nortti
/
sortix-mirror
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
sortix-mirror/kernel/signal.cpp

1006 lines
31 KiB
C++
Raw Permalink Normal View History

Relicense Sortix to the ISC license. I hereby relicense all my work on Sortix under the ISC license as below. All Sortix contributions by other people are already under this license, are not substantial enough to be copyrightable, or have been removed. All imported code from other projects is compatible with this license. All GPL licensed code from other projects had previously been removed. Copyright 2011-2016 Jonas 'Sortie' Termansen and contributors. 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.
2016-03-02 22:38:16 +00:00
/*
Fix sigsuspend(2) not running signal handlers with the temporary mask.
2021-02-13 21:32:21 +00:00
* Copyright (c) 2011-2016, 2018, 2021 Jonas 'Sortie' Termansen.
Relicense Sortix to the ISC license. I hereby relicense all my work on Sortix under the ISC license as below. All Sortix contributions by other people are already under this license, are not substantial enough to be copyrightable, or have been removed. All imported code from other projects is compatible with this license. All GPL licensed code from other projects had previously been removed. Copyright 2011-2016 Jonas 'Sortie' Termansen and contributors. 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.
2016-03-02 22:38:16 +00:00
*
* 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.
*
* signal.cpp
* Asynchronous user-space thread interruption.
*/
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sys/types.h>
Remove <sortix/kernel/platform.h>.
2013-10-27 00:42:10 +00:00
#include <assert.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <errno.h>
Remove <sortix/kernel/platform.h>.
2013-10-27 00:42:10 +00:00
#include <string.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <stdint.h>
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
#include <stdlib.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <signal.h>
Remove <sortix/kernel/platform.h>.
2013-10-27 00:42:10 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sortix/sigaction.h>
Refactor kernel signal.h header.
2013-05-12 22:52:58 +00:00
#include <sortix/signal.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sortix/sigset.h>
#include <sortix/stack.h>
#include <sortix/ucontext.h>
Refactor kernel signal.h header.
2013-05-12 22:52:58 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sortix/kernel/copy.h>
Refactor kernel interrupt API.
2013-01-09 22:30:36 +00:00
#include <sortix/kernel/interrupt.h>
Remove <sortix/kernel/platform.h>.
2013-10-27 00:42:10 +00:00
#include <sortix/kernel/kernel.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sortix/kernel/process.h>
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
#include <sortix/kernel/ptable.h>
Refactor kernel signal.h header.
2013-05-12 22:52:58 +00:00
#include <sortix/kernel/signal.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sortix/kernel/syscall.h>
Refactor kernel process.h and thread.h headers.
2013-05-12 22:41:30 +00:00
#include <sortix/kernel/thread.h>
Refactor kernel interrupt API.
2013-01-09 22:30:36 +00:00
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
namespace Sortix {
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
sigset_t default_ignored_signals;
sigset_t default_stop_signals;
sigset_t unblockable_signals;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
// A per-cpu value whether a signal is pending in the running task.
extern "C" { volatile unsigned long asm_signal_is_pending = 0; }
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
static
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
void UpdatePendingSignals(Thread* thread) // thread->process->signal_lock held
{
struct sigaction* signal_actions = thread->process->signal_actions;
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Determine which signals wouldn't be ignored if received.
sigset_t handled_signals;
sigemptyset(&handled_signals);
for ( int i = 1; i < SIG_MAX_NUM; i++ )
{
if ( signal_actions[i].sa_handler == SIG_IGN )
continue;
if ( signal_actions[i].sa_handler == SIG_DFL &&
sigismember(&default_ignored_signals, i) )
continue;
// TODO: A process that is a member of an orphaned process group shall
// not be allowed to stop in response to the SIGTSTP, SIGTTIN, or
// SIGTTOU signals. In cases where delivery of one of these
// signals would stop such a process, the signal shall be
// discarded.
if ( /* is member of an orphaned process group */ false &&
signal_actions[i].sa_handler == SIG_DFL &&
sigismember(&default_stop_signals, i) )
continue;
sigaddset(&handled_signals, i);
}
// TODO: Handle that signals can be pending process-wide!
// Discard all requested signals that would be ignored if delivered.
sigandset(&thread->signal_pending, &thread->signal_pending, &handled_signals);
// Determine which signals are not blocked.
sigset_t permitted_signals;
signotset(&permitted_signals, &thread->signal_mask);
sigorset(&permitted_signals, &permitted_signals, &unblockable_signals);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Determine which signals can currently be delivered to this thread.
sigset_t deliverable_signals;
sigandset(&deliverable_signals, &permitted_signals, &thread->signal_pending);
// Determine whether any signals can be delivered.
unsigned long is_pending = !sigisemptyset(&deliverable_signals) ? 1 : 0;
Fix untimely delivery of signals during userfs reference count messages.
2015-08-14 16:24:40 +00:00
if ( thread->force_no_signals )
is_pending = 0;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Store whether a signal is pending in the virtual register.
if ( thread == CurrentThread() )
asm_signal_is_pending = is_pending;
else
Implement threading primitives that truly sleep. The idle thread is now actually run when the system is idle because it truly goes idle. The idle thread is made power efficient by using the hlt instruction rather than a busy loop. The new futex(2) system call is used to implement fast user-space mutexes, condition variables, and semaphores. The same backend and design is used as kutexes for truly sleeping kernel mutexes and condition variables. The new exit_thread(2) flag EXIT_THREAD_FUTEX_WAKE wakes a futex. Sleeping on clocks in the kernel now uses timers for true sleep. The interrupt worker thread now truly sleeps when idle. Kernel threads are now named. This is a compatible ABI change.
2021-02-06 20:23:43 +00:00
Scheduler::SetSignalPending(thread, is_pending);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
}
Fix untimely delivery of signals during userfs reference count messages.
2015-08-14 16:24:40 +00:00
void Thread::DoUpdatePendingSignal()
{
ScopedLock lock(&process->signal_lock);
UpdatePendingSignals(this);
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
int sys_sigaction(int signum,
const struct sigaction* user_newact,
struct sigaction* user_oldact)
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( signum < 0 || signum == 0 /* null signal */ || SIG_MAX_NUM <= signum )
Fix system calls returning errno values instead of setting errno.
2023-04-06 14:39:15 +00:00
return errno = EINVAL, -1;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
Process* process = CurrentProcess();
ScopedLock lock(&process->signal_lock);
struct sigaction* kact = &process->signal_actions[signum];
// Let the caller know the previous action.
if ( user_oldact )
{
if ( !CopyToUser(user_oldact, kact, sizeof(struct sigaction)) )
return -1;
}
// Retrieve and validate the new signal action.
if ( user_newact )
{
struct sigaction newact;
if ( !CopyFromUser(&newact, user_newact, sizeof(struct sigaction)) )
return -1;
if ( newact.sa_flags & ~__SA_SUPPORTED_FLAGS )
return errno = EINVAL, -1;
if ( newact.sa_handler == SIG_ERR )
return errno = EINVAL, -1;
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 )
UpdatePendingSignals(t);
}
return 0;
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
int sys_sigaltstack(const stack_t* user_newstack, stack_t* user_oldstack)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
Thread* thread = CurrentThread();
if ( user_oldstack )
{
if ( !CopyToUser(user_oldstack, &thread->signal_stack, sizeof(stack_t)) )
return -1;
}
if ( user_newstack )
{
stack_t newstack;
if ( !CopyFromUser(&newstack, user_newstack, sizeof(stack_t)) )
return -1;
if ( newstack.ss_flags & ~__SS_SUPPORTED_FLAGS )
return errno = EINVAL, -1;
memcpy(&thread->signal_stack, &newstack, sizeof(stack_t));
}
return 0;
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
int sys_sigpending(sigset_t* set)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
Process* process = CurrentProcess();
Thread* thread = CurrentThread();
ScopedLock lock(&process->signal_lock);
// TODO: What about process-wide signals?
return CopyToUser(set, &thread->signal_pending, sizeof(sigset_t)) ? 0 : -1;
}
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
namespace Signal {
void UpdateMask(int how, const sigset_t* set, sigset_t* oldset)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
Process* process = CurrentProcess();
Thread* thread = CurrentThread();
// TODO: Signal masks are a per-thread property, perhaps this should be
// locked in another manner?
ScopedLock lock(&process->signal_lock);
// Let the caller know the previous signal mask.
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
if ( oldset )
memcpy(oldset, &thread->signal_mask, sizeof(sigset_t));
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Update the current signal mask according to how.
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
if ( set )
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
switch ( how )
{
case SIG_BLOCK:
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
sigorset(&thread->signal_mask, &thread->signal_mask, set);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
break;
case SIG_UNBLOCK:
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
{
sigset_t notset;
signotset(&notset, set);
sigandset(&thread->signal_mask, &thread->signal_mask, &notset);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
break;
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
case SIG_SETMASK:
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
memcpy(&thread->signal_mask, set, sizeof(sigset_t));
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
break;
};
UpdatePendingSignals(thread);
}
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
}
} // namespace Signal
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
int sys_sigprocmask(int how, const sigset_t* user_set, sigset_t* user_oldset)
{
if ( how != SIG_BLOCK && how != SIG_UNBLOCK && how != SIG_SETMASK )
return errno = EINVAL, -1;
sigset_t set, oldset;
if ( user_set && !CopyFromUser(&set, user_set, sizeof(sigset_t)) )
return -1;
Signal::UpdateMask(
how, user_set ? &set : NULL, user_oldset ? &oldset : NULL);
if ( user_oldset && !CopyToUser(user_oldset, &oldset, sizeof(sigset_t)) )
return -1;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
return 0;
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
int sys_sigsuspend(const sigset_t* set)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
Process* process = CurrentProcess();
Thread* thread = CurrentThread();
Silence maybe initialized warning in sigsuspend.
2015-11-21 23:42:10 +00:00
sigset_t old_signal_mask; sigemptyset(&old_signal_mask);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
ScopedLock lock(&process->signal_lock);
// Only accept signals from the user-provided set if given.
if ( set )
{
Fix sigsuspend(2) not running signal handlers with the temporary mask.
2021-02-13 21:32:21 +00:00
sigset_t new_signal_mask;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( !CopyFromUser(&new_signal_mask, set, sizeof(sigset_t)) )
return -1;
memcpy(&old_signal_mask, &thread->signal_mask, sizeof(sigset_t));
memcpy(&thread->signal_mask, &new_signal_mask, sizeof(sigset_t));
UpdatePendingSignals(thread);
}
// Wait for a signal to happen or otherwise never halt.
kthread_cond_t never_triggered = KTHREAD_COND_INITIALIZER;
while ( !Signal::IsPending() )
kthread_cond_wait_signal(&never_triggered, &process->signal_lock);
Fix sigsuspend(2) not running signal handlers with the temporary mask.
2021-02-13 21:32:21 +00:00
// The pending signal might only be pending with the temporary signal mask,
// so don't restore it. Instead ask for the real signal mask to be restored
// after the signal has been processed.
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( set )
{
Fix sigsuspend(2) not running signal handlers with the temporary mask.
2021-02-13 21:32:21 +00:00
thread->has_saved_signal_mask = true;
memcpy(&thread->saved_signal_mask, &old_signal_mask, sizeof(sigset_t));
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
}
// The system call never halts or it halts because a signal interrupted it.
return errno = EINTR, -1;
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
int sys_kill(pid_t pid, int signum)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
// Protect the kernel process.
if ( !pid )
return errno = EPERM, -1;
// TODO: Implement that pid == -1 means all processes!
bool process_group = pid < 0 ? (pid = -pid, true) : false;
Add support for sessions. This change refactors the process group implementation and adds support for sessions. The setsid(2) and getsid(2) system calls were added. psctl(2) now has PSCTL_TTYNAME, which lets you get the name of a process's terminal, and ps(1) now uses it. The initial terminal is now called /dev/tty1. /dev/tty is now a factory for the current terminal. A global lock now protects the process hierarchy which makes it safe to access other processes. This refactor removes potential vulnerabilities and increases system robustness. A number of terminal ioctls have been added. This is a compatible ABI change.
2016-11-15 20:08:41 +00:00
ScopedLock lock(&process_family_lock);
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
Process* process = CurrentProcess()->GetPTable()->Get(pid);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( !process )
return errno = ESRCH, -1;
// TODO: Protect init?
// TODO: Check for permission.
// TODO: Check for zombies.
if ( process_group )
{
if ( !process->DeliverGroupSignal(signum) && errno != ESIGPENDING )
return -1;
return errno = 0, 0;
}
if ( !process->DeliverSignal(signum) && errno != ESIGPENDING )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
return -1;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
return errno = 0, 0;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Add support for sessions. This change refactors the process group implementation and adds support for sessions. The setsid(2) and getsid(2) system calls were added. psctl(2) now has PSCTL_TTYNAME, which lets you get the name of a process's terminal, and ps(1) now uses it. The initial terminal is now called /dev/tty1. /dev/tty is now a factory for the current terminal. A global lock now protects the process hierarchy which makes it safe to access other processes. This refactor removes potential vulnerabilities and increases system robustness. A number of terminal ioctls have been added. This is a compatible ABI change.
2016-11-15 20:08:41 +00:00
bool Process::DeliverGroupSignal(int signum) // process_family_lock held
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( !groupfirst )
return errno = ESRCH, false;
for ( Process* iter = groupfirst; iter; iter = iter->groupnext )
{
int saved_errno = errno;
if ( !iter->DeliverSignal(signum) && errno != ESIGPENDING )
{
// This is not currently an error condition.
}
errno = saved_errno;
}
return true;
}
Add support for sessions. This change refactors the process group implementation and adds support for sessions. The setsid(2) and getsid(2) system calls were added. psctl(2) now has PSCTL_TTYNAME, which lets you get the name of a process's terminal, and ps(1) now uses it. The initial terminal is now called /dev/tty1. /dev/tty is now a factory for the current terminal. A global lock now protects the process hierarchy which makes it safe to access other processes. This refactor removes potential vulnerabilities and increases system robustness. A number of terminal ioctls have been added. This is a compatible ABI change.
2016-11-15 20:08:41 +00:00
bool Process::DeliverSessionSignal(int signum) // process_family_lock held
{
if ( !sessionfirst )
return errno = ESRCH, false;
for ( Process* iter = sessionfirst; iter; iter = iter->sessionnext )
{
int saved_errno = errno;
if ( !iter->DeliverSignal(signum) && errno != ESIGPENDING )
{
// This is not currently an error condition.
}
errno = saved_errno;
}
return true;
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
bool Process::DeliverSignal(int signum)
{
ScopedLock lock(&threadlock);
if ( !firstthread )
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 )
{
if ( !t->DeliverSignal(signum) && errno != ESIGPENDING )
{
// This is not currently an error condition.
}
}
errno = saved_errno;
return true;
}
// Route the signal to a suitable thread that accepts it.
// 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);
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
int sys_raise(int signum)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
if ( !CurrentThread()->DeliverSignal(signum) && errno != ESIGPENDING )
return -1;
return errno = 0, 0;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
bool Thread::DeliverSignal(int signum)
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
ScopedLock lock(&process->signal_lock);
return DeliverSignalUnlocked(signum);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
bool Thread::DeliverSignalUnlocked(int signum) // thread->process->signal_lock held
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( signum <= 0 || SIG_MAX_NUM <= signum )
return errno = EINVAL, false;
// Discard the null signal, which does error checking, but doesn't actually
// deliver a signal to the process or thread.
if ( signum == 0 )
return true;
if ( sigismember(&signal_pending, signum) )
return errno = ESIGPENDING, false;
sigaddset(&signal_pending, signum);
if ( signum == SIGSTOP || signum == SIGTSTP ||
signum == SIGTTIN || signum == SIGTTOU )
sigdelset(&signal_pending, SIGCONT);
if ( signum == SIGCONT )
{
sigdelset(&signal_pending, SIGSTOP);
sigdelset(&signal_pending, SIGTSTP);
sigdelset(&signal_pending, SIGTTIN);
sigdelset(&signal_pending, SIGTTOU);
}
UpdatePendingSignals(this);
return true;
}
static int PickImportantSignal(const sigset_t* set)
{
if ( sigismember(set, SIGKILL) )
return SIGKILL;
if ( sigismember(set, SIGSTOP) )
return SIGSTOP;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
for ( int i = 1; i < SIG_MAX_NUM; i++ )
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( sigismember(set, i) )
return i;
return 0;
}
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
static void EncodeMachineContext(mcontext_t* mctx,
const struct thread_registers* regs,
const struct interrupt_context* intctx)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
memset(mctx, 0, sizeof(*mctx));
#if defined(__i386__)
// TODO: REG_GS
// TODO: REG_FS
// TODO: REG_ES
// TODO: REG_DS
mctx->gregs[REG_EDI] = regs->edi;
mctx->gregs[REG_ESI] = regs->esi;
mctx->gregs[REG_EBP] = regs->ebp;
mctx->gregs[REG_ESP] = regs->esp;
mctx->gregs[REG_EBX] = regs->ebx;
mctx->gregs[REG_EDX] = regs->edx;
mctx->gregs[REG_ECX] = regs->ecx;
mctx->gregs[REG_EAX] = regs->eax;
mctx->gregs[REG_EIP] = regs->eip;
// TODO: REG_CS
mctx->gregs[REG_EFL] = regs->eflags & 0x0000FFFF;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
mctx->gregs[REG_CR2] = intctx->cr2;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// TODO: REG_SS
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
memcpy(mctx->fpuenv, regs->fpuenv, 512);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#elif defined(__x86_64__)
mctx->gregs[REG_R8] = regs->r8;
mctx->gregs[REG_R9] = regs->r9;
mctx->gregs[REG_R10] = regs->r10;
mctx->gregs[REG_R11] = regs->r11;
mctx->gregs[REG_R12] = regs->r12;
mctx->gregs[REG_R13] = regs->r13;
mctx->gregs[REG_R14] = regs->r14;
mctx->gregs[REG_R15] = regs->r15;
mctx->gregs[REG_RDI] = regs->rdi;
mctx->gregs[REG_RSI] = regs->rsi;
mctx->gregs[REG_RBP] = regs->rbp;
mctx->gregs[REG_RBX] = regs->rbx;
mctx->gregs[REG_RDX] = regs->rdx;
mctx->gregs[REG_RAX] = regs->rax;
mctx->gregs[REG_RCX] = regs->rcx;
mctx->gregs[REG_RSP] = regs->rsp;
mctx->gregs[REG_RIP] = regs->rip;
mctx->gregs[REG_EFL] = regs->rflags & 0x000000000000FFFF;
// TODO: REG_CSGSFS.
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
mctx->gregs[REG_CR2] = intctx->cr2;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
mctx->gregs[REG_FSBASE] = 0x0;
mctx->gregs[REG_GSBASE] = 0x0;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
memcpy(mctx->fpuenv, regs->fpuenv, 512);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#error "You need to implement conversion to mcontext"
#endif
}
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
static void DecodeMachineContext(const mcontext_t* mctx,
struct thread_registers* regs)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
#if defined(__i386__) || defined(__x86_64__)
unsigned long user_flags = FLAGS_CARRY | FLAGS_PARITY | FLAGS_AUX
| FLAGS_ZERO | FLAGS_SIGN | FLAGS_DIRECTION
| FLAGS_OVERFLOW;
#endif
#if defined(__i386__)
regs->edi = mctx->gregs[REG_EDI];
regs->esi = mctx->gregs[REG_ESI];
regs->ebp = mctx->gregs[REG_EBP];
regs->esp = mctx->gregs[REG_ESP];
regs->ebx = mctx->gregs[REG_EBX];
regs->edx = mctx->gregs[REG_EDX];
regs->ecx = mctx->gregs[REG_ECX];
regs->eax = mctx->gregs[REG_EAX];
regs->eip = mctx->gregs[REG_EIP];
regs->eflags &= ~user_flags;
regs->eflags |= mctx->gregs[REG_EFL] & user_flags;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
memcpy(regs->fpuenv, mctx->fpuenv, 512);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#elif defined(__x86_64__)
regs->r8 = mctx->gregs[REG_R8];
regs->r9 = mctx->gregs[REG_R9];
regs->r10 = mctx->gregs[REG_R10];
regs->r11 = mctx->gregs[REG_R11];
regs->r12 = mctx->gregs[REG_R12];
regs->r13 = mctx->gregs[REG_R13];
regs->r14 = mctx->gregs[REG_R14];
regs->r15 = mctx->gregs[REG_R15];
regs->rdi = mctx->gregs[REG_RDI];
regs->rsi = mctx->gregs[REG_RSI];
regs->rbp = mctx->gregs[REG_RBP];
regs->rbx = mctx->gregs[REG_RBX];
regs->rdx = mctx->gregs[REG_RDX];
regs->rax = mctx->gregs[REG_RAX];
regs->rcx = mctx->gregs[REG_RCX];
regs->rsp = mctx->gregs[REG_RSP];
regs->rip = mctx->gregs[REG_RIP];
regs->rflags &= ~user_flags;
regs->rflags |= mctx->gregs[REG_EFL] & user_flags;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
memcpy(regs->fpuenv, mctx->fpuenv, 512);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#error "You need to implement conversion to mcontext"
#endif
}
#if defined(__i386__)
struct stack_frame
{
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
uintptr_t misalignment[3];
uintptr_t sigreturn;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
int signum_param;
siginfo_t* siginfo_param;
ucontext_t* ucontext_param;
void* cookie_param;
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
uintptr_t canary;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
ucontext_t ucontext;
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
siginfo_t siginfo;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
};
#elif defined(__x86_64__)
struct stack_frame
{
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
uintptr_t misalignment[1];
uintptr_t sigreturn;
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
uintptr_t canary;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
ucontext_t ucontext;
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
siginfo_t siginfo;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
};
#else
#error "You need to implement struct stack_frame"
#endif
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
void Thread::HandleSignal(struct interrupt_context* intctx)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
assert(Interrupt::IsEnabled());
assert(this == CurrentThread());
ScopedLock lock(&process->signal_lock);
retry_another_signal:
// Determine which signals are not blocked.
sigset_t permitted_signals;
signotset(&permitted_signals, &signal_mask);
sigorset(&permitted_signals, &permitted_signals, &unblockable_signals);
// Determine which signals can currently be delivered to this thread.
sigset_t deliverable_signals;
sigandset(&deliverable_signals, &permitted_signals, &signal_pending);
// Decide which signal to deliver to the thread.
int signum = PickImportantSignal(&deliverable_signals);
if ( !signum )
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
{
if ( has_saved_signal_mask )
{
memcpy(&signal_mask, &saved_signal_mask, sizeof(sigset_t));
has_saved_signal_mask = false;
UpdatePendingSignals(this);
goto retry_another_signal;
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
return;
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Unmark the selected signal as pending.
sigdelset(&signal_pending, signum);
UpdatePendingSignals(this);
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
intctx->signal_pending = asm_signal_is_pending;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Destroy the current thread if the signal is critical.
if ( signum == SIGKILL )
{
lock.Reset();
kthread_exit();
}
struct sigaction* action = &process->signal_actions[signum];
// Stop the current thread upon receipt of a stop signal that isn't handled
// or cannot be handled (SIGSTOP).
if ( (action->sa_handler == SIG_DFL &&
sigismember(&default_stop_signals, signum) ) ||
signum == SIGSTOP )
{
Log::PrintF("%s:%u: `%s' FIXME SIGSTOP\n", __FILE__, __LINE__, __PRETTY_FUNCTION__);
// TODO: Stop the current process.
// TODO: Deliver SIGCHLD to the parent except if SA_NOCLDSTOP is set in
// the parent's SIGCHLD sigaction.
// TODO: SIGCHLD should not be delivered until all the threads in the
// process has received SIGSTOP and stopped?
// TODO: SIGKILL must still be deliverable to a stopped process.
}
// Resume the current thread upon receipt of SIGCONT.
if ( signum == SIGCONT )
{
Log::PrintF("%s:%u: `%s' FIXME SIGCONT\n", __FILE__, __LINE__, __PRETTY_FUNCTION__);
// TODO: Resume the current process.
// TODO: Can SIGCONT be masked?
// TODO: Can SIGCONT be handled?
// TODO: Can SIGCONT be ignored?
// TODO: Deliver SIGCHLD to the parent except if SA_NOCLDSTOP is set in
// the parent's SIGCHLD sigaction.
}
// Signals that would be ignored are already filtered away at this point.
assert(action->sa_handler != SIG_IGN);
assert(action->sa_handler != SIG_DFL || !sigismember(&default_ignored_signals, signum));
// The default action must be to terminate the process. Signals that are
Fix signal dispatch if sigreturn is NULL.
2018-08-06 19:42:27 +00:00
// ignored by default got discarded earlier. If execve() failed, sigreturn
// may be NULL and the process isn't able to properly process signals.
if ( action->sa_handler == SIG_DFL || !process->sigreturn )
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
kthread_mutex_unlock(&process->signal_lock);
process->ExitThroughSignal(signum);
kthread_mutex_lock(&process->signal_lock);
goto retry_another_signal;
}
// At this point we have to attempt to invoke the user-space signal handler,
// which will then return control to us through sigreturn. However, we can't
// save the kernel state because 1) we can't trust the user-space stack 2)
// we can't rely on the kernel stack being intact as the signal handler may
// invoke system calls. For those reasons, we'll have to modify the saved
// registers so they restore a user-space state. We can do this because
// threads in the kernel cannot be delivered signals except when returning
// from a system call, so we'll simply save the state that would have been
// returned to user-space had no signal occured.
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
if ( !InUserspace(intctx) )
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
#if defined(__i386__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
uint32_t* params = (uint32_t*) intctx->ebx;
intctx->eip = params[0];
intctx->eflags = params[2];
intctx->esp = params[3];
intctx->cs = UCS | URPL;
intctx->ds = UDS | URPL;
intctx->ss = UDS | URPL;
Clear registers after system calls.
2015-05-16 17:22:46 +00:00
intctx->ebx = 0;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#elif defined(__x86_64__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
intctx->rip = intctx->rdi;
intctx->rflags = intctx->rsi;
intctx->rsp = intctx->r8;
intctx->cs = UCS | URPL;
intctx->ds = UDS | URPL;
intctx->ss = UDS | URPL;
Clear registers after system calls.
2015-05-16 17:22:46 +00:00
intctx->rdi = 0;
intctx->rsi = 0;
intctx->r8 = 0;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#error "You may need to fix the registers"
#endif
}
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
struct thread_registers stopped_regs;
Scheduler::SaveInterruptedContext(intctx, &stopped_regs);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
sigset_t new_signal_mask;
memcpy(&new_signal_mask, &action->sa_mask, sizeof(sigset_t));
sigorset(&new_signal_mask, &new_signal_mask, &signal_mask);
// Prevent signals from interrupting themselves by default.
if ( !(action->sa_flags & SA_NODEFER) )
sigaddset(&new_signal_mask, signum);
// Determine whether we use an alternate signal stack.
bool signal_uses_altstack = action->sa_flags & SA_ONSTACK;
bool usable_altstack = !(signal_stack.ss_flags & (SS_DISABLE | SS_ONSTACK));
bool use_altstack = signal_uses_altstack && usable_altstack;
// Determine which signal stack to use and what to save.
stack_t old_signal_stack, new_signal_stack;
uintptr_t stack_location;
if ( use_altstack )
{
old_signal_stack = signal_stack;
new_signal_stack = signal_stack;
new_signal_stack.ss_flags |= SS_ONSTACK;
#if defined(__i386__) || defined(__x86_64__)
stack_location = (uintptr_t) signal_stack.ss_sp + signal_stack.ss_size;
#else
#error "You need to implement getting the alternate stack pointer"
#endif
}
else
{
old_signal_stack.ss_sp = NULL;
old_signal_stack.ss_flags = SS_DISABLE;
old_signal_stack.ss_size = 0;
new_signal_stack = signal_stack;
#if defined(__i386__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
stack_location = (uintptr_t) stopped_regs.esp;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#elif defined(__x86_64__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
stack_location = (uintptr_t) stopped_regs.rsp;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#error "You need to implement getting the user-space stack pointer"
#endif
}
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
struct thread_registers handler_regs;
memcpy(&handler_regs, &stopped_regs, sizeof(handler_regs));
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
struct stack_frame stack_frame;
memset(&stack_frame, 0, sizeof(stack_frame));
void* handler_ptr = action->sa_flags & SA_COOKIE ?
(void*) action->sa_sigaction_cookie :
action->sa_flags & SA_SIGINFO ?
(void*) action->sa_sigaction :
(void*) action->sa_handler;
#if defined(__i386__)
stack_location -= sizeof(stack_frame);
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
stack_location &= ~(16UL-1UL); /* 16-byte align */
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
struct stack_frame* stack = (struct stack_frame*) stack_location;
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
stack_frame.sigreturn = (uintptr_t) process->sigreturn;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
stack_frame.signum_param = signum;
stack_frame.siginfo_param = &stack->siginfo;
stack_frame.ucontext_param = &stack->ucontext;
stack_frame.cookie_param = action->sa_cookie;
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
handler_regs.esp = (unsigned long) stack + sizeof(stack->misalignment);
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
handler_regs.eip = (unsigned long) handler_ptr;
handler_regs.eflags &= ~FLAGS_DIRECTION;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#elif defined(__x86_64__)
stack_location -= 128; /* Red zone. */
stack_location -= sizeof(stack_frame);
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
stack_location &= ~(16UL-1UL); /* 16-byte align */
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
struct stack_frame* stack = (struct stack_frame*) stack_location;
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
stack_frame.sigreturn = (uintptr_t) process->sigreturn;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
handler_regs.rdi = (unsigned long) signum;
handler_regs.rsi = (unsigned long) &stack->siginfo;
handler_regs.rdx = (unsigned long) &stack->ucontext;
handler_regs.rcx = (unsigned long) action->sa_cookie;
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
handler_regs.rsp = (unsigned long) stack + sizeof(stack->misalignment);
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
handler_regs.rip = (unsigned long) handler_ptr;
handler_regs.rflags &= ~FLAGS_DIRECTION;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#error "You need to format the stack frame"
#endif
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
// Store a canary so it can later be verified this is a real signal being
// returned from.
if ( signal_count == 0 )
arc4random_buf(&signal_canary, sizeof(signal_canary));
stack_frame.canary = signal_canary ^ (uintptr_t) stack;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Format the siginfo into the stack frame.
stack_frame.siginfo.si_signo = signum;
#if defined(__i386__) || defined(__x86_64__)
// TODO: Is this cr2 value trustworthy? I don't think it is.
if ( signum == SIGSEGV )
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
stack_frame.siginfo.si_addr = (void*) intctx->cr2;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#warning "You need to tell user-space where it crashed"
#endif
// Format the ucontext into the stack frame.
stack_frame.ucontext.uc_link = NULL;
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
if ( has_saved_signal_mask )
{
// If a system call temporarily set another signal mask, it wants us to
// deliver signals for that temporary signal masks, however we must
// restore the original saved signal mask in that case.
memcpy(&stack_frame.ucontext.uc_sigmask, &saved_signal_mask,
sizeof(saved_signal_mask));
// 'has_saved_signal_mask' is set to false below when it's actually
// delivered. This handles the case where the signal delivery fails and
// and instead turns into another (we still want to restore the right
// saved mask in that case).
}
else
memcpy(&stack_frame.ucontext.uc_sigmask, &signal_mask, sizeof(signal_mask));
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
memcpy(&stack_frame.ucontext.uc_stack, &signal_stack, sizeof(signal_stack));
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
EncodeMachineContext(&stack_frame.ucontext.uc_mcontext, &stopped_regs, intctx);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( !CopyToUser(stack, &stack_frame, sizeof(stack_frame)) )
{
// Self-destruct if we crashed during delivering the crash signal.
if ( signum == SIGSEGV )
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
kthread_mutex_unlock(&process->signal_lock);
process->ExitThroughSignal(signum);
kthread_mutex_lock(&process->signal_lock);
goto retry_another_signal;
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Deliver SIGSEGV if we could not deliver the signal on the stack.
// TODO: Is it possible to block SIGSEGV here?
kthread_mutex_unlock(&process->signal_lock);
DeliverSignal(SIGSEGV);
kthread_mutex_lock(&process->signal_lock);
goto retry_another_signal;
}
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
// Update the current signal mask. UpdatePendingSignals isn't called because
// sa_mask was or'd onto the current signal mask, only blocking signals, so
// nothing new can be delivered.
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
memcpy(&signal_mask, &new_signal_mask, sizeof(sigset_t));
// Update the current alternate signal stack.
signal_stack = new_signal_stack;
// Update the current registers.
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
Scheduler::LoadInterruptedContext(intctx, &handler_regs);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
Implement sigaction(2) SA_RESETHAND.
2015-08-19 12:54:19 +00:00
// Reset the signal handler if this signal handler is once only.
if ( action->sa_flags & SA_RESETHAND )
{
// POSIX mandates SA_RESETHAND is silently ignored for these signals.
if ( signum != SIGILL && signum != SIGTRAP )
{
action->sa_flags &= ~SA_RESETHAND;
action->sa_handler = SIG_DFL;
}
}
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
// Know for sure when there's no signal still being handled. This is an
// over-approximation as programs may do absolutely awful things such as
// longjmp(3)ing out of signal handlers, so each delivered signal does not
// nessesarily mean a sigreturn. This will work correctly in reasonable
// programs though and will harden those programs. Remember the stack frame
// as well for verification if there is no recursive signal handling.
if ( signal_count != SIZE_MAX )
signal_count++;
if ( (signal_single = signal_count == 1) )
signal_single_frame = (uintptr_t) stack;
Add signal mask support to ppoll(2).
2018-10-20 10:57:31 +00:00
// If there was a saved signal mask, it will be restored once the signal
// handler complete, so forget it was ever saved.
has_saved_signal_mask = false;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
// Run the signal handler by returning to user-space.
return;
}
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
void Thread::HandleSigreturn(struct interrupt_context* intctx)
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
assert(Interrupt::IsEnabled());
assert(this == CurrentThread());
struct stack_frame stack_frame;
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
struct stack_frame* user_stack_frame;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#if defined(__i386__)
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
user_stack_frame = (struct stack_frame*)
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
(intctx->esp - offsetof(struct stack_frame, sigreturn) - 4);
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
bool wrong_ip = intctx->eip != (uintptr_t) process->sigreturn + 2;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#elif defined(__x86_64__)
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
user_stack_frame = (struct stack_frame*)
Fix signal delivery stack alignment.
2016-05-13 20:21:42 +00:00
(intctx->rsp - offsetof(struct stack_frame, sigreturn) - 8);
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
bool wrong_ip = intctx->rip != (uintptr_t) process->sigreturn + 2;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#else
#error "You need to locate the stack we passed the signal handler"
#endif
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
// Protect against sigreturn oriented programming (SROP) as described in
// "Framing Signals - A Return to Portable Shellcode" (Bosman and Bos 2014).
// If the we're not called from the kernel sigreturn page, it is not a valid
// sigreturn.
if ( wrong_ip )
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
{
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
process->ExitThroughSignal(SIGABRT);
Log::PrintF("%s[%ji]: sigreturn smashing detected: Bypassed kernel sigreturn page\n",
process->program_image_path,
(intmax_t) process->pid);
// TODO: Allow debugging this event (see scram(2)).
kthread_exit();
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
}
Add protection against sigreturn oriented programming (SROP). This change hardens against invalid calls to sigreturn, which is a very useful gadget when compromising a process. The system call now verifies it is a real return from a signal and aborts the process otherwise. This should render such attacks impossible in threads that are not servicing a signal, and infeasible in threads that are handling signals they are yet to return from. The kernel now keeps track for each thread how many signals are being handled but haven't returned yet. Each thread now has a random signal value. It is re-randomized when the thread handles a signal and the current signal counter is zero. This is xorred with the context address and used as canary on the stack during signal dispatch, protecting the saved context on the stack. This works mostly like the regular stack protector. The kernel now keeps track of the stack pointer for a single handled signal per thread. It doesn't seem worth it to keep track of multiple handled signals, as more than one is rare. Note that each delivered signal will not necessarily result in a sigreturn because it is valid for a thread to longjmp(3) out of a signal handler to a valid jmp_buf. The sigreturn system call will abort if either: - It was not called from the kernel sigreturn page. - The thread is not currently processing a signal. - The thread is processing a single signal, and the stack pointer did not have the expected value. - It fails to read the context on the stack. - The canary is wrong.
2016-05-13 23:14:26 +00:00
// If no signals are being serviced by this thread at the moment, it is not
// a valid sigreturn.
if ( signal_count == 0 )
{
process->ExitThroughSignal(SIGABRT);
Log::PrintF("%s[%ji]: sigreturn smashing detected: Thread wasn't servicing a signal\n",
process->program_image_path,
(intmax_t) process->pid);
// TODO: Allow debugging this event (see scram(2)).
kthread_exit();
}
// If a single signal is being serviced by this thread at the moment, the
// stack pointer must be what we expect it to be. If there's multiple, we
// don't know which one is the correct. (We could keep track of them and
// ensure it's one of them, but that's not really worth it. The list of such
// delivered signals could grow without bound because it's valid to longjmp
// out of a signal handler)
if ( signal_single && (uintptr_t) user_stack_frame != signal_single_frame )
{
process->ExitThroughSignal(SIGABRT);
Log::PrintF("%s[%ji]: sigreturn smashing detected: Stack pointer was wrong\n",
process->program_image_path,
(intmax_t) process->pid);
// TODO: Allow debugging this event (see scram(2)).
kthread_exit();
}
// If we couldn't read the frame, the sigreturn is certainly bad.
if ( !CopyFromUser(&stack_frame, user_stack_frame, sizeof(stack_frame)) )
{
process->ExitThroughSignal(SIGABRT);
Log::PrintF("%s[%ji]: sigreturn smashing detected: Couldn't read stack frame: %m\n",
process->program_image_path,
(intmax_t) process->pid);
// TODO: Allow debugging this event (see scram(2)).
kthread_exit();
}
ZeroUser(user_stack_frame, sizeof(*user_stack_frame));
// If the random canary isn't correct, the sigreturn is certianly bad.
if ( stack_frame.canary != (signal_canary ^ (uintptr_t) user_stack_frame) )
{
process->ExitThroughSignal(SIGABRT);
Log::PrintF("%s[%ji]: sigreturn smashing detected: Verification value was incorrect\n",
process->program_image_path,
(intmax_t) process->pid);
// TODO: Allow debugging this event (see scram(2)).
kthread_exit();
}
ScopedLock lock(&process->signal_lock);
memcpy(&signal_mask, &stack_frame.ucontext.uc_sigmask, sizeof(signal_mask));
memcpy(&signal_stack, &stack_frame.ucontext.uc_stack, sizeof(signal_stack));
signal_stack.ss_flags &= __SS_SUPPORTED_FLAGS;
struct thread_registers resume_regs;
Scheduler::SaveInterruptedContext(intctx, &resume_regs);
DecodeMachineContext(&stack_frame.ucontext.uc_mcontext, &resume_regs);
Scheduler::LoadInterruptedContext(intctx, &resume_regs);
if ( signal_count != SIZE_MAX )
signal_count--;
signal_single = false;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
UpdatePendingSignals(this);
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
intctx->signal_pending = asm_signal_is_pending;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
lock.Reset();
Add assertions to Signal::DispatchHandler callers.
2016-03-28 15:37:37 +00:00
assert(Interrupt::IsEnabled());
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
HandleSignal(intctx);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
namespace Signal {
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
void DispatchHandler(struct interrupt_context* intctx, void* /*user*/)
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Add assertions to Signal::DispatchHandler callers.
2016-03-28 15:37:37 +00:00
assert(Interrupt::IsEnabled());
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
return CurrentThread()->HandleSignal(intctx);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
void ReturnHandler(struct interrupt_context* intctx, void* /*user*/)
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Add assertions to Signal::DispatchHandler callers.
2016-03-28 15:37:37 +00:00
assert(Interrupt::IsEnabled());
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
return CurrentThread()->HandleSigreturn(intctx);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
void Init()
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
sigemptyset(&default_ignored_signals);
sigaddset(&default_ignored_signals, SIGCHLD);
sigaddset(&default_ignored_signals, SIGURG);
sigaddset(&default_ignored_signals, SIGPWR);
sigaddset(&default_ignored_signals, SIGWINCH);
sigemptyset(&default_stop_signals);
sigaddset(&default_stop_signals, SIGTSTP);
sigaddset(&default_stop_signals, SIGTTIN);
sigaddset(&default_stop_signals, SIGTTOU);
sigemptyset(&unblockable_signals);
sigaddset(&unblockable_signals, SIGKILL);
sigaddset(&unblockable_signals, SIGSTOP);
Initial signal support. Please squash improvements into this commit.
2011-09-15 20:38:40 +00:00
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
} // namespace Signal
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
} // namespace Sortix