32_t AllocateCluster();
+	void FreeCluster(uint32_t cluster);
+	uint32_t ReadFAT(uint32_t cluster);
+	bool WriteFAT(uint32_t cluster, uint32_t value);
+	uint32_t CalculateFreeCount();
+	void BeginWrite();
+	void FinishWrite();
+	void Sync();
+
+};
+
+#endif

fat/fsmarshall.cpp

@@ -0,0 +1,858 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2016, 2022, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fsmarshall.cpp
+ * Sortix fsmarshall frontend.
+ */
+
+#if defined(__sortix__)
+
+#include <sys/mman.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <err.h>
+#include <errno.h>
+#include <dirent.h>
+#include <fcntl.h>
+#include <ioleast.h>
+#include <signal.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <timespec.h>
+#include <unistd.h>
+
+#include <sortix/dirent.h>
+
+#include <fsmarshall.h>
+
+#include "block.h"
+#include "device.h"
+#include "fat.h"
+#include "fatfs.h"
+#include "filesystem.h"
+#include "fsmarshall.h"
+#include "fuse.h"
+#include "inode.h"
+
+bool RespondData(int chl, const void* ptr, size_t count)
+{
+	return writeall(chl, ptr, count) == count;
+}
+
+bool RespondHeader(int chl, size_t type, size_t size)
+{
+	struct fsm_msg_header hdr;
+	hdr.msgtype = type;
+	hdr.msgsize = size;
+	return RespondData(chl, &hdr, sizeof(hdr));
+}
+
+bool RespondMessage(int chl, size_t type, const void* ptr, size_t count)
+{
+	return RespondHeader(chl, type, count) &&
+	       RespondData(chl, ptr, count);
+}
+
+bool RespondError(int chl, int errnum)
+{
+	struct fsm_resp_error body;
+	body.errnum = errnum;
+	//fprintf(stderr, "fatfs: sending error %i (%s)\n", errnum, strerror(errnum));
+	return RespondMessage(chl, FSM_RESP_ERROR, &body, sizeof(body));
+}
+
+bool RespondSuccess(int chl)
+{
+	struct fsm_resp_success body;
+	return RespondMessage(chl, FSM_RESP_SUCCESS, &body, sizeof(body));
+}
+
+bool RespondStat(int chl, struct stat* st)
+{
+	struct fsm_resp_stat body;
+	body.st = *st;
+	return RespondMessage(chl, FSM_RESP_STAT, &body, sizeof(body));
+}
+
+bool RespondStatVFS(int chl, struct statvfs* stvfs)
+{
+	struct fsm_resp_statvfs body;
+	body.stvfs = *stvfs;
+	return RespondMessage(chl, FSM_RESP_STATVFS, &body, sizeof(body));
+}
+
+bool RespondSeek(int chl, off_t offset)
+{
+	struct fsm_resp_lseek body;
+	body.offset = offset;
+	return RespondMessage(chl, FSM_RESP_LSEEK, &body, sizeof(body));
+}
+
+bool RespondRead(int chl, const uint8_t* buf, size_t count)
+{
+	struct fsm_resp_read body;
+	body.count = count;
+	return RespondMessage(chl, FSM_RESP_READ, &body, sizeof(body)) &&
+	       RespondData(chl, buf, count);
+}
+
+bool RespondReadlink(int chl, const uint8_t* buf, size_t count)
+{
+	struct fsm_resp_readlink body;
+	body.targetlen = count;
+	return RespondMessage(chl, FSM_RESP_READLINK, &body, sizeof(body)) &&
+	       RespondData(chl, buf, count);
+}
+
+bool RespondWrite(int chl, size_t count)
+{
+	struct fsm_resp_write body;
+	body.count = count;
+	return RespondMessage(chl, FSM_RESP_WRITE, &body, sizeof(body));
+}
+
+bool RespondOpen(int chl, ino_t ino, mode_t type)
+{
+	struct fsm_resp_open body;
+	body.ino = ino;
+	body.type = type;
+	return RespondMessage(chl, FSM_RESP_OPEN, &body, sizeof(body));
+}
+
+bool RespondMakeDir(int chl, ino_t ino)
+{
+	struct fsm_resp_mkdir body;
+	body.ino = ino;
+	return RespondMessage(chl, FSM_RESP_MKDIR, &body, sizeof(body));
+}
+
+bool RespondReadDir(int chl, struct dirent* dirent)
+{
+	struct fsm_resp_readdirents body;
+	body.ino = dirent->d_ino;
+	body.type = dirent->d_type;
+	body.namelen = dirent->d_namlen;
+	return RespondMessage(chl, FSM_RESP_READDIRENTS, &body, sizeof(body)) &&
+	       RespondData(chl, dirent->d_name, dirent->d_namlen);
+}
+
+bool RespondTCGetBlob(int chl, const void* data, size_t data_size)
+{
+	struct fsm_resp_tcgetblob body;
+	body.count = data_size;
+	return RespondMessage(chl, FSM_RESP_TCGETBLOB, &body, sizeof(body)) &&
+	       RespondData(chl, data, data_size);
+}
+
+Inode* SafeGetInode(Filesystem* fs, ino_t ino)
+{
+	if ( (uint32_t) ino != ino )
+		return errno = EBADF, (Inode*) NULL;
+	return fs->GetInode((uint32_t) ino);
+}
+
+void HandleRefer(int chl, struct fsm_req_refer* msg, Filesystem* fs)
+{
+	(void) chl;
+	if ( Inode* inode = SafeGetInode(fs, (uint32_t) msg->ino) )
+	{
+		if ( inode->implied_reference )
+			inode->implied_reference--;
+		else
+			inode->RemoteRefer();
+		inode->Unref();
+	}
+}
+
+void HandleUnref(int chl, struct fsm_req_unref* msg, Filesystem* fs)
+{
+	(void) chl;
+	if ( Inode* inode = SafeGetInode(fs, (uint32_t) msg->ino) )
+	{
+		inode->RemoteUnref();
+		inode->Unref();
+	}
+}
+
+void HandleSync(int chl, struct fsm_req_sync* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	inode->Sync();
+	inode->Unref();
+	RespondSuccess(chl);
+}
+
+void HandleStat(int chl, struct fsm_req_stat* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	struct stat st;
+	StatInode(inode, &st);
+	inode->Unref();
+	RespondStat(chl, &st);
+}
+
+void HandleChangeMode(int chl, struct fsm_req_chmod* msg, Filesystem* fs)
+{
+	if ( !fs->device->write ) { RespondError(chl, EROFS); return; }
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( !inode->ChangeMode(msg->mode) )
+		RespondError(chl, errno);
+	else
+		RespondSuccess(chl);
+	inode->Unref();
+}
+
+void HandleChangeOwner(int chl, struct fsm_req_chown* msg, Filesystem* fs)
+{
+	if ( !fs->device->write ) { RespondError(chl, EROFS); return; }
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( !inode->ChangeOwner(msg->uid, msg->gid) )
+		RespondError(chl, errno);
+	else
+		RespondSuccess(chl);
+	inode->Unref();
+}
+
+void HandleUTimens(int chl, struct fsm_req_utimens* msg, Filesystem* fs)
+{
+	if ( !fs->device->write ) { RespondError(chl, EROFS); return; }
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	inode->UTimens(msg->times);
+	inode->Unref();
+	RespondSuccess(chl);
+}
+
+void HandleTruncate(int chl, struct fsm_req_truncate* msg, Filesystem* fs)
+{
+	if ( !fs->device->write ) { RespondError(chl, EROFS); return; }
+	if ( msg->size < 0 ) { RespondError(chl, EINVAL); return; }
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	inode->Truncate((uint64_t) msg->size);
+	inode->Unref();
+	RespondSuccess(chl);
+}
+
+void HandleSeek(int chl, struct fsm_req_lseek* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( msg->whence == SEEK_SET )
+		RespondSeek(chl, msg->offset);
+	else if ( msg->whence == SEEK_END )
+	{
+		off_t inode_size = inode->Size();
+		if ( (msg->offset < 0 && inode_size + msg->offset < 0) ||
+		     (0 <= msg->offset && OFF_MAX - inode_size < msg->offset) )
+			RespondError(chl, EOVERFLOW);
+		else
+			RespondSeek(chl, msg->offset + inode_size);
+	}
+	else
+		RespondError(chl, EINVAL);
+	inode->Unref();
+}
+
+void HandleReadAt(int chl, struct fsm_req_pread* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	uint8_t* buf = (uint8_t*) malloc(msg->count);
+	if ( !buf ) { inode->Unref(); RespondError(chl, errno); return; }
+	ssize_t amount = inode->ReadAt(buf, msg->count, msg->offset);
+	inode->Unref();
+	if ( amount < 0 ) { free(buf); RespondError(chl, errno); return; }
+	RespondRead(chl, buf, amount);
+	free(buf);
+}
+
+void HandleWriteAt(int chl, struct fsm_req_pwrite* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	const uint8_t* buf = (const uint8_t*) &msg[1];
+	ssize_t amount = inode->WriteAt(buf, msg->count, msg->offset);
+	inode->Unref();
+	if ( amount < 0 ) { RespondError(chl, errno); return; }
+	RespondWrite(chl, amount);
+}
+
+void HandleOpen(int chl, struct fsm_req_open* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	Inode* result = inode->Open(path, msg->flags, msg->mode & 07777);
+
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondOpen(chl, result->inode_id, result->Mode() & S_IFMT);
+	// TODO: Unfortunately Open does not implicitly imply RemoteRefer so we need
+	// to try and pretend that it does so the inode isn't destroyed early.
+	result->implied_reference++;
+	result->RemoteRefer();
+	result->Unref();
+}
+
+void HandleMakeDir(int chl, struct fsm_req_mkdir* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	Inode* result = inode->CreateDirectory(path, msg->mode & 07777);
+
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondMakeDir(chl, result->inode_id);
+	result->Unref();
+}
+
+// TODO: Encapsulate.
+void HandleReadDir(int chl, struct fsm_req_readdirents* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( !S_ISDIR(inode->Mode()) )
+	{
+		inode->Unref();
+		RespondError(chl, ENOTDIR);
+		return;
+	}
+	union
+	{
+		struct dirent kernel_entry;
+		// TODO: Adjust with LFN's real limit.
+		uint8_t padding[sizeof(struct dirent) + 256];
+	};
+	memset(&kernel_entry, 0, sizeof(kernel_entry));
+
+	if ( inode->inode_id == inode->filesystem->root_inode_id )
+	{
+		if ( msg->rec_num < 2 )
+		{
+			const char* name = msg->rec_num ? ".." : ".";
+			size_t name_len = strlen(name);
+			kernel_entry.d_reclen = sizeof(kernel_entry) + name_len;
+			kernel_entry.d_ino = inode->inode_id;
+			kernel_entry.d_dev = 0;
+			kernel_entry.d_type = DT_DIR;
+			kernel_entry.d_namlen = name_len;
+			memcpy(kernel_entry.d_name, name, name_len);
+			size_t dname_offset = offsetof(struct dirent, d_name);
+			padding[dname_offset + kernel_entry.d_namlen] = '\0';
+			RespondReadDir(chl, &kernel_entry);
+			return;
+		}
+		msg->rec_num -= 2;
+	}
+
+	uint32_t cluster = inode->first_cluster;
+	uint8_t sector = 0;
+	uint16_t offset = 0;
+	Block* block = NULL;
+	while ( inode->Iterate(&block, &cluster, &sector, &offset) )
+	{
+		const uint8_t* block_data = block->block_data + offset;
+		const struct fat_dirent* entry = (const struct fat_dirent*) block_data;
+		if ( !entry->name[0] )
+			break;
+		if ( (unsigned char) entry->name[0] != 0xE5 &&
+		     !(entry->attributes & FAT_ATTRIBUTE_VOLUME_ID) &&
+		     !(msg->rec_num--) )
+		{
+			char name[8 + 1 + 3 + 1];
+			decode_8_3(entry->name, name);
+			size_t name_len = strnlen(entry->name, sizeof(entry->name));
+			uint8_t file_type =
+				entry->attributes & FAT_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG;
+			kernel_entry.d_reclen = sizeof(kernel_entry) + name_len;
+			kernel_entry.d_ino = entry->cluster_low | entry->cluster_high << 16;
+			kernel_entry.d_dev = 0;
+			kernel_entry.d_type = file_type;
+			kernel_entry.d_namlen = name_len;
+			memcpy(kernel_entry.d_name, name, name_len);
+			size_t dname_offset = offsetof(struct dirent, d_name);
+			padding[dname_offset + kernel_entry.d_namlen] = '\0';
+			block->Unref();
+			inode->Unref();
+			RespondReadDir(chl, &kernel_entry);
+			return;
+		}
+		offset += sizeof(struct fat_dirent);
+	}
+	int errnum = errno;
+	if ( block )
+		block->Unref();
+	inode->Unref();
+	if ( errnum )
+	{
+		RespondError(chl, errnum);
+		return;
+	}
+
+	kernel_entry.d_reclen = sizeof(kernel_entry);
+	RespondReadDir(chl, &kernel_entry);
+}
+
+void HandleIsATTY(int chl, struct fsm_req_isatty* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	RespondError(chl, ENOTTY);
+	inode->Unref();
+}
+
+void HandleUnlink(int chl, struct fsm_req_unlink* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->Unlink(path, false);
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleRemoveDir(int chl, struct fsm_req_rmdir* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->RemoveDirectory(path);
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleLink(int chl, struct fsm_req_link* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	Inode* dest = SafeGetInode(fs, msg->linkino);
+	if ( !dest ) { inode->Unref(); RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->Link(path, dest, false);
+
+	free(path);
+	dest->Unref();
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleSymlink(int chl, struct fsm_req_symlink* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* dest_raw = (char*) &(msg[1]);
+	char* dest = (char*) malloc(msg->targetlen + 1);
+	if ( !dest )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(dest, dest_raw, msg->targetlen);
+	dest[msg->targetlen] = '\0';
+
+	char* path_raw = (char*) dest_raw + msg->targetlen;
+	char* path = (char*) malloc(msg->namelen + 1);
+	if ( !path )
+	{
+		free(dest);
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, path_raw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->Symlink(path, dest);
+
+	free(path);
+	free(dest);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleReadlink(int chl, struct fsm_req_readlink* msg, Filesystem* fs)
+{
+	(void) msg;
+	(void) fs;
+	RespondError(chl, EINVAL);
+}
+
+void HandleRename(int chl, struct fsm_req_rename* msg, Filesystem* fs)
+{
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->oldnamelen+1 + msg->newnamelen+1);
+	if ( !path ) { RespondError(chl, errno); return; }
+	memcpy(path, pathraw, msg->oldnamelen);
+	path[msg->oldnamelen] = '\0';
+	memcpy(path + msg->oldnamelen + 1, pathraw + msg->oldnamelen, msg->newnamelen);
+	path[msg->oldnamelen + 1 + msg->newnamelen] = '\0';
+
+	const char* oldname = path;
+	const char* newname = path + msg->oldnamelen + 1;
+
+	Inode* olddir = SafeGetInode(fs, msg->olddirino);
+	if ( !olddir ) { free(path); RespondError(chl, errno); return; }
+	Inode* newdir = SafeGetInode(fs, msg->newdirino);
+	if ( !newdir ) { olddir->Unref(); free(path); RespondError(chl, errno); return; }
+
+	bool result = newdir->Rename(olddir, oldname, newname);
+
+	newdir->Unref();
+	olddir->Unref();
+	free(path);
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleStatVFS(int chl, struct fsm_req_statvfs* msg, Filesystem* fs)
+{
+	(void) msg;
+	struct statvfs stvfs;
+	stvfs.f_bsize = fs->cluster_size;
+	stvfs.f_frsize = fs->cluster_size;
+	stvfs.f_blocks = fs->cluster_count;
+	// TODO: Locate FsInfo and count on FAT12/FAT16.
+	stvfs.f_bfree = fs->CalculateFreeCount();
+	stvfs.f_bavail = stvfs.f_bfree;
+	stvfs.f_files = stvfs.f_blocks;
+	stvfs.f_ffree = stvfs.f_bfree;
+	stvfs.f_favail = stvfs.f_files;
+	stvfs.f_fsid = 0;
+	stvfs.f_flag = 0;
+	if ( !fs->device->write )
+		stvfs.f_flag |= ST_RDONLY;
+	stvfs.f_namemax = 8 + 3; // TODO: Long file name support.
+	RespondStatVFS(chl, &stvfs);
+}
+
+void HandleTCGetBlob(int chl, struct fsm_req_tcgetblob* msg, Filesystem* fs)
+{
+	char* nameraw = (char*) &(msg[1]);
+	char* name = (char*) malloc(msg->namelen + 1);
+	if ( !name )
+		return (void) RespondError(chl, errno);
+	memcpy(name, nameraw, msg->namelen);
+	name[msg->namelen] = '\0';
+
+	static const char index[] =
+		"device-path\0filesystem-type\0filesystem-uuid\0mount-path\0"
+		"fat-size\0volume-id\0volume-label\0";
+	if ( !strcmp(name, "") )
+		RespondTCGetBlob(chl, index, sizeof(index) - 1);
+	else if ( !strcmp(name, "device-path") )
+		RespondTCGetBlob(chl, fs->device->path, strlen(fs->device->path));
+	else if ( !strcmp(name, "filesystem-type") )
+		RespondTCGetBlob(chl, "fat", strlen("fat"));
+	else if ( !strcmp(name, "fat-size") )
+	{
+		const char* str = fs->fat_type == 32 ? "32" :
+		                  fs->fat_type == 16 ? "16" : "12";
+		RespondTCGetBlob(chl, str, strlen(str));
+	}
+	else if ( !strcmp(name, "filesystem-uuid") )
+	{
+		unsigned char uuid[16];
+		if ( fs->fat_type == 32 )
+		{
+			memcpy(uuid, &fs->bpb->fat32_volume_id, 4);
+			memcpy(uuid + 4, &fs->bpb->fat32_volume_label, 11);
+		}
+		else
+		{
+			memcpy(uuid, &fs->bpb->fat12_volume_id, 4);
+			memcpy(uuid + 4, &fs->bpb->fat12_volume_label, 11);
+		}
+		uuid[15] = '\0';
+		RespondTCGetBlob(chl, uuid, sizeof(uuid));
+	}
+	else if ( !strcmp(name, "volume-id") )
+	{
+		if ( fs->fat_type == 32 )
+			RespondTCGetBlob(chl, &fs->bpb->fat32_volume_id, 4);
+		else
+			RespondTCGetBlob(chl, &fs->bpb->fat12_volume_id, 4);
+	}
+	else if ( !strcmp(name, "volume-label") )
+	{
+		if ( fs->fat_type == 32 )
+			RespondTCGetBlob(chl, &fs->bpb->fat32_volume_label, 11);
+		else
+			RespondTCGetBlob(chl, &fs->bpb->fat12_volume_label, 11);
+	}
+	else if ( !strcmp(name, "mount-path") )
+		RespondTCGetBlob(chl, fs->mount_path, strlen(fs->mount_path));
+	else
+		RespondError(chl, ENOENT);
+
+	free(name);
+}
+
+void HandleIncomingMessage(int chl, struct fsm_msg_header* hdr, Filesystem* fs)
+{
+	request_uid = hdr->uid;
+	request_gid = hdr->gid;
+	typedef void (*handler_t)(int, void*, Filesystem*);
+	handler_t handlers[FSM_MSG_NUM] = { NULL };
+	handlers[FSM_REQ_SYNC] = (handler_t) HandleSync;
+	handlers[FSM_REQ_STAT] = (handler_t) HandleStat;
+	handlers[FSM_REQ_CHMOD] = (handler_t) HandleChangeMode;
+	handlers[FSM_REQ_CHOWN] = (handler_t) HandleChangeOwner;
+	handlers[FSM_REQ_TRUNCATE] = (handler_t) HandleTruncate;
+	handlers[FSM_REQ_LSEEK] = (handler_t) HandleSeek;
+	handlers[FSM_REQ_PREAD] = (handler_t) HandleReadAt;
+	handlers[FSM_REQ_OPEN] = (handler_t) HandleOpen;
+	handlers[FSM_REQ_READDIRENTS] = (handler_t) HandleReadDir;
+	handlers[FSM_REQ_PWRITE] = (handler_t) HandleWriteAt;
+	handlers[FSM_REQ_ISATTY] = (handler_t) HandleIsATTY;
+	handlers[FSM_REQ_UTIMENS] = (handler_t) HandleUTimens;
+	handlers[FSM_REQ_MKDIR] = (handler_t) HandleMakeDir;
+	handlers[FSM_REQ_RMDIR] = (handler_t) HandleRemoveDir;
+	handlers[FSM_REQ_UNLINK] = (handler_t) HandleUnlink;
+	handlers[FSM_REQ_LINK] = (handler_t) HandleLink;
+	handlers[FSM_REQ_SYMLINK] = (handler_t) HandleSymlink;
+	handlers[FSM_REQ_READLINK] = (handler_t) HandleReadlink;
+	handlers[FSM_REQ_RENAME] = (handler_t) HandleRename;
+	handlers[FSM_REQ_REFER] = (handler_t) HandleRefer;
+	handlers[FSM_REQ_UNREF] = (handler_t) HandleUnref;
+	handlers[FSM_REQ_STATVFS] = (handler_t) HandleStatVFS;
+	handlers[FSM_REQ_TCGETBLOB] = (handler_t) HandleTCGetBlob;
+	if ( FSM_MSG_NUM <= hdr->msgtype || !handlers[hdr->msgtype] )
+	{
+		warn("message type %zu not supported\n", hdr->msgtype);
+		RespondError(chl, ENOTSUP);
+		return;
+	}
+	uint8_t body_buffer[65536];
+	uint8_t* body = body_buffer;
+	if ( sizeof(body_buffer) < hdr->msgsize )
+	{
+		body = (uint8_t*) mmap(NULL, hdr->msgsize, PROT_READ | PROT_WRITE,
+		                       MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+		if ( (void*) body == MAP_FAILED )
+		{
+			RespondError(chl, errno);
+			return;
+		}
+	}
+	if ( readall(chl, body, hdr->msgsize) == hdr->msgsize )
+		handlers[hdr->msgtype](chl, body, fs);
+	else
+		RespondError(chl, errno);
+	if ( sizeof(body_buffer) < hdr->msgsize )
+		munmap(body, hdr->msgsize);
+}
+static volatile bool should_terminate = false;
+
+void TerminationHandler(int)
+{
+	should_terminate = true;
+}
+
+static void ready(void)
+{
+	const char* readyfd_env = getenv("READYFD");
+	if ( !readyfd_env )
+		return;
+	int readyfd = atoi(readyfd_env);
+	char c = '\n';
+	write(readyfd, &c, 1);
+	close(readyfd);
+	unsetenv("READYFD");
+}
+
+int fsmarshall_main(const char* argv0,
+                    const char* mount_path,
+                    bool foreground,
+                    Filesystem* fs,
+                    Device* dev)
+{
+	(void) argv0;
+
+	// Stat the root directory;
+	struct stat root_dir_st;
+	memset(&root_dir_st, 0, sizeof(root_dir_st));
+	root_dir_st.st_ino = fs->root_inode_id;
+	root_dir_st.st_mode = S_IFDIR | 0755;
+
+	// Create a filesystem server connected to the kernel that we'll listen on.
+	int serverfd = fsm_mountat(AT_FDCWD, mount_path, &root_dir_st, 0);
+	if ( serverfd < 0 )
+		err(1, "%s", mount_path);
+
+	// Make sure the server isn't unexpectedly killed and data is lost.
+	signal(SIGINT, TerminationHandler);
+	signal(SIGTERM, TerminationHandler);
+	signal(SIGQUIT, TerminationHandler);
+
+	// Become a background process in its own process group by default.
+	if ( !foreground )
+	{
+		pid_t child_pid = fork();
+		if ( child_pid < 0 )
+			err(1, "fork");
+		if ( child_pid )
+			exit(0);
+		setpgid(0, 0);
+	}
+	else
+		ready();
+
+	dev->SpawnSyncThread();
+
+	// Listen for filesystem messages and sync the filesystem every few seconds.
+	struct timespec last_sync_at;
+	clock_gettime(CLOCK_MONOTONIC, &last_sync_at);
+	int channel;
+	while ( 0 <= (channel = accept(serverfd, NULL, NULL)) )
+	{
+		if ( should_terminate )
+			break;
+		struct fsm_msg_header hdr;
+		size_t amount;
+		if ( (amount = readall(channel, &hdr, sizeof(hdr))) != sizeof(hdr) )
+		{
+			//warn("incomplete header: got %zi of %zu bytes", amount, sizeof(hdr));
+			errno = 0;
+			continue;
+		}
+		HandleIncomingMessage(channel, &hdr, fs);
+		close(channel);
+
+		if ( dev->write && !dev->has_sync_thread )
+		{
+			struct timespec now;
+			clock_gettime(CLOCK_MONOTONIC, &now);
+
+			if ( 5 <= timespec_sub(now, last_sync_at).tv_sec )
+			{
+				fs->Sync();
+				last_sync_at = now;
+			}
+		}
+	}
+
+	// TODO: Replace with FAT concept.
+	// Garbage collect all open inode references.
+	while ( fs->mru_inode )
+	{
+		Inode* inode = fs->mru_inode;
+		if ( inode->remote_reference_count )
+			inode->RemoteUnref();
+		else if ( inode->reference_count )
+			inode->Unref();
+	}
+
+	// Sync the filesystem before shutting down.
+	if ( dev->write )
+		fs->Sync();
+
+	close(serverfd);
+
+	delete fs;
+	delete dev;
+
+	return 0;
+}
+
+#endif

fat/fsmarshall.h

@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fsmarshall.h
+ * Sortix fsmarshall frontend.
+ */
+
+#ifndef FSMARSHALL_H
+#define FSMARSHALL_H
+
+class Device;
+class Filesystem;
+
+int fsmarshall_main(const char* argv0,
+                    const char* mount_path,
+                    bool foreground,
+                    Filesystem* fs,
+                    Device* dev);
+#endif

fat/fuse.cpp

@@ -0,0 +1,618 @@
+/*
+ * Copyright (c) 2013, 2014, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fuse.cpp
+ * FUSE frontend.
+ */
+
+#if !defined(__sortix__)
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <limits.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define FUSE_USE_VERSION 26
+#include <fuse.h>
+
+#include "ext-constants.h"
+#include "ext-structs.h"
+
+#include "blockgroup.h"
+#include "block.h"
+#include "device.h"
+#include "fatfs.h"
+#include "filesystem.h"
+#include "fuse.h"
+#include "inode.h"
+
+struct fat_fuse_ctx
+{
+	Device* dev;
+	Filesystem* fs;
+};
+
+#ifndef S_SETABLE
+#define S_SETABLE 02777
+#endif
+
+#define FUSE_FS (((struct fat_fuse_ctx*) (fuse_get_context()->private_data))->fs)
+
+void* fat_fuse_init(struct fuse_conn_info* /*conn*/)
+{
+	return fuse_get_context()->private_data;
+}
+
+void fat_fuse_destroy(void* fs_private)
+{
+	struct fat_fuse_ctx* fat_fuse_ctx = (struct fat_fuse_ctx*) fs_private;
+	while ( fat_fuse_ctx->fs->mru_inode )
+	{
+		Inode* inode = fat_fuse_ctx->fs->mru_inode;
+		if ( inode->remote_reference_count )
+			inode->RemoteUnref();
+		else if ( inode->reference_count )
+			inode->Unref();
+	}
+	fat_fuse_ctx->fs->Sync();
+	fat_fuse_ctx->dev->Sync();
+	delete fat_fuse_ctx->fs; fat_fuse_ctx->fs = NULL;
+	delete fat_fuse_ctx->dev; fat_fuse_ctx->dev = NULL;
+}
+
+Inode* fat_fuse_resolve_path(const char* path)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode(FAT_ROOT_INO);
+	if ( !inode )
+		return (Inode*) NULL;
+	while ( path[0] )
+	{
+		if ( *path == '/' )
+		{
+			if ( !FAT_S_ISDIR(inode->Mode()) )
+				return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+			path++;
+			continue;
+		}
+		size_t elem_len = strcspn(path, "/");
+		char* elem = strndup(path, elem_len);
+		if ( !elem )
+			return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+		path += elem_len;
+		Inode* next = inode->Open(elem, O_RDONLY, 0);
+		free(elem);
+		inode->Unref();
+		if ( !next )
+			return NULL;
+		inode = next;
+	}
+	return inode;
+}
+
+// Assumes that the path doesn't end with / unless it's the root directory.
+Inode* fat_fuse_parent_dir(const char** path_ptr)
+{
+	const char* path = *path_ptr;
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode(FAT_ROOT_INO);
+	if ( !inode )
+		return (Inode*) NULL;
+	while ( strchr(path, '/') )
+	{
+		if ( *path == '/' )
+		{
+			if ( !FAT_S_ISDIR(inode->Mode()) )
+				return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+			path++;
+			continue;
+		}
+		size_t elem_len = strcspn(path, "/");
+		char* elem = strndup(path, elem_len);
+		if ( !elem )
+			return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+		path += elem_len;
+		Inode* next = inode->Open(elem, O_RDONLY, 0);
+		free(elem);
+		inode->Unref();
+		if ( !next )
+			return (Inode*) NULL;
+		inode = next;
+	}
+	*path_ptr = *path ? path : ".";
+	assert(!strchr(*path_ptr, '/'));
+	return inode;
+}
+
+int fat_fuse_getattr(const char* path, struct stat* st)
+{
+	Inode* inode = fat_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	StatInode(inode, st);
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_fgetattr(const char* /*path*/, struct stat* st,
+                       struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	StatInode(inode, st);
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_readlink(const char* path, char* buf, size_t bufsize)
+{
+	Inode* inode = fat_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	if ( !FAT_S_ISLNK(inode->Mode()) )
+		return inode->Unref(), -(errno = EINVAL);
+	if ( !bufsize )
+		return inode->Unref(), -(errno = EINVAL);
+	ssize_t amount = inode->ReadAt((uint8_t*) buf, bufsize, 0);
+	if ( amount < 0 )
+		return inode->Unref(), -errno;
+	buf[(size_t) amount < bufsize ? (size_t) amount : bufsize - 1] = '\0';
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_mknod(const char* path, mode_t mode, dev_t dev)
+{
+	(void) path;
+	(void) mode;
+	(void) dev;
+	return -(errno = ENOSYS);
+}
+
+int fat_fuse_mkdir(const char* path, mode_t mode)
+{
+	Inode* inode = fat_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	Inode* newdir = inode->CreateDirectory(path, ExtModeFromHostMode(mode));
+	inode->Unref();
+	if ( !newdir )
+		return -errno;
+	newdir->Unref();
+	return 0;
+}
+
+int fat_fuse_unlink(const char* path)
+{
+	Inode* inode = fat_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	bool success = inode->Unlink(path, false);
+	inode->Unref();
+	return success ? 0 : -errno;
+}
+
+int fat_fuse_rmdir(const char* path)
+{
+	Inode* inode = fat_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	bool success = inode->RemoveDirectory(path);
+	inode->Unref();
+	return success ? 0 : -errno;
+}
+
+int fat_fuse_symlink(const char* oldname, const char* newname)
+{
+	Inode* newdir = fat_fuse_parent_dir(&newname);
+	if ( !newdir )
+		return -errno;
+	bool success = newdir->Symlink(newname, oldname);
+	newdir->Unref();
+	return success ? 0 : -errno;
+}
+
+int fat_fuse_rename(const char* oldname, const char* newname)
+{
+	Inode* olddir = fat_fuse_parent_dir(&oldname);
+	if ( !olddir )
+		return -errno;
+	Inode* newdir = fat_fuse_parent_dir(&newname);
+	if ( !newdir )
+		return olddir->Unref(), -errno;
+	bool success = newdir->Rename(olddir, oldname, newname);
+	newdir->Unref();
+	olddir->Unref();
+	return success ? 0 : -errno;
+}
+
+int fat_fuse_link(const char* oldname, const char* newname)
+{
+	Inode* inode = fat_fuse_resolve_path(oldname);
+	if ( !inode )
+		return -errno;
+	Inode* newdir = fat_fuse_parent_dir(&newname);
+	if ( !newdir )
+		return inode->Unref(), -errno;
+	bool success = inode->Link(newname, inode, false);
+	newdir->Unref();
+	inode->Unref();
+	return success ? 0 : -errno;
+}
+
+int fat_fuse_chmod(const char* path, mode_t mode)
+{
+	Inode* inode = fat_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	if ( !FUSE_FS->device->write )
+		return inode->Unref(), -(errno = EROFS);
+	uint32_t req_mode = ExtModeFromHostMode(mode);
+	uint32_t old_mode = inode->Mode();
+	uint32_t new_mode = (old_mode & ~S_SETABLE) | (req_mode & S_SETABLE);
+	inode->SetMode(new_mode);
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_chown(const char* path, uid_t owner, gid_t group)
+{
+	Inode* inode = fat_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	if ( !FUSE_FS->device->write )
+		return inode->Unref(), -(errno = EROFS);
+	inode->SetUserId((uint32_t) owner);
+	inode->SetGroupId((uint32_t) group);
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_truncate(const char* path, off_t size)
+{
+	Inode* inode = fat_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	if ( !FUSE_FS->device->write )
+		return inode->Unref(), -(errno = EROFS);
+	inode->Truncate((uint64_t) size);
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_ftruncate(const char* /*path*/, off_t size,
+                        struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	if ( !FUSE_FS->device->write )
+		return inode->Unref(), -(errno = EROFS);
+	inode->Truncate((uint64_t) size);
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_open(const char* path, struct fuse_file_info* fi)
+{
+	int flags = fi->flags;
+	Inode* dir = fat_fuse_parent_dir(&path);
+	if ( !dir )
+		return -errno;
+	Inode* result = dir->Open(path, flags, 0);
+	dir->Unref();
+	if ( !result )
+		return -errno;
+	fi->fh = (uint64_t) result->inode_id;
+	fi->keep_cache = 1;
+	result->RemoteRefer();
+	result->Unref();
+	return 0;
+}
+
+int fat_fuse_access(const char* path, int mode)
+{
+	Inode* dir = fat_fuse_parent_dir(&path);
+	if ( !dir )
+		return -errno;
+	Inode* result = dir->Open(path, O_RDONLY, 0);
+	dir->Unref();
+	if ( !result )
+		return -errno;
+	(void) mode;
+	result->Unref();
+	return 0;
+}
+
+int fat_fuse_create(const char* path, mode_t mode, struct fuse_file_info* fi)
+{
+	int flags = fi->flags | O_CREAT;
+	Inode* inode = fat_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	Inode* result = inode->Open(path, flags, ExtModeFromHostMode(mode));
+	inode->Unref();
+	if ( !result )
+		return -errno;
+	fi->fh = (uint64_t) result->inode_id;
+	fi->keep_cache = 1;
+	result->RemoteRefer();
+	result->Unref();
+	return 0;
+}
+
+int fat_fuse_opendir(const char* path, struct fuse_file_info* fi)
+{
+	return fat_fuse_open(path, fi);
+}
+
+int fat_fuse_read(const char* /*path*/, char* buf, size_t count, off_t offset,
+                   struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	if ( INT_MAX < count )
+		count = INT_MAX;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	ssize_t result = inode->ReadAt((uint8_t*) buf, count, offset);
+	inode->Unref();
+	return 0 <= result ? (int) result : -errno;
+}
+
+int fat_fuse_write(const char* /*path*/, const char* buf, size_t count,
+                    off_t offset, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	if ( INT_MAX < count )
+		count = INT_MAX;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	ssize_t result = inode->WriteAt((const uint8_t*) buf, count, offset);
+	inode->Unref();
+	return 0 <= result ? (int) result : -errno;
+}
+
+int fat_fuse_statfs(const char* /*path*/, struct statvfs* stvfs)
+{
+	memset(stvfs, 0, sizeof(*stvfs));
+	Filesystem* fs = FUSE_FS;
+	stvfs->f_bsize = fs->block_size;
+	stvfs->f_frsize = fs->block_size;
+	stvfs->f_blocks = fs->num_blocks;
+	stvfs->f_bfree = fs->sb->s_free_blocks_count;
+	stvfs->f_bavail = fs->sb->s_free_blocks_count;
+	stvfs->f_files = fs->num_inodes;
+	stvfs->f_ffree = fs->sb->s_free_inodes_count;
+	stvfs->f_favail = fs->sb->s_free_inodes_count;
+	stvfs->f_ffree = fs->sb->s_free_inodes_count;
+	stvfs->f_fsid = 0;
+	stvfs->f_flag = 0;
+	if ( !fs->device->write )
+		stvfs->f_flag |= ST_RDONLY;
+	stvfs->f_namemax = 255;
+	return 0;
+}
+
+int fat_fuse_flush(const char* /*path*/, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->Sync();
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_release(const char* /*path*/, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->RemoteUnref();
+	inode->Unref();
+	return 0;
+}
+
+int fat_fuse_releasedir(const char* path, struct fuse_file_info* fi)
+{
+	return fat_fuse_release(path, fi);
+}
+
+int fat_fuse_fsync(const char* /*path*/, int data, struct fuse_file_info* fi)
+{
+	(void) data;
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->Sync();
+	inode->Unref();
+	return 0;
+}
+
+/*int fat_fuse_syncdir(const char* path, int data, struct fuse_file_info* fi)
+{
+	return fat_fuse_sync(path, data, fi);
+}*/
+
+/*int fat_fuse_setxattr(const char *, const char *, const char *, size_t, int)
+{
+	return -(errno = ENOSYS);
+}*/
+
+/*int fat_fuse_getxattr(const char *, const char *, char *, size_t)
+{
+	return -(errno = ENOSYS);
+}*/
+
+/*int fat_fuse_listxattr(const char *, char *, size_t)
+{
+	return -(errno = ENOSYS);
+}*/
+
+/*int fat_fuse_removexattr(const char *, const char *)
+{
+	return -(errno = ENOSYS);
+}*/
+
+int fat_fuse_readdir(const char* /*path*/, void* buf, fuse_fill_dir_t filler,
+                      off_t rec_num, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	if ( !S_ISDIR(inode->Mode()) )
+		return inode->Unref(), -(errno = ENOTDIR);
+
+	uint64_t file_size = inode->Size();
+	uint64_t offset = 0;
+	Block* block = NULL;
+	uint64_t block_id = 0;
+	while ( offset < file_size )
+	{
+		uint64_t entry_block_id = offset / fs->block_size;
+		uint64_t entry_block_offset = offset % fs->block_size;
+		if ( block && block_id != entry_block_id )
+			block->Unref(),
+			block = NULL;
+		if ( !block && !(block = inode->GetBlock(block_id = entry_block_id)) )
+			return inode->Unref(), -errno;
+		const uint8_t* block_data = block->block_data + entry_block_offset;
+		const struct ext_dirent* entry = (const struct ext_dirent*) block_data;
+		if ( entry->inode && entry->name_len && (!rec_num || !rec_num--) )
+		{
+			char* entry_name = strndup(entry->name, entry->name_len);
+			if ( !entry_name )
+				return block->Unref(), inode->Unref(), -errno;
+			memcpy(entry_name, entry->name, entry->name_len);
+			bool full = filler(buf, entry_name, NULL, 0);
+			free(entry_name);
+			if ( full )
+			{
+				block->Unref();
+				inode->Unref();
+				return 0;
+			}
+		}
+		offset += entry->reclen;
+	}
+	if ( block )
+		block->Unref();
+
+	inode->Unref();
+	return 0;
+}
+
+/*int fat_fuse_lock(const char*, struct fuse_file_info*, int, struct flock*)
+{
+	return -(errno = ENOSYS);
+}*/
+
+int fat_fuse_utimens(const char* path, const struct timespec tv[2])
+{
+	Inode* inode = fat_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	if ( !FUSE_FS->device->write )
+		return inode->Unref(), -(errno = EROFS);
+	inode->BeginWrite();
+	inode->data->i_atime = tv[0].tv_sec;
+	inode->data->i_mtime = tv[1].tv_sec;
+	inode->FinishWrite();
+	inode->Unref();
+	return 0;
+}
+
+/*int fat_fuse_bmap(const char*, size_t blocksize, uint64_t* idx)
+{
+	return -(errno = ENOSYS);
+}*/
+
+int fat_fuse_main(const char* argv0,
+                   const char* mount_path,
+                   bool foreground,
+                   Filesystem* fs,
+                   Device* dev)
+{
+	struct fuse_operations operations;
+	memset(&operations, 0, sizeof(operations));
+
+	operations.access = fat_fuse_access;
+	operations.chmod = fat_fuse_chmod;
+	operations.chown = fat_fuse_chown;
+	operations.create = fat_fuse_create;
+	operations.destroy = fat_fuse_destroy;
+	operations.fgetattr = fat_fuse_fgetattr;
+	operations.flush = fat_fuse_flush;
+	operations.fsync = fat_fuse_fsync;
+	operations.ftruncate = fat_fuse_ftruncate;
+	operations.getattr = fat_fuse_getattr;
+	operations.init = fat_fuse_init;
+	operations.link = fat_fuse_link;
+	operations.mkdir = fat_fuse_mkdir;
+	operations.mknod = fat_fuse_mknod;
+	operations.opendir = fat_fuse_opendir;
+	operations.open = fat_fuse_open;
+	operations.readdir = fat_fuse_readdir;
+	operations.read = fat_fuse_read;
+	operations.readlink = fat_fuse_readlink;
+	operations.releasedir = fat_fuse_releasedir;
+	operations.release = fat_fuse_release;
+	operations.rename = fat_fuse_rename;
+	operations.rmdir = fat_fuse_rmdir;
+	operations.statfs = fat_fuse_statfs;
+	operations.symlink = fat_fuse_symlink;
+	operations.truncate = fat_fuse_truncate;
+	operations.unlink = fat_fuse_unlink;
+	operations.utimens = fat_fuse_utimens;
+	operations.write = fat_fuse_write;
+
+	operations.flag_nullpath_ok = 1;
+	operations.flag_nopath = 1;
+
+	char* argv_fuse[] =
+	{
+		(char*) argv0,
+		(char*) "-s",
+		(char*) mount_path,
+		(char*) NULL,
+	};
+
+	int argc_fuse = sizeof(argv_fuse) / sizeof(argv_fuse[0]) - 1;
+
+	struct fat_fuse_ctx fat_fuse_ctx;
+	fat_fuse_ctx.fs = fs;
+	fat_fuse_ctx.dev = dev;
+
+	(void) foreground;
+
+	return fuse_main(argc_fuse, argv_fuse, &operations, &fat_fuse_ctx);
+}
+
+#endif

fat/fuse.h

@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fuse.h
+ * FUSE frontend.
+ */
+
+#ifndef FUSE_H
+#define FUSE_H
+
+class Device;
+class Filesystem;
+
+int fat_fuse_main(const char* argv0,
+                   const char* mount_path,
+                   bool foreground,
+                   Filesystem* fs,
+                   Device* dev);
+
+#endif

fat/inode.h

@@ -0,0 +1,90 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * inode.h
+ * Filesystem inode.
+ */
+
+#ifndef INODE_H
+#define INODE_H
+
+class Block;
+class Filesystem;
+
+class Inode
+{
+public:
+	Inode(Filesystem* filesystem, uint32_t inode_id);
+	~Inode();
+
+public:
+	Inode* prev_inode;
+	Inode* next_inode;
+	Inode* prev_hashed;
+	Inode* next_hashed;
+	Inode* prev_dirty;
+	Inode* next_dirty;
+	Block* data_block;
+	struct fat_dirent* dirent;
+	struct fat_dirent deleted_dirent;
+	uint32_t first_cluster;
+	Filesystem* filesystem;
+	size_t reference_count;
+	size_t remote_reference_count;
+	size_t implied_reference;
+	uint32_t inode_id;
+	bool dirty;
+	bool deleted;
+
+public:
+	uint32_t Mode();
+	uint32_t UserId();
+	uint32_t GroupId();
+	uint64_t Size();
+	void UTimens(const struct timespec times[2]);
+	bool ChangeMode(mode_t mode);
+	bool ChangeOwner(uid_t uid, gid_t gid);
+	bool Truncate(uint64_t new_size);
+	Block* GetClusterSector(uint32_t cluster, uint8_t sector);
+	bool Iterate(Block** block_ptr, uint32_t* cluster_ptr,
+	             uint8_t* sector_ptr, uint16_t* offset);
+	uint32_t SeekCluster(uint32_t cluster_id);
+	Inode* Open(const char* elem, int flags, mode_t mode);
+	bool Link(const char* elem, Inode* dest, bool directories);
+	bool Symlink(const char* elem, const char* dest);
+	bool Unlink(const char* elem, bool directories, bool force=false);
+	Inode* UnlinkKeep(const char* elem, bool directories, bool force=false);
+	ssize_t ReadAt(uint8_t* buffer, size_t count, off_t offset);
+	ssize_t WriteAt(const uint8_t* buffer, size_t count, off_t offset);
+	bool Rename(Inode* olddir, const char* oldname, const char* newname);
+	Inode* CreateDirectory(const char* path, mode_t mode);
+	bool RemoveDirectory(const char* path);
+	bool IsEmptyDirectory();
+	void Refer();
+	void Unref();
+	void RemoteRefer();
+	void RemoteUnref();
+	void Sync();
+	void BeginWrite();
+	void FinishWrite();
+	void Modified();
+	void Use();
+	void Unlink();
+	void Prelink();
+	void Delete();
+
+};
+
+#endif

fat/ioleast.h

@@ -0,0 +1,150 @@
+/*
+ * Copyright (c) 2012, 2013, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * ioleast.h
+ * Versions of {,p}{read,write} that don't return until it has returned as much
+ * data as requested, end of file, or an error occurs. This is sometimes needed
+ * as read(2) and write(2) is not always guaranteed to fill up the entire
+ * buffer or write it all.
+ */
+
+#ifndef SORTIX_COMPATIBILITY_INCLUDE_IOLEAST_H
+#define SORTIX_COMPATIBILITY_INCLUDE_IOLEAST_H
+
+#if defined(__sortix__) || defined(__sortix_libc__)
+
+#include_next <ioleast.h>
+
+#else
+
+#include <sys/types.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <unistd.h>
+
+#if !defined(EEOF) && defined(EIO)
+#define EEOF EIO
+#endif
+
+__attribute__((unused)) static inline
+size_t readleast(int fd, void* buf_ptr, size_t least, size_t max)
+{
+	assert(least <= max);
+	unsigned char* buf = (unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = read(fd, buf + done, max - done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t writeleast(int fd, const void* buf_ptr, size_t least, size_t max)
+{
+	assert(least <= max);
+	const unsigned char* buf = (const unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = write(fd, buf + done, max - done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t preadleast(int fd, void* buf_ptr, size_t least, size_t max, off_t off)
+{
+	assert(least <= max);
+	unsigned char* buf = (unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = pread(fd, buf + done, max - done, off + done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t pwriteleast(int fd, const void* buf_ptr, size_t least, size_t max, off_t off)
+{
+	assert(least <= max);
+	const unsigned char* buf = (const unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = pwrite(fd, buf + done, max - done, off + done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t readall(int fd, void* buf, size_t count)
+{
+	return readleast(fd, buf, count, count);
+}
+
+__attribute__((unused)) static inline
+size_t writeall(int fd, const void* buf, size_t count)
+{
+	return writeleast(fd, buf, count, count);
+}
+
+__attribute__((unused)) static inline
+size_t preadall(int fd, void* buf, size_t count, off_t off)
+{
+	return preadleast(fd, buf, count, count, off);
+}
+
+__attribute__((unused)) static inline
+size_t pwriteall(int fd, const void* buf, size_t count, off_t off)
+{
+	return pwriteleast(fd, buf, count, count, off);
+}
+
+#endif
+
+#endif

fat/util.h

@@ -0,0 +1,49 @@
+/*
+ * Copyright (c) 2013, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * util.h
+ * Utility functions for the filesystem implementation.
+ */
+
+#ifndef UTIL_H
+#define UTIL_H
+
+template <class T> T divup(T a, T b)
+{
+	return a/b + (a % b ? 1 : 0);
+}
+
+template <class T> T roundup(T a, T b)
+{
+	return a % b ? a + b - a % b : a;
+}
+
+static inline bool checkbit(const uint8_t* bitmap, size_t bit)
+{
+	uint8_t bits = bitmap[bit / 8UL];
+	return bits & (1U << (bit % 8UL));
+}
+
+static inline void setbit(uint8_t* bitmap, size_t bit)
+{
+	bitmap[bit / 8UL] |= 1U << (bit % 8UL);
+}
+
+static inline void clearbit(uint8_t* bitmap, size_t bit)
+{
+	bitmap[bit / 8UL] &= ~(1U << (bit % 8UL));
+}
+
+#endif

init/.gitignore

@@ -1,2 +1,3 @@
 init
+service
 *.o

init/Makefile

@@ -8,26 +8,26 @@ CFLAGS?=$(OPTLEVEL)
 
 CFLAGS:=$(CFLAGS) -Wall -Wextra
 
-BINARY=init
+BINARIES=\
+init \
+service \
 
-OBJS=\
-init.o \
+MANPAGES8=\
+init.8 \
+service.8 \
 
-all: $(BINARY)
+all: $(BINARIES)
 
 .PHONY: all install clean
 
-$(BINARY): $(OBJS)
-	$(CC) $(CFLAGS) $(OBJS) -o $(BINARY) -lmount $(LIBS)
-
-%.o: %.c
-	$(CC) -std=gnu11 $(CFLAGS) $(CPPFLAGS) -c $< -o $@
+%: %.c
+	$(CC) -std=gnu11 $(CFLAGS) $(CPPFLAGS) $< -o $@ -lmount $(LIBS)
 
 install: all
 	mkdir -p $(DESTDIR)$(SBINDIR)
-	install $(BINARY) $(DESTDIR)$(SBINDIR)
+	install $(BINARIES) $(DESTDIR)$(SBINDIR)
 	mkdir -p $(DESTDIR)$(MANDIR)/man8
-	cp init.8 $(DESTDIR)$(MANDIR)/man8/init.8
+	install $(MANPAGES8) $(DESTDIR)$(MANDIR)/man8
 
 clean:
-	rm -f $(BINARY) $(OBJS) *.o
+	rm -f $(BINARIES) $(OBJS) *.o

init/init.8

@@ -213,12 +213,9 @@ daemon is meant to start the installation's local daemon requirements.
 .Sh ENVIRONMENT
 .Nm
 sets the following environment variables.
-.Bl -tag -width "INIT_PID"
+.Bl -tag -width "LOGNAME"
 .It Ev HOME
 root's home directory
-.It Ev INIT_PID
-.Nm Ns 's
-process id
 .It Ev LOGNAME
 root
 .It Ev PATH
@@ -301,6 +298,7 @@ exits with the same exit status as its target session if it terminates normally.
 .Xr login 8 ,
 .Xr poweroff 8 ,
 .Xr reboot 8 ,
+.Xr service 8 ,
 .Xr sysmerge 8 ,
 .Xr update-initrd 8
 .Sh SECURITY CONSIDERATIONS

init/service.8

@@ -0,0 +1,391 @@
+.Dd May 24, 2024
+.Dt SERVICE 8
+.Os
+.Sh NAME
+.Nm service
+.Nd daemon maintenance
+.Sh SYNOPSIS
+.Nm service
+.Op Fl lr
+.Op Fl \-s Ns "=" Ns source-daemon
+.Op Fl \-exit-code
+.Op Fl \-no-await
+.Op Fl \-no-optional
+.Op Fl \-raw
+.Op Fl \-source Ns "=" Ns source-daemon
+.Ar daemon
+.Oo
+.Sy dependents
+|
+.Sy disable
+|
+.Sy edges
+|
+.Sy enable
+|
+.Sy exit-code
+|
+.Sy kill
+|
+.Sy pid
+|
+.Sy reconfigure
+|
+.Sy reload
+|
+.Sy requirements
+|
+.Sy restart
+|
+.Sy signal
+|
+.Sy start
+|
+.Sy state
+|
+.Sy status
+|
+.Sy stop
+|
+.Sy terminate
+.Oc
+.Sh DESCRIPTION
+.Nm
+performs maintenance of daemons run by
+.Xr init 8
+as configured in
+.Xr init 5 .
+The daemons are serviced by connecting to the
+.Nm init
+process and writing the requested command to the
+.Pa /var/run/init
+filesytem socket.
+.Pp
+The options are as follows:
+.Bl -tag -width "12345678"
+.It Fl s , Fl \-source-daemon Ns "=" Ns Ar source-daemon
+When modifying a dependency using the
+.Sy enable , disable , start
+and
+.Sy stop
+commands, use the
+.Ar source-daemon
+as the source daemon in the dependency on the target
+.Ar daemon .
+The default the
+.Sy local
+daemon which is the parent of the locally configured daemons.
+.It Fl \-exit-code
+Set the
+.Sy exit-code
+flag on the dependency created in the
+.Sy enable
+and
+.Sy start
+commands.
+.It Fl \-no-await
+Set the
+.Sy no-await
+flag on the dependency created in the
+.Sy enable
+and
+.Sy start
+commands.
+.It Fl \-no-optional
+Unset the
+.Sy optional
+flag on the dependency created in the
+.Sy enable
+and
+.Sy start
+commands.
+The default is to set the
+.Sy optional
+flag, which is the opposite of the
+.Xr init 5
+.Sy require
+declaration where dependencies are mandatory by default.
+.It Fl l , \-list
+Write a table containing the status of every loaded daemon in the format of the
+.Sy status
+command.
+.It Fl r , \-raw
+Write the command and additional operands as a raw message without verification
+on the
+.Nm init
+socket
+and output the raw reply sent from
+.Nm init.
+.El
+.Pp
+The commands are as follows:
+.Bl -tag -width "requirements"
+.It Sy dependents
+Write the incoming dependencies on the
+.Ar daemon
+in the format of the
+.Sy edges
+command, which explains why a daemon is running.
+.It Sy disable
+Permanently disable the
+.Ar daemon
+by removing the dependency on it from the configuration of the
+.Ar source-daemon
+(the
+.Sy local
+daemon by default, see
+.Xr init 5 )
+and then stopping it using the
+.Sy stop
+command.
+The daemon will continue to run if any other daemon depends on it, which can
+be diagnosed using the
+.Sy dependents
+command.
+.It Sy edges
+Write the incoming dependencies on the
+.Ar daemon
+and its outgoing dependencies in the format of the
+.Sy require
+declaration in
+.Xr init 5
+with an extra
+.Ar source
+operand:
+.Pp
+.Sy require
+.Ar source
+.Ar target
+[exit-code]
+[no-await]
+[optional]
+.It Sy enable
+Permanently enable the
+.Ar daemon
+by adding it to the configuration of the
+.Ar source-daemon
+(the
+.Sy local
+daemon by default, see
+.Xr init 5 )
+with the dependency flags per
+.Fl \-exit-code ,
+.Fl \-no-await ,
+and
+.Fl \-no-optional
+and starting it by sending the
+.Sy start
+command.
+The daemon only starts if the
+.Ar source-daemon
+is running.
+.It Sy exit-code
+Write the exit code of the
+.Ar daemon ,
+or the empty string if it has not exited.
+.It Sy kill
+Kill the
+.Ar daemon
+by sending the
+.Sy SIGKILL
+signal as a last resort that may cause data loss.
+.It Sy pid
+Write the process id of the
+.Ar daemon
+if it is running, or the empty output if it
+does not have a process.
+Process ids can be recycled and are subject to inherent race conditions.
+Prefer to use the other commands in
+.Nm
+that addresses the daemon by its symbolic name as
+.Xr init 8
+will ensure the command operates on the correct process.
+.It Sy reconfigure
+Reread the
+.Xr init 5
+configuration for the
+.Ar daemon
+and apply it after restarting the daemon.
+.It Sy reload
+Request the
+.Ar daemon
+gracefully reload its own configuration sending
+the reload signal (usually
+.Sy SIGHUP )
+without restarting the daemon and without reloading the
+.Xr init 5
+configuration.
+.It Sy requirements
+Write the outgoing dependencies from the
+.Ar daemon
+in the format of the
+.Sy edges
+command, which explains what daemons the daemon needs.
+.It Sy restart
+Restart the
+.Ar daemon
+by terminating it and starting it up again afterwards.
+.It Sy signal Ar signal
+Send the
+.Ar signal
+in the symbolic signal name format (e.g.
+.Sy SIGUSR1 )
+to the
+.Ar daemon .
+.It Sy start
+Start the
+.Ar daemon
+by asking
+.Xr init 8
+to add a runtime dependency from the
+.Ar source-daemon
+(the
+.Sy local
+daemon by default, see
+.Xr init 5 )
+to the daemon with the dependency flags per
+.Fl \-exit-code ,
+.Fl \-no-await ,
+and
+.Fl \-no-optional
+and starting it by sending the
+.Sy start
+command.
+The daemon only starts if the
+.Ar source-daemon
+is running.
+.It Sy state
+Write which the state the
+.Ar daemon
+is in.
+.It Sy status
+Write the status of the
+.Ar daemon
+as a single table row in the format:
+.Pp
+.Ar name
+.Ar state
+.Li pid Ns "=" Ns Ar pid
+.Li exit Ns "=" Ns Ar exit-code
+.Pp
+The
+.Ar state
+is one of
+.Sy terminated ,
+.Sy scheduled ,
+.Sy waiting ,
+.Sy satisfied ,
+.Sy starting ,
+.Sy running ,
+.Sy terminating ,
+.Sy finishing ,
+.Sy finished ,
+or
+.Sy failed .
+The
+.Ar pid
+is the process id if any, or
+.Li 0
+otherwise.
+The
+.Ar exit-code
+is the exit code of the daemon if it has exited, or the name of a signal that
+killed it, or
+.Li n/a
+if the daemon has not exited.
+.It Sy stop
+Stop the
+.Ar daemon
+by asking
+.Xr init 8
+to remove the dependency from the
+.Ar source-daemon
+(the
+.Sy local
+daemon by default, see
+.Xr init 5 )
+on the daemon.
+The daemon will continue to run as long if other daemon depends on it, which can
+be diagnosed using the
+.Sy dependents
+command.
+.It Sy terminate
+Terminate the
+.Ar daemon
+gracefully by sending the
+.Sy SIGTERM
+signal and
+.Sy SIGKILL
+after a timeout.
+Prefer the
+.Sy stop
+command if possible as the
+.Sy terminate
+command bypasses the reference count and may cause data loss if other daemons
+malfunction when the daemon is unexpectedly terminated.
+.El
+.Sh ENVIRONMENT
+.Bl -tag -width "INIT_SOCKET"
+.It Ev INIT_SOCKET
+.Xr init 8 Ns 's
+filesystem socket for communication,
+.Pa /var/run/init
+by default.
+.El
+.Sh FILES
+.Bl -tag -width "/etc/init/local" -compact
+.It Pa /etc/init/
+Daemon configuration for the local system (first in search path) (see
+.Xr init 5 )
+.It Pa /etc/init/local
+Configuration for the
+.Sy local
+daemon (see
+.Xr init 5 )
+.El
+.Sh EXIT STATUS
+.Nm
+will exit 0 on success and non-zero otherwise.
+.Sh EXAMPLES
+Permanently enable the sshd daemon:
+.Bd -literal
+$ service sshd enable
+.Ed
+.Pp
+Permanently disable the ntpd daemon:
+.Bd -literal
+$ service ntpd disable
+.Ed
+.Pp
+Temporarily start the nginx daemon without changing
+.Pa /etc/init/local :
+.Bd -literal
+$ service nginx start
+.Ed
+.Pp
+Temporarily disable the sshd daemon without changing
+.Pa /etc/init/local :
+.Bd -literal
+$ service sshd stop
+.Ed
+.Pp
+Temporarily stop the ntpd daemon and diagnose why it kept running due to the
+.Sy time
+daemon depending on it:
+.Bd -literal
+$ service ntpd stop
+$ service ntpd state
+running
+$ service ntpd dependents
+require time ntpd exit-code
+$ service --source-daemon=time ntpd stop
+$ service ntpd state
+finished
+.Ed
+.Sh SEE ALSO
+.Xr kill 1 ,
+.Xr init 5 ,
+.Xr halt 8 ,
+.Xr init 8 ,
+.Xr poweroff 8 ,
+.Xr reboot 8

init/service.c

@@ -0,0 +1,594 @@
+/*
+ * Copyright (c) 2024 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * service.c
+ * Start and stop services.
+ */
+
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/un.h>
+
+#include <ctype.h>
+#include <err.h>
+#include <errno.h>
+#include <getopt.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <wchar.h>
+
+static bool array_add(void*** array_ptr,
+                      size_t* used_ptr,
+                      size_t* length_ptr,
+                      void* value)
+{
+	void** array;
+	memcpy(&array, array_ptr, sizeof(array)); // Strict aliasing.
+
+	if ( *used_ptr == *length_ptr )
+	{
+		size_t length = *length_ptr;
+		if ( !length )
+			length = 4;
+		void** new_array = reallocarray(array, length, 2 * sizeof(void*));
+		if ( !new_array )
+			return false;
+		array = new_array;
+		memcpy(array_ptr, &array, sizeof(array)); // Strict aliasing.
+		*length_ptr = length * 2;
+	}
+
+	memcpy(array + (*used_ptr)++, &value, sizeof(value)); // Strict aliasing.
+
+	return true;
+}
+
+static char** tokenize(size_t* out_tokens_used, const char* string)
+{
+	size_t tokens_used = 0;
+	size_t tokens_length = 0;
+	char** tokens = malloc(sizeof(char*));
+	if ( !tokens )
+		return NULL;
+	bool failed = false;
+	bool invalid = false;
+	while ( *string )
+	{
+		if ( isspace((unsigned char) *string) )
+		{
+			string++;
+			continue;
+		}
+		if ( *string == '#' )
+			break;
+		char* token;
+		size_t token_size;
+		FILE* fp = open_memstream(&token, &token_size);
+		if ( !fp )
+		{
+			failed = true;
+			break;
+		}
+		bool singly = false;
+		bool doubly = false;
+		bool escaped = false;
+		while ( *string )
+		{
+			char c = *string++;
+			if ( !escaped && !singly && !doubly && isspace((unsigned char) c) )
+				break;
+			if ( !escaped && !doubly && c == '\'' )
+			{
+				singly = !singly;
+				continue;
+			}
+			if ( !escaped && !singly && c == '"' )
+			{
+				doubly = !doubly;
+				continue;
+			}
+			if ( !singly && !escaped && c == '\\' )
+			{
+				escaped = true;
+				continue;
+			}
+			if ( escaped )
+			{
+				switch ( c )
+				{
+				case 'a': c = '\a'; break;
+				case 'b': c = '\b'; break;
+				case 'e': c = '\e'; break;
+				case 'f': c = '\f'; break;
+				case 'n': c = '\n'; break;
+				case 'r': c = '\r'; break;
+				case 't': c = '\t'; break;
+				case 'v': c = '\v'; break;
+				default: break;
+				};
+			}
+			escaped = false;
+			if ( fputc((unsigned char) c, fp) == EOF )
+			{
+				failed = true;
+				break;
+			}
+		}
+		if ( singly || doubly || escaped )
+		{
+			fclose(fp);
+			free(token);
+			invalid = true;
+			break;
+		}
+		if ( fflush(fp) == EOF )
+		{
+			fclose(fp);
+			free(token);
+			failed = true;
+			break;
+		}
+		fclose(fp);
+		if ( !array_add((void***) &tokens, &tokens_used, &tokens_length,
+		                token) )
+		{
+			free(token);
+			failed = true;
+			break;
+		}
+	}
+	if ( failed || invalid )
+	{
+		for ( size_t i = 0; i < tokens_used; i++ )
+			free(tokens[i]);
+		free(tokens);
+		if ( invalid )
+			errno = 0;
+		return NULL;
+	}
+	char** new_tokens = reallocarray(tokens, tokens_used, sizeof(char*));
+	if ( new_tokens )
+		tokens = new_tokens;
+	*out_tokens_used = tokens_used;
+	return tokens;
+}
+
+static char** receive(FILE* fp, size_t* out_tokens_used)
+{
+	char* line = NULL;
+	size_t line_size;
+	if ( getline(&line, &line_size, fp) < 0 )
+	{
+		if ( ferror(fp) )
+			errx(1, "receiving reply: Unexpected end of connection");
+		else
+			err(1, "receiving reply");
+	}
+	char** result = tokenize(out_tokens_used, line);
+	free(line);
+	if ( !result )
+	{
+		if ( errno )
+			errx(1, "invalid reply: %s", line);
+		else
+			errx(1, "failed to parse replyt");
+	}
+	if ( !*out_tokens_used )
+			errx(1, "invalid empty reply");
+	if ( !strcmp(result[0], "ok") )
+		return result;
+	else if ( !strcmp(result[0], "error") )
+		errx(1, "error: %s", 2 <= *out_tokens_used ? result[1] : "Unknown");
+	else
+		errx(1, "unknown reply: %s", result[0]);
+}
+
+static int open_local_client_socket(const char* path, int flags)
+{
+	size_t path_length = strlen(path);
+	size_t addr_size = offsetof(struct sockaddr_un, sun_path) + path_length + 1;
+	struct sockaddr_un* sockaddr = malloc(addr_size);
+	if ( !sockaddr )
+		return -1;
+	sockaddr->sun_family = AF_LOCAL;
+	strcpy(sockaddr->sun_path, path);
+	int fd = socket(AF_LOCAL, SOCK_STREAM | flags, 0);
+	if ( fd < 0 )
+		return free(sockaddr), -1;
+	if ( connect(fd, (const struct sockaddr*) sockaddr, addr_size) < 0 )
+		return close(fd), free(sockaddr), -1;
+	free(sockaddr);
+	return fd;
+}
+
+static void rewrite(const char* path, const char* daemon, const char* flags)
+{
+	FILE* fp = fopen(path, "r");
+	if ( !fp && errno != ENOENT )
+		err(1, "%s", path);
+	char* out_path;
+	if ( asprintf(&out_path, "%s.XXXXXX", path) < 0 )
+		err(1, "malloc");
+	int out_fd = mkstemp(out_path);
+	if ( out_fd < 0 )
+		err(1, "mkstemp: %s.XXXXXX", path);
+	FILE* out = fdopen(out_fd, "w");
+	if ( !out )
+	{
+		unlink(out_path);
+		err(1, "fdopen");
+	}
+	bool found = false;
+	char* line = NULL;
+	size_t line_size = 0;
+	ssize_t line_length;
+	off_t line_number = 0;
+	while ( fp && 0 < (line_length = getline(&line, &line_size, fp)) )
+	{
+		line_number++;
+		size_t tokenc;
+		char** tokens = tokenize(&tokenc, line);
+		if ( !tokens )
+		{
+			unlink(out_path);
+			if ( errno )
+				err(1, "%s", path);
+			else
+				errx(1, "%s:%ji: Syntax error", path, (intmax_t) line_number);
+		}
+		if ( 2 <= tokenc &&
+		     !strcmp(tokens[0], "require") && !strcmp(tokens[1], daemon) )
+		{
+			found = true;
+			if ( flags )
+				fprintf(out, "require %s%s\n", daemon, flags);
+		}
+		else
+			fputs(line, out);
+	}
+	free(line);
+	if ( !found && flags )
+		fprintf(out, "require %s%s\n", daemon, flags);
+	if ( (fp && ferror(fp)) || ferror(out) || fflush(out) == EOF )
+	{
+		unlink(out_path);
+		err(1, "%s", path);
+	}
+	if ( fp )
+	{
+		struct stat st;
+		fstat(fileno(fp), &st);
+		fchmod(out_fd, st.st_mode & 07777);
+		fchown(out_fd, st.st_uid, st.st_gid);
+		fclose(fp);
+	}
+	else
+		fchmod(out_fd, 0666 & ~getumask());
+	if ( rename(out_path, path) < 0 )
+	{
+		unlink(out_path);
+		err(1, "rename: %s -> %s", out_path, path);
+	}
+	fclose(out);
+}
+
+static bool check_daemon_exists_in_dir(const char* dir, const char* daemon)
+{
+	char* path;
+	if ( asprintf(&path, "%s/%s", dir, daemon) < 0 )
+		err(1, "malloc");
+	bool result = !access(path, F_OK);
+	free(path);
+	return result;
+}
+
+
+static void check_daemon_exists(const char* daemon)
+{
+	if ( !check_daemon_exists_in_dir("/etc/init", daemon) &&
+	     !check_daemon_exists_in_dir("/share/init", daemon) )
+		errx(1, "%s: Daemon does not exist", daemon);
+}
+
+static size_t string_display_length(const char* str)
+{
+	size_t display_length = 0;
+	mbstate_t ps;
+	memset(&ps, 0, sizeof(ps));
+	while ( true )
+	{
+		wchar_t wc;
+		size_t amount = mbrtowc(&wc, str, SIZE_MAX, &ps);
+		if ( amount == 0 )
+			break;
+		if ( amount == (size_t) -1 || amount == (size_t) -2 )
+		{
+			display_length++;
+			str++;
+			memset(&ps, 0, sizeof(ps));
+			continue;
+		}
+		int width = wcwidth(wc);
+		if ( width < 0 )
+			width = 0;
+		if ( SIZE_MAX - display_length < (size_t) width )
+			display_length = SIZE_MAX;
+		else
+			display_length += (size_t) width;
+		str += amount;
+	}
+	return display_length;
+}
+
+static void pad(const char* string, size_t padding)
+{
+	fputs(string, stdout);
+	for ( size_t length = string_display_length(string);
+	      length < padding; padding-- )
+		putchar(' ');
+}
+
+static void format_statuses(char** tokens, size_t count)
+{
+	size_t daemon_length = 0;
+	size_t state_length = 10;
+	for ( size_t i = 0; i < count; i++ )
+	{
+		if ( !strncmp(tokens[i], "daemon=", strlen("daemon=")) )
+		{
+			const char* arg = tokens[i] + strlen("daemon=");
+			size_t length = string_display_length	(arg);
+			if ( daemon_length < length )
+				daemon_length = length;
+		}
+		else if ( !strncmp(tokens[i], "state=", strlen("state=")) )
+		{
+			const char* arg = tokens[i] + strlen("state=");
+			size_t length = string_display_length(arg);
+			if ( state_length < length )
+				state_length = length;
+		}
+	}
+	size_t i = 0;
+	while ( i < count )
+	{
+		size_t next = i;
+		while ( next < count && strcmp(tokens[next], ",") )
+			next++;
+		const char* daemon = NULL;
+		const char* state = NULL;
+		for ( size_t n = i; n < next; n++ )
+		{
+			if ( !strncmp(tokens[n], "state=", strlen("state=")) )
+				state = tokens[n] + strlen("state=");
+			else if ( !strncmp(tokens[n], "daemon=", strlen("daemon=")) )
+				daemon = tokens[n] + strlen("daemon=");
+		}
+		if ( !state || !daemon )
+			errx(1, "missing information in reply");
+		pad(daemon, daemon_length + 2);
+		pad(state, state_length);
+		for ( size_t n = i; n < next; n++ )
+		{
+			if ( strncmp(tokens[n], "state=", strlen("state=")) &&
+			     strncmp(tokens[n], "daemon=", strlen("daemon=")) )
+				printf("  %s", tokens[n]);
+		}
+		putchar('\n');
+		i = next;
+		if ( i < count && !strcmp(tokens[i], ",") )
+			i++;
+	}
+}
+
+int main(int argc, char* argv[])
+{
+	const char* init_socket = getenv("INIT_SOCKET");
+	if ( !init_socket )
+		init_socket = "/var/run/init";
+
+	bool exit_code = false;
+	bool list = false;
+	bool no_await = false;
+	bool optional = true;
+	bool raw = false;
+	const char* source = "local";
+
+	const struct option longopts[] =
+	{
+		{"exit-code", no_argument, NULL, 256},
+		{"list", no_argument, NULL, 'l'},
+		{"no-await", no_argument, NULL, 257},
+		{"no-optional", no_argument, NULL, 258},
+		{"source", required_argument, NULL, 's'},
+		{"raw", no_argument, NULL, 'r'},
+		{0, 0, 0, 0}
+	};
+	const char* opts = "lrs:";
+	int opt;
+	while ( (opt = getopt_long(argc, argv, opts, longopts, NULL)) != -1 )
+	{
+		switch ( opt )
+		{
+		case 'l': list = true; break;
+		case 'r': raw = true; break;
+		case 's': source = optarg; break;
+		case 256: exit_code = true; break;
+		case 257: no_await = true; break;
+		case 258: optional = false; break;
+		default: return 2;
+		}
+	}
+
+	int fd = open_local_client_socket(init_socket, 0);
+	if ( fd < 0 )
+		err(1, "%s", init_socket);
+	FILE* fp = fdopen(fd, "r+");
+	if ( !fp )
+		err(1, "fdopen");
+
+	if ( raw )
+	{
+		for ( int i = optind; i < argc; i++ )
+		{
+			if ( fprintf(fp, "%s%c", argv[i], i + 1 == argc ? '\n' : ' ') < 0 )
+				err(1, "%s", init_socket);
+		}
+		size_t tokens_count;
+		char** tokens = receive(fp, &tokens_count);
+		for ( size_t i = 0; i < tokens_count; i++ )
+			printf("%s%c", tokens[i], i + 1 == tokens_count ? '\n' : ' ');
+		return 0;
+	}
+
+	char flags[sizeof(" optional no-await exit-code")];
+	snprintf(flags, sizeof(flags), "%s%s%s",
+	         optional ? " optional" : "",
+	         no_await ? " no-await" : "",
+	         exit_code ? " exit-code" : "");
+	char* source_path;
+	if ( asprintf(&source_path, "/etc/init/%s", source) < 0 )
+		err(1, "malloc");
+
+	bool no_command = list;
+
+	if ( no_command && 0 < argc - optind  )
+		errx(1, "unexpected extra operand: %s", argv[optind]);
+	else if ( !no_command && argc - optind < 2 )
+		errx(1, "usage: <daemon> <command>");
+
+	const char* daemon = !no_command ? argv[optind++] : NULL;
+	const char* command = !no_command ? argv[optind++] : NULL;
+
+	if ( command && strcmp(command, "signal") && 0 < argc - optind )
+		errx(1, "unexpected extra operand: %s", argv[optind]);
+
+	if ( list )
+		fprintf(fp, "list\n");
+	else if ( !strcmp(command, "enable") )
+	{
+		check_daemon_exists(daemon);
+		rewrite(source_path, daemon, flags);
+		fprintf(fp, "require %s %s %s start\n", source, daemon, flags);
+	}
+	else if ( !strcmp(command, "disable") )
+	{
+		rewrite(source_path, daemon, NULL);
+		fprintf(fp, "unrequire %s %s\n", source, daemon);
+	}
+	else if ( !strcmp(command, "start") )
+		fprintf(fp, "require %s %s %s start\n", source, daemon, flags);
+	else if ( !strcmp(command, "stop") )
+		fprintf(fp, "unrequire %s %s\n", source, daemon);
+	else if ( !strcmp(command, "restart") )
+		fprintf(fp, "restart %s\n", daemon);
+	else if ( !strcmp(command, "reload") )
+		fprintf(fp, "reload %s\n", daemon);
+	else if ( !strcmp(command, "reconfigure") )
+		fprintf(fp, "reconfigure %s\n", daemon);
+	else if ( !strcmp(command, "terminate") )
+		fprintf(fp, "terminate %s\n", daemon);
+	else if ( !strcmp(command, "kill") )
+		fprintf(fp, "kill %s\n", daemon);
+	else if ( !strcmp(command, "signal") )
+	{
+		if ( argc - optind < 1 )
+			errx(1, "expected signal name");
+		const char* signal = argv[optind++];
+		if ( 0 < argc - optind )
+		errx(1, "unexpected extra operand: %s", argv[optind]);
+		fprintf(fp, "signal %s %s\n", daemon, signal);
+	}
+	else if ( !strcmp(command, "status") )
+		fprintf(fp, "status %s\n", daemon);
+	else if ( !strcmp(command, "state") )
+		fprintf(fp, "status %s\n", daemon);
+	else if ( !strcmp(command, "pid") )
+		fprintf(fp, "status %s\n", daemon);
+	else if ( !strcmp(command, "exit-code") )
+		fprintf(fp, "status %s\n", daemon);
+	else if ( !strcmp(command, "requirements") ||
+	          !strcmp(command, "dependents") ||
+	          !strcmp(command, "edges") )
+		fprintf(fp, "%s %s\n", command, daemon);
+	else
+		errx(1, "unknown command: %s", command);
+
+	if ( ferror(fp) || fflush(fp) == EOF )
+		err(1, "%s", init_socket);
+
+	size_t tokens_count;
+	char** tokens = receive(fp, &tokens_count);
+
+	if ( list )
+		format_statuses(tokens + 1, tokens_count - 1);
+	else if ( !strcmp(command, "status") )
+		format_statuses(tokens + 1, tokens_count - 1);
+	else if ( !strcmp(command, "state") )
+	{
+		for ( size_t i = 1; i < tokens_count; i++ )
+		{
+			if ( !strncmp(tokens[i], "state=", strlen("state=")) )
+			{
+				puts(tokens[i] + strlen("state="));
+				break;
+			}
+		}
+	}
+	else if ( !strcmp(command, "pid") )
+	{
+		for ( size_t i = 1; i < tokens_count; i++ )
+		{
+			if ( !strncmp(tokens[i], "pid=", strlen("pid=")) )
+			{
+				const char* arg = tokens[i] + strlen("pid=");
+				if ( strcmp(arg, "0") )
+					puts(arg);
+				break;
+			}
+		}
+	}
+	else if ( !strcmp(command, "exit-code") )
+	{
+		for ( size_t i = 1; i < tokens_count; i++ )
+		{
+			if ( !strncmp(tokens[i], "exit=", strlen("exit=")) )
+			{
+				const char* arg = tokens[i] + strlen("exit=");
+				if ( strcmp(arg, "n/a") )
+					puts(arg);
+				break;
+			}
+		}
+	}
+	else if ( !strcmp(command, "requirements") ||
+	          !strcmp(command, "dependents") ||
+	          !strcmp(command, "edges") )
+	{
+		for ( size_t i = 1; i < tokens_count; i++ )
+		{
+			if ( !strcmp(tokens[i], ",") )
+				continue;
+			size_t eol = i + 1 == tokens_count || !strcmp(tokens[i + 1], ",");
+			printf("%s%c", tokens[i], eol ? '\n' : ' ');
+		}
+	}
+
+	return 0;
+}

iso9660/.gitignore

@@ -0,0 +1,2 @@
+iso9660fs
+*.o

iso9660/Makefile

@@ -0,0 +1,33 @@
+include ../build-aux/platform.mak
+include ../build-aux/compiler.mak
+include ../build-aux/version.mak
+include ../build-aux/dirs.mak
+
+OPTLEVEL?=$(DEFAULT_OPTLEVEL)
+CXXFLAGS?=$(OPTLEVEL)
+
+CPPFLAGS:=$(CPPFLAGS) -DVERSIONSTR=\"$(VERSION)\"
+CXXFLAGS:=$(CXXFLAGS) -Wall -Wextra -fno-exceptions -fno-rtti -fcheck-new
+
+LIBS:=$(LIBS)
+
+ifeq ($(HOST_IS_SORTIX),0)
+  LIBS:=$(LIBS) -lfuse
+  CPPFLAGS:=$(CPPFLAGS) -D_FILE_OFFSET_BITS=64
+endif
+
+BINARIES:=iso9660fs
+
+all: $(BINARIES)
+
+.PHONY: all install clean
+
+install: all
+	mkdir -p $(DESTDIR)$(SBINDIR)
+	install $(BINARIES) $(DESTDIR)$(SBINDIR)
+
+iso9660fs: *.cpp *.h
+	$(CXX) -std=gnu++11 $(CPPFLAGS) $(CXXFLAGS) *.cpp -o $@ $(LIBS)
+
+clean:
+	rm -f $(BINARIES) *.o

iso9660/block.cpp

@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * block.cpp
+ * Blocks in the filesystem.
+ */
+
+#include <sys/types.h>
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#include "block.h"
+#include "device.h"
+#include "ioleast.h"
+
+Block::Block()
+{
+	this->block_data = NULL;
+}
+
+Block::Block(Device* device, uint32_t block_id)
+{
+	Construct(device, block_id);
+}
+
+void Block::Construct(Device* device, uint32_t block_id)
+{
+	this->prev_block = NULL;
+	this->next_block = NULL;
+	this->prev_hashed = NULL;
+	this->next_hashed = NULL;
+	this->device = device;
+	this->reference_count = 1;
+	this->block_id = block_id;
+}
+
+Block::~Block()
+{
+	Destruct();
+	delete[] block_data;
+}
+
+void Block::Destruct()
+{
+	Unlink();
+}
+
+void Block::Refer()
+{
+	reference_count++;
+}
+
+void Block::Unref()
+{
+	if ( !--reference_count )
+	{
+#if 0
+		device->block_count--;
+		delete this;
+#endif
+	}
+}
+
+void Block::Use()
+{
+	Unlink();
+	Prelink();
+}
+
+void Block::Unlink()
+{
+	(prev_block ? prev_block->next_block : device->mru_block) = next_block;
+	(next_block ? next_block->prev_block : device->lru_block) = prev_block;
+	size_t bin = block_id % DEVICE_HASH_LENGTH;
+	(prev_hashed ? prev_hashed->next_hashed : device->hash_blocks[bin]) = next_hashed;
+	if ( next_hashed ) next_hashed->prev_hashed = prev_hashed;
+}
+
+void Block::Prelink()
+{
+	prev_block = NULL;
+	next_block = device->mru_block;
+	if ( device->mru_block )
+		device->mru_block->prev_block = this;
+	device->mru_block = this;
+	if ( !device->lru_block )
+		device->lru_block = this;
+	size_t bin = block_id % DEVICE_HASH_LENGTH;
+	prev_hashed = NULL;
+	next_hashed = device->hash_blocks[bin];
+	device->hash_blocks[bin] = this;
+	if ( next_hashed )
+		next_hashed->prev_hashed = this;
+}

iso9660/block.h

@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * block.h
+ * Blocks in the filesystem.
+ */
+
+#ifndef BLOCK_H
+#define BLOCK_H
+
+class Device;
+
+class Block
+{
+public:
+	Block();
+	Block(Device* device, uint32_t block_id);
+	~Block();
+	void Construct(Device* device, uint32_t block_id);
+	void Destruct();
+
+public:
+	Block* prev_block;
+	Block* next_block;
+	Block* prev_hashed;
+	Block* next_hashed;
+	Device* device;
+	size_t reference_count;
+	uint32_t block_id;
+	uint8_t* block_data;
+
+public:
+	void Refer();
+	void Unref();
+	void Use();
+	void Unlink();
+	void Prelink();
+
+};
+
+#endif

iso9660/device.cpp

@@ -0,0 +1,107 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * device.cpp
+ * Block device.
+ */
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "block.h"
+#include "device.h"
+#include "ioleast.h"
+
+Device::Device(int fd, const char* path, uint32_t block_size)
+{
+	this->mru_block = NULL;
+	this->lru_block = NULL;
+	for ( size_t i = 0; i < DEVICE_HASH_LENGTH; i++ )
+		hash_blocks[i] = NULL;
+	struct stat st;
+	fstat(fd, &st);
+	this->device_size = st.st_size;
+	this->path = path;
+	this->block_size = block_size;
+	this->fd = fd;
+	this->block_count = 0;
+#ifdef __sortix__
+	// TODO: This isn't scaleable if there's multiple filesystems mounted.
+	size_t memory;
+	memstat(NULL, &memory);
+	this->block_limit = (memory / 10) / block_size;
+#else
+	this->block_limit = 32768;
+#endif
+}
+
+Device::~Device()
+{
+	while ( mru_block )
+		delete mru_block;
+	close(fd);
+}
+
+Block* Device::AllocateBlock()
+{
+	if ( block_limit <= block_count )
+	{
+		for ( Block* block = lru_block; block; block = block->prev_block )
+		{
+			if ( block->reference_count )
+				continue;
+			block->Destruct(); // Syncs.
+			return block;
+		}
+	}
+	uint8_t* data = new uint8_t[block_size];
+	if ( !data ) // TODO: Use operator new nothrow!
+		return NULL;
+	Block* block = new Block();
+	if ( !block ) // TODO: Use operator new nothrow!
+		return delete[] data, (Block*) NULL;
+	block->block_data = data;
+	block_count++;
+	return block;
+}
+
+Block* Device::GetBlock(uint32_t block_id)
+{
+	if ( Block* block = GetCachedBlock(block_id) )
+		return block;
+	Block* block = AllocateBlock();
+	if ( !block )
+		return NULL;
+	block->Construct(this, block_id);
+	off_t file_offset = (off_t) block_size * (off_t) block_id;
+	preadall(fd, block->block_data, block_size, file_offset);
+	block->Prelink();
+	return block;
+}
+
+Block* Device::GetCachedBlock(uint32_t block_id)
+{
+	size_t bin = block_id % DEVICE_HASH_LENGTH;
+	for ( Block* iter = hash_blocks[bin]; iter; iter = iter->next_hashed )
+		if ( iter->block_id == block_id )
+			return iter->Refer(), iter;
+	return NULL;
+}

iso9660/device.h

@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * device.h
+ * Block device.
+ */
+
+#ifndef DEVICE_H
+#define DEVICE_H
+
+class Block;
+
+static const size_t DEVICE_HASH_LENGTH = 1 << 16;
+
+class Device
+{
+public:
+	Device(int fd, const char* path, uint32_t block_size);
+	~Device();
+
+public:
+	Block* mru_block;
+	Block* lru_block;
+	Block* hash_blocks[DEVICE_HASH_LENGTH];
+	off_t device_size;
+	const char* path;
+	uint32_t block_size;
+	int fd;
+	size_t block_count;
+	size_t block_limit;
+
+public:
+	Block* AllocateBlock();
+	Block* GetBlock(uint32_t block_id);
+	Block* GetCachedBlock(uint32_t block_id);
+
+};
+
+#endif

iso9660/filesystem.cpp

@@ -0,0 +1,87 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * filesystem.cpp
+ * ISO 9660 filesystem implementation.
+ */
+
+#include <sys/types.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <time.h>
+
+#include <stdio.h> // DEBUG
+
+#include "block.h"
+#include "device.h"
+#include "filesystem.h"
+#include "inode.h"
+#include "util.h"
+
+Filesystem::Filesystem(Device* device, const char* mount_path,
+                       uint64_t pvd_offset)
+{
+	// TODO: This should be replaced by the . entry within the root directory
+	//       so the Rock Ridge extensions are available.
+	uint32_t pvd_block_id = pvd_offset / device->block_size;
+	this->pvd_block = device->GetBlock(pvd_block_id);
+	assert(pvd_block); // TODO: This can fail.
+	this->pvd = (struct iso9660_pvd*)
+	            (pvd_block->block_data + pvd_offset % device->block_size);
+	this->root_ino = pvd_offset + offsetof(struct iso9660_pvd, root_dirent);
+	this->device = device;
+	this->mount_path = mount_path;
+	this->block_size = device->block_size;
+	this->mru_inode = NULL;
+	this->lru_inode = NULL;
+	for ( size_t i = 0; i < INODE_HASH_LENGTH; i++ )
+		this->hash_inodes[i] = NULL;
+}
+
+Filesystem::~Filesystem()
+{
+	while ( mru_inode )
+		delete mru_inode;
+	pvd_block->Unref();
+}
+
+Inode* Filesystem::GetInode(iso9660_ino_t inode_id)
+{
+	size_t bin = inode_id % INODE_HASH_LENGTH;
+	for ( Inode* iter = hash_inodes[bin]; iter; iter = iter->next_hashed )
+		if ( iter->inode_id == inode_id )
+			return iter->Refer(), iter;
+
+	uint32_t block_id = inode_id / block_size;
+	uint32_t offset = inode_id % block_size;
+
+	Block* block = device->GetBlock(block_id);
+	if ( !block )
+		return (Inode*) NULL;
+	Inode* inode = new Inode(this, inode_id);
+	if ( !inode )
+		return block->Unref(), (Inode*) NULL;
+	inode->data_block = block;
+	uint8_t* buf = inode->data_block->block_data + offset;
+	inode->data = (struct iso9660_dirent*) buf;
+	inode->Prelink();
+	inode->Parse();
+
+	return inode;
+}

iso9660/filesystem.h

@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * filesystem.h
+ * ISO 9660 filesystem implementation.
+ */
+
+#ifndef FILESYSTEM_H
+#define FILESYSTEM_H
+
+#include "iso9660.h"
+
+class Device;
+class Inode;
+
+static const size_t INODE_HASH_LENGTH = 1 << 16;
+
+class Filesystem
+{
+public:
+	Filesystem(Device* device, const char* mount_path, uint64_t pvd_offset);
+	~Filesystem();
+
+public:
+	Block* pvd_block;
+	struct iso9660_pvd* pvd;
+	Device* device;
+	const char* mount_path;
+	iso9660_ino_t root_ino;
+	uint32_t block_size;
+	Inode* mru_inode;
+	Inode* lru_inode;
+	Inode* hash_inodes[INODE_HASH_LENGTH];
+
+public:
+	Inode* GetInode(iso9660_ino_t inode_id);
+
+};
+
+#endif

iso9660/fsmarshall.cpp

@@ -0,0 +1,701 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2016, 2022, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fsmarshall.cpp
+ * Sortix fsmarshall frontend.
+ */
+
+#if defined(__sortix__)
+
+#include <sys/mman.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <err.h>
+#include <errno.h>
+#include <dirent.h>
+#include <fcntl.h>
+#include <ioleast.h>
+#include <signal.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <timespec.h>
+#include <unistd.h>
+
+#include <sortix/dirent.h>
+
+#include <fsmarshall.h>
+
+#include "block.h"
+#include "device.h"
+#include "filesystem.h"
+#include "fsmarshall.h"
+#include "fuse.h"
+#include "inode.h"
+#include "iso9660fs.h"
+
+bool RespondData(int chl, const void* ptr, size_t count)
+{
+	return writeall(chl, ptr, count) == count;
+}
+
+bool RespondHeader(int chl, size_t type, size_t size)
+{
+	struct fsm_msg_header hdr;
+	hdr.msgtype = type;
+	hdr.msgsize = size;
+	return RespondData(chl, &hdr, sizeof(hdr));
+}
+
+bool RespondMessage(int chl, size_t type, const void* ptr, size_t count)
+{
+	return RespondHeader(chl, type, count) &&
+	       RespondData(chl, ptr, count);
+}
+
+bool RespondError(int chl, int errnum)
+{
+	struct fsm_resp_error body;
+	body.errnum = errnum;
+	return RespondMessage(chl, FSM_RESP_ERROR, &body, sizeof(body));
+}
+
+bool RespondSuccess(int chl)
+{
+	struct fsm_resp_success body;
+	return RespondMessage(chl, FSM_RESP_SUCCESS, &body, sizeof(body));
+}
+
+bool RespondStat(int chl, struct stat* st)
+{
+	struct fsm_resp_stat body;
+	body.st = *st;
+	return RespondMessage(chl, FSM_RESP_STAT, &body, sizeof(body));
+}
+
+bool RespondStatVFS(int chl, struct statvfs* stvfs)
+{
+	struct fsm_resp_statvfs body;
+	body.stvfs = *stvfs;
+	return RespondMessage(chl, FSM_RESP_STATVFS, &body, sizeof(body));
+}
+
+bool RespondSeek(int chl, off_t offset)
+{
+	struct fsm_resp_lseek body;
+	body.offset = offset;
+	return RespondMessage(chl, FSM_RESP_LSEEK, &body, sizeof(body));
+}
+
+bool RespondRead(int chl, const uint8_t* buf, size_t count)
+{
+	struct fsm_resp_read body;
+	body.count = count;
+	return RespondMessage(chl, FSM_RESP_READ, &body, sizeof(body)) &&
+	       RespondData(chl, buf, count);
+}
+
+bool RespondReadlink(int chl, const uint8_t* buf, size_t count)
+{
+	struct fsm_resp_readlink body;
+	body.targetlen = count;
+	return RespondMessage(chl, FSM_RESP_READLINK, &body, sizeof(body)) &&
+	       RespondData(chl, buf, count);
+}
+
+bool RespondWrite(int chl, size_t count)
+{
+	struct fsm_resp_write body;
+	body.count = count;
+	return RespondMessage(chl, FSM_RESP_WRITE, &body, sizeof(body));
+}
+
+bool RespondOpen(int chl, ino_t ino, mode_t type)
+{
+	struct fsm_resp_open body;
+	body.ino = ino;
+	body.type = type;
+	return RespondMessage(chl, FSM_RESP_OPEN, &body, sizeof(body));
+}
+
+bool RespondMakeDir(int chl, ino_t ino)
+{
+	struct fsm_resp_mkdir body;
+	body.ino = ino;
+	return RespondMessage(chl, FSM_RESP_MKDIR, &body, sizeof(body));
+}
+
+bool RespondReadDir(int chl, struct dirent* dirent)
+{
+	struct fsm_resp_readdirents body;
+	body.ino = dirent->d_ino;
+	body.type = dirent->d_type;
+	body.namelen = dirent->d_namlen;
+	return RespondMessage(chl, FSM_RESP_READDIRENTS, &body, sizeof(body)) &&
+	       RespondData(chl, dirent->d_name, dirent->d_namlen);
+}
+
+bool RespondTCGetBlob(int chl, const void* data, size_t data_size)
+{
+	struct fsm_resp_tcgetblob body;
+	body.count = data_size;
+	return RespondMessage(chl, FSM_RESP_TCGETBLOB, &body, sizeof(body)) &&
+	       RespondData(chl, data, data_size);
+}
+
+Inode* SafeGetInode(Filesystem* fs, ino_t ino)
+{
+	if ( (iso9660_ino_t) ino != ino )
+		return errno = EBADF, (Inode*) ino;
+	return fs->GetInode((iso9660_ino_t) ino);
+}
+
+void HandleRefer(int chl, struct fsm_req_refer* msg, Filesystem* fs)
+{
+	(void) chl;
+	if ( Inode* inode = SafeGetInode(fs, msg->ino) )
+	{
+		inode->RemoteRefer();
+		inode->Unref();
+	}
+}
+
+void HandleUnref(int chl, struct fsm_req_unref* msg, Filesystem* fs)
+{
+	(void) chl;
+	if ( Inode* inode = SafeGetInode(fs, msg->ino) )
+	{
+		inode->RemoteUnref();
+		inode->Unref();
+	}
+}
+
+void HandleSync(int chl, struct fsm_req_sync* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	inode->Unref();
+	RespondSuccess(chl);
+}
+
+void HandleStat(int chl, struct fsm_req_stat* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	struct stat st;
+	StatInode(inode, &st);
+	inode->Unref();
+	RespondStat(chl, &st);
+}
+
+void HandleChangeMode(int chl, struct fsm_req_chmod* /*msg*/, Filesystem* /*fs*/)
+{
+	RespondError(chl, EROFS);
+}
+
+void HandleChangeOwner(int chl, struct fsm_req_chown* /*msg*/, Filesystem* /*fs*/)
+{
+	RespondError(chl, EROFS);
+}
+
+void HandleUTimens(int chl, struct fsm_req_utimens* /*msg*/, Filesystem* /*fs*/)
+{
+	RespondError(chl, EROFS);
+}
+
+void HandleTruncate(int chl, struct fsm_req_truncate* /*msg*/, Filesystem* /*fs*/)
+{
+	RespondError(chl, EROFS);
+}
+
+void HandleSeek(int chl, struct fsm_req_lseek* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( msg->whence == SEEK_SET )
+		RespondSeek(chl, msg->offset);
+	else if ( msg->whence == SEEK_END )
+	{
+		off_t inode_size = inode->Size();
+		if ( (msg->offset < 0 && inode_size + msg->offset < 0) ||
+		     (0 <= msg->offset && OFF_MAX - inode_size < msg->offset) )
+			RespondError(chl, EOVERFLOW);
+		else
+			RespondSeek(chl, msg->offset + inode_size);
+	}
+	else
+		RespondError(chl, EINVAL);
+	inode->Unref();
+}
+
+void HandleReadAt(int chl, struct fsm_req_pread* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	uint8_t* buf = (uint8_t*) malloc(msg->count);
+	if ( !buf ) { inode->Unref(); RespondError(chl, errno); return; }
+	ssize_t amount = inode->ReadAt(buf, msg->count, msg->offset);
+	inode->Unref();
+	if ( amount < 0 ) { free(buf); RespondError(chl, errno); return; }
+	RespondRead(chl, buf, amount);
+	free(buf);
+}
+
+void HandleWriteAt(int chl, struct fsm_req_pwrite* /*msg*/, Filesystem* /*fs*/)
+{
+	RespondError(chl, EROFS);
+}
+
+void HandleOpen(int chl, struct fsm_req_open* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	Inode* result = inode->Open(path, msg->flags, FsModeFromHostMode(msg->mode));
+
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondOpen(chl, result->inode_id, result->Mode() & S_IFMT);
+	result->Unref();
+}
+
+void HandleMakeDir(int chl, struct fsm_req_mkdir* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	inode->Unref();
+	RespondError(chl, EROFS);
+}
+
+void HandleReadDir(int chl, struct fsm_req_readdirents* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( !S_ISDIR(inode->Mode()) )
+	{
+		inode->Unref();
+		RespondError(chl, ENOTDIR);
+		return;
+	}
+	union
+	{
+		struct dirent kernel_entry;
+		uint8_t padding[sizeof(struct dirent) + 256];
+	};
+	memset(&kernel_entry, 0, sizeof(kernel_entry));
+
+	uint64_t offset = 0;
+	Block* block = NULL;
+	uint64_t block_id = 0;
+	char name[256];
+	uint8_t file_type;
+	iso9660_ino_t inode_id;
+	while ( inode->ReadDirectory(&offset, &block, &block_id,
+	                             msg->rec_num ? NULL : name, &file_type,
+	                             &inode_id) )
+	{
+		if ( !(msg->rec_num--) )
+		{
+			size_t name_len = strlen(name);
+			kernel_entry.d_reclen = sizeof(kernel_entry) + name_len;
+			kernel_entry.d_ino = inode_id;
+			kernel_entry.d_dev = 0;
+			kernel_entry.d_type = HostDTFromFsDT(file_type);
+			kernel_entry.d_namlen = name_len;
+			memcpy(kernel_entry.d_name, name, name_len);
+			size_t dname_offset = offsetof(struct dirent, d_name);
+			padding[dname_offset + kernel_entry.d_namlen] = '\0';
+			block->Unref();
+			inode->Unref();
+			RespondReadDir(chl, &kernel_entry);
+			return;
+		}
+	}
+	int errnum = errno;
+	if ( block )
+		block->Unref();
+	inode->Unref();
+
+	if ( errnum )
+	{
+		RespondError(chl, errnum);
+		return;
+	}
+
+	kernel_entry.d_reclen = sizeof(kernel_entry);
+	RespondReadDir(chl, &kernel_entry);
+}
+
+void HandleIsATTY(int chl, struct fsm_req_isatty* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	RespondError(chl, ENOTTY);
+	inode->Unref();
+}
+
+void HandleUnlink(int chl, struct fsm_req_unlink* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->Unlink(path, false);
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleRemoveDir(int chl, struct fsm_req_rmdir* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->RemoveDirectory(path);
+	free(path);
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleLink(int chl, struct fsm_req_link* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	Inode* dest = SafeGetInode(fs, msg->linkino);
+	if ( !dest ) { inode->Unref(); RespondError(chl, errno); return; }
+
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->namelen+1);
+	if ( !path )
+	{
+		RespondError(chl, errno);
+		inode->Unref();
+		return;
+	}
+	memcpy(path, pathraw, msg->namelen);
+	path[msg->namelen] = '\0';
+
+	bool result = inode->Link(path, dest);
+
+	free(path);
+	dest->Unref();
+	inode->Unref();
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleSymlink(int chl, struct fsm_req_symlink* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->dirino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	inode->Unref();
+	RespondError(chl, EROFS);
+}
+
+void HandleReadlink(int chl, struct fsm_req_readlink* msg, Filesystem* fs)
+{
+	Inode* inode = SafeGetInode(fs, msg->ino);
+	if ( !inode ) { RespondError(chl, errno); return; }
+	if ( !ISO9660_S_ISLNK(inode->Mode()) ) { inode->Unref(); RespondError(chl, EINVAL); return; }
+	size_t count = inode->Size();
+	uint8_t* buf = (uint8_t*) malloc(count);
+	if ( !buf ) { inode->Unref(); RespondError(chl, errno); return; }
+	ssize_t amount = inode->ReadLink(buf, count);
+	inode->Unref();
+	if ( amount < 0 )  { RespondError(chl, errno); return; }
+	RespondReadlink(chl, buf, amount);
+	free(buf);
+}
+
+void HandleRename(int chl, struct fsm_req_rename* msg, Filesystem* fs)
+{
+	char* pathraw = (char*) &(msg[1]);
+	char* path = (char*) malloc(msg->oldnamelen+1 + msg->newnamelen+1);
+	if ( !path ) { RespondError(chl, errno); return; }
+	memcpy(path, pathraw, msg->oldnamelen);
+	path[msg->oldnamelen] = '\0';
+	memcpy(path + msg->oldnamelen + 1, pathraw + msg->oldnamelen, msg->newnamelen);
+	path[msg->oldnamelen + 1 + msg->newnamelen] = '\0';
+
+	const char* oldname = path;
+	const char* newname = path + msg->oldnamelen + 1;
+
+	Inode* olddir = SafeGetInode(fs, msg->olddirino);
+	if ( !olddir ) { free(path); RespondError(chl, errno); return; }
+	Inode* newdir = SafeGetInode(fs, msg->newdirino);
+	if ( !newdir ) { olddir->Unref(); free(path); RespondError(chl, errno); return; }
+
+	bool result = newdir->Rename(olddir, oldname, newname);
+
+	newdir->Unref();
+	olddir->Unref();
+	free(path);
+
+	if ( !result ) { RespondError(chl, errno); return; }
+
+	RespondSuccess(chl);
+}
+
+void HandleStatVFS(int chl, struct fsm_req_statvfs* msg, Filesystem* fs)
+{
+	(void) msg;
+	struct statvfs stvfs;
+	stvfs.f_bsize = fs->block_size;
+	stvfs.f_frsize = fs->block_size;
+	stvfs.f_blocks = fs->device->device_size / fs->block_size;
+	stvfs.f_bfree = 0;
+	stvfs.f_bavail = 0;
+	stvfs.f_files = 0;
+	stvfs.f_ffree = 0;
+	stvfs.f_favail = 0;
+	stvfs.f_ffree = 0;
+	stvfs.f_fsid = 0;
+	stvfs.f_flag = ST_RDONLY;
+	stvfs.f_namemax = 255;
+	RespondStatVFS(chl, &stvfs);
+}
+
+void HandleTCGetBlob(int chl, struct fsm_req_tcgetblob* msg, Filesystem* fs)
+{
+	char* nameraw = (char*) &(msg[1]);
+	char* name = (char*) malloc(msg->namelen + 1);
+	if ( !name )
+		return (void) RespondError(chl, errno);
+	memcpy(name, nameraw, msg->namelen);
+	name[msg->namelen] = '\0';
+
+	//static const char index[] = "device-path\0filesystem-type\0filesystem-uuid\0mount-path\0";
+	static const char index[] = "device-path\0filesystem-type\0mount-path\0";
+	if ( !strcmp(name, "") )
+		RespondTCGetBlob(chl, index, sizeof(index) - 1);
+	else if ( !strcmp(name, "device-path") )
+		RespondTCGetBlob(chl, fs->device->path, strlen(fs->device->path));
+	else if ( !strcmp(name, "filesystem-type") )
+		RespondTCGetBlob(chl, "iso9660", strlen("iso9660"));
+	// TODO: Some kind of unique id.
+	//else if ( !strcmp(name, "filesystem-uuid") )
+	//	RespondTCGetBlob(chl, fs->sb->s_uuid, sizeof(fs->sb->s_uuid));
+	else if ( !strcmp(name, "mount-path") )
+		RespondTCGetBlob(chl, fs->mount_path, strlen(fs->mount_path));
+	else
+		RespondError(chl, ENOENT);
+
+	free(name);
+}
+
+void HandleIncomingMessage(int chl, struct fsm_msg_header* hdr, Filesystem* fs)
+{
+	request_uid = hdr->uid;
+	request_gid = hdr->gid;
+	if ( (uint16_t) request_uid != request_uid ||
+	     (uint16_t) request_gid != request_gid )
+	{
+		fprintf(stderr, "extfs: id exceeded 16-bit: uid=%ju gid=%ju\n",
+		        (uintmax_t) request_uid, (uintmax_t) request_gid);
+		RespondError(chl, EOVERFLOW);
+		return;
+	}
+	typedef void (*handler_t)(int, void*, Filesystem*);
+	handler_t handlers[FSM_MSG_NUM] = { NULL };
+	handlers[FSM_REQ_SYNC] = (handler_t) HandleSync;
+	handlers[FSM_REQ_STAT] = (handler_t) HandleStat;
+	handlers[FSM_REQ_CHMOD] = (handler_t) HandleChangeMode;
+	handlers[FSM_REQ_CHOWN] = (handler_t) HandleChangeOwner;
+	handlers[FSM_REQ_TRUNCATE] = (handler_t) HandleTruncate;
+	handlers[FSM_REQ_LSEEK] = (handler_t) HandleSeek;
+	handlers[FSM_REQ_PREAD] = (handler_t) HandleReadAt;
+	handlers[FSM_REQ_OPEN] = (handler_t) HandleOpen;
+	handlers[FSM_REQ_READDIRENTS] = (handler_t) HandleReadDir;
+	handlers[FSM_REQ_PWRITE] = (handler_t) HandleWriteAt;
+	handlers[FSM_REQ_ISATTY] = (handler_t) HandleIsATTY;
+	handlers[FSM_REQ_UTIMENS] = (handler_t) HandleUTimens;
+	handlers[FSM_REQ_MKDIR] = (handler_t) HandleMakeDir;
+	handlers[FSM_REQ_RMDIR] = (handler_t) HandleRemoveDir;
+	handlers[FSM_REQ_UNLINK] = (handler_t) HandleUnlink;
+	handlers[FSM_REQ_LINK] = (handler_t) HandleLink;
+	handlers[FSM_REQ_SYMLINK] = (handler_t) HandleSymlink;
+	handlers[FSM_REQ_READLINK] = (handler_t) HandleReadlink;
+	handlers[FSM_REQ_RENAME] = (handler_t) HandleRename;
+	handlers[FSM_REQ_REFER] = (handler_t) HandleRefer;
+	handlers[FSM_REQ_UNREF] = (handler_t) HandleUnref;
+	handlers[FSM_REQ_STATVFS] = (handler_t) HandleStatVFS;
+	handlers[FSM_REQ_TCGETBLOB] = (handler_t) HandleTCGetBlob;
+	if ( FSM_MSG_NUM <= hdr->msgtype || !handlers[hdr->msgtype] )
+	{
+		fprintf(stderr, "extfs: message type %zu not supported\n", hdr->msgtype);
+		RespondError(chl, ENOTSUP);
+		return;
+	}
+	uint8_t body_buffer[65536];
+	uint8_t* body = body_buffer;
+	if ( sizeof(body_buffer) < hdr->msgsize )
+	{
+		body = (uint8_t*) mmap(NULL, hdr->msgsize, PROT_READ | PROT_WRITE,
+		                       MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+		if ( (void*) body == MAP_FAILED )
+		{
+			RespondError(chl, errno);
+			return;
+		}
+	}
+	if ( readall(chl, body, hdr->msgsize) == hdr->msgsize )
+		handlers[hdr->msgtype](chl, body, fs);
+	else
+		RespondError(chl, errno);
+	if ( sizeof(body_buffer) < hdr->msgsize )
+		munmap(body, hdr->msgsize);
+}
+static volatile bool should_terminate = false;
+
+void TerminationHandler(int)
+{
+	should_terminate = true;
+}
+
+static void ready(void)
+{
+	const char* readyfd_env = getenv("READYFD");
+	if ( !readyfd_env )
+		return;
+	int readyfd = atoi(readyfd_env);
+	char c = '\n';
+	write(readyfd, &c, 1);
+	close(readyfd);
+	unsetenv("READYFD");
+}
+
+int fsmarshall_main(const char* argv0,
+                    const char* mount_path,
+                    bool foreground,
+                    Filesystem* fs,
+                    Device* dev)
+{
+	(void) argv0;
+
+	// Stat the root inode.
+	struct stat root_inode_st;
+	Inode* root_inode = fs->GetInode(fs->root_ino);
+	if ( !root_inode )
+		err(1, "GetInode");
+	StatInode(root_inode, &root_inode_st);
+	root_inode->Unref();
+
+	// Create a filesystem server connected to the kernel that we'll listen on.
+	int serverfd = fsm_mountat(AT_FDCWD, mount_path, &root_inode_st, 0);
+	if ( serverfd < 0 )
+		err(1, "%s", mount_path);
+
+	// Make sure the server isn't unexpectedly killed and data is lost.
+	signal(SIGINT, TerminationHandler);
+	signal(SIGTERM, TerminationHandler);
+	signal(SIGQUIT, TerminationHandler);
+
+	// Become a background process in its own process group by default.
+	if ( !foreground )
+	{
+		pid_t child_pid = fork();
+		if ( child_pid < 0 )
+			err(1, "fork");
+		if ( child_pid )
+			exit(0);
+		setpgid(0, 0);
+	}
+	else
+		ready();
+
+	// Listen for filesystem messages.
+	int channel;
+	while ( 0 <= (channel = accept(serverfd, NULL, NULL)) )
+	{
+		if ( should_terminate )
+			break;
+		struct fsm_msg_header hdr;
+		size_t amount;
+		if ( (amount = readall(channel, &hdr, sizeof(hdr))) != sizeof(hdr) )
+		{
+			//warn("incomplete header: got %zi of %zu bytes", amount, sizeof(hdr));
+			errno = 0;
+			continue;
+		}
+		HandleIncomingMessage(channel, &hdr, fs);
+		close(channel);
+	}
+
+	// Garbage collect all open inode references.
+	while ( fs->mru_inode )
+	{
+		Inode* inode = fs->mru_inode;
+		if ( inode->remote_reference_count )
+			inode->RemoteUnref();
+		else if ( inode->reference_count )
+			inode->Unref();
+	}
+
+	close(serverfd);
+
+	delete fs;
+	delete dev;
+
+	return 0;
+}
+
+#endif

iso9660/fsmarshall.h

@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fsmarshall.h
+ * Sortix fsmarshall frontend.
+ */
+
+#ifndef FSMARSHALL_H
+#define FSMARSHALL_H
+
+class Device;
+class Filesystem;
+
+int fsmarshall_main(const char* argv0,
+                    const char* mount_path,
+                    bool foreground,
+                    Filesystem* fs,
+                    Device* dev);
+#endif

iso9660/fuse.cpp

@@ -0,0 +1,572 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fuse.cpp
+ * FUSE frontend.
+ */
+
+#if !defined(__sortix__)
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <ctype.h>
+#include <errno.h>
+#include <limits.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define FUSE_USE_VERSION 26
+#include <fuse.h>
+
+#include "block.h"
+#include "device.h"
+#include "filesystem.h"
+#include "fuse.h"
+#include "inode.h"
+#include "iso9660fs.h"
+
+#include <stdio.h> // DEBUG
+
+struct iso9660_fuse_ctx
+{
+	Device* dev;
+	Filesystem* fs;
+};
+
+#ifndef S_SETABLE
+#define S_SETABLE 02777
+#endif
+
+#define FUSE_FS (((struct iso9660_fuse_ctx*) (fuse_get_context()->private_data))->fs)
+
+void* iso9660_fuse_init(struct fuse_conn_info* /*conn*/)
+{
+	return fuse_get_context()->private_data;
+}
+
+void iso9660_fuse_destroy(void* fs_private)
+{
+	struct iso9660_fuse_ctx* iso9660_fuse_ctx =
+		(struct iso9660_fuse_ctx*) fs_private;
+	while ( iso9660_fuse_ctx->fs->mru_inode )
+	{
+		Inode* inode = iso9660_fuse_ctx->fs->mru_inode;
+		if ( inode->remote_reference_count )
+			inode->RemoteUnref();
+		else if ( inode->reference_count )
+			inode->Unref();
+	}
+	delete iso9660_fuse_ctx->fs; iso9660_fuse_ctx->fs = NULL;
+	delete iso9660_fuse_ctx->dev; iso9660_fuse_ctx->dev = NULL;
+}
+
+Inode* iso9660_fuse_resolve_path(const char* path)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode(fs->root_ino);
+	if ( !inode )
+		return (Inode*) NULL;
+	while ( path[0] )
+	{
+		if ( *path == '/' )
+		{
+			if ( !ISO9660_S_ISDIR(inode->Mode()) )
+				return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+			path++;
+			continue;
+		}
+		size_t elem_len = strcspn(path, "/");
+		char* elem = strndup(path, elem_len);
+		if ( !elem )
+			return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+		path += elem_len;
+		Inode* next = inode->Open(elem, O_RDONLY, 0);
+		free(elem);
+		inode->Unref();
+		if ( !next )
+			return NULL;
+		inode = next;
+	}
+	return inode;
+}
+
+// Assumes that the path doesn't end with / unless it's the root directory.
+Inode* iso9660_fuse_parent_dir(const char** path_ptr)
+{
+	const char* path = *path_ptr;
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode(fs->root_ino);
+	if ( !inode )
+		return (Inode*) NULL;
+	while ( strchr(path, '/') )
+	{
+		if ( *path == '/' )
+		{
+			if ( !ISO9660_S_ISDIR(inode->Mode()) )
+				return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+			path++;
+			continue;
+		}
+		size_t elem_len = strcspn(path, "/");
+		char* elem = strndup(path, elem_len);
+		if ( !elem )
+			return inode->Unref(), errno = ENOTDIR, (Inode*) NULL;
+		path += elem_len;
+		Inode* next = inode->Open(elem, O_RDONLY, 0);
+		free(elem);
+		inode->Unref();
+		if ( !next )
+			return (Inode*) NULL;
+		inode = next;
+	}
+	*path_ptr = *path ? path : ".";
+	assert(!strchr(*path_ptr, '/'));
+	return inode;
+}
+
+int iso9660_fuse_getattr(const char* path, struct stat* st)
+{
+	Inode* inode = iso9660_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	StatInode(inode, st);
+	inode->Unref();
+	return 0;
+}
+
+int iso9660_fuse_fgetattr(const char* /*path*/, struct stat* st,
+                       struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	StatInode(inode, st);
+	inode->Unref();
+	return 0;
+}
+
+int iso9660_fuse_readlink(const char* path, char* buf, size_t bufsize)
+{
+	Inode* inode = iso9660_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	if ( !ISO9660_S_ISLNK(inode->Mode()) )
+		return inode->Unref(), -(errno = EINVAL);
+	if ( !bufsize )
+		return inode->Unref(), -(errno = EINVAL);
+	ssize_t amount = inode->ReadLink((uint8_t*) buf, bufsize);
+	if ( amount < 0 )
+		return inode->Unref(), -errno;
+	buf[(size_t) amount < bufsize ? (size_t) amount : bufsize - 1] = '\0';
+	inode->Unref();
+	return 0;
+}
+
+int iso9660_fuse_mknod(const char* path, mode_t mode, dev_t dev)
+{
+	(void) path;
+	(void) mode;
+	(void) dev;
+	return -(errno = ENOSYS);
+}
+
+int iso9660_fuse_mkdir(const char* path, mode_t /*mode*/)
+{
+	Inode* inode = iso9660_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	inode->Unref();
+	return -(errno = EROFS);
+}
+
+int iso9660_fuse_unlink(const char* path)
+{
+	Inode* inode = iso9660_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	bool success = inode->Unlink(path, false);
+	inode->Unref();
+	return success ? 0 : -errno;
+}
+
+int iso9660_fuse_rmdir(const char* path)
+{
+	Inode* inode = iso9660_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	bool success = inode->RemoveDirectory(path);
+	inode->Unref();
+	return success ? 0 : -errno;
+}
+
+int iso9660_fuse_symlink(const char* /*oldname*/, const char* newname)
+{
+	Inode* newdir = iso9660_fuse_parent_dir(&newname);
+	if ( !newdir )
+		return -errno;
+	newdir->Unref();
+	return -(errno = EROFS);
+}
+
+int iso9660_fuse_rename(const char* oldname, const char* newname)
+{
+	Inode* olddir = iso9660_fuse_parent_dir(&oldname);
+	if ( !olddir )
+		return -errno;
+	Inode* newdir = iso9660_fuse_parent_dir(&newname);
+	if ( !newdir )
+		return olddir->Unref(), -errno;
+	bool success = newdir->Rename(olddir, oldname, newname);
+	newdir->Unref();
+	olddir->Unref();
+	return success ? 0 : -errno;
+}
+
+int iso9660_fuse_link(const char* oldname, const char* newname)
+{
+	Inode* inode = iso9660_fuse_resolve_path(oldname);
+	if ( !inode )
+		return -errno;
+	Inode* newdir = iso9660_fuse_parent_dir(&newname);
+	if ( !newdir )
+		return inode->Unref(), -errno;
+	bool success = inode->Link(newname, inode);
+	newdir->Unref();
+	inode->Unref();
+	return success ? 0 : -errno;
+}
+
+int iso9660_fuse_chmod(const char* path, mode_t mode)
+{
+	Inode* inode = iso9660_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	(void) mode;
+	return inode->Unref(), -(errno = EROFS);
+}
+
+int iso9660_fuse_chown(const char* path, uid_t owner, gid_t group)
+{
+	Inode* inode = iso9660_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	(void) owner;
+	(void) group;
+	return inode->Unref(), -(errno = EROFS);
+}
+
+int iso9660_fuse_truncate(const char* path, off_t size)
+{
+	Inode* inode = iso9660_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	(void) size;
+	return inode->Unref(), -(errno = EROFS);
+}
+
+int iso9660_fuse_ftruncate(const char* /*path*/, off_t size,
+                        struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	(void) size;
+	return inode->Unref(), -(errno = EROFS);
+}
+
+int iso9660_fuse_open(const char* path, struct fuse_file_info* fi)
+{
+	int flags = fi->flags;
+	Inode* dir = iso9660_fuse_parent_dir(&path);
+	if ( !dir )
+		return -errno;
+	Inode* result = dir->Open(path, flags, 0);
+	dir->Unref();
+	if ( !result )
+		return -errno;
+	fi->fh = (uint64_t) result->inode_id;
+	fi->keep_cache = 1;
+	result->RemoteRefer();
+	result->Unref();
+	return 0;
+}
+
+int iso9660_fuse_access(const char* path, int mode)
+{
+	Inode* dir = iso9660_fuse_parent_dir(&path);
+	if ( !dir )
+		return -errno;
+	Inode* result = dir->Open(path, O_RDONLY, 0);
+	dir->Unref();
+	if ( !result )
+		return -errno;
+	(void) mode;
+	result->Unref();
+	return 0;
+}
+
+int iso9660_fuse_create(const char* path, mode_t mode, struct fuse_file_info* fi)
+{
+	int flags = fi->flags | O_CREAT;
+	Inode* inode = iso9660_fuse_parent_dir(&path);
+	if ( !inode )
+		return -errno;
+	Inode* result = inode->Open(path, flags, FsModeFromHostMode(mode));
+	inode->Unref();
+	if ( !result )
+		return -errno;
+	fi->fh = (uint64_t) result->inode_id;
+	fi->keep_cache = 1;
+	result->RemoteRefer();
+	result->Unref();
+	return 0;
+}
+
+int iso9660_fuse_opendir(const char* path, struct fuse_file_info* fi)
+{
+	return iso9660_fuse_open(path, fi);
+}
+
+int iso9660_fuse_read(const char* /*path*/, char* buf, size_t count,
+                      off_t offset, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	if ( INT_MAX < count )
+		count = INT_MAX;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	ssize_t result = inode->ReadAt((uint8_t*) buf, count, offset);
+	inode->Unref();
+	return 0 <= result ? (int) result : -errno;
+}
+
+int iso9660_fuse_write(const char* /*path*/, const char* /*buf*/,
+                       size_t /*count*/, off_t /*offset*/,
+                       struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->Unref();
+	return -(errno = EROFS);
+}
+
+int iso9660_fuse_statfs(const char* /*path*/, struct statvfs* stvfs)
+{
+	memset(stvfs, 0, sizeof(*stvfs));
+	Filesystem* fs = FUSE_FS;
+	stvfs->f_bsize = fs->block_size;
+	stvfs->f_frsize = fs->block_size;
+	stvfs->f_blocks = fs->device->device_size / fs->block_size;
+	stvfs->f_bfree = 0;
+	stvfs->f_bavail = 0;
+	stvfs->f_files = 0;
+	stvfs->f_ffree = 0;
+	stvfs->f_favail = 0;
+	stvfs->f_ffree = 0;
+	stvfs->f_fsid = 0;
+	stvfs->f_flag = ST_RDONLY;
+	stvfs->f_namemax = 255;
+	return 0;
+}
+
+int iso9660_fuse_flush(const char* /*path*/, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->Unref();
+	return 0;
+}
+
+int iso9660_fuse_release(const char* /*path*/, struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->RemoteUnref();
+	inode->Unref();
+	return 0;
+}
+
+int iso9660_fuse_releasedir(const char* path, struct fuse_file_info* fi)
+{
+	return iso9660_fuse_release(path, fi);
+}
+
+int iso9660_fuse_fsync(const char* /*path*/, int data, struct fuse_file_info* fi)
+{
+	(void) data;
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((uint32_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	inode->Unref();
+	return 0;
+}
+
+/*int iso9660_fuse_syncdir(const char* path, int data, struct fuse_file_info* fi)
+{
+	return iso9660_fuse_sync(path, data, fi);
+}*/
+
+/*int iso9660_fuse_setxattr(const char *, const char *, const char *, size_t, int)
+{
+	return -(errno = ENOSYS);
+}*/
+
+/*int iso9660_fuse_getxattr(const char *, const char *, char *, size_t)
+{
+	return -(errno = ENOSYS);
+}*/
+
+/*int iso9660_fuse_listxattr(const char *, char *, size_t)
+{
+	return -(errno = ENOSYS);
+}*/
+
+/*int iso9660_fuse_removexattr(const char *, const char *)
+{
+	return -(errno = ENOSYS);
+}*/
+
+int iso9660_fuse_readdir(const char* /*path*/, void* buf,
+                         fuse_fill_dir_t filler, off_t rec_num,
+                         struct fuse_file_info* fi)
+{
+	Filesystem* fs = FUSE_FS;
+	Inode* inode = fs->GetInode((iso9660_ino_t) fi->fh);
+	if ( !inode )
+		return -errno;
+	if ( !S_ISDIR(inode->Mode()) )
+		return inode->Unref(), -(errno = ENOTDIR);
+	uint64_t offset = 0;
+	Block* block = NULL;
+	uint64_t block_id = 0;
+	char name[256];
+	uint8_t file_type;
+	iso9660_ino_t inode_id;
+	while ( inode->ReadDirectory(&offset, &block, &block_id,
+	                             rec_num ? NULL : name, &file_type, &inode_id) )
+	{
+		if ( !rec_num || !rec_num-- )
+		{
+			if ( filler(buf, name, NULL, 0) )
+			{
+				block->Unref();
+				inode->Unref();
+				return 0;
+			}
+		}
+	}
+	int errnum = errno;
+	if ( block )
+		block->Unref();
+	inode->Unref();
+	if ( errnum )
+		return -errnum;
+	return 0;
+}
+
+/*int iso9660_fuse_lock(const char*, struct fuse_file_info*, int, struct flock*)
+{
+	return -(errno = ENOSYS);
+}*/
+
+int iso9660_fuse_utimens(const char* path, const struct timespec tv[2])
+{
+	Inode* inode = iso9660_fuse_resolve_path(path);
+	if ( !inode )
+		return -errno;
+	(void) tv;
+	return inode->Unref(), -(errno = EROFS);
+}
+
+/*int iso9660_fuse_bmap(const char*, size_t blocksize, uint64_t* idx)
+{
+	return -(errno = ENOSYS);
+}*/
+
+int iso9660_fuse_main(const char* argv0,
+                   const char* mount_path,
+                   bool foreground,
+                   Filesystem* fs,
+                   Device* dev)
+{
+	struct fuse_operations operations;
+	memset(&operations, 0, sizeof(operations));
+
+	operations.access = iso9660_fuse_access;
+	operations.chmod = iso9660_fuse_chmod;
+	operations.chown = iso9660_fuse_chown;
+	operations.create = iso9660_fuse_create;
+	operations.destroy = iso9660_fuse_destroy;
+	operations.fgetattr = iso9660_fuse_fgetattr;
+	operations.flush = iso9660_fuse_flush;
+	operations.fsync = iso9660_fuse_fsync;
+	operations.ftruncate = iso9660_fuse_ftruncate;
+	operations.getattr = iso9660_fuse_getattr;
+	operations.init = iso9660_fuse_init;
+	operations.link = iso9660_fuse_link;
+	operations.mkdir = iso9660_fuse_mkdir;
+	operations.mknod = iso9660_fuse_mknod;
+	operations.opendir = iso9660_fuse_opendir;
+	operations.open = iso9660_fuse_open;
+	operations.readdir = iso9660_fuse_readdir;
+	operations.read = iso9660_fuse_read;
+	operations.readlink = iso9660_fuse_readlink;
+	operations.releasedir = iso9660_fuse_releasedir;
+	operations.release = iso9660_fuse_release;
+	operations.rename = iso9660_fuse_rename;
+	operations.rmdir = iso9660_fuse_rmdir;
+	operations.statfs = iso9660_fuse_statfs;
+	operations.symlink = iso9660_fuse_symlink;
+	operations.truncate = iso9660_fuse_truncate;
+	operations.unlink = iso9660_fuse_unlink;
+	operations.utimens = iso9660_fuse_utimens;
+	operations.write = iso9660_fuse_write;
+
+	operations.flag_nullpath_ok = 1;
+	operations.flag_nopath = 1;
+
+	char* argv_fuse[] =
+	{
+		(char*) argv0,
+		(char*) "-s",
+		(char*) (foreground ? "-f" : mount_path),
+		(char*) (foreground ? mount_path : NULL),
+		(char*) NULL,
+	};
+
+	int argc_fuse = sizeof(argv_fuse) / sizeof(argv_fuse[0]) - 1;
+
+	struct iso9660_fuse_ctx iso9660_fuse_ctx;
+	iso9660_fuse_ctx.fs = fs;
+	iso9660_fuse_ctx.dev = dev;
+
+	return fuse_main(argc_fuse, argv_fuse, &operations, &iso9660_fuse_ctx);
+}
+
+#endif

iso9660/fuse.h

@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * fuse.h
+ * FUSE frontend.
+ */
+
+#ifndef FUSE_H
+#define FUSE_H
+
+class Device;
+class Filesystem;
+
+int iso9660_fuse_main(const char* argv0,
+                      const char* mount_path,
+                      bool foreground,
+                      Filesystem* fs,
+                      Device* dev);
+
+#endif

iso9660/inode.cpp

@@ -0,0 +1,614 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2018, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * inode.cpp
+ * Filesystem inode.
+ */
+
+#define __STDC_CONSTANT_MACROS
+#define __STDC_LIMIT_MACROS
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <ctype.h>
+#include <errno.h>
+#include <endian.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+#include <time.h>
+
+#include "block.h"
+#include "device.h"
+#include "filesystem.h"
+#include "inode.h"
+#include "iso9660fs.h"
+#include "util.h"
+
+#ifndef S_SETABLE
+#define S_SETABLE 02777
+#endif
+#ifndef O_WRITE
+#define O_WRITE (O_WRONLY | O_RDWR)
+#endif
+
+Inode::Inode(Filesystem* filesystem, iso9660_ino_t inode_id)
+{
+	this->prev_inode = NULL;
+	this->next_inode = NULL;
+	this->prev_hashed = NULL;
+	this->next_hashed = NULL;
+	this->data_block = NULL;
+	this->data = NULL;
+	this->filesystem = filesystem;
+	this->reference_count = 1;
+	this->remote_reference_count = 0;
+	this->inode_id = inode_id;
+	this->parent_inode_id = 0;
+}
+
+Inode::~Inode()
+{
+	if ( data_block )
+		data_block->Unref();
+	Unlink();
+}
+
+void Inode::Parse()
+{
+	const uint8_t* block_data = (const uint8_t*) data;
+	uid = 0;
+	gid = 0;
+	time = 0;
+	uint8_t file_flags = block_data[25];
+	bool is_directory = file_flags & ISO9660_DIRENT_FLAG_DIR;
+	mode = 0555 | (is_directory ? ISO9660_S_IFDIR : ISO9660_S_IFREG);
+	uint32_t u32;
+	memcpy(&u32, block_data + 10, sizeof(u32));
+	size = le32toh(u32);
+	nlink = 1;
+	const uint8_t* time_bytes = block_data + 18;
+	struct tm tm;
+	memset(&tm, 0, sizeof(tm));
+	tm.tm_year = time_bytes[0];
+	tm.tm_mon = time_bytes[1] - 1;
+	tm.tm_mday = time_bytes[2];
+	tm.tm_hour = time_bytes[3];
+	tm.tm_min = time_bytes[4];
+	tm.tm_sec = time_bytes[5];
+	// TODO: Is this accurate?
+	time_t tz_offset = (-48 + time_bytes[6]) * 15 * 60;
+	// TODO: The timezone offset should've been mixed in with mktime, somehow.
+	time = mktime(&tm) + tz_offset;
+	uint8_t dirent_len = block_data[0];
+	uint8_t name_len = block_data[32];
+	size_t extended_off = 33 + name_len + !(name_len & 1);
+	for ( size_t i = extended_off; i < dirent_len && 3 <= dirent_len - i; )
+	{
+		const uint8_t* field = block_data + i;
+		uint8_t len = field[2];
+		if ( !len || dirent_len - i < len )
+			break;
+		i += len;
+		if ( len == 36 && field[0] == 'P' && field[1] == 'X' && field[3] == 1 )
+		{
+			uint32_t bits;
+			memcpy(&bits, field + 4, sizeof(bits));
+			mode = le32toh(bits) & 0xffff;
+			memcpy(&bits, field + 12, sizeof(bits));
+			nlink = le32toh(bits);
+			memcpy(&bits, field + 20, sizeof(bits));
+			uid = le32toh(bits);
+			memcpy(&bits, field + 28, sizeof(bits));
+			gid = le32toh(bits);
+		}
+	}
+	if ( ISO9660_S_ISLNK(mode) )
+	{
+		uint8_t buf[256];
+		size = ReadLink(buf, sizeof(buf));
+	}
+}
+
+uint32_t Inode::Mode()
+{
+	return mode;
+}
+
+uint32_t Inode::UserId()
+{
+	return uid;
+}
+
+uint32_t Inode::GroupId()
+{
+	return gid;
+}
+
+uint64_t Inode::Size()
+{
+	return size;
+}
+
+uint64_t Inode::Time()
+{
+	return time;
+}
+
+Block* Inode::GetBlock(uint32_t offset)
+{
+	uint32_t lba;
+	memcpy(&lba, (uint8_t*) data + 2, sizeof(lba));
+	lba = le32toh(lba);
+	uint32_t block_id = lba + offset;
+	return filesystem->device->GetBlock(block_id);
+}
+
+bool Inode::FindParentInode(uint64_t parent_lba, uint64_t parent_size)
+{
+	if ( inode_id == filesystem->root_ino )
+		return parent_inode_id = inode_id, true;
+	Block* block = NULL;
+	uint32_t block_size = filesystem->block_size;
+	uint64_t parent_offset = parent_lba * block_size;
+	uint64_t block_id = 0;
+	uint64_t offset = 0;
+	while ( offset < parent_size )
+	{
+		uint64_t entry_block_id = (parent_offset + offset) / block_size;
+		uint64_t entry_block_offset = (parent_offset + offset) % block_size;
+		if ( block && block_id != entry_block_id )
+		{
+			block->Unref();
+			block = NULL;
+		}
+		if ( !block && !(block = filesystem->device->GetBlock(entry_block_id)) )
+			return false;
+		const uint8_t* block_data = block->block_data + entry_block_offset;
+		uint8_t dirent_len = block_data[0];
+		if ( !dirent_len )
+		{
+			offset = (entry_block_id + 1) * block_size;
+			continue;
+		}
+		uint8_t name_len = block_data[32];
+		const uint8_t* name_data = block_data + 33;
+		if ( name_len == 0 || !name_data[0] )
+		{
+			parent_inode_id = parent_offset + offset;
+			return true;
+		}
+		// TODO: Can dirent_len be misaligned?
+		uint64_t reclen = dirent_len + (dirent_len & 1);
+		if ( !reclen )
+			return errno = EINVAL, false;
+		offset += reclen;
+	}
+	return errno = EINVAL, false;
+}
+
+bool Inode::ReadDirectory(uint64_t* offset_inout,
+                          Block** block_inout,
+                          uint64_t* block_id_inout,
+                          char* name,
+                          uint8_t* file_type_out,
+                          iso9660_ino_t* inode_id_out)
+{
+	uint64_t offset = *offset_inout;
+next_block:
+	uint64_t filesize = Size();
+	if ( filesize <= offset )
+		return errno = 0, false;
+	uint64_t entry_block_id = offset / filesystem->block_size;
+	uint64_t entry_block_offset = offset % filesystem->block_size;
+	if ( *block_inout && *block_id_inout != entry_block_id )
+	{
+		(*block_inout)->Unref();
+		(*block_inout) = NULL;
+	}
+	if ( !*block_inout &&
+	     !(*block_inout = GetBlock(*block_id_inout = entry_block_id)) )
+		return false;
+	const uint8_t* block_data = (*block_inout)->block_data + entry_block_offset;
+	uint8_t dirent_len = block_data[0];
+	if ( !dirent_len )
+	{
+		offset = (entry_block_id + 1) * filesystem->block_size;
+		goto next_block;
+	}
+	if ( name )
+	{
+		uint8_t name_len = block_data[32];
+		const uint8_t* name_data = block_data + 33;
+		size_t extended_off = 33 + name_len + !(name_len & 1);
+		iso9660_ino_t entry_inode_id =
+			((*block_inout)->block_id * filesystem->block_size) +
+			entry_block_offset;
+		// TODO: The root directory inode should be that of its . entry.
+		if ( name_len == 0 || !name_data[0] )
+		{
+			name[0] = '.';
+			name[1] = '\0';
+			name_len = 1;
+			entry_inode_id = inode_id;
+		}
+		else if ( name_len == 1 && name_data[0] == 1 )
+		{
+			name[0] = '.';
+			name[1] = '.';
+			name[2] = '\0';
+			name_len = 2;
+			if ( !parent_inode_id )
+			{
+				uint32_t parent_lba;
+				memcpy(&parent_lba, block_data + 2, sizeof(parent_lba));
+				parent_lba = le32toh(parent_lba);
+				uint32_t parent_size;
+				memcpy(&parent_size, block_data + 10, sizeof(parent_size));
+				parent_size = le32toh(parent_size);
+				if ( !FindParentInode(parent_lba, parent_size) )
+					return false;
+			}
+			entry_inode_id = parent_inode_id;
+		}
+		else
+		{
+			for ( size_t i = 0; i < name_len; i++ )
+			{
+				if ( name_data[i] == ';' )
+				{
+					name_len = i;
+					break;
+				}
+				name[i] = tolower(name_data[i]);
+			}
+			name[name_len] = '\0';
+		}
+		for ( size_t i = extended_off; i < dirent_len && 3 <= dirent_len - i; )
+		{
+			const uint8_t* field = block_data + i;
+			uint8_t len = field[2];
+			if ( !len || dirent_len - i < len )
+				break;
+			i += len;
+			if ( 5 <= len && field[0] == 'N' && field[1] == 'M' &&
+				 field[3] == 1 )
+			{
+				uint8_t nm_flags = field[4];
+				if ( nm_flags & (1 << 0) ) // TODO: Continued names.
+					break;
+				name_len = len - 5;
+				memcpy(name, field + 5, name_len);
+				name[name_len] = '\0';
+			}
+			// TODO: Other extensions.
+		}
+		uint8_t file_flags = block_data[25];
+		// TODO: Rock Ridge.
+		bool is_directory = file_flags & ISO9660_DIRENT_FLAG_DIR;
+		uint8_t file_type = is_directory ? ISO9660_FT_DIR : ISO9660_FT_REG_FILE;
+		*file_type_out = file_type;
+		*inode_id_out = entry_inode_id;
+	}
+	// TODO: Can dirent_len be misaligned?
+	uint64_t reclen = dirent_len + (dirent_len & 1);
+	if ( !reclen )
+		return errno = EINVAL, false;
+	offset += reclen;
+	*offset_inout = offset;
+	return true;
+}
+
+Inode* Inode::Open(const char* elem, int flags, mode_t mode)
+{
+	if ( !ISO9660_S_ISDIR(Mode()) )
+		return errno = ENOTDIR, (Inode*) NULL;
+	size_t elem_length = strlen(elem);
+	if ( elem_length == 0 )
+		return errno = ENOENT, (Inode*) NULL;
+	uint64_t offset = 0;
+	Block* block = NULL;
+	uint64_t block_id = 0;
+	char name[256];
+	uint8_t file_type;
+	iso9660_ino_t inode_id;
+	while ( ReadDirectory(&offset, &block, &block_id, name, &file_type,
+	                      &inode_id) )
+	{
+		size_t name_len = strlen(name);
+		if ( name_len == elem_length && memcmp(elem, name, elem_length) == 0 )
+		{
+			block->Unref();
+			if ( (flags & O_CREAT) && (flags & O_EXCL) )
+				return errno = EEXIST, (Inode*) NULL;
+			if ( (flags & O_DIRECTORY) &&
+			     file_type != ISO9660_FT_UNKNOWN &&
+			     file_type != ISO9660_FT_DIR &&
+			     file_type != ISO9660_FT_SYMLINK )
+				return errno = ENOTDIR, (Inode*) NULL;
+			Inode* inode = filesystem->GetInode(inode_id);
+			if ( !inode )
+				return (Inode*) NULL;
+			if ( flags & O_DIRECTORY &&
+			     !ISO9660_S_ISDIR(inode->Mode()) &&
+			     !ISO9660_S_ISLNK(inode->Mode()) )
+			{
+				inode->Unref();
+				return errno = ENOTDIR, (Inode*) NULL;
+			}
+			if ( flags & O_WRITE )
+			{
+				inode->Unref();
+				return errno = EROFS, (Inode*) NULL;
+			}
+			return inode;
+		}
+	}
+	if ( block )
+		block->Unref();
+	if ( flags & O_CREAT )
+	{
+		(void) mode;
+		return errno = EROFS, (Inode*) NULL;
+	}
+	return errno = ENOENT, (Inode*) NULL;
+}
+
+bool Inode::Link(const char* elem, Inode* dest)
+{
+	if ( !ISO9660_S_ISDIR(Mode()) )
+		return errno = ENOTDIR, false;
+	if ( ISO9660_S_ISDIR(dest->Mode()) )
+		return errno = EISDIR, false;
+
+	size_t elem_length = strlen(elem);
+	if ( elem_length == 0 )
+		return errno = ENOENT, false;
+	uint64_t offset = 0;
+	Block* block = NULL;
+	uint64_t block_id = 0;
+	char name[256];
+	uint8_t file_type;
+	iso9660_ino_t inode_id;
+	while ( ReadDirectory(&offset, &block, &block_id, name, &file_type,
+	                      &inode_id) )
+	{
+		size_t name_len = strlen(name);
+		if ( name_len == elem_length && memcmp(elem, name, elem_length) == 0 )
+		{
+			block->Unref();
+			return errno = EEXIST, false;
+		}
+	}
+	if ( block )
+		block->Unref();
+	return errno = EROFS, false;
+}
+
+Inode* Inode::UnlinkKeep(const char* elem, bool directories, bool force)
+{
+	if ( !ISO9660_S_ISDIR(Mode()) )
+		return errno = ENOTDIR, (Inode*) NULL;
+	size_t elem_length = strlen(elem);
+	if ( elem_length == 0 )
+		return errno = ENOENT, (Inode*) NULL;
+	uint64_t offset = 0;
+	Block* block = NULL;
+	uint64_t block_id = 0;
+	char name[256];
+	uint8_t file_type;
+	iso9660_ino_t inode_id;
+	while ( ReadDirectory(&offset, &block, &block_id, name, &file_type,
+	                      &inode_id) )
+	{
+		size_t name_len = strlen(name);
+		if ( name_len == elem_length && memcmp(elem, name, elem_length) == 0 )
+		{
+			(void) directories;
+			(void) force;
+			block->Unref();
+			return errno = EROFS, (Inode*) NULL;
+		}
+	}
+	if ( block )
+		block->Unref();
+	return errno = ENOENT, (Inode*) NULL;
+}
+
+bool Inode::Unlink(const char* elem, bool directories, bool force)
+{
+	Inode* result = UnlinkKeep(elem, directories, force);
+	if ( !result )
+		return false;
+	result->Unref();
+	return true;
+}
+
+ssize_t Inode::ReadLink(uint8_t* buf, size_t bufsize)
+{
+	size_t result = 0;
+	const uint8_t* block_data = (const uint8_t*) data;
+	uint8_t dirent_len = block_data[0];
+	uint8_t name_len = block_data[32];
+	size_t extended_off = 33 + name_len + !(name_len & 1);
+	bool continued = true;
+	for ( size_t i = extended_off; i < dirent_len && 3 <= dirent_len - i; )
+	{
+		const uint8_t* field = block_data + i;
+		uint8_t len = field[2];
+		if ( !len || dirent_len - i < len )
+			break;
+		i += len;
+		if ( 5 <= len && field[0] == 'S' && field[1] == 'L' && field[3] == 1 )
+		{
+			for ( size_t n = 5; n < len && 2 <= len - n; )
+			{
+				uint8_t comp_flags = field[n + 0];
+				uint8_t comp_len = field[n + 1];
+				if ( len - (n + 2) < comp_len )
+					break;
+				const char* data = (const char*) (field + n + 2);
+				size_t datalen = comp_len;
+				// TODO: How is a trailing slash encoded?
+				if ( !continued || (comp_flags & (1 << 3) /* root */) )
+				{
+					buf[result++] = '/';
+					if ( result == bufsize )
+						return (ssize_t) result;
+				}
+				if ( comp_flags & (1 << 1) )
+				{
+					data = ".";
+					datalen = 1;
+				}
+				else if ( comp_flags & (1 << 2) )
+				{
+					data = "..";
+					datalen = 2;
+				}
+				size_t possible = bufsize - result;
+				size_t count = datalen < possible ? datalen : possible;
+				memcpy(buf + result, data, count);
+				result += count;
+				if ( result == bufsize )
+					return (ssize_t) result;
+				continued = comp_flags & (1 << 0);
+				n += 2 + comp_len;
+			}
+		}
+	}
+	return (ssize_t) result;
+}
+
+ssize_t Inode::ReadAt(uint8_t* buf, size_t s_count, off_t o_offset)
+{
+	if ( !ISO9660_S_ISREG(Mode()) && !ISO9660_S_ISLNK(Mode()) )
+		return errno = EISDIR, -1;
+	if ( o_offset < 0 )
+		return errno = EINVAL, -1;
+	if ( SSIZE_MAX < s_count )
+		s_count = SSIZE_MAX;
+	uint64_t sofar = 0;
+	uint64_t count = (uint64_t) s_count;
+	uint64_t offset = (uint64_t) o_offset;
+	uint64_t file_size = Size();
+	if ( file_size <= offset )
+		return 0;
+	if ( file_size - offset < count )
+		count = file_size - offset;
+	while ( sofar < count )
+	{
+		uint64_t block_id = offset / filesystem->block_size;
+		uint32_t block_offset = offset % filesystem->block_size;
+		uint32_t block_left = filesystem->block_size - block_offset;
+		Block* block = GetBlock(block_id);
+		if ( !block )
+			return sofar ? sofar : -1;
+		size_t amount = count - sofar < block_left ? count - sofar : block_left;
+		memcpy(buf + sofar, block->block_data + block_offset, amount);
+		sofar += amount;
+		offset += amount;
+		block->Unref();
+	}
+	return (ssize_t) sofar;
+}
+
+bool Inode::Rename(Inode* olddir, const char* oldname, const char* newname)
+{
+	if ( !strcmp(oldname, ".") || !strcmp(oldname, "..") ||
+	     !strcmp(newname, ".") || !strcmp(newname, "..") )
+		return errno = EPERM, false;
+	Inode* src_inode = olddir->Open(oldname, O_RDONLY, 0);
+	if ( !src_inode )
+		return false;
+	if ( Inode* dst_inode = Open(newname, O_RDONLY, 0) )
+	{
+		bool same_inode = src_inode->inode_id == dst_inode->inode_id;
+		dst_inode->Unref();
+		if ( same_inode )
+			return src_inode->Unref(), true;
+	}
+	src_inode->Unref();
+	return errno = EROFS, false;
+}
+
+bool Inode::RemoveDirectory(const char* path)
+{
+	return UnlinkKeep(path, true);
+}
+
+void Inode::Refer()
+{
+	reference_count++;
+}
+
+void Inode::Unref()
+{
+	assert(0 < reference_count);
+	reference_count--;
+	if ( !reference_count && !remote_reference_count )
+		delete this;
+}
+
+void Inode::RemoteRefer()
+{
+	remote_reference_count++;
+}
+
+void Inode::RemoteUnref()
+{
+	assert(0 < remote_reference_count);
+	remote_reference_count--;
+	if ( !reference_count && !remote_reference_count )
+		delete this;
+}
+
+void Inode::Use()
+{
+	data_block->Use();
+	Unlink();
+	Prelink();
+}
+
+void Inode::Unlink()
+{
+	(prev_inode ? prev_inode->next_inode : filesystem->mru_inode) = next_inode;
+	(next_inode ? next_inode->prev_inode : filesystem->lru_inode) = prev_inode;
+	size_t bin = inode_id % INODE_HASH_LENGTH;
+	(prev_hashed ? prev_hashed->next_hashed : filesystem->hash_inodes[bin]) = next_hashed;
+	if ( next_hashed ) next_hashed->prev_hashed = prev_hashed;
+}
+
+void Inode::Prelink()
+{
+	prev_inode = NULL;
+	next_inode = filesystem->mru_inode;
+	if ( filesystem->mru_inode )
+		filesystem->mru_inode->prev_inode = this;
+	filesystem->mru_inode = this;
+	if ( !filesystem->lru_inode )
+		filesystem->lru_inode = this;
+	size_t bin = inode_id % INODE_HASH_LENGTH;
+	prev_hashed = NULL;
+	next_hashed = filesystem->hash_inodes[bin];
+	filesystem->hash_inodes[bin] = this;
+	if ( next_hashed )
+		next_hashed->prev_hashed = this;
+}

iso9660/inode.h

@@ -0,0 +1,83 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * inode.h
+ * Filesystem inode.
+ */
+
+#ifndef INODE_H
+#define INODE_H
+
+#include "iso9660.h"
+
+class Block;
+class Filesystem;
+
+class Inode
+{
+public:
+	Inode(Filesystem* filesystem, iso9660_ino_t inode_id);
+	~Inode();
+
+public:
+	Inode* prev_inode;
+	Inode* next_inode;
+	Inode* prev_hashed;
+	Inode* next_hashed;
+	Block* data_block;
+	struct iso9660_dirent* data;
+	Filesystem* filesystem;
+	size_t reference_count;
+	size_t remote_reference_count;
+	iso9660_ino_t inode_id;
+	iso9660_ino_t parent_inode_id;
+	uint32_t uid;
+	uint32_t gid;
+	uint64_t size;
+	uint64_t time;
+	uint32_t mode;
+	uint32_t nlink;
+
+public:
+	void Parse();
+	uint32_t Mode();
+	uint32_t UserId();
+	uint32_t GroupId();
+	uint64_t Size();
+	uint64_t Time();
+	Block* GetBlock(uint32_t offset);
+	bool FindParentInode(uint64_t parent_lba, uint64_t parent_size);
+	bool ReadDirectory(uint64_t* offset_inout, Block** block_inout,
+	                   uint64_t* block_id_inout, char* name,
+	                   uint8_t* file_type_out, iso9660_ino_t* inode_id_out);
+	Inode* Open(const char* elem, int flags, mode_t mode);
+	bool Link(const char* elem, Inode* dest);
+	bool Unlink(const char* elem, bool directories, bool force=false);
+	Inode* UnlinkKeep(const char* elem, bool directories, bool force=false);
+	ssize_t ReadLink(uint8_t* buf, size_t bufsize);
+	ssize_t ReadAt(uint8_t* buffer, size_t count, off_t offset);
+	bool Rename(Inode* olddir, const char* oldname, const char* newname);
+	bool RemoveDirectory(const char* path);
+	void Refer();
+	void Unref();
+	void RemoteRefer();
+	void RemoteUnref();
+	void Use();
+	void Unlink();
+	void Prelink();
+
+};
+
+#endif

iso9660/ioleast.h

@@ -0,0 +1,150 @@
+/*
+ * Copyright (c) 2012, 2013, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * ioleast.h
+ * Versions of {,p}{read,write} that don't return until it has returned as much
+ * data as requested, end of file, or an error occurs. This is sometimes needed
+ * as read(2) and write(2) is not always guaranteed to fill up the entire
+ * buffer or write it all.
+ */
+
+#ifndef SORTIX_COMPATIBILITY_INCLUDE_IOLEAST_H
+#define SORTIX_COMPATIBILITY_INCLUDE_IOLEAST_H
+
+#if defined(__sortix__) || defined(__sortix_libc__)
+
+#include_next <ioleast.h>
+
+#else
+
+#include <sys/types.h>
+
+#include <assert.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <unistd.h>
+
+#if !defined(EEOF) && defined(EIO)
+#define EEOF EIO
+#endif
+
+__attribute__((unused)) static inline
+size_t readleast(int fd, void* buf_ptr, size_t least, size_t max)
+{
+	assert(least <= max);
+	unsigned char* buf = (unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = read(fd, buf + done, max - done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t writeleast(int fd, const void* buf_ptr, size_t least, size_t max)
+{
+	assert(least <= max);
+	const unsigned char* buf = (const unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = write(fd, buf + done, max - done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t preadleast(int fd, void* buf_ptr, size_t least, size_t max, off_t off)
+{
+	assert(least <= max);
+	unsigned char* buf = (unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = pread(fd, buf + done, max - done, off + done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t pwriteleast(int fd, const void* buf_ptr, size_t least, size_t max, off_t off)
+{
+	assert(least <= max);
+	const unsigned char* buf = (const unsigned char*) buf_ptr;
+	size_t done = 0;
+	do
+	{
+		ssize_t amount = pwrite(fd, buf + done, max - done, off + done);
+		if ( amount < 0 )
+			return done;
+		if ( !amount && done < least )
+			return errno = EEOF, done;
+		if ( !amount )
+			break;
+		done += amount;
+	} while ( done < least );
+	return done;
+}
+
+__attribute__((unused)) static inline
+size_t readall(int fd, void* buf, size_t count)
+{
+	return readleast(fd, buf, count, count);
+}
+
+__attribute__((unused)) static inline
+size_t writeall(int fd, const void* buf, size_t count)
+{
+	return writeleast(fd, buf, count, count);
+}
+
+__attribute__((unused)) static inline
+size_t preadall(int fd, void* buf, size_t count, off_t off)
+{
+	return preadleast(fd, buf, count, count, off);
+}
+
+__attribute__((unused)) static inline
+size_t pwriteall(int fd, const void* buf, size_t count, off_t off)
+{
+	return pwriteleast(fd, buf, count, count, off);
+}
+
+#endif
+
+#endif

iso9660/iso9660.h

@@ -0,0 +1,128 @@
+/*
+ * Copyright (c) 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * iso9660.h
+ * Data structures for the ISO 9660 filesystem.
+ */
+
+#ifndef ISO9660_H
+#define ISO9660_H
+
+#include <stdint.h>
+
+const uint8_t TYPE_BOOT_RECORD = 0x00;
+const uint8_t TYPE_PRIMARY_VOLUME_DESCRIPTOR = 0x01;
+const uint8_t TYPE_VOLUME_DESCRIPTOR_SET_TERMINATOR = 0xFF;
+
+struct iso9660_pvd /* primary volume descriptor */
+{
+	uint8_t type;
+	char standard_identifier[5];
+	uint8_t version;
+	uint8_t unused1;
+	char system_identifier[32];
+	char volume_identifier[32];
+	uint8_t unused2[8];
+	uint32_t volume_space_size_le;
+	uint32_t volume_space_size_be;
+	uint8_t unused3[32];
+	uint16_t volume_set_size_le;
+	uint16_t volume_set_size_be;
+	uint16_t volume_sequence_number_le;
+	uint16_t volume_sequence_number_be;
+	uint16_t logical_block_size_le;
+	uint16_t logical_block_size_be;
+	uint32_t path_table_size_le;
+	uint32_t path_table_size_be;
+	uint32_t path_table_lba_le;
+	uint32_t path_table_opt_lba_le;
+	uint32_t path_table_lba_be;
+	uint32_t path_table_opt_lba_be;
+	uint8_t root_dirent[34];
+	char volume_set_identifier[128];
+	char publisher_identifier[128];
+	char data_preparer_identifier[128];
+	char application_identifier[128];
+	char copyright_file_identifier[37];
+	char abstract_file_identifier[37];
+	char bibliographic_file_identifier[37];
+	char creation_datetime[17];
+	char modification_datetime[17];
+	char expiration_datetime[17];
+	char effective_datetime[17];
+	uint8_t file_structure_version;
+	uint8_t unused4;
+	uint8_t application_use[512];
+	uint8_t reserved[653];
+};
+
+static_assert(sizeof(struct iso9660_pvd) == 2048,
+              "sizeof(struct iso9660_pvd) == 2048");
+
+typedef uint64_t iso9660_ino_t;
+
+struct iso9660_dirent
+{
+	uint8_t unused;
+};
+
+#define ISO9660_DIRENT_FLAG_DIR (1 << 1)
+
+#define ISO9660_S_IFMT (0170000)
+#define ISO9660_S_IFIFO (0010000)
+#define ISO9660_S_IFCHR (0020000)
+#define ISO9660_S_IFDIR (0040000)
+#define ISO9660_S_IFBLK (0060000)
+#define ISO9660_S_IFREG (0100000)
+#define ISO9660_S_IFLNK (0120000)
+#define ISO9660_S_IFSOCK (0140000)
+
+#define ISO9660_S_ISSOCK(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFSOCK)
+#define ISO9660_S_ISLNK(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFLNK)
+#define ISO9660_S_ISREG(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFREG)
+#define ISO9660_S_ISBLK(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFBLK)
+#define ISO9660_S_ISDIR(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFDIR)
+#define ISO9660_S_ISCHR(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFCHR)
+#define ISO9660_S_ISFIFO(mode) (((mode) & ISO9660_S_IFMT) == ISO9660_S_IFIFO)
+
+static const uint8_t ISO9660_FT_UNKNOWN = 0;
+static const uint8_t ISO9660_FT_REG_FILE = 1;
+static const uint8_t ISO9660_FT_DIR = 2;
+static const uint8_t ISO9660_FT_CHRDEV = 3;
+static const uint8_t ISO9660_FT_BLKDEV = 4;
+static const uint8_t ISO9660_FT_FIFO = 5;
+static const uint8_t ISO9660_FT_SOCK = 6;
+static const uint8_t ISO9660_FT_SYMLINK = 7;
+
+static inline uint8_t ISO9660_FT_OF_MODE(mode_t mode)
+{
+	if ( ISO9660_S_ISREG(mode) )
+		return ISO9660_FT_REG_FILE;
+	if ( ISO9660_S_ISDIR(mode) )
+		return ISO9660_FT_DIR;
+	if ( ISO9660_S_ISCHR(mode) )
+		return ISO9660_FT_CHRDEV;
+	if ( ISO9660_S_ISBLK(mode) )
+		return ISO9660_FT_BLKDEV;
+	if ( ISO9660_S_ISFIFO(mode) )
+		return ISO9660_FT_FIFO;
+	if ( ISO9660_S_ISSOCK(mode) )
+		return ISO9660_FT_SOCK;
+	if ( ISO9660_S_ISLNK(mode) )
+		return ISO9660_FT_SYMLINK;
+	return ISO9660_FT_UNKNOWN;
+}
+
+#endif

iso9660/iso9660fs.cpp

@@ -0,0 +1,278 @@
+/*
+ * Copyright (c) 2013, 2014, 2015, 2016, 2022, 2023 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * iso9660fs.cpp
+ * Implementation of the ISO 9660 filesystem.
+ */
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <assert.h>
+#include <dirent.h>
+#include <endian.h>
+#include <err.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#if defined(__sortix__)
+#include "fsmarshall.h"
+#else
+#include "fuse.h"
+#endif
+
+#include "block.h"
+#include "device.h"
+#include "filesystem.h"
+#include "inode.h"
+#include "ioleast.h"
+#include "iso9660.h"
+#include "iso9660fs.h"
+#include "util.h"
+
+uid_t request_uid;
+uid_t request_gid;
+
+mode_t HostModeFromFsMode(uint32_t mode)
+{
+	mode_t hostmode = mode & 07777;
+	if ( ISO9660_S_ISSOCK(mode) ) hostmode |= S_IFSOCK;
+	if ( ISO9660_S_ISLNK(mode) ) hostmode |= S_IFLNK;
+	if ( ISO9660_S_ISREG(mode) ) hostmode |= S_IFREG;
+	if ( ISO9660_S_ISBLK(mode) ) hostmode |= S_IFBLK;
+	if ( ISO9660_S_ISDIR(mode) ) hostmode |= S_IFDIR;
+	if ( ISO9660_S_ISCHR(mode) ) hostmode |= S_IFCHR;
+	if ( ISO9660_S_ISFIFO(mode) ) hostmode |= S_IFIFO;
+	return hostmode;
+}
+
+uint32_t FsModeFromHostMode(mode_t hostmode)
+{
+	uint32_t mode = hostmode & 07777;
+	if ( S_ISSOCK(hostmode) ) mode |= ISO9660_S_IFSOCK;
+	if ( S_ISLNK(hostmode) ) mode |= ISO9660_S_IFLNK;
+	if ( S_ISREG(hostmode) ) mode |= ISO9660_S_IFREG;
+	if ( S_ISBLK(hostmode) ) mode |= ISO9660_S_IFBLK;
+	if ( S_ISDIR(hostmode) ) mode |= ISO9660_S_IFDIR;
+	if ( S_ISCHR(hostmode) ) mode |= ISO9660_S_IFCHR;
+	if ( S_ISFIFO(hostmode) ) mode |= ISO9660_S_IFIFO;
+	return mode;
+}
+
+uint8_t HostDTFromFsDT(uint8_t fsdt)
+{
+	switch ( fsdt )
+	{
+	case ISO9660_FT_UNKNOWN: return DT_UNKNOWN;
+	case ISO9660_FT_REG_FILE: return DT_REG;
+	case ISO9660_FT_DIR: return DT_DIR;
+	case ISO9660_FT_CHRDEV: return DT_CHR;
+	case ISO9660_FT_BLKDEV: return DT_BLK;
+	case ISO9660_FT_FIFO: return DT_FIFO;
+	case ISO9660_FT_SOCK: return DT_SOCK;
+	case ISO9660_FT_SYMLINK: return DT_LNK;
+	}
+	return DT_UNKNOWN;
+}
+
+void StatInode(Inode* inode, struct stat* st)
+{
+	memset(st, 0, sizeof(*st));
+	st->st_ino = inode->inode_id;
+	st->st_mode = HostModeFromFsMode(inode->Mode());
+	st->st_nlink = inode->nlink;
+	st->st_uid = inode->uid;
+	st->st_gid = inode->gid;
+	st->st_size = inode->size;
+	// TODO: Rock Ridge.
+	st->st_atim.tv_sec = inode->Time();
+	st->st_atim.tv_nsec = 0;
+	// TODO: Rock Ridge.
+	st->st_ctim.tv_sec = inode->Time();
+	st->st_ctim.tv_nsec = 0;
+	// TODO: Rock Ridge.
+	st->st_mtim.tv_sec = inode->Time();
+	st->st_mtim.tv_nsec = 0;
+	st->st_blksize = inode->filesystem->block_size;
+	st->st_blocks = divup(st->st_size, (off_t) 512);
+}
+
+static void compact_arguments(int* argc, char*** argv)
+{
+	for ( int i = 0; i < *argc; i++ )
+	{
+		while ( i < *argc && !(*argv)[i] )
+		{
+			for ( int n = i; n < *argc; n++ )
+				(*argv)[n] = (*argv)[n+1];
+			(*argc)--;
+		}
+	}
+}
+
+static void help(FILE* fp, const char* argv0)
+{
+	fprintf(fp, "Usage: %s [OPTION]... DEVICE [MOUNT-POINT]\n", argv0);
+}
+
+static void version(FILE* fp, const char* argv0)
+{
+	fprintf(fp, "%s (Sortix) %s\n", argv0, VERSIONSTR);
+}
+
+int main(int argc, char* argv[])
+{
+	const char* argv0 = argv[0];
+	const char* pretend_mount_path = NULL;
+	bool foreground = false;
+	for ( int i = 1; i < argc; i++ )
+	{
+		const char* arg = argv[i];
+		if ( arg[0] != '-' || !arg[1] )
+			continue;
+		argv[i] = NULL;
+		if ( !strcmp(arg, "--") )
+			break;
+		if ( arg[1] != '-' )
+		{
+			while ( char c = *++arg ) switch ( c )
+			{
+			case 'r': break;
+			default:
+				fprintf(stderr, "%s: unknown option -- '%c'\n", argv0, c);
+				help(stderr, argv0);
+				exit(1);
+			}
+		}
+		else if ( !strcmp(arg, "--help") )
+			help(stdout, argv0), exit(0);
+		else if ( !strcmp(arg, "--version") )
+			version(stdout, argv0), exit(0);
+		else if ( !strcmp(arg, "--background") )
+			foreground = false;
+		else if ( !strcmp(arg, "--foreground") )
+			foreground = true;
+		else if ( !strcmp(arg, "--read") )
+		{
+		}
+		else if ( !strncmp(arg, "--pretend-mount-path=", strlen("--pretend-mount-path=")) )
+			pretend_mount_path = arg + strlen("--pretend-mount-path=");
+		else if ( !strcmp(arg, "--pretend-mount-path") )
+		{
+			if ( i+1 == argc )
+			{
+				fprintf(stderr, "%s: --pretend-mount-path: Missing operand\n", argv0);
+				exit(1);
+			}
+			pretend_mount_path = argv[++i];
+			argv[i] = NULL;
+		}
+		else
+		{
+			fprintf(stderr, "%s: unknown option: %s\n", argv0, arg);
+			help(stderr, argv0);
+			exit(1);
+		}
+	}
+
+	if ( argc == 1 )
+	{
+		help(stdout, argv0);
+		exit(0);
+	}
+
+	compact_arguments(&argc, &argv);
+
+	const char* device_path = 2 <= argc ? argv[1] : NULL;
+	const char* mount_path = 3 <= argc ? argv[2] : NULL;
+
+	if ( !device_path )
+	{
+		help(stderr, argv0);
+		exit(1);
+	}
+
+	if ( !pretend_mount_path )
+		pretend_mount_path = mount_path;
+
+	int fd = open(device_path, O_RDONLY);
+	if ( fd < 0 )
+		err(1, "%s", device_path);
+
+	// Search for the Primary Volume Descriptor.
+	off_t block_size = 2048;
+	struct iso9660_pvd pvd;
+	off_t pvd_offset;
+	for ( uint32_t i = 16; true; i++ )
+	{
+		pvd_offset = block_size * i;
+		if ( preadall(fd, &pvd, sizeof(pvd), pvd_offset) != sizeof(pvd) )
+		{
+			if ( errno == EEOF )
+				errx(1, "Not a valid ISO 9660 filesystem: %s", device_path);
+			else
+				err(1, "read: %s", device_path);
+		}
+		if ( memcmp(pvd.standard_identifier, "CD001", 5) != 0 )
+			errx(1, "Not a valid ISO 9660 filesystem: %s", device_path);
+		if ( pvd.type == TYPE_PRIMARY_VOLUME_DESCRIPTOR )
+			break;
+		if ( pvd.type == TYPE_VOLUME_DESCRIPTOR_SET_TERMINATOR )
+			errx(1, "Not a valid ISO 9660 filesystem: %s", device_path);
+	}
+
+	if ( pvd.version != 1 || pvd.file_structure_version != 1 )
+		errx(1, "Unsupported ISO 9660 filesystem version: %s", device_path);
+	// TODO: Test if appropriate power of two and allow.
+	if ( le32toh(pvd.logical_block_size_le) != block_size )
+		errx(1, "Unsupported ISO 9660 block size: %s", device_path);
+	block_size = le32toh(pvd.logical_block_size_le);
+
+	Device* dev = new Device(fd, device_path, block_size);
+	if ( !dev )
+		err(1, "malloc");
+	Filesystem* fs = new Filesystem(dev, pretend_mount_path, pvd_offset);
+	if ( !fs )
+		err(1, "%s", device_path);
+
+	// Change the root inode to be its own . entry instead of the pvd record, so
+	// parent directories correctly .. to it, and so Rock Ridge extensions work.
+	Inode* root = fs->GetInode(fs->root_ino);
+	if ( !root )
+		err(1, "%s: GetInode", device_path);
+	fs->root_ino = 0;
+	uint32_t root_lba;
+	uint32_t root_size;
+	memcpy(&root_lba, pvd.root_dirent + 2, sizeof(root_lba));
+	memcpy(&root_size, pvd.root_dirent + 10, sizeof(root_size));
+	root->FindParentInode(root_lba, root_size);
+	fs->root_ino = root->parent_inode_id;
+	root->Unref();
+
+	if ( !mount_path )
+		return 0;
+
+#if defined(__sortix__)
+	return fsmarshall_main(argv0, mount_path, foreground, fs, dev);
+#else
+	return iso9660_fuse_main(argv0, mount_path, foreground, fs, dev);
+#endif
+}

iso9660/iso9660fs.h

@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2022 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * iso9660fs.h
+ * Implementation of the ISO 9660 filesystem.
+ */
+
+#ifndef ISO9660FS_H
+#define ISO9660FS_H
+
+extern uid_t request_uid;
+extern gid_t request_gid;
+
+class Inode;
+
+mode_t HostModeFromFsMode(uint32_t fsmode);
+uint32_t FsModeFromHostMode(mode_t hostmode);
+uint8_t HostDTFromFsDT(uint8_t fsdt);
+void StatInode(Inode* inode, struct stat* st);
+
+#endif

iso9660/util.h

@@ -0,0 +1,49 @@
+/*
+ * Copyright (c) 2013, 2015 Jonas 'Sortie' Termansen.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * util.h
+ * Utility functions for the filesystem implementation.
+ */
+
+#ifndef UTIL_H
+#define UTIL_H
+
+template <class T> T divup(T a, T b)
+{
+	return a/b + (a % b ? 1 : 0);
+}
+
+template <class T> T roundup(T a, T b)
+{
+	return a % b ? a + b - a % b : a;
+}
+
+static inline bool checkbit(const uint8_t* bitmap, size_t bit)
+{
+	uint8_t bits = bitmap[bit / 8UL];
+	return bits & (1U << (bit % 8UL));
+}
+
+static inline void setbit(uint8_t* bitmap, size_t bit)
+{
+	bitmap[bit / 8UL] |= 1U << (bit % 8UL);
+}
+
+static inline void clearbit(uint8_t* bitmap, size_t bit)
+{
+	bitmap[bit / 8UL] &= ~(1U << (bit % 8UL));
+}
+
+#endif

kernel/addralloc.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2013, 2014 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2013, 2014, 2022 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -29,6 +29,7 @@
 namespace Sortix {
 
 size_t aux_allocated = 0;
+size_t aux_leaked = 0;
 size_t heap_allocated = 0;
 kthread_mutex_t alloc_lock = KTHREAD_MUTEX_INITIALIZER;
 
@@ -69,6 +70,8 @@ void FreeKernelAddress(addralloc_t* alloc)
 	addr_t aux_reached = kmem_from + kmem_size - aux_allocated;
 	if ( alloc->from == aux_reached )
 		aux_allocated -= alloc->size;
+	else
+		aux_leaked += alloc->size;
 
 	alloc->from = 0;
 	alloc->size = 0;
@@ -98,6 +101,17 @@ void ShrinkHeap(size_t decrease)
 	heap_allocated -= decrease;
 }
 
+void KernelAddressStatistics(size_t* heap, size_t* aux, size_t* leaked,
+                             size_t* total)
+{
+	ScopedLock lock(&alloc_lock);
+	addr_t kmem_from;
+	Memory::GetKernelVirtualArea(&kmem_from, total);
+	*heap = heap_allocated;
+	*aux = aux_allocated - aux_leaked;
+	*leaked = aux_leaked;
+}
+
 // No need for locks in these three functions, since only the heap calls these
 // and it already uses an internal lock, and heap_allocated will not change
 // unless the heap calls ExpandHeap.

kernel/com.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011, 2012, 2014, 2015, 2016 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011, 2012, 2014, 2015, 2016, 2023 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -17,6 +17,8 @@
  * Handles communication to COM serial ports.
  */
 
+#include <sys/ioctl.h>
+
 #include <errno.h>
 #include <stddef.h>
 #include <stdint.h>
@@ -38,6 +40,7 @@
 #include <sortix/kernel/thread.h>
 
 #include "com.h"
+#include "tty.h"
 
 extern "C" unsigned char nullpage[4096];
 
@@ -136,17 +139,30 @@ static inline bool CanWriteByte(uint16_t port)
 	return inport8(port + LSR) & LSR_THRE;
 }
 
-class DevCOMPort : public AbstractInode
+class DevCOMPort : public TTY
 {
 public:
-	DevCOMPort(dev_t dev, uid_t owner, gid_t group, mode_t mode, uint16_t port);
+	DevCOMPort(dev_t dev, uid_t owner, gid_t group, mode_t mode, uint16_t port,
+	           const char* name);
 	virtual ~DevCOMPort();
+	virtual int ioctl(ioctx_t* ctx, int cmd, uintptr_t arg);
 	virtual int sync(ioctx_t* ctx);
-	virtual ssize_t read(ioctx_t* ctx, uint8_t* buf, size_t count);
-	virtual ssize_t write(ioctx_t* ctx, const uint8_t* buf, size_t count);
+	virtual void tty_output(const unsigned char* buffer, size_t length);
+
+public:
+	bool Initialize(int interrupt);
+
+private:
+	static void InterruptHandler(struct interrupt_context*, void*);
+	static void InterruptWorkHandler(void* context);
+	void OnInterrupt();
+	void InterruptWork();
 
 private:
 	kthread_mutex_t port_lock;
+	struct interrupt_handler irq_registration;
+	struct interrupt_work interrupt_work;
+	struct winsize ws;
 	uint16_t port;
 	uint8_t pending_input_byte;
 	bool has_pending_input_byte;
@@ -154,24 +170,93 @@ private:
 };
 
 DevCOMPort::DevCOMPort(dev_t dev, uid_t owner, gid_t group, mode_t mode,
-                       uint16_t port)
+                       uint16_t port, const char* name) : TTY(dev, ino, mode,
+                       owner, group, name)
 {
-	inode_type = INODE_TYPE_STREAM;
 	this->port = port;
 	this->port_lock = KTHREAD_MUTEX_INITIALIZER;
-	this->stat_uid = owner;
-	this->stat_gid = group;
-	this->type = S_IFCHR;
-	this->stat_mode = (mode & S_SETABLE) | this->type;
-	this->dev = dev;
-	this->ino = (ino_t) this;
 	this->has_pending_input_byte = false;
+	interrupt_work.handler = InterruptWorkHandler;
+	interrupt_work.context = this;
 }
 
 DevCOMPort::~DevCOMPort()
 {
 }
 
+bool DevCOMPort::Initialize(int interrupt)
+{
+	uint8_t interrupts = 1;
+	// TODO: This was 9600.
+	tio.c_ispeed = B19200;
+	tio.c_ospeed = B19200;
+	uint16_t divisor = 115200 / tio.c_ispeed;
+	outport8(port + FCR, 0);
+	outport8(port + LCR, LCR_DLAB);
+	outport8(port + DLL, divisor & 0xFF);
+	outport8(port + DLM, divisor >> 8);
+	outport8(port + LCR, LCR_WLEN8); // 8n1
+	outport8(port + MCR, 0x1 /* DTR */ | 0x2 /* RTS */);
+	outport8(port + IER, interrupts);
+	irq_registration.handler = DevCOMPort::InterruptHandler;
+	irq_registration.context = this;
+	Interrupt::RegisterHandler(interrupt, &irq_registration);
+	return true;
+}
+
+void DevCOMPort::InterruptHandler(struct interrupt_context*, void* user)
+{
+	((DevCOMPort*) user)->OnInterrupt();
+}
+
+void DevCOMPort::OnInterrupt()
+{
+	if ( !IsLineReady(port) )
+		return;
+	Interrupt::ScheduleWork(&interrupt_work);
+}
+
+void DevCOMPort::InterruptWorkHandler(void* context)
+{
+	((DevCOMPort*) context)->InterruptWork();
+}
+
+void DevCOMPort::InterruptWork()
+{
+	ScopedLock lock1(&termlock);
+	ScopedLock lock2(&port_lock);
+	while ( IsLineReady(port) )
+	{
+		unsigned char byte = inport8(port + RXR);
+		if ( tio.c_cflag & CREAD )
+			ProcessByte(byte);
+	}
+}
+
+int DevCOMPort::ioctl(ioctx_t* ctx, int cmd, uintptr_t arg)
+{
+	ScopedLock lock(&termlock);
+	if ( hungup )
+		return errno = EIO, -1;
+	if ( cmd == TIOCGWINSZ )
+	{
+		struct winsize* user_ws = (struct winsize*) arg;
+		if ( !ctx->copy_to_dest(user_ws, &ws, sizeof(ws)) )
+			return -1;
+		return 0;
+	}
+	else if ( cmd == TIOCSWINSZ )
+	{
+		const struct winsize* user_ws = (const struct winsize*) arg;
+		if ( !ctx->copy_from_src(&ws, user_ws, sizeof(ws)) )
+			return -1;
+		winch();
+		return 0;
+	}
+	lock.Reset();
+	return TTY::ioctl(ctx, cmd, arg);
+}
+
 int DevCOMPort::sync(ioctx_t* /*ctx*/)
 {
 	ScopedLock lock(&port_lock);
@@ -179,57 +264,9 @@ int DevCOMPort::sync(ioctx_t* /*ctx*/)
 	return 0;
 }
 
-ssize_t DevCOMPort::read(ioctx_t* ctx, uint8_t* dest, size_t count)
+void DevCOMPort::tty_output(const unsigned char* buffer, size_t length)
 {
-	ScopedLock lock(&port_lock);
-
-	for ( size_t i = 0; i < count; i++ )
-	{
-		unsigned long attempt = 0;
-		while ( !has_pending_input_byte && !IsLineReady(port) )
-		{
-			attempt++;
-			if ( attempt <= 10 )
-				continue;
-			if ( attempt <= 15 && !(ctx->dflags & O_NONBLOCK) )
-			{
-				kthread_mutex_unlock(&port_lock);
-				kthread_yield();
-				kthread_mutex_lock(&port_lock);
-				continue;
-			}
-			if ( i )
-				return (ssize_t) i;
-			if ( ctx->dflags & O_NONBLOCK )
-				return errno = EWOULDBLOCK, -1;
-			if ( Signal::IsPending() )
-				return errno = EINTR, -1;
-			kthread_mutex_unlock(&port_lock);
-			kthread_yield();
-			kthread_mutex_lock(&port_lock);
-		}
-
-		uint8_t value = has_pending_input_byte ?
-		                pending_input_byte :
-		                inport8(port + RXR);
-		if ( !ctx->copy_to_dest(dest + i, &value, sizeof(value)) )
-		{
-			has_pending_input_byte = true;
-			pending_input_byte = value;
-			return i ? (ssize_t) i : -1;
-		}
-
-		has_pending_input_byte = false;
-	}
-
-	return (ssize_t) count;
-}
-
-ssize_t DevCOMPort::write(ioctx_t* ctx, const uint8_t* src, size_t count)
-{
-	ScopedLock lock(&port_lock);
-
-	for ( size_t i = 0; i < count; i++ )
+	for ( size_t i = 0; i < length; i++ )
 	{
 		unsigned long attempt = 0;
 		while ( !CanWriteByte(port) )
@@ -237,7 +274,7 @@ ssize_t DevCOMPort::write(ioctx_t* ctx, const uint8_t* src, size_t count)
 			attempt++;
 			if ( attempt <= 10 )
 				continue;
-			if ( attempt <= 15 && !(ctx->dflags & O_NONBLOCK) )
+			if ( attempt <= 15 )
 			{
 				kthread_mutex_unlock(&port_lock);
 				kthread_yield();
@@ -245,20 +282,16 @@ ssize_t DevCOMPort::write(ioctx_t* ctx, const uint8_t* src, size_t count)
 				continue;
 			}
 			if ( i )
-				return (ssize_t) i;
-			if ( ctx->dflags & O_NONBLOCK )
-				return errno = EWOULDBLOCK, -1;
+				return;
+			// TODO: This is problematic.
 			if ( Signal::IsPending() )
-				return errno = EINTR, -1;
+			{
+				errno = EINTR;
+				return;
+			}
 		}
-
-		uint8_t value;
-		if ( !ctx->copy_from_src(&value, src + i, sizeof(value)) )
-			return i ? (ssize_t) i : -1;
-		outport8(port + TXR, value);
+		outport8(port + TXR, buffer[i]);
 	}
-
-	return (ssize_t) count;
 }
 
 static Ref<DevCOMPort> com_devices[1 + NUM_COM_PORTS];
@@ -282,21 +315,6 @@ void Init(const char* devpath, Ref<Descriptor> slashdev)
 
 	ioctx_t ctx; SetupKernelIOCtx(&ctx);
 
-	for ( size_t i = 1; i <= NUM_COM_PORTS; i++ )
-	{
-		uint16_t port = com_ports[i];
-		if ( !port )
-			continue;
-		uint8_t interrupts = 0;
-		outport8(port + FCR, 0);
-		outport8(port + LCR, 0x80);
-		outport8(port + DLL, 0xC);
-		outport8(port + DLM, 0x0);
-		outport8(port + LCR, 0x3); // 8n1
-		outport8(port + MCR, 0x3); // DTR + RTS
-		outport8(port + IER, interrupts);
-	}
-
 	for ( size_t i = 1; i <= NUM_COM_PORTS; i++ )
 	{
 		if ( !com_ports[i] )
@@ -304,13 +322,17 @@ void Init(const char* devpath, Ref<Descriptor> slashdev)
 			com_devices[i] = Ref<DevCOMPort>();
 			continue;
 		}
-		com_devices[i] = Ref<DevCOMPort>(new DevCOMPort(slashdev->dev, 0, 0, 0660, com_ports[i]));
-		if ( !com_devices[i] )
+		char ttyname[TTY_NAME_MAX+1];
+		snprintf(ttyname, sizeof(ttyname), "com%zu", i);
+		Ref<DevCOMPort> com(
+			new DevCOMPort(slashdev->dev, 0, 0, 0660, com_ports[i], ttyname));
+		if ( !com )
 			PanicF("Unable to allocate device for COM port %zu", i);
-		char name[3 + sizeof(size_t) * 3];
-		snprintf(name, sizeof(name), "com%zu", i);
-		if ( LinkInodeInDir(&ctx, slashdev, name, com_devices[i]) != 0 )
-			PanicF("Unable to link %s/%s to COM port driver.", devpath, name);
+		com_devices[i] = com;
+		int interrupt = i == 1 || i == 3 ? Interrupt::IRQ4 : Interrupt::IRQ3;
+		com->Initialize(interrupt);
+		if ( LinkInodeInDir(&ctx, slashdev, ttyname, com) != 0 )
+			PanicF("Unable to link %s/%s to COM port driver.", devpath, ttyname);
 	}
 }
 

kernel/descriptor.cpp

@@ -679,9 +679,9 @@ Ref<Descriptor> Descriptor::open(ioctx_t* ctx, const char* filename, int flags,
                                  mode_t mode)
 {
 	Process* process = CurrentProcess();
-	kthread_mutex_lock(&process->idlock);
+	kthread_mutex_lock(&process->id_lock);
 	mode &= ~process->umask;
-	kthread_mutex_unlock(&process->idlock);
+	kthread_mutex_unlock(&process->id_lock);
 
 	if ( !filename[0] )
 		return errno = ENOENT, Ref<Descriptor>();
@@ -840,9 +840,9 @@ Ref<Descriptor> Descriptor::open_elem(ioctx_t* ctx, const char* filename,
 int Descriptor::mkdir(ioctx_t* ctx, const char* filename, mode_t mode)
 {
 	Process* process = CurrentProcess();
-	kthread_mutex_lock(&process->idlock);
+	kthread_mutex_lock(&process->id_lock);
 	mode &= ~process->umask;
-	kthread_mutex_unlock(&process->idlock);
+	kthread_mutex_unlock(&process->id_lock);
 
 	char* final;
 	Ref<Descriptor> dir = OpenDirContainingPath(ctx, Ref<Descriptor>(this),

kernel/disk/ata/hba.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011, 2012, 2013, 2014, 2015, 2016 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2016, 2022 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -49,11 +49,17 @@ static unsigned long AllocateDiskNumber()
 	return InterlockedIncrement(&next_disk_number).o;
 }
 
-static void sleep_400_nanoseconds()
+static void sleep_400_nanoseconds(uint16_t port_base)
 {
+	// TODO: The clock granularity of 10 ms slows down the early boot.
+#if 0
 	struct timespec delay = timespec_make(0, 400);
 	Clock* clock = Time::GetClock(CLOCK_BOOTTIME);
 	clock->SleepDelay(delay);
+#else
+	for ( int i = 0; i < 14; i++ )
+		inport8(port_base + REG_STATUS);
+#endif
 }
 
 Channel::Channel()
@@ -105,7 +111,7 @@ void Channel::SelectDrive(unsigned int drive_index) // hw_lock locked
 	outport8(port_base + REG_DRIVE_SELECT, value);
 	//outport8(port_control, value); // TODO: Or is it port_control we use?
 
-	sleep_400_nanoseconds();
+	sleep_400_nanoseconds(port_base);
 
 	// TODO: Do we need to wait for non-busy now? Can this operation fail?
 
@@ -229,7 +235,7 @@ bool Channel::Initialize(Ref<Descriptor> dev, const char* devpath)
 
 	busmaster_base = busmasterbar.addr() + 8 * channel_index;
 
-	current_drive = (unsigned int) -1; // We don't know.
+	current_drive = (inport8(port_base + REG_DRIVE_SELECT) >> 4) & 1;
 
 	for ( unsigned int i = 0; i < 2; i++ )
 	{

kernel/disk/ata/port.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011-2016, 2018, 2021 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2016, 2018, 2021, 2022 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -56,11 +56,17 @@ static void copy_ata_string(char* dest, const char* src, size_t length)
 	dest[length] = '\0';
 }
 
-static void sleep_400_nanoseconds()
+static void sleep_400_nanoseconds(uint16_t port_base)
 {
+	// TODO: The clock granularity of 10 ms slows down the early boot.
+#if 0
 	struct timespec delay = timespec_make(0, 400);
 	Clock* clock = Time::GetClock(CLOCK_BOOTTIME);
 	clock->SleepDelay(delay);
+#else
+	for ( int i = 0; i < 14; i++ )
+		inport8(port_base + REG_STATUS);
+#endif
 }
 
 Port::Port(Channel* channel, unsigned int port_index)
@@ -176,7 +182,7 @@ retry_identify_packet:
 	outport8(channel->port_base + REG_COMMAND,
 	         is_packet_interface ? CMD_IDENTIFY_PACKET : CMD_IDENTIFY);
 
-	sleep_400_nanoseconds();
+	sleep_400_nanoseconds(channel->port_base);
 
 	// TODO: The status polling logic should be double-checked against some
 	//       formal specification telling how this should properly be done.

kernel/identity.cpp

@@ -29,14 +29,14 @@ namespace Sortix {
 uid_t sys_getuid()
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	return process->uid;
 }
 
 int sys_setuid(uid_t uid)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	// TODO: Implement security checks in many place across the operating system
 	//       and until then allow anyone to do this to not pretend to be secure.
 	process->uid = uid;
@@ -47,14 +47,14 @@ int sys_setuid(uid_t uid)
 gid_t sys_getgid()
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	return process->gid;
 }
 
 int sys_setgid(gid_t gid)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	// TODO: Implement security checks in many place across the operating system
 	//       and until then allow anyone to do this to not pretend to be secure.
 	process->gid = gid;
@@ -65,14 +65,14 @@ int sys_setgid(gid_t gid)
 uid_t sys_geteuid()
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	return process->euid;
 }
 
 int sys_seteuid(uid_t euid)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	// TODO: Implement security checks in many place across the operating system
 	//       and until then allow anyone to do this to not pretend to be secure.
 	process->euid = euid;
@@ -82,14 +82,14 @@ int sys_seteuid(uid_t euid)
 gid_t sys_getegid()
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	return process->egid;
 }
 
 int sys_setegid(gid_t egid)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	// TODO: Implement security checks in many place across the operating system
 	//       and until then allow anyone to do this to not pretend to be secure.
 	process->egid = egid;

kernel/include/sortix/ioctl.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2016, 2017, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -39,6 +39,7 @@
 #define TIOCGNAME __IOCTL(6, __IOCTL_TYPE_PTR)
 #define TIOCGPTN __IOCTL(7, __IOCTL_TYPE_PTR)
 #define TIOCGDISPLAYS __IOCTL(8, __IOCTL_TYPE_PTR)
+#define TIOCUCTTY __IOCTL(9, __IOCTL_TYPE_INT)
 
 #define IOC_TYPE(x) ((x) >> 0 & 0xFF)
 #define IOC_TYPE_BLOCK_DEVICE 1

kernel/include/sortix/kernel/addralloc.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2013 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2013, 2022 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -39,6 +39,7 @@ void ShrinkHeap(size_t decrease);
 addr_t GetHeapLower();
 addr_t GetHeapUpper();
 size_t GetHeapSize();
+void KernelAddressStatistics(size_t* heap, size_t* aux, size_t* leaked, size_t* total);
 
 } // namespace Sortix
 

kernel/include/sortix/kernel/decl.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011, 2012, 2013, 2014 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011, 2012, 2013, 2014, 2022 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -36,4 +36,35 @@ typedef uintptr_t addr_t;
 	#define CPU X64
 #endif
 
+#ifdef __TRACE_ALLOCATION_SITES
+
+struct kernel_allocation_site
+{
+	struct __allocation_site allocation_site;
+	struct kernel_allocation_site* next;
+	bool registered;
+};
+
+extern struct kernel_allocation_site* first_kernel_allocation_site;
+
+#undef ALLOCATION_SITE
+#define ALLOCATION_SITE \
+({ \
+	static struct kernel_allocation_site site = \
+		{ { __FILE__, __LINE__, __func__, 0, 0 }, NULL, false }; \
+	if ( !site.registered ) \
+	{ \
+		site.registered = true; \
+		site.next = first_kernel_allocation_site; \
+		first_kernel_allocation_site = &site; \
+	} \
+	&site.allocation_site; \
+})
+
+#include <stddef.h>
+void* operator new(size_t size, struct __allocation_site* allocation_site);
+void* operator new[](size_t size, struct __allocation_site* allocation_site);
+#define new new (ALLOCATION_SITE)
+#endif
+
 #endif

kernel/include/sortix/kernel/process.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011-2016, 2021 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2016, 2021, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -66,17 +66,17 @@ public:
 	pid_t pid;
 
 public:
-	kthread_mutex_t nicelock;
+	kthread_mutex_t nice_lock;
 	int nice;
 
 public:
-	kthread_mutex_t idlock;
+	kthread_mutex_t id_lock;
 	uid_t uid, euid;
 	gid_t gid, egid;
 	mode_t umask;
 
 private:
-	kthread_mutex_t ptrlock;
+	kthread_mutex_t ptr_lock;
 	Ref<Descriptor> tty;
 	Ref<Descriptor> root;
 	Ref<Descriptor> cwd;
@@ -112,29 +112,34 @@ public:
 
 public:
 	Process* parent;
-	Process* prevsibling;
-	Process* nextsibling;
-	Process* firstchild;
-	Process* zombiechild;
+	Process* prev_sibling;
+	Process* next_sibling;
+	Process* first_child;
+	Process* zombie_child;
 	Process* group;
-	Process* groupprev;
-	Process* groupnext;
-	Process* groupfirst;
+	Process* group_prev;
+	Process* group_next;
+	Process* group_first;
 	Process* session;
-	Process* sessionprev;
-	Process* sessionnext;
-	Process* sessionfirst;
-	kthread_mutex_t childlock;
-	kthread_mutex_t parentlock;
-	kthread_cond_t zombiecond;
-	bool iszombie;
-	bool nozombify;
+	Process* session_prev;
+	Process* session_next;
+	Process* session_first;
+	Process* init;
+	Process* init_prev;
+	Process* init_next;
+	Process* init_first;
+	kthread_mutex_t child_lock;
+	kthread_mutex_t parent_lock;
+	kthread_cond_t zombie_cond;
+	bool is_zombie;
+	bool no_zombify;
 	bool limbo;
+	bool is_init_exiting;
 	int exit_code;
 
 public:
-	Thread* firstthread;
-	kthread_mutex_t threadlock;
+	Thread* first_thread;
+	kthread_mutex_t thread_lock;
 	size_t threads_not_exiting_count;
 	bool threads_exiting;
 
@@ -178,6 +183,7 @@ public:
 	                int prot);
 	void GroupRemoveMember(Process* child);
 	void SessionRemoveMember(Process* child);
+	void InitRemoveMember(Process* child);
 
 public:
 	Process* Fork();

kernel/include/sortix/kernel/scheduler.h

@@ -41,8 +41,6 @@ void SetThreadState(Thread* thread, ThreadState state, bool wake_only = false);
 void SetSignalPending(Thread* thread, unsigned long is_pending);
 ThreadState GetThreadState(Thread* thread);
 void SetIdleThread(Thread* thread);
-void SetInitProcess(Process* init);
-Process* GetInitProcess();
 Process* GetKernelProcess();
 void InterruptYieldCPU(struct interrupt_context* intctx, void* user);
 void ThreadExitCPU(struct interrupt_context* intctx, void* user);

kernel/include/sortix/kernel/syscall.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011-2016, 2021, 2022, 2023 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2016, 2021-2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -97,6 +97,7 @@ int sys_getentropy(void*, size_t);
 uid_t sys_geteuid(void);
 gid_t sys_getgid(void);
 int sys_gethostname(char*, size_t);
+pid_t sys_getinit(pid_t);
 size_t sys_getpagesize(void);
 int sys_getpeername(int, void*, size_t*);
 pid_t sys_getpgid(pid_t);
@@ -151,6 +152,7 @@ int sys_setegid(gid_t);
 int sys_seteuid(uid_t);
 int sys_setgid(gid_t);
 int sys_sethostname(const char*, size_t);
+int sys_setinit(void);
 int sys_setpgid(pid_t, pid_t);
 int sys_setpriority(int, id_t, int);
 pid_t sys_setsid(void);

kernel/include/sortix/kernel/thread.h

@@ -79,8 +79,8 @@ public:
 	struct thread_registers registers;
 	size_t id;
 	Process* process;
-	Thread* prevsibling;
-	Thread* nextsibling;
+	Thread* prev_sibling;
+	Thread* next_sibling;
 	Thread* scheduler_list_prev;
 	Thread* scheduler_list_next;
 	volatile ThreadState state;
@@ -88,12 +88,12 @@ public:
 	sigset_t signal_mask;
 	sigset_t saved_signal_mask;
 	stack_t signal_stack;
-	addr_t kernelstackpos;
-	size_t kernelstacksize;
+	addr_t kernel_stack_pos;
+	size_t kernel_stack_size;
 	size_t signal_count;
 	uintptr_t signal_single_frame;
 	uintptr_t signal_canary;
-	bool kernelstackmalloced;
+	bool kernel_stack_malloced;
 	bool pledged_destruction;
 	bool force_no_signals;
 	bool signal_single;

kernel/include/sortix/syscall.h

@@ -190,6 +190,8 @@
 #define SYSCALL_SETDNSCONFIG 167
 #define SYSCALL_FUTEX 168
 #define SYSCALL_MEMUSAGE 169
-#define SYSCALL_MAX_NUM 170 /* index of highest constant + 1 */
+#define SYSCALL_GETINIT 170
+#define SYSCALL_SETINIT 171
+#define SYSCALL_MAX_NUM 172 /* index of highest constant + 1 */
 
 #endif

kernel/ioctx.cpp

@@ -27,7 +27,7 @@ namespace Sortix {
 void SetupUserIOCtx(ioctx_t* ctx)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	ctx->uid = ctx->auth_uid = process->uid;
 	ctx->gid = ctx->auth_gid = process->gid;
 	ctx->copy_to_dest = CopyToUser;
@@ -39,7 +39,7 @@ void SetupUserIOCtx(ioctx_t* ctx)
 void SetupKernelIOCtx(ioctx_t* ctx)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	ctx->uid = ctx->auth_uid = process->uid;
 	ctx->gid = ctx->auth_gid = process->gid;
 	ctx->copy_to_dest = CopyToKernel;

kernel/kernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011-2018, 2021-2022 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2018, 2021-2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -103,6 +103,8 @@
 #include "x86-family/vbox.h"
 #endif
 
+struct kernel_allocation_site* first_kernel_allocation_site = NULL;
+
 // Keep the stack size aligned with $CPU/boot.s
 const size_t STACK_SIZE = 64*1024;
 extern "C" { __attribute__((aligned(16))) size_t stack[STACK_SIZE / sizeof(size_t)]; }
@@ -231,6 +233,13 @@ extern "C" void KernelInit(unsigned long magic, multiboot_info_t* bootinfo_p)
 
 	if ( !(kernel_options = strdup(cmdline ? cmdline : "")) )
 		Panic("Failed to allocate kernel command line");
+#if defined(__i386__) || defined(__x86_64__)
+	// TODO: Detect EFI.
+	kernel_firmware = "bios";
+#else
+	#warning "Name your system firmware here"
+	kernel_firmware = "unknown";
+#endif
 
 	int argmax = 1;
 	argv = new char*[argmax + 1];
@@ -286,6 +295,20 @@ extern "C" void KernelInit(unsigned long magic, multiboot_info_t* bootinfo_p)
 			enable_network_drivers = true;
 		else if ( !strcmp(arg, "--no-random-seed") )
 			no_random_seed = true;
+		else if ( !strncmp(arg, "--firmware=", strlen("--firmware=")) )
+		{
+			const char* firmware = arg + strlen("--firmware=");
+#if defined(__i386__) || defined(__x86_64__)
+			if ( !strcmp(firmware, "bios") || !strcmp(firmware, "pc") )
+				kernel_firmware = "bios";
+			else if ( !strcmp(firmware, "efi") )
+				kernel_firmware = "efi";
+			else
+#endif
+			{
+				PanicF("Unsupported firmware option: %s", firmware);
+			}
+		}
 		else
 		{
 			Log::PrintF("\r\e[J");
@@ -351,13 +374,17 @@ extern "C" void KernelInit(unsigned long magic, multiboot_info_t* bootinfo_p)
 	ptable.Reset();
 	system->addrspace = Memory::GetAddressSpace();
 	system->group = system;
-	system->groupprev = NULL;
-	system->groupnext = NULL;
-	system->groupfirst = system;
+	system->group_prev = NULL;
+	system->group_next = NULL;
+	system->group_first = system;
 	system->session = system;
-	system->sessionprev = NULL;
-	system->sessionnext = NULL;
-	system->sessionfirst = system;
+	system->session_prev = NULL;
+	system->session_next = NULL;
+	system->session_first = system;
+	system->init = NULL;
+	system->init_prev = NULL;
+	system->init_next = NULL;
+	system->init_first = NULL;
 
 	if ( !(system->program_image_path = String::Clone("<kernel process>")) )
 		Panic("Unable to clone string for system process name");
@@ -369,10 +396,10 @@ extern "C" void KernelInit(unsigned long magic, multiboot_info_t* bootinfo_p)
 	Thread* idlethread = new Thread();
 	idlethread->name = "idle";
 	idlethread->process = system;
-	idlethread->kernelstackpos = (addr_t) stack;
-	idlethread->kernelstacksize = STACK_SIZE;
-	idlethread->kernelstackmalloced = false;
-	system->firstthread = idlethread;
+	idlethread->kernel_stack_pos = (addr_t) stack;
+	idlethread->kernel_stack_size = STACK_SIZE;
+	idlethread->kernel_stack_malloced = false;
+	system->first_thread = idlethread;
 	system->threads_not_exiting_count = 1;
 	Scheduler::SetIdleThread(idlethread);
 
@@ -640,19 +667,23 @@ static void BootThread(void* /*user*/)
 	kthread_mutex_lock(&process_family_lock);
 	Process* kernel_process = CurrentProcess();
 	init->parent = kernel_process;
-	init->nextsibling = kernel_process->firstchild;
-	init->prevsibling = NULL;
-	if ( kernel_process->firstchild )
-		kernel_process->firstchild->prevsibling = init;
-	kernel_process->firstchild = init;
+	init->next_sibling = kernel_process->first_child;
+	init->prev_sibling = NULL;
+	if ( kernel_process->first_child )
+		kernel_process->first_child->prev_sibling = init;
+	kernel_process->first_child = init;
 	init->group = init;
-	init->groupprev = NULL;
-	init->groupnext = NULL;
-	init->groupfirst = init;
+	init->group_prev = NULL;
+	init->group_next = NULL;
+	init->group_first = init;
 	init->session = init;
-	init->sessionprev = NULL;
-	init->sessionnext = NULL;
-	init->sessionfirst = init;
+	init->session_prev = NULL;
+	init->session_next = NULL;
+	init->session_first = init;
+	init->init = init;
+	init->init_prev = NULL;
+	init->init_next = NULL;
+	init->init_first = init;
 	kthread_mutex_unlock(&process_family_lock);
 
 	// TODO: Why don't we fork from pid=0 and this is done for us?
@@ -662,7 +693,6 @@ static void BootThread(void* /*user*/)
 	mtable.Reset();
 	init->BootstrapDirectories(droot);
 	init->addrspace = initaddrspace;
-	Scheduler::SetInitProcess(init);
 
 	Thread* initthread = RunKernelThread(init, InitThread, NULL, "main");
 	if ( !initthread )

kernel/kernelinfo.cpp

@@ -25,6 +25,7 @@
 #include <sortix/kernel/syscall.h>
 
 #include "kernelinfo.h"
+#include "net/tcp.h"
 
 #ifndef VERSIONSTR
 #define VERSIONSTR "unknown"
@@ -33,20 +34,17 @@
 namespace Sortix {
 
 char* kernel_options;
+const char* kernel_firmware;
 
 static const char* KernelInfo(const char* req)
 {
-	if ( strcmp(req, "name") == 0 ) { return BRAND_KERNEL_NAME; }
-	if ( strcmp(req, "version") == 0 ) { return VERSIONSTR; }
-	if ( strcmp(req, "tagline") == 0 ) { return BRAND_RELEASE_TAGLINE; }
-	if ( strcmp(req, "options") == 0 ) { return kernel_options; }
-	if ( strcmp(req, "builddate") == 0 ) { return __DATE__; }
-	if ( strcmp(req, "buildtime") == 0 ) { return __TIME__; }
-#if defined(__i386__) || defined(__x86_64__)
-	if ( strcmp(req, "firmware") == 0 ) { return "bios"; }
-#else
-	#warning "Name your system firmware here"
-#endif
+	if ( strcmp(req, "name") == 0 ) return BRAND_KERNEL_NAME;
+	if ( strcmp(req, "version") == 0 ) return VERSIONSTR;
+	if ( strcmp(req, "tagline") == 0 ) return BRAND_RELEASE_TAGLINE;
+	if ( strcmp(req, "options") == 0 ) return kernel_options;
+	if ( strcmp(req, "builddate") == 0 ) return __DATE__;
+	if ( strcmp(req, "buildtime") == 0 ) return __TIME__;
+	if ( strcmp(req, "firmware") == 0 ) return kernel_firmware;
 	return NULL;
 }
 
@@ -55,6 +53,77 @@ ssize_t sys_kernelinfo(const char* user_req, char* user_resp, size_t resplen)
 	char* req = GetStringFromUser(user_req);
 	if ( !req )
 		return -1;
+	// DEBUG
+	if ( !strcmp(req, "virtual-usage") )
+	{
+		delete[] req;
+		size_t heap;
+		size_t aux;
+		size_t leaked;
+		size_t total;
+		KernelAddressStatistics(&heap, &aux, &leaked, &total);
+		char str[4 * (20 + 3 * sizeof(size_t) + 1 + 8)];
+		snprintf(str, sizeof(str),
+		         "%20zu B heap\n%20zu B aux\n%20zu B leaked\n%20zu B total",
+		         heap, aux, leaked, total);
+		size_t stringlen = strlen(str);
+		if ( resplen < stringlen + 1 )
+			return errno = ERANGE, (ssize_t) stringlen;
+		if ( !CopyToUser(user_resp, str, sizeof(char) * (stringlen + 1)) )
+			return -1;
+		return 0;
+	}
+	// DEBUG
+	if ( !strcmp(req, "tcp") )
+	{
+		delete[] req;
+		return TCP::Info(user_resp, resplen);
+	}
+#ifdef __TRACE_ALLOCATION_SITES
+	if ( !strcmp(req, "allocations") )
+	{
+		delete[] req;
+		bool exhausted = false;
+		size_t total_needed = 0;
+		for ( struct kernel_allocation_site* site = first_kernel_allocation_site;
+			  site;
+			  site = site->next )
+		{
+			char str[256];
+			snprintf(str, sizeof(str), "%20zu B %zu allocations %s:%zu %s %c",
+			         site->allocation_site.current_size,
+			         site->allocation_site.allocations,
+			         site->allocation_site.file,
+			         site->allocation_site.line,
+			         site->allocation_site.function,
+			         site->next ? '\n' : '\0');
+			size_t stringlen = strlen(str);
+			total_needed += stringlen;
+			if ( exhausted )
+				continue;
+			if ( resplen < stringlen )
+			{
+				exhausted = true;
+				continue;
+			}
+			if ( !CopyToUser(user_resp, str, sizeof(char) * stringlen) )
+				return -1;
+			user_resp += stringlen;
+			resplen -= stringlen;
+		}
+		if ( !exhausted && !resplen )
+			exhausted = true;
+		if ( !exhausted )
+		{
+			char zero = '\0';
+			if ( !CopyToUser(user_resp, &zero, 1) )
+				return -1;
+		}
+		if ( exhausted )
+			return errno = ERANGE, (ssize_t) total_needed;
+		return 0;
+	}
+#endif
 	const char* str = KernelInfo(req);
 	delete[] req;
 	if ( !str )

kernel/kernelinfo.h

@@ -23,6 +23,7 @@
 namespace Sortix {
 
 extern char* kernel_options;
+extern const char* kernel_firmware;
 
 } // namespace Sortix
 

kernel/kthread.cpp

@@ -258,9 +258,9 @@ void kthread_exit()
 	// only threads in this process, except the initial thread. Otherwise more
 	// threads may appear, and we can't conclude whether this is the last thread
 	// in the process to exit.
-	kthread_mutex_lock(&process->threadlock);
+	kthread_mutex_lock(&process->thread_lock);
 	bool is_last_to_exit = --process->threads_not_exiting_count == 0;
-	kthread_mutex_unlock(&process->threadlock);
+	kthread_mutex_unlock(&process->thread_lock);
 	// All other threads in the process have committed to exiting, though they
 	// might not have exited yet. However, we know they are only running the
 	// below code that schedules thread termination. It's therefore safe to run

kernel/linebuffer.cpp

@@ -50,6 +50,7 @@ LineBuffer::~LineBuffer()
 
 bool LineBuffer::Push(uint32_t unicode)
 {
+	// TODO: An infinite line buffer seems like a bad idea.
 	// Check if we need to allocate or resize the circular queue.
 	if ( bufferused == bufferlength )
 	{

kernel/net/fs.cpp

@@ -52,13 +52,11 @@ namespace NetFS {
 class Manager;
 class StreamSocket;
 
-class Manager : public AbstractInode
+class Manager : public Refcountable
 {
 public:
-	Manager(uid_t owner, gid_t group, mode_t mode);
-	virtual ~Manager() { }
-	virtual Ref<Inode> open(ioctx_t* ctx, const char* filename, int flags,
-	                        mode_t mode);
+	Manager();
+	virtual ~Manager();
 
 public:
 	bool Bind(StreamSocket* socket, struct sockaddr_un* addr, size_t addrsize);
@@ -114,6 +112,7 @@ public:
 public: /* For use by Manager. */
 	PollChannel accept_poll_channel;
 	Ref<Manager> manager;
+	Ref<Descriptor> root;
 	PipeEndpoint incoming;
 	PipeEndpoint outgoing;
 	StreamSocket* prev_socket;
@@ -189,6 +188,7 @@ StreamSocket::StreamSocket(uid_t owner, gid_t group, mode_t mode,
 	this->is_connected = false;
 	this->is_refused = false;
 	this->manager = manager;
+	this->root = CurrentProcess()->GetRoot();
 	this->socket_lock = KTHREAD_MUTEX_INITIALIZER;
 	this->pending_cond = KTHREAD_COND_INITIALIZER;
 	this->accepted_cond = KTHREAD_COND_INITIALIZER;
@@ -476,20 +476,19 @@ int StreamSocket::getsockname(ioctx_t* ctx, uint8_t* addr, size_t* addrsize)
 	return 0;
 }
 
-Manager::Manager(uid_t owner, gid_t group, mode_t mode)
+Manager::Manager()
 {
-	inode_type = INODE_TYPE_UNKNOWN;
-	dev = (dev_t) this;
-	ino = 0;
-	this->type = S_IFDIR;
-	this->stat_uid = owner;
-	this->stat_gid = group;
-	this->stat_mode = (mode & S_SETABLE) | this->type;
 	this->manager_lock = KTHREAD_MUTEX_INITIALIZER;
 	this->first_server = NULL;
 	this->last_server = NULL;
 }
 
+Manager::~Manager()
+{
+	assert(!first_server);
+	assert(!last_server);
+}
+
 static int CompareAddress(const struct sockaddr_un* a,
                           const struct sockaddr_un* b)
 {
@@ -498,8 +497,10 @@ static int CompareAddress(const struct sockaddr_un* a,
 
 StreamSocket* Manager::LookupServer(struct sockaddr_un* address)
 {
+	Ref<Descriptor> root = CurrentProcess()->GetRoot();
 	for ( StreamSocket* iter = first_server; iter; iter = iter->next_socket )
-		if ( CompareAddress(iter->name, address) == 0 )
+		if ( CompareAddress(iter->name, address) == 0 &&
+		     iter->root->dev == root->dev && iter->root->ino == root->ino )
 			return iter;
 	return NULL;
 }
@@ -664,24 +665,11 @@ bool Manager::Connect(StreamSocket* socket,
 	return true;
 }
 
-// TODO: Support a poll method in Manager.
-
-Ref<Inode> Manager::open(ioctx_t* /*ctx*/, const char* filename,
-                             int /*flags*/, mode_t /*mode*/)
-{
-	if ( !strcmp(filename, "stream") )
-	{
-		StreamSocket* socket = new StreamSocket(0, 0, 0666, Ref<Manager>(this));
-		return Ref<StreamSocket>(socket);
-	}
-	return errno = ENOENT, Ref<Inode>(NULL);
-}
-
 static Ref<Manager> manager;
 
 void Init()
 {
-	manager = Ref<Manager>(new Manager(0, 0, 0600));
+	manager = Ref<Manager>(new Manager());
 }
 
 Ref<Inode> Socket(int type, int protocol)

kernel/net/ping.cpp

@@ -202,10 +202,10 @@ PingSocket::PingSocket(int af)
 	dev = (dev_t) this;
 	ino = (ino_t) this;
 	type = S_IFSOCK;
-	kthread_mutex_lock(&process->idlock);
+	kthread_mutex_lock(&process->id_lock);
 	stat_uid = process->uid;
 	stat_gid = process->gid;
-	kthread_mutex_unlock(&process->idlock);
+	kthread_mutex_unlock(&process->id_lock);
 	stat_mode = 0600 | this->type;
 	supports_iovec = true;
 	socket_lock = KTHREAD_MUTEX_INITIALIZER;

kernel/net/tcp.cpp

@@ -36,6 +36,7 @@
 #include <sys/socket.h>
 #include <sys/stat.h>
 
+#include <arpa/inet.h>
 #include <assert.h>
 #include <endian.h>
 #include <errno.h>
@@ -120,6 +121,9 @@ static kthread_mutex_t tcp_lock = KTHREAD_MUTEX_INITIALIZER;
 static TCPSocket** bindings_v4;
 static TCPSocket** bindings_v6;
 
+static TCPSocket* all_first_socket;
+static TCPSocket* all_last_socket;
+
 void Init()
 {
 	if ( !(bindings_v4 = new TCPSocket*[65536]) ||
@@ -218,6 +222,7 @@ public:
 	int getsockname(ioctx_t* ctx, uint8_t* addr, size_t* addrsize);
 
 public:
+	size_t Describe(char* buf, size_t buflen);
 	void Unreference();
 	void ProcessPacket(Ref<Packet> pkt, union tcp_sockaddr* pkt_src,
 	                   union tcp_sockaddr* pkt_dst);
@@ -278,6 +283,12 @@ public:
 	// The listening socket this socket is in the listening queue for.
 	TCPSocket* connecting_parent;
 
+	// DEBUG: The previous socket of all sockets.
+	TCPSocket* all_prev_socket;
+
+	// DEBUG: The next socket of all sockets.
+	TCPSocket* all_next_socket;
+
 	// Condition variable that is signaled when new data can be received.
 	kthread_cond_t receive_cond;
 
@@ -461,7 +472,7 @@ void TCPSocket__OnTimer(Clock* /*clock*/, Timer* /*timer*/, void* user)
 	((TCPSocket*) user)->OnTimer();
 }
 
-TCPSocket::TCPSocket(int af)
+TCPSocket::TCPSocket(int af) // DEBUG: tcp_lock taken
 {
 	prev_socket = NULL;
 	next_socket = NULL;
@@ -517,9 +528,15 @@ TCPSocket::TCPSocket(int af)
 	shutdown_receive = false;
 	memset(incoming, 0, sizeof(incoming));
 	memset(outgoing, 0, sizeof(outgoing));
+	// DEBUG
+	all_prev_socket = all_last_socket;
+	all_next_socket = NULL;
+	(all_last_socket ?
+	 all_last_socket->all_next_socket : all_first_socket) = this;
+	all_last_socket = this;
 }
 
-TCPSocket::~TCPSocket()
+TCPSocket::~TCPSocket() // DEBUG: tcp_lock taken
 {
 	assert(state == TCP_STATE_CLOSED);
 	assert(!bound);
@@ -539,6 +556,52 @@ TCPSocket::~TCPSocket()
 		receive_queue = packet->next;
 		packet->next.Reset();
 	}
+	// DEBUG
+	(all_prev_socket ?
+	 all_prev_socket->all_next_socket : all_first_socket) = all_next_socket;
+	(all_next_socket ?
+	 all_next_socket->all_prev_socket : all_last_socket) = all_prev_socket;
+	all_prev_socket = NULL;
+	all_next_socket = NULL;
+}
+
+// DEBUG
+size_t TCPSocket::Describe(char* buf, size_t buflen) // tcp_lock taken
+{
+	const char* const STATE_NAMES[] =
+	{
+		"CLOSED",
+		"LISTEN",
+		"SYN_SENT",
+		"SYN_RECV",
+		"ESTAB",
+		"FIN_WAIT_1",
+		"FIN_WAIT_2",
+		"CLOSE_WAIT",
+		"CLOSING",
+		"LAST_ACK",
+		"TIME_WAIT",
+	};
+	const char* state_name = STATE_NAMES[state];
+	char local_str[INET_ADDRSTRLEN];
+	char remote_str[INET_ADDRSTRLEN];
+	inet_ntop(AF_INET, &local.in.sin_addr, local_str, sizeof(local_str));
+	inet_ntop(AF_INET, &remote.in.sin_addr, remote_str, sizeof(remote_str));
+	char timeout[64] = "none";
+	if ( timer_armed )
+	{
+		struct itimerspec its;
+		timer.Get(&its);
+		snprintf(timeout, sizeof(timeout), "%ji.%09li",
+		         (intmax_t) its.it_value.tv_sec, its.it_value.tv_nsec);
+	}
+	return snprintf(buf, buflen,
+	                "%s %s %u -> %s %u"
+	                " timeout=%s resends=%u sockerr=%i transmit=%i refed=%i\n",
+	                state_name, local_str, be16toh(local.in.sin_port),
+	                remote_str, be16toh(remote.in.sin_port), timeout,
+	                retransmissions, sockerr, transmit_scheduled,
+	                is_referenced);
 }
 
 void TCPSocket::Unreference()
@@ -2354,10 +2417,10 @@ TCPSocketNode::TCPSocketNode(TCPSocket* socket)
 	dev = (dev_t) this;
 	ino = (ino_t) this;
 	type = S_IFSOCK;
-	kthread_mutex_lock(&process->idlock);
+	kthread_mutex_lock(&process->id_lock);
 	stat_uid = process->uid;
 	stat_gid = process->gid;
-	kthread_mutex_unlock(&process->idlock);
+	kthread_mutex_unlock(&process->id_lock);
 	stat_mode = 0600 | this->type;
 }
 
@@ -2558,6 +2621,7 @@ Ref<Inode> Socket(int af)
 {
 	if ( !IsSupportedAddressFamily(af) )
 		return errno = EAFNOSUPPORT, Ref<Inode>(NULL);
+	ScopedLock lock(&tcp_lock); // DEBUG
 	TCPSocket* socket = new TCPSocket(af);
 	if ( !socket )
 		return Ref<Inode>();
@@ -2567,5 +2631,45 @@ Ref<Inode> Socket(int af)
 	return result;
 }
 
+// DEBUG
+ssize_t Info(char* user_resp, size_t resplen)
+{
+	ScopedLock lock(&tcp_lock); // DEBUG
+	bool exhausted = false;
+	size_t total_needed = 0;
+	for ( TCPSocket* socket = all_first_socket;
+	      socket;
+	      socket = socket->all_next_socket )
+	{
+		char str[256];
+		size_t stringlen = socket->Describe(str, sizeof(str));
+		if ( !socket->all_next_socket && stringlen )
+			stringlen--;
+		total_needed += stringlen;
+		if ( exhausted )
+			continue;
+		if ( resplen < stringlen )
+		{
+			exhausted = true;
+			continue;
+		}
+		if ( !CopyToUser(user_resp, str, sizeof(char) * stringlen) )
+			return -1;
+		user_resp += stringlen;
+		resplen -= stringlen;
+	}
+	if ( !exhausted && !resplen )
+		exhausted = true;
+	if ( !exhausted )
+	{
+		char zero = '\0';
+		if ( !CopyToUser(user_resp, &zero, 1) )
+			return -1;
+	}
+	if ( exhausted )
+		return errno = ERANGE, (ssize_t) total_needed;
+	return 0;
+}
+
 } // namespace TCP
 } // namespace Sortix

kernel/net/tcp.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2016, 2017, 2022 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -34,6 +34,7 @@ void HandleIP(Ref<Packet> pkt,
               const struct in_addr* dst,
               bool dst_broadcast);
 Ref<Inode> Socket(int af);
+ssize_t Info(char* user_resp, size_t resplen);
 
 } // namespace TCP
 } // namespace Sortix

kernel/net/udp.cpp

@@ -175,10 +175,10 @@ UDPSocket::UDPSocket(int af)
 	dev = (dev_t) this;
 	ino = (ino_t) this;
 	type = S_IFSOCK;
-	kthread_mutex_lock(&process->idlock);
+	kthread_mutex_lock(&process->id_lock);
 	stat_uid = process->uid;
 	stat_gid = process->gid;
-	kthread_mutex_unlock(&process->idlock);
+	kthread_mutex_unlock(&process->id_lock);
 	stat_mode = 0600 | this->type;
 	supports_iovec = true;
 	socket_lock = KTHREAD_MUTEX_INITIALIZER;

kernel/op-new.cpp

@@ -24,6 +24,19 @@
 #warning "security: -fcheck-new might not work on clang"
 #endif
 
+#ifdef __TRACE_ALLOCATION_SITES
+#undef new
+
+void* operator new(size_t size, struct __allocation_site* allocation_site)
+{
+	return malloc_trace(allocation_site, size);
+}
+
+void* operator new[](size_t size, struct __allocation_site* allocation_site)
+{
+	return malloc_trace(allocation_site, size);
+}
+#else
 void* operator new(size_t size)
 {
 	return malloc(size);
@@ -33,6 +46,7 @@ void* operator new[](size_t size)
 {
 	return malloc(size);
 }
+#endif
 
 void operator delete(void* addr)
 {

kernel/poll.cpp

@@ -309,6 +309,7 @@ int sys_ppoll(struct pollfd* user_fds, size_t nfds,
 	{
 		if ( !kthread_cond_wait_signal(&wakeup_cond, &wakeup_mutex) )
 		{
+			ret = -1;
 			errno = -EINTR;
 			self_woken = true;
 			deliver_signal = true;
@@ -327,7 +328,7 @@ int sys_ppoll(struct pollfd* user_fds, size_t nfds,
 		timer.Detach();
 	}
 
-	if ( !unexpected_error )
+	if ( !deliver_signal && !unexpected_error )
 	{
 		int num_events = 0;
 		for ( size_t i = 0; i < reqs; i++ )
@@ -346,6 +347,9 @@ int sys_ppoll(struct pollfd* user_fds, size_t nfds,
 	if ( 0 <= ret )
 		errno = 0;
 
+	if ( Signal::IsPending() )
+		ret = -1, errno = EINTR, deliver_signal = true;
+
 	if ( user_sigmask )
 	{
 		if ( !deliver_signal )

kernel/process.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011-2016, 2021-2022 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2016, 2021-2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -73,23 +73,21 @@ kthread_mutex_t process_family_lock = KTHREAD_MUTEX_INITIALIZER;
 
 // The system is shutting down and creation of additional processes and threads
 // should be prevented. Protected by process_family_lock.
-static bool is_init_exiting = false;
-
 Process::Process()
 {
 	program_image_path = NULL;
 	addrspace = 0;
 	pid = 0;
 
-	nicelock = KTHREAD_MUTEX_INITIALIZER;
+	nice_lock = KTHREAD_MUTEX_INITIALIZER;
 	nice = 0;
 
-	idlock = KTHREAD_MUTEX_INITIALIZER;
+	id_lock = KTHREAD_MUTEX_INITIALIZER;
 	uid = euid = 0;
 	gid = egid = 0;
 	umask = 0022;
 
-	ptrlock = KTHREAD_MUTEX_INITIALIZER;
+	ptr_lock = KTHREAD_MUTEX_INITIALIZER;
 	// tty set to null reference in the member constructor.
 	// root set to null reference in the member constructor.
 	// cwd set to null reference in the member constructor.
@@ -118,26 +116,31 @@ Process::Process()
 	sigreturn = NULL;
 
 	parent = NULL;
-	prevsibling = NULL;
-	nextsibling = NULL;
-	firstchild = NULL;
-	zombiechild = NULL;
+	prev_sibling = NULL;
+	next_sibling = NULL;
+	first_child = NULL;
+	zombie_child = NULL;
 	group = NULL;
-	groupprev = NULL;
-	groupnext = NULL;
-	groupfirst = NULL;
+	group_prev = NULL;
+	group_next = NULL;
+	group_first = NULL;
 	session = NULL;
-	sessionprev = NULL;
-	sessionnext = NULL;
-	sessionfirst = NULL;
-	zombiecond = KTHREAD_COND_INITIALIZER;
-	iszombie = false;
-	nozombify = false;
+	session_prev = NULL;
+	session_next = NULL;
+	session_first = NULL;
+	init = NULL;
+	init_prev = NULL;
+	init_next = NULL;
+	init_first = NULL;
+	zombie_cond = KTHREAD_COND_INITIALIZER;
+	is_zombie = false;
+	no_zombify = false;
 	limbo = false;
+	is_init_exiting = false;
 	exit_code = -1;
 
-	firstthread = NULL;
-	threadlock = KTHREAD_MUTEX_INITIALIZER;
+	first_thread = NULL;
+	thread_lock = KTHREAD_MUTEX_INITIALIZER;
 	threads_not_exiting_count = 0;
 	threads_exiting = false;
 
@@ -167,13 +170,15 @@ Process::~Process() // process_family_lock taken
 	if ( alarm_timer.IsAttached() )
 		alarm_timer.Detach();
 	delete[] program_image_path;
-	assert(!zombiechild);
+	assert(!zombie_child);
+	assert(!init);
 	assert(!session);
 	assert(!group);
 	assert(!parent);
-	assert(!sessionfirst);
-	assert(!groupfirst);
-	assert(!firstchild);
+	assert(!init_first);
+	assert(!session_first);
+	assert(!group_first);
+	assert(!first_child);
 	assert(!addrspace);
 	assert(!segments);
 	assert(!dtable);
@@ -190,7 +195,7 @@ Process::~Process() // process_family_lock taken
 
 void Process::BootstrapTables(Ref<DescriptorTable> dtable, Ref<MountTable> mtable)
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(!this->dtable);
 	assert(!this->mtable);
 	this->dtable = dtable;
@@ -199,7 +204,7 @@ void Process::BootstrapTables(Ref<DescriptorTable> dtable, Ref<MountTable> mtabl
 
 void Process::BootstrapDirectories(Ref<Descriptor> root)
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(!this->root);
 	assert(!this->cwd);
 	this->root = root;
@@ -208,9 +213,6 @@ void Process::BootstrapDirectories(Ref<Descriptor> root)
 
 void Process::OnLastThreadExit()
 {
-	Process* init = Scheduler::GetInitProcess();
-	assert(init);
-
 	// Child processes can't be reparented away if we're init. The system is
 	// about to shut down, so broadcast SIGKILL every process and wait for every
 	// single process to exit. The operating system is finished when init has
@@ -221,36 +223,43 @@ void Process::OnLastThreadExit()
 		// Forbid any more processes and threads from being created, so this
 		// loop will always terminate.
 		is_init_exiting = true;
-		kthread_mutex_lock(&ptrlock);
-		for ( pid_t pid = ptable->Next(0); 0 < pid; pid = ptable->Next(pid) )
+		Process* process = first_child;
+		while ( process )
 		{
-			Process* process = ptable->Get(pid);
-			if ( process->pid != 0 && process != init )
+			if ( process->pid != 0 )
 				process->DeliverSignal(SIGKILL);
+			if ( process->init == process )
+				process->is_init_exiting = true;
+			if ( process->first_child )
+				process = process->first_child;
+			while ( process && process != this && !process->next_sibling )
+				process = process->parent;
+			if ( process == this )
+				break;
+			process = process->next_sibling;
 		}
-		kthread_mutex_unlock(&ptrlock);
-		// NotifyChildExit always signals zombiecond for init when
+		// NotifyChildExit always signals zombie_cond for init when
 		// is_init_exiting is true.
-		while ( firstchild )
-			kthread_cond_wait(&zombiecond, &process_family_lock);
+		while ( first_child )
+			kthread_cond_wait(&zombie_cond, &process_family_lock);
 	}
 }
 
 void Process::OnThreadDestruction(Thread* thread)
 {
 	assert(thread->process == this);
-	kthread_mutex_lock(&threadlock);
-	if ( thread->prevsibling )
-		thread->prevsibling->nextsibling = thread->nextsibling;
-	if ( thread->nextsibling )
-		thread->nextsibling->prevsibling = thread->prevsibling;
-	if ( thread == firstthread )
-		firstthread = thread->nextsibling;
-	if ( firstthread )
-		firstthread->prevsibling = NULL;
-	thread->prevsibling = thread->nextsibling = NULL;
-	bool threadsleft = firstthread;
-	kthread_mutex_unlock(&threadlock);
+	kthread_mutex_lock(&thread_lock);
+	if ( thread->prev_sibling )
+		thread->prev_sibling->next_sibling = thread->next_sibling;
+	if ( thread->next_sibling )
+		thread->next_sibling->prev_sibling = thread->prev_sibling;
+	if ( thread == first_thread )
+		first_thread = thread->next_sibling;
+	if ( first_thread )
+		first_thread->prev_sibling = NULL;
+	thread->prev_sibling = thread->next_sibling = NULL;
+	bool threadsleft = first_thread;
+	kthread_mutex_unlock(&thread_lock);
 
 	// We are called from the threads destructor, let it finish before we
 	// we handle the situation by killing ourselves.
@@ -266,7 +275,7 @@ void Process__AfterLastThreadExit(void* user)
 void Process::ScheduleDeath()
 {
 	// All our threads must have exited at this point.
-	assert(!firstthread);
+	assert(!first_thread);
 	Worker::Schedule(Process__AfterLastThreadExit, this);
 }
 
@@ -275,7 +284,7 @@ void Process::ScheduleDeath()
 // process after this call as another thread may garbage collect it.
 void Process::AbortConstruction()
 {
-	nozombify = true;
+	no_zombify = true;
 	ScheduleDeath();
 }
 
@@ -302,7 +311,7 @@ void Process::LastPrayer()
 {
 	assert(this);
 	// This must never be called twice.
-	assert(!iszombie);
+	assert(!is_zombie);
 
 	// This must be called from a thread using another address space as the
 	// address space of this process is about to be destroyed.
@@ -310,7 +319,7 @@ void Process::LastPrayer()
 	assert(curthread->process != this);
 
 	// This can't be called if the process is still alive.
-	assert(!firstthread);
+	assert(!first_thread);
 
 	// Disarm and detach all the timers in the process.
 	DeleteTimers();
@@ -339,45 +348,43 @@ void Process::LastPrayer()
 
 	ScopedLock family_lock(&process_family_lock);
 
-	iszombie = true;
+	is_zombie = true;
 
 	// Init is nice and will gladly raise our orphaned children and zombies.
-	Process* init = Scheduler::GetInitProcess();
-	assert(init);
-
 	// Child processes can't be reparented away if we're init. OnLastThreadExit
 	// must have already killed all the child processes and prevented more from
 	// being created.
+	assert(init);
 	if ( init == this )
 	{
 		assert(is_init_exiting);
-		assert(!firstchild);
+		assert(!first_child);
 	}
 
-	while ( firstchild )
+	while ( first_child )
 	{
-		Process* process = firstchild;
-		firstchild = process->nextsibling;
+		Process* process = first_child;
+		first_child = process->next_sibling;
 		process->parent = init;
-		process->prevsibling = NULL;
-		process->nextsibling = init->firstchild;
-		if ( init->firstchild )
-			init->firstchild->prevsibling = process;
-		init->firstchild = process;
-		process->nozombify = true;
+		process->prev_sibling = NULL;
+		process->next_sibling = init->first_child;
+		if ( init->first_child )
+			init->first_child->prev_sibling = process;
+		init->first_child = process;
+		process->no_zombify = true;
 	}
 	// Since we have no more children (they are with init now), we don't
 	// have to worry about new zombie processes showing up, so just collect
-	// those that are left. Then we satisfiy the invariant !zombiechild that
+	// those that are left. Then we satisfiy the invariant !zombie_child that
 	// applies on process termination.
-	while ( zombiechild )
+	while ( zombie_child )
 	{
-		Process* zombie = zombiechild;
-		zombiechild = zombie->nextsibling;
-		zombie->nextsibling = NULL;
-		if ( zombiechild )
-			zombiechild->prevsibling = NULL;
-		zombie->nozombify = true;
+		Process* zombie = zombie_child;
+		zombie_child = zombie->next_sibling;
+		zombie->next_sibling = NULL;
+		if ( zombie_child )
+			zombie_child->prev_sibling = NULL;
+		zombie->no_zombify = true;
 		zombie->WaitedFor();
 	}
 	// Remove ourself from our process group.
@@ -386,8 +393,11 @@ void Process::LastPrayer()
 	// Remove ourself from our session.
 	if ( session )
 		session->SessionRemoveMember(this);
+	// Remove ourself from our init.
+	if ( init )
+		init->InitRemoveMember(this);
 
-	bool zombify = !nozombify;
+	bool zombify = !no_zombify;
 
 	// This class instance will be destroyed by our parent process when it
 	// has received and acknowledged our death.
@@ -403,7 +413,7 @@ void Process::WaitedFor() // process_family_lock taken
 {
 	parent = NULL;
 	limbo = false;
-	if ( groupfirst || sessionfirst )
+	if ( group_first || session_first || init_first )
 		limbo = true;
 	if ( !limbo )
 		delete this;
@@ -429,12 +439,12 @@ void Process::ResetAddressSpace()
 void Process::GroupRemoveMember(Process* child) // process_family_lock taken
 {
 	assert(child->group == this);
-	if ( child->groupprev )
-		child->groupprev->groupnext = child->groupnext;
+	if ( child->group_prev )
+		child->group_prev->group_next = child->group_next;
 	else
-		groupfirst = child->groupnext;
-	if ( child->groupnext )
-		child->groupnext->groupprev = child->groupprev;
+		group_first = child->group_next;
+	if ( child->group_next )
+		child->group_next->group_prev = child->group_prev;
 	child->group = NULL;
 	if ( IsLimboDone() )
 		delete this;
@@ -443,47 +453,61 @@ void Process::GroupRemoveMember(Process* child) // process_family_lock taken
 void Process::SessionRemoveMember(Process* child) // process_family_lock taken
 {
 	assert(child->session == this);
-	if ( child->sessionprev )
-		child->sessionprev->sessionnext = child->sessionnext;
+	if ( child->session_prev )
+		child->session_prev->session_next = child->session_next;
 	else
-		sessionfirst = child->sessionnext;
-	if ( child->sessionnext )
-		child->sessionnext->sessionprev = child->sessionprev;
+		session_first = child->session_next;
+	if ( child->session_next )
+		child->session_next->session_prev = child->session_prev;
 	child->session = NULL;
-	if ( !sessionfirst )
+	if ( !session_first )
 	{
 		// Remove reference to tty when session is empty.
-		ScopedLock lock(&ptrlock);
+		ScopedLock lock(&ptr_lock);
 		tty.Reset();
 	}
 	if ( IsLimboDone() )
 		delete this;
 }
 
+void Process::InitRemoveMember(Process* child) // process_family_lock taken
+{
+	assert(child->init == this);
+	if ( child->init_prev )
+		child->init_prev->init_next = child->init_next;
+	else
+		init_first = child->init_next;
+	if ( child->init_next )
+		child->init_next->init_prev = child->init_prev;
+	child->init = NULL;
+	if ( IsLimboDone() )
+		delete this;
+}
+
 bool Process::IsLimboDone() // process_family_lock taken
 {
-	return limbo && !groupfirst && !sessionfirst;
+	return limbo && !group_first && !session_first && !init_first;
 }
 
 // process_family_lock taken
 void Process::NotifyChildExit(Process* child, bool zombify)
 {
-	if ( child->prevsibling )
-		child->prevsibling->nextsibling = child->nextsibling;
-	if ( child->nextsibling )
-		child->nextsibling->prevsibling = child->prevsibling;
-	if ( firstchild == child )
-		firstchild = child->nextsibling;
-	if ( firstchild )
-		firstchild->prevsibling = NULL;
+	if ( child->prev_sibling )
+		child->prev_sibling->next_sibling = child->next_sibling;
+	if ( child->next_sibling )
+		child->next_sibling->prev_sibling = child->prev_sibling;
+	if ( first_child == child )
+		first_child = child->next_sibling;
+	if ( first_child )
+		first_child->prev_sibling = NULL;
 
 	if ( zombify )
 	{
-		if ( zombiechild )
-			zombiechild->prevsibling = child;
-		child->prevsibling = NULL;
-		child->nextsibling = zombiechild;
-		zombiechild = child;
+		if ( zombie_child )
+			zombie_child->prev_sibling = child;
+		child->prev_sibling = NULL;
+		child->next_sibling = zombie_child;
+		zombie_child = child;
 	}
 
 	// Notify this parent process about the child exiting if it's meant to
@@ -491,8 +515,8 @@ void Process::NotifyChildExit(Process* child, bool zombify)
 	// when init is exiting, because OnLastThreadExit needs to be able to catch
 	// every child exiting.
 	DeliverSignal(SIGCHLD);
-	if ( zombify || (is_init_exiting && Scheduler::GetInitProcess() == this) )
-		kthread_cond_broadcast(&zombiecond);
+	if ( zombify || is_init_exiting )
+		kthread_cond_broadcast(&zombie_cond);
 }
 
 pid_t Process::Wait(pid_t thepid, int* status_ptr, int options)
@@ -504,7 +528,7 @@ pid_t Process::Wait(pid_t thepid, int* status_ptr, int options)
 	ScopedLock lock(&process_family_lock);
 
 	// A process can only wait if it has children.
-	if ( !firstchild && !zombiechild )
+	if ( !first_child && !zombie_child )
 		return errno = ECHILD, -1;
 
 	// Processes can only wait for their own children to exit.
@@ -513,11 +537,11 @@ pid_t Process::Wait(pid_t thepid, int* status_ptr, int options)
 		// TODO: This is a slow but multithread safe way to verify that the
 		// target process has the correct parent.
 		bool found = false;
-		for ( Process* p = firstchild; !found && p; p = p->nextsibling )
-			if ( p->pid == thepid && !p->nozombify )
+		for ( Process* p = first_child; !found && p; p = p->next_sibling )
+			if ( p->pid == thepid && !p->no_zombify )
 				found = true;
-		for ( Process* p = zombiechild; !found && p; p = p->nextsibling )
-			if ( p->pid == thepid && !p->nozombify )
+		for ( Process* p = zombie_child; !found && p; p = p->next_sibling )
+			if ( p->pid == thepid && !p->no_zombify )
 				found = true;
 		if ( !found )
 			return errno = ECHILD, -1;
@@ -526,26 +550,26 @@ pid_t Process::Wait(pid_t thepid, int* status_ptr, int options)
 	Process* zombie = NULL;
 	while ( !zombie )
 	{
-		for ( zombie = zombiechild; zombie; zombie = zombie->nextsibling )
-			if ( (thepid == -1 || thepid == zombie->pid) && !zombie->nozombify )
+		for ( zombie = zombie_child; zombie; zombie = zombie->next_sibling )
+			if ( (thepid == -1 || thepid == zombie->pid) && !zombie->no_zombify )
 				break;
 		if ( zombie )
 			break;
 		if ( options & WNOHANG )
 			return 0;
-		if ( !kthread_cond_wait_signal(&zombiecond, &process_family_lock) )
+		if ( !kthread_cond_wait_signal(&zombie_cond, &process_family_lock) )
 			return errno = EINTR, -1;
 	}
 
 	// Remove from the list of zombies.
-	if ( zombie->prevsibling )
-		zombie->prevsibling->nextsibling = zombie->nextsibling;
-	if ( zombie->nextsibling )
-		zombie->nextsibling->prevsibling = zombie->prevsibling;
-	if ( zombiechild == zombie )
-		zombiechild = zombie->nextsibling;
-	if ( zombiechild )
-		zombiechild->prevsibling = NULL;
+	if ( zombie->prev_sibling )
+		zombie->prev_sibling->next_sibling = zombie->next_sibling;
+	if ( zombie->next_sibling )
+		zombie->next_sibling->prev_sibling = zombie->prev_sibling;
+	if ( zombie_child == zombie )
+		zombie_child = zombie->next_sibling;
+	if ( zombie_child )
+		zombie_child->prev_sibling = NULL;
 
 	thepid = zombie->pid;
 
@@ -584,7 +608,7 @@ void Process::ExitThroughSignal(int signal)
 
 void Process::ExitWithCode(int requested_exit_code)
 {
-	ScopedLock lock(&threadlock);
+	ScopedLock lock(&thread_lock);
 	if ( exit_code == -1 )
 		exit_code = requested_exit_code;
 
@@ -592,74 +616,74 @@ void Process::ExitWithCode(int requested_exit_code)
 	// of process termination. We simply can't stop the threads as they may
 	// be running in kernel mode doing dangerous stuff. This thread will be
 	// destroyed by SIGKILL once the system call returns.
-	for ( Thread* t = firstthread; t; t = t->nextsibling )
+	for ( Thread* t = first_thread; t; t = t->next_sibling )
 		t->DeliverSignal(SIGKILL);
 }
 
 Ref<MountTable> Process::GetMTable()
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(mtable);
 	return mtable;
 }
 
 Ref<DescriptorTable> Process::GetDTable()
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(dtable);
 	return dtable;
 }
 
 Ref<ProcessTable> Process::GetPTable()
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(ptable);
 	return ptable;
 }
 
 Ref<Descriptor> Process::GetTTY()
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	return tty;
 }
 
 Ref<Descriptor> Process::GetRoot()
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(root);
 	return root;
 }
 
 Ref<Descriptor> Process::GetCWD()
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(cwd);
 	return cwd;
 }
 
 void Process::SetTTY(Ref<Descriptor> newtty)
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	tty = newtty;
 }
 
 void Process::SetRoot(Ref<Descriptor> newroot)
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(newroot);
 	root = newroot;
 }
 
 void Process::SetCWD(Ref<Descriptor> newcwd)
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(newcwd);
 	cwd = newcwd;
 }
 
 Ref<Descriptor> Process::GetDescriptor(int fd)
 {
-	ScopedLock lock(&ptrlock);
+	ScopedLock lock(&ptr_lock);
 	assert(dtable);
 	return dtable->Get(fd);
 }
@@ -704,7 +728,7 @@ Process* Process::Fork()
 	kthread_mutex_lock(&process_family_lock);
 
 	// Forbid the creation of new processes if init has exited.
-	if ( is_init_exiting )
+	if ( init->is_init_exiting )
 	{
 		kthread_mutex_unlock(&process_family_lock);
 		clone->AbortConstruction();
@@ -721,25 +745,32 @@ Process* Process::Fork()
 
 	// Remember the relation to the child process.
 	clone->parent = this;
-	clone->nextsibling = firstchild;
-	clone->prevsibling = NULL;
-	if ( firstchild )
-		firstchild->prevsibling = clone;
-	firstchild = clone;
+	clone->next_sibling = first_child;
+	clone->prev_sibling = NULL;
+	if ( first_child )
+		first_child->prev_sibling = clone;
+	first_child = clone;
 
 	// Add the new process to the current process group.
 	clone->group = group;
-	clone->groupprev = NULL;
-	if ( (clone->groupnext = group->groupfirst) )
-		group->groupfirst->groupprev = clone;
-	group->groupfirst = clone;
+	clone->group_prev = NULL;
+	if ( (clone->group_next = group->group_first) )
+		group->group_first->group_prev = clone;
+	group->group_first = clone;
 
 	// Add the new process to the current session.
 	clone->session = session;
-	clone->sessionprev = NULL;
-	if ( (clone->sessionnext = session->sessionfirst) )
-		session->sessionfirst->sessionprev = clone;
-	session->sessionfirst = clone;
+	clone->session_prev = NULL;
+	if ( (clone->session_next = session->session_first) )
+		session->session_first->session_prev = clone;
+	session->session_first = clone;
+
+	// Add the new process to the current init.
+	clone->init = init;
+	clone->init_prev = NULL;
+	if ( (clone->init_next = init->init_first) )
+		init->init_first->init_prev = clone;
+	init->init_first = clone;
 
 	kthread_mutex_unlock(&process_family_lock);
 
@@ -749,22 +780,22 @@ Process* Process::Fork()
 		clone->resource_limits[i] = resource_limits[i];
 	kthread_mutex_unlock(&resource_limits_lock);
 
-	kthread_mutex_lock(&nicelock);
+	kthread_mutex_lock(&nice_lock);
 	clone->nice = nice;
-	kthread_mutex_unlock(&nicelock);
+	kthread_mutex_unlock(&nice_lock);
 
-	kthread_mutex_lock(&ptrlock);
+	kthread_mutex_lock(&ptr_lock);
 	clone->root = root;
 	clone->cwd = cwd;
-	kthread_mutex_unlock(&ptrlock);
+	kthread_mutex_unlock(&ptr_lock);
 
-	kthread_mutex_lock(&idlock);
+	kthread_mutex_lock(&id_lock);
 	clone->uid = uid;
 	clone->gid = gid;
 	clone->euid = euid;
 	clone->egid = egid;
 	clone->umask = umask;
-	kthread_mutex_unlock(&idlock);
+	kthread_mutex_unlock(&id_lock);
 
 	kthread_mutex_lock(&signal_lock);
 	memcpy(&clone->signal_actions, &signal_actions, sizeof(signal_actions));
@@ -775,13 +806,13 @@ Process* Process::Fork()
 	// Initialize things that can fail and abort if needed.
 	bool failure = false;
 
-	kthread_mutex_lock(&ptrlock);
+	kthread_mutex_lock(&ptr_lock);
 	if ( !(clone->dtable = dtable->Fork()) )
 		failure = true;
 	//if ( !(clone->mtable = mtable->Fork()) )
 	//	failure = true;
 	clone->mtable = mtable;
-	kthread_mutex_unlock(&ptrlock);
+	kthread_mutex_unlock(&ptr_lock);
 
 	if ( !(clone->program_image_path = String::Clone(program_image_path)) )
 		failure = true;
@@ -1475,26 +1506,24 @@ pid_t sys_tfork(int flags, struct tfork* user_regs)
 
 	// TODO: Is it a hack to create a new kernel stack here?
 	Thread* curthread = CurrentThread();
-	size_t newkernelstacksize = curthread->kernelstacksize;
-	uint8_t* newkernelstack = new uint8_t[newkernelstacksize + stack_alignment];
+	size_t newkernel_stack_size = curthread->kernel_stack_size;
+	uint8_t* newkernelstack = new uint8_t[newkernel_stack_size + stack_alignment];
 	if ( !newkernelstack )
 		return -1;
 
 	uintptr_t stack_aligned = (uintptr_t) newkernelstack;
-	size_t stack_aligned_size = newkernelstacksize;
+	size_t stack_aligned_size = newkernel_stack_size;
 
 	if ( ((uintptr_t) stack_aligned) & (stack_alignment-1) )
 		stack_aligned = (stack_aligned + 16) & ~(stack_alignment-1);
 	stack_aligned_size &= 0xFFFFFFF0;
 
+	Process* parent_process = CurrentProcess();
 	Process* child_process;
 	if ( making_thread )
-		child_process = CurrentProcess();
-	else if ( !(child_process = CurrentProcess()->Fork()) )
-	{
-		delete[] newkernelstack;
-		return -1;
-	}
+		child_process = parent_process;
+	else if ( !(child_process = parent_process->Fork()) )
+		return delete[] newkernelstack, -1;
 
 	struct thread_registers cpuregs;
 	memset(&cpuregs, 0, sizeof(cpuregs));
@@ -1550,7 +1579,7 @@ pid_t sys_tfork(int flags, struct tfork* user_regs)
 
 	// Forbid the creation of new threads if init has exited.
 	ScopedLock process_family_lock_lock(&process_family_lock);
-	if ( is_init_exiting )
+	if ( child_process->init->is_init_exiting )
 		return errno = EPERM, -1;
 
 	// If the thread could not be created, make the process commit suicide
@@ -1565,9 +1594,9 @@ pid_t sys_tfork(int flags, struct tfork* user_regs)
 		return -1;
 	}
 
-	thread->kernelstackpos = (addr_t) newkernelstack;
-	thread->kernelstacksize = newkernelstacksize;
-	thread->kernelstackmalloced = true;
+	thread->kernel_stack_pos = (addr_t) newkernelstack;
+	thread->kernel_stack_size = newkernel_stack_size;
+	thread->kernel_stack_malloced = true;
 	memcpy(&thread->signal_mask, &regs.sigmask, sizeof(sigset_t));
 	memcpy(&thread->signal_stack, &regs.altstack, sizeof(stack_t));
 
@@ -1604,7 +1633,8 @@ pid_t sys_getpgid(pid_t pid)
 pid_t sys_getsid(pid_t pid)
 {
 	ScopedLock lock(&process_family_lock);
-	Process* process = !pid ? CurrentProcess() : CurrentProcess()->GetPTable()->Get(pid);
+	Process* process =
+		!pid ? CurrentProcess() : CurrentProcess()->GetPTable()->Get(pid);
 	if ( !process )
 		return errno = ESRCH, -1;
 	if ( !process->session )
@@ -1612,6 +1642,16 @@ pid_t sys_getsid(pid_t pid)
 	return process->session->pid;
 }
 
+pid_t sys_getinit(pid_t pid)
+{
+	ScopedLock lock(&process_family_lock);
+	Process* process =
+		!pid ? CurrentProcess() : CurrentProcess()->GetPTable()->Get(pid);
+	if ( !process->init )
+		return errno = ESRCH, -1;
+	return process->init->pid;
+}
+
 int sys_setpgid(pid_t pid, pid_t pgid)
 {
 	if ( pid < 0 || pgid < 0 )
@@ -1649,13 +1689,13 @@ int sys_setpgid(pid_t pid, pid_t pgid)
 		return errno = EPERM, -1;
 	// The process must not be a process group leader.
 	// TODO: Maybe POSIX actually allows this.
-	if ( process->groupfirst )
+	if ( process->group_first )
 		return errno = EPERM, -1;
 	// The process must not be a session leader.
-	if ( process->sessionfirst )
+	if ( process->session_first )
 		return errno = EPERM, -1;
 	// The group must either exist or be the process itself.
-	if ( !group->groupfirst && group != process )
+	if ( !group->group_first && group != process )
 		return errno = EPERM, -1;
 
 	// Exit early if this is a noop.
@@ -1667,11 +1707,11 @@ int sys_setpgid(pid_t pid, pid_t pgid)
 		process->group->GroupRemoveMember(process);
 
 	// Insert the process into its new process group.
-	process->groupprev = NULL;
-	process->groupnext = group->groupfirst;
-	if ( group->groupfirst )
-		group->groupfirst->groupprev = process;
-	group->groupfirst = process;
+	process->group_prev = NULL;
+	process->group_next = group->group_first;
+	if ( group->group_first )
+		group->group_first->group_prev = process;
+	group->group_first = process;
 	process->group = group;
 
 	return 0;
@@ -1696,15 +1736,58 @@ pid_t sys_setsid(void)
 		process->session->SessionRemoveMember(process);
 
 	// Insert the process into its new session.
-	process->sessionprev = NULL;
-	process->sessionnext = NULL;
-	process->sessionfirst = process;
+	process->session_prev = NULL;
+	process->session_next = NULL;
+	process->session_first = process;
 	process->session = process;
 
 	// Insert the process into its new process group.
-	process->groupprev = NULL;
-	process->groupnext = NULL;
-	process->groupfirst = process;
+	process->group_prev = NULL;
+	process->group_next = NULL;
+	process->group_first = process;
+	process->group = process;
+
+	return process->pid;
+}
+
+int sys_setinit(void)
+{
+	Process* process = CurrentProcess();
+
+	ScopedLock lock(&process_family_lock);
+
+	// Test if already a process group leader.
+	if ( process->group == process )
+		return errno = EPERM, -1;
+
+	// Remove the process from its current process group.
+	if ( process->group )
+		process->group->GroupRemoveMember(process);
+
+	// Remove the process from its current session.
+	if ( process->session )
+		process->session->SessionRemoveMember(process);
+
+	// Remove the process from its current init.
+	if ( process->init )
+		process->init->InitRemoveMember(process);
+
+	// Insert the process into its new init.
+	process->init_prev = NULL;
+	process->init_next = NULL;
+	process->init_first = process;
+	process->init = process;
+
+	// Insert the process into its new session.
+	process->session_prev = NULL;
+	process->session_next = NULL;
+	process->session_first = process;
+	process->session = process;
+
+	// Insert the process into its new process group.
+	process->group_prev = NULL;
+	process->group_next = NULL;
+	process->group_first = process;
 	process->group = process;
 
 	return process->pid;
@@ -1718,7 +1801,7 @@ size_t sys_getpagesize(void)
 mode_t sys_umask(mode_t newmask)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	mode_t oldmask = process->umask;
 	process->umask = newmask & 0666;
 	return oldmask;
@@ -1727,7 +1810,7 @@ mode_t sys_umask(mode_t newmask)
 mode_t sys_getumask(void)
 {
 	Process* process = CurrentProcess();
-	ScopedLock lock(&process->idlock);
+	ScopedLock lock(&process->id_lock);
 	return process->umask;
 }
 

kernel/psctl.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015, 2016, 2022 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2015, 2016, 2022, 2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -72,8 +72,8 @@ int sys_psctl(pid_t pid, int request, void* ptr)
 		{
 			Process* parent = process->parent;
 			psst.ppid = parent->pid;
-			psst.ppid_prev = process->prevsibling ? process->prevsibling->pid : -1;
-			psst.ppid_next = process->nextsibling ? process->nextsibling->pid : -1;
+			psst.ppid_prev = process->prev_sibling ? process->prev_sibling->pid : -1;
+			psst.ppid_next = process->next_sibling ? process->next_sibling->pid : -1;
 		}
 		else
 		{
@@ -81,13 +81,13 @@ int sys_psctl(pid_t pid, int request, void* ptr)
 			psst.ppid_prev = -1;
 			psst.ppid_next = -1;
 		}
-		psst.ppid_first = process->firstchild ? process->firstchild->pid : -1;
+		psst.ppid_first = process->first_child ? process->first_child->pid : -1;
 		if ( process->group )
 		{
 			Process* group = process->group;
 			psst.pgid = group->pid;
-			psst.pgid_prev = process->groupprev ? process->groupprev->pid : -1;
-			psst.pgid_next = process->groupnext ? process->groupnext->pid : -1;
+			psst.pgid_prev = process->group_prev ? process->group_prev->pid : -1;
+			psst.pgid_next = process->group_next ? process->group_next->pid : -1;
 		}
 		else
 		{
@@ -95,13 +95,13 @@ int sys_psctl(pid_t pid, int request, void* ptr)
 			psst.pgid_prev = -1;
 			psst.pgid_next = -1;
 		}
-		psst.pgid_first = process->groupfirst ? process->groupfirst->pid : -1;
+		psst.pgid_first = process->group_first ? process->group_first->pid : -1;
 		if ( process->session )
 		{
 			Process* session = process->session;
 			psst.sid = session->pid;
-			psst.sid_prev = process->sessionprev ? process->sessionprev->pid : -1;
-			psst.sid_next = process->sessionnext ? process->sessionnext->pid : -1;
+			psst.sid_prev = process->session_prev ? process->session_prev->pid : -1;
+			psst.sid_next = process->session_next ? process->session_next->pid : -1;
 		}
 		else
 		{
@@ -109,24 +109,34 @@ int sys_psctl(pid_t pid, int request, void* ptr)
 			psst.sid_prev = -1;
 			psst.sid_next = -1;
 		}
-		psst.sid_first = process->sessionfirst ? process->sessionfirst->pid : -1;
-		// TODO: Implement init groupings.
-		psst.init = 1;
-		psst.init_prev = ptable->Prev(pid);
-		psst.init_next = ptable->Next(pid);
-		psst.init_first = pid == 1 ? 1 : -1;
-		kthread_mutex_lock(&process->idlock);
+		psst.sid_first = process->session_first ? process->session_first->pid : -1;
+
+		if ( process->init )
+		{
+			Process* init = process->init;
+			psst.init = init->pid;
+			psst.init_prev = process->init_prev ? process->init_prev->pid : -1;
+			psst.init_next = process->init_next ? process->init_next->pid : -1;
+		}
+		else
+		{
+			psst.init = -1;
+			psst.init_prev = -1;
+			psst.init_next = -1;
+		}
+		psst.init_first = process->init_first ? process->init_first->pid : -1;
+		kthread_mutex_lock(&process->id_lock);
 		psst.uid = process->uid;
 		psst.euid = process->euid;
 		psst.gid = process->gid;
 		psst.egid = process->egid;
-		kthread_mutex_unlock(&process->idlock);
-		kthread_mutex_lock(&process->threadlock);
+		kthread_mutex_unlock(&process->id_lock);
+		kthread_mutex_lock(&process->thread_lock);
 		psst.status = process->exit_code;
-		kthread_mutex_unlock(&process->threadlock);
-		kthread_mutex_lock(&process->nicelock);
+		kthread_mutex_unlock(&process->thread_lock);
+		kthread_mutex_lock(&process->nice_lock);
 		psst.nice = process->nice;
-		kthread_mutex_unlock(&process->nicelock);
+		kthread_mutex_unlock(&process->nice_lock);
 		kthread_mutex_lock(&process->segment_lock);
 		// TODO: Cache these.
 		for ( size_t i = 0; i < process->segments_used; i++ )

kernel/pty.cpp

@@ -477,6 +477,8 @@ private:
 	size_t output_offset;
 	size_t output_used;
 	static const size_t output_size = 4096;
+	// TODO: This is not safe because ^W can produce unbounded output.
+	static const size_t output_probably_safe = 2048;
 	uint8_t output[output_size];
 	int ptynum;
 
@@ -538,6 +540,16 @@ ssize_t PTY::master_write(ioctx_t* ctx, const uint8_t* buf, size_t count)
 	ScopedLockSignal lock(&termlock);
 	if ( !lock.IsAcquired() )
 		return errno = EINTR, -1;
+	// TODO: Work around deadlock by refusing writes when the buffer is starting
+	//       to be too full. This can block / "deadlock" too if the caller
+	//       didn't try to read the data written by a previous call.
+	while ( output_probably_safe <= output_used )
+	{
+		if ( ctx->dflags & O_NONBLOCK )
+			return errno = EWOULDBLOCK, -1;
+		if ( !kthread_cond_wait_signal(&output_possible_cond, &termlock) )
+			return errno = EINTR, -1;
+	}
 	size_t sofar = 0;
 	while ( sofar < count )
 	{
@@ -550,9 +562,13 @@ ssize_t PTY::master_write(ioctx_t* ctx, const uint8_t* buf, size_t count)
 		{
 			if ( Signal::IsPending() )
 				return sofar ? (ssize_t) sofar : (errno = EINTR, -1);
+			// TODO: Unfortunately sequences like ^W can cause an unbounded
+			//       number of tty_output data causing a deadlock.
 			ProcessByte(input[i]);
+			sofar++;
+			if ( output_probably_safe <= output_used )
+				return (ssize_t) sofar;
 		}
-		sofar += amount;
 	}
 	return (ssize_t) sofar;
 }
@@ -599,7 +615,7 @@ short PTY::PollMasterEventStatus()
 	short status = 0;
 	if ( output_used )
 		status |= POLLIN | POLLRDNORM;
-	if ( true /* can always write */ )
+	if ( output_used < output_probably_safe )
 		status |= POLLOUT | POLLWRNORM;
 	return status;
 }
@@ -656,6 +672,7 @@ int PTY::master_ioctl(ioctx_t* ctx, int cmd, uintptr_t arg)
 		const struct winsize* user_ws = (const struct winsize*) arg;
 		if ( !ctx->copy_from_src(&ws, user_ws, sizeof(ws)) )
 			return -1;
+		winch();
 		return 0;
 	}
 	return ioctl(ctx, cmd, arg);

kernel/resource.cpp

@@ -39,7 +39,7 @@ static int GetProcessPriority(pid_t who)
 	Process* process = who ? CurrentProcess()->GetPTable()->Get(who) : CurrentProcess();
 	if ( !process )
 		return errno = ESRCH, -1;
-	ScopedLock lock(&process->nicelock);
+	ScopedLock lock(&process->nice_lock);
 	return process->nice;
 }
 
@@ -50,7 +50,7 @@ static int SetProcessPriority(pid_t who, int prio)
 	Process* process = who ? CurrentProcess()->GetPTable()->Get(who) : CurrentProcess();
 	if ( !process )
 		return errno = ESRCH, -1;
-	ScopedLock lock(&process->nicelock);
+	ScopedLock lock(&process->nice_lock);
 	process->nice = prio;
 	return 0;
 }
@@ -68,9 +68,9 @@ static int GetProcessGroupPriority(pid_t who)
 	if ( !group )
 		return errno = ESRCH, -1;
 	int lowest = INT_MAX;
-	for ( Process* process = group->groupfirst; process; process = process->groupnext )
+	for ( Process* process = group->group_first; process; process = process->group_next )
 	{
-		ScopedLock lock(&process->nicelock);
+		ScopedLock lock(&process->nice_lock);
 		if ( process->nice < lowest )
 			lowest = process->nice;
 	}
@@ -84,9 +84,9 @@ static int SetProcessGroupPriority(pid_t who, int prio)
 	Process* group = who ? CurrentProcess()->GetPTable()->Get(who) : CurrentProcessGroup();
 	if ( !group )
 		return errno = ESRCH, -1;
-	for ( Process* process = group->groupfirst; process; process = process->groupnext )
+	for ( Process* process = group->group_first; process; process = process->group_next )
 	{
-		ScopedLock lock(&process->nicelock);
+		ScopedLock lock(&process->nice_lock);
 		process->nice = prio;
 	}
 	return 0;

kernel/scheduler.cpp

@@ -181,8 +181,8 @@ extern "C" void fake_interrupt(void);
 static void FakeInterruptedContext(struct interrupt_context* intctx, int int_no)
 {
 #if defined(__i386__)
-	uintptr_t stack = current_thread->kernelstackpos +
-	                  current_thread->kernelstacksize;
+	uintptr_t stack = current_thread->kernel_stack_pos +
+	                  current_thread->kernel_stack_size;
 	stack -= sizeof(struct interrupt_context);
 	struct interrupt_context* fakectx = (struct interrupt_context*) stack;
 	memcpy(fakectx, intctx, sizeof(struct interrupt_context));
@@ -209,8 +209,8 @@ static void FakeInterruptedContext(struct interrupt_context* intctx, int int_no)
 	intctx->esp = stack;
 	intctx->ss = KDS | KRPL;
 #elif defined(__x86_64__)
-	uintptr_t stack = current_thread->kernelstackpos +
-	                  current_thread->kernelstacksize;
+	uintptr_t stack = current_thread->kernel_stack_pos +
+	                  current_thread->kernel_stack_size;
 	stack -= sizeof(struct interrupt_context);
 	struct interrupt_context* fakectx = (struct interrupt_context*) stack;
 	memcpy(fakectx, intctx, sizeof(struct interrupt_context));
@@ -268,7 +268,6 @@ static void SwitchRegisters(struct interrupt_context* intctx,
 static Thread* idle_thread;
 static Thread* first_runnable_thread;
 static Thread* true_current_thread;
-static Process* init_process;
 
 static void SwitchThread(struct interrupt_context* intctx,
                          Thread* old_thread,
@@ -407,16 +406,6 @@ void SetIdleThread(Thread* thread)
 	true_current_thread = thread;
 }
 
-void SetInitProcess(Process* init)
-{
-	init_process = init;
-}
-
-Process* GetInitProcess()
-{
-	return init_process;
-}
-
 Process* GetKernelProcess()
 {
 	if ( !idle_thread )

kernel/signal.cpp

@@ -144,8 +144,8 @@ int sys_sigaction(int signum,
 		memcpy(kact, &newact, sizeof(struct sigaction));
 
 		// Signals may become discarded because of the new handler.
-		ScopedLock threads_lock(&process->threadlock);
-		for ( Thread* t = process->firstthread; t; t = t->nextsibling )
+		ScopedLock threads_lock(&process->thread_lock);
+		for ( Thread* t = process->first_thread; t; t = t->next_sibling )
 			UpdatePendingSignals(t);
 	}
 
@@ -315,9 +315,9 @@ int sys_kill(pid_t pid, int signum)
 
 bool Process::DeliverGroupSignal(int signum) // process_family_lock held
 {
-	if ( !groupfirst )
+	if ( !group_first )
 		return errno = ESRCH, false;
-	for ( Process* iter = groupfirst; iter; iter = iter->groupnext )
+	for ( Process* iter = group_first; iter; iter = iter->group_next )
 	{
 		int saved_errno = errno;
 		if ( !iter->DeliverSignal(signum) && errno != ESIGPENDING )
@@ -331,9 +331,9 @@ bool Process::DeliverGroupSignal(int signum) // process_family_lock held
 
 bool Process::DeliverSessionSignal(int signum) // process_family_lock held
 {
-	if ( !sessionfirst )
+	if ( !session_first )
 		return errno = ESRCH, false;
-	for ( Process* iter = sessionfirst; iter; iter = iter->sessionnext )
+	for ( Process* iter = session_first; iter; iter = iter->session_next )
 	{
 		int saved_errno = errno;
 		if ( !iter->DeliverSignal(signum) && errno != ESIGPENDING )
@@ -347,16 +347,16 @@ bool Process::DeliverSessionSignal(int signum) // process_family_lock held
 
 bool Process::DeliverSignal(int signum)
 {
-	ScopedLock lock(&threadlock);
+	ScopedLock lock(&thread_lock);
 
-	if ( !firstthread )
+	if ( !first_thread )
 		return errno = EINIT, false;
 
 	// Broadcast particular signals to all the threads in the process.
 	if ( signum == SIGCONT || signum == SIGSTOP || signum == SIGKILL )
 	{
 		int saved_errno = errno;
-		for ( Thread* t = firstthread; t; t = t->nextsibling )
+		for ( Thread* t = first_thread; t; t = t->next_sibling )
 		{
 			if ( !t->DeliverSignal(signum) && errno != ESIGPENDING )
 			{
@@ -371,7 +371,7 @@ bool Process::DeliverSignal(int signum)
 	// TODO: This isn't how signals should be routed to a particular thread.
 	if ( CurrentThread()->process == this )
 		return CurrentThread()->DeliverSignal(signum);
-	return firstthread->DeliverSignal(signum);
+	return first_thread->DeliverSignal(signum);
 }
 
 int sys_raise(int signum)

kernel/syscall.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011-2016, 2021-2022 Jonas 'Sortie' Termansen.
+ * Copyright (c) 2011-2016, 2021-2024 Jonas 'Sortie' Termansen.
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -204,6 +204,8 @@ void* syscall_list[SYSCALL_MAX_NUM + 1] =
 	[SYSCALL_SETDNSCONFIG] = (void*) sys_setdnsconfig,
 	[SYSCALL_FUTEX] = (void*) sys_futex,
 	[SYSCALL_MEMUSAGE] = (void*) sys_memusage,
+	[SYSCALL_GETINIT] = (void*) sys_getinit,
+	[SYSCALL_SETINIT] = (void*) sys_setinit,
 	[SYSCALL_MAX_NUM] = (void*) sys_bad_syscall,
 };
 } /* extern "C" */

kernel/thread.cpp

@@ -59,18 +59,18 @@ Thread::Thread()
 	yield_to_tid = 0;
 	id = 0; // TODO: Make a thread id.
 	process = NULL;
-	prevsibling = NULL;
-	nextsibling = NULL;
+	prev_sibling = NULL;
+	next_sibling = NULL;
 	scheduler_list_prev = NULL;
 	scheduler_list_next = NULL;
 	state = NONE;
 	memset(&registers, 0, sizeof(registers));
-	kernelstackpos = 0;
-	kernelstacksize = 0;
+	kernel_stack_pos = 0;
+	kernel_stack_size = 0;
 	signal_count = 0;
 	signal_single_frame = 0;
 	signal_canary = 0;
-	kernelstackmalloced = false;
+	kernel_stack_malloced = false;
 	pledged_destruction = false;
 	force_no_signals = false;
 	signal_single = false;
@@ -99,8 +99,8 @@ Thread::~Thread()
 	if ( process )
 		process->OnThreadDestruction(this);
 	assert(CurrentThread() != this);
-	if ( kernelstackmalloced )
-		delete[] (uint8_t*) kernelstackpos;
+	if ( kernel_stack_malloced )
+		delete[] (uint8_t*) kernel_stack_pos;
 }
 
 Thread* CreateKernelThread(Process* process,
@@ -116,7 +116,7 @@ Thread* CreateKernelThread(Process* process,
 		return errno = EINVAL, (Thread*) NULL;
 #endif
 
-	kthread_mutex_lock(&process->threadlock);
+	kthread_mutex_lock(&process->thread_lock);
 
 	// Note: Only allow the process itself to make threads, except the initial
 	// thread. This requirement is because kthread_exit() needs to know when
@@ -124,7 +124,7 @@ Thread* CreateKernelThread(Process* process,
 	// and that no more threads will appear, so it can run some final process
 	// termination steps without any interference. It's always allowed to create
 	// threads in the kernel process as it never exits.
-	assert(!process->firstthread ||
+	assert(!process->first_thread ||
 	       process == CurrentProcess() ||
 	       process == Scheduler::GetKernelProcess());
 
@@ -137,14 +137,14 @@ Thread* CreateKernelThread(Process* process,
 
 	// Create the family tree.
 	thread->process = process;
-	Thread* firsty = process->firstthread;
+	Thread* firsty = process->first_thread;
 	if ( firsty )
-		firsty->prevsibling = thread;
-	thread->nextsibling = firsty;
-	process->firstthread = thread;
+		firsty->prev_sibling = thread;
+	thread->next_sibling = firsty;
+	process->first_thread = thread;
 	process->threads_not_exiting_count++;
 
-	kthread_mutex_unlock(&process->threadlock);
+	kthread_mutex_unlock(&process->thread_lock);
 
 	return thread;
 }
@@ -260,9 +260,9 @@ Thread* CreateKernelThread(Process* process, void (*entry)(void*), void* user,
 	Thread* thread = CreateKernelThread(process, &regs, name);
 	if ( !thread ) { delete[] stack; return NULL; }
 
-	thread->kernelstackpos = (uintptr_t) stack;
-	thread->kernelstacksize = stacksize;
-	thread->kernelstackmalloced = true;
+	thread->kernel_stack_pos = (uintptr_t) stack;
+	thread->kernel_stack_size = stacksize;
+	thread->kernel_stack_malloced = true;
 
 	return thread;
 }
@@ -335,11 +335,11 @@ int sys_exit_thread(int requested_exit_code,
 
 	extended.unmap_size = Page::AlignUp(extended.unmap_size);
 
-	kthread_mutex_lock(&thread->process->threadlock);
+	kthread_mutex_lock(&thread->process->thread_lock);
 	bool is_others = false;
-	for ( Thread* iter = thread->process->firstthread;
+	for ( Thread* iter = thread->process->first_thread;
 	      !is_others && iter;
-	     iter = iter->nextsibling )
+	     iter = iter->next_sibling )
 	{
 		if ( iter == thread )
 			continue;
@@ -355,7 +355,7 @@ int sys_exit_thread(int requested_exit_code,
 		process->threads_exiting = true;
 	else if ( process->threads_exiting )
 		are_threads_exiting = true;
-	kthread_mutex_unlock(&thread->process->threadlock);
+	kthread_mutex_unlock(&thread->process->thread_lock);
 
 	// Self-destruct if another thread began exiting the process.
 	if ( are_threads_exiting )