cosmopolitan/libc/sock/sendfile.c

224 lines
8.7 KiB
C
Raw Normal View History

2020-06-15 14:18:57 +00:00
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
vi: set net ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi
Copyright 2020 Justine Alexandra Roberts Tunney
2020-12-28 01:18:44 +00:00
Permission to use, copy, modify, and/or 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.
2020-06-15 14:18:57 +00:00
2020-12-28 01:18:44 +00:00
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.
2020-06-15 14:18:57 +00:00
*/
#include "libc/assert.h"
#include "libc/calls/bo.internal.h"
2020-06-15 14:18:57 +00:00
#include "libc/calls/calls.h"
#include "libc/calls/internal.h"
#include "libc/calls/sig.internal.h"
2022-05-23 22:06:11 +00:00
#include "libc/calls/syscall-sysv.internal.h"
#include "libc/calls/syscall_support-nt.internal.h"
2020-06-15 14:18:57 +00:00
#include "libc/dce.h"
#include "libc/errno.h"
2022-09-13 11:32:29 +00:00
#include "libc/intrin/asan.internal.h"
#include "libc/intrin/describeflags.internal.h"
#include "libc/intrin/safemacros.internal.h"
2022-09-13 11:32:29 +00:00
#include "libc/intrin/strace.internal.h"
#include "libc/macros.internal.h"
#include "libc/nt/enum/filetype.h"
#include "libc/nt/enum/wait.h"
#include "libc/nt/errors.h"
#include "libc/nt/files.h"
#include "libc/nt/struct/byhandlefileinformation.h"
2020-06-15 14:18:57 +00:00
#include "libc/nt/winsock.h"
#include "libc/sock/internal.h"
2022-05-23 22:06:11 +00:00
#include "libc/sock/sendfile.internal.h"
2020-06-15 14:18:57 +00:00
#include "libc/str/str.h"
#include "libc/sysv/errfuns.h"
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
#include "libc/thread/posixthread.internal.h"
2020-06-15 14:18:57 +00:00
// sendfile() isn't specified as raising eintr
static textwindows int SendfileBlock(int64_t handle,
struct NtOverlapped *overlapped) {
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
struct PosixThread *pt;
uint32_t i, got, flags = 0;
if (WSAGetLastError() != kNtErrorIoPending &&
WSAGetLastError() != WSAEINPROGRESS) {
NTTRACE("TransmitFile failed %lm");
return __winsockerr();
}
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
pt = _pthread_self();
pt->abort_errno = 0;
pt->ioverlap = overlapped;
pt->iohandle = handle;
for (;;) {
i = WSAWaitForMultipleEvents(1, &overlapped->hEvent, true,
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
__SIG_IO_INTERVAL_MS, true);
if (i == kNtWaitFailed) {
NTTRACE("WSAWaitForMultipleEvents failed %lm");
return __winsockerr();
} else if (i == kNtWaitTimeout || i == kNtWaitIoCompletion) {
if (_check_interrupts(kSigOpRestartable)) return -1;
#if _NTTRACE
POLLTRACE("WSAWaitForMultipleEvents...");
#endif
} else {
break;
}
}
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
pt->ioverlap = 0;
pt->iohandle = 0;
if (WSAGetOverlappedResult(handle, overlapped, &got, false, &flags)) {
return got;
} else {
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
if (WSAGetLastError() == kNtErrorOperationAborted) {
errno = pt->abort_errno;
}
return -1;
}
}
static dontinline textwindows ssize_t sys_sendfile_nt(
int outfd, int infd, int64_t *opt_in_out_inoffset, uint32_t uptobytes) {
ssize_t rc;
int64_t ih, oh, pos, eof, offset;
struct NtByHandleFileInformation wst;
if (!__isfdkind(infd, kFdFile)) return ebadf();
if (!__isfdkind(outfd, kFdSocket)) return ebadf();
ih = g_fds.p[infd].handle;
oh = g_fds.p[outfd].handle;
if (!SetFilePointerEx(ih, 0, &pos, SEEK_CUR)) {
return __winerr();
2020-06-15 14:18:57 +00:00
}
if (opt_in_out_inoffset) {
offset = *opt_in_out_inoffset;
} else {
offset = pos;
}
if (GetFileInformationByHandle(ih, &wst)) {
// TransmitFile() returns EINVAL if `uptobytes` goes past EOF.
eof = (uint64_t)wst.nFileSizeHigh << 32 | wst.nFileSizeLow;
if (offset + uptobytes > eof) {
uptobytes = eof - offset;
}
} else {
return ebadf();
}
struct NtOverlapped ov = {
Make improvements - Every unit test now passes on Apple Silicon. The final piece of this puzzle was porting our POSIX threads cancelation support, since that works differently on ARM64 XNU vs. AMD64. Our semaphore support on Apple Silicon is also superior now compared to AMD64, thanks to the grand central dispatch library which lets *NSYNC locks go faster. - The Cosmopolitan runtime is now more stable, particularly on Windows. To do this, thread local storage is mandatory at all runtime levels, and the innermost packages of the C library is no longer being built using ASAN. TLS is being bootstrapped with a 128-byte TIB during the process startup phase, and then later on the runtime re-allocates it either statically or dynamically to support code using _Thread_local. fork() and execve() now do a better job cooperating with threads. We can now check how much stack memory is left in the process or thread when functions like kprintf() / execve() etc. call alloca(), so that ENOMEM can be raised, reduce a buffer size, or just print a warning. - POSIX signal emulation is now implemented the same way kernels do it with pthread_kill() and raise(). Any thread can interrupt any other thread, regardless of what it's doing. If it's blocked on read/write then the killer thread will cancel its i/o operation so that EINTR can be returned in the mark thread immediately. If it's doing a tight CPU bound operation, then that's also interrupted by the signal delivery. Signal delivery works now by suspending a thread and pushing context data structures onto its stack, and redirecting its execution to a trampoline function, which calls SetThreadContext(GetCurrentThread()) when it's done. - We're now doing a better job managing locks and handles. On NetBSD we now close semaphore file descriptors in forked children. Semaphores on Windows can now be canceled immediately, which means mutexes/condition variables will now go faster. Apple Silicon semaphores can be canceled too. We're now using Apple's pthread_yield() funciton. Apple _nocancel syscalls are now used on XNU when appropriate to ensure pthread_cancel requests aren't lost. The MbedTLS library has been updated to support POSIX thread cancelations. See tool/build/runitd.c for an example of how it can be used for production multi-threaded tls servers. Handles on Windows now leak less often across processes. All i/o operations on Windows are now overlapped, which means file pointers can no longer be inherited across dup() and fork() for the time being. - We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4() which means, for example, that posix_spawn() now goes 3x faster. POSIX spawn is also now more correct. Like Musl, it's now able to report the failure code of execve() via a pipe although our approach favors using shared memory to do that on systems that have a true vfork() function. - We now spawn a thread to deliver SIGALRM to threads when setitimer() is used. This enables the most precise wakeups the OS makes possible. - The Cosmopolitan runtime now uses less memory. On NetBSD for example, it turned out the kernel would actually commit the PT_GNU_STACK size which caused RSS to be 6mb for every process. Now it's down to ~4kb. On Apple Silicon, we reduce the mandatory upstream thread size to the smallest possible size to reduce the memory overhead of Cosmo threads. The examples directory has a program called greenbean which can spawn a web server on Linux with 10,000 worker threads and have the memory usage of the process be ~77mb. The 1024 byte overhead of POSIX-style thread-local storage is now optional; it won't be allocated until the pthread_setspecific/getspecific functions are called. On Windows, the threads that get spawned which are internal to the libc implementation use reserve rather than commit memory, which shaves a few hundred kb. - sigaltstack() is now supported on Windows, however it's currently not able to be used to handle stack overflows, since crash signals are still generated by WIN32. However the crash handler will still switch to the alt stack, which is helpful in environments with tiny threads. - Test binaries are now smaller. Many of the mandatory dependencies of the test runner have been removed. This ensures many programs can do a better job only linking the the thing they're testing. This caused the test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb - long double is no longer used in the implementation details of libc, except in the APIs that define it. The old code that used long double for time (instead of struct timespec) has now been thoroughly removed. - ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing backtraces itself, it'll just print a command you can run on the shell using our new `cosmoaddr2line` program to view the backtrace. - Crash report signal handling now works in a much better way. Instead of terminating the process, it now relies on SA_RESETHAND so that the default SIG_IGN behavior can terminate the process if necessary. - Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
.Pointer = offset,
.hEvent = WSACreateEvent(),
};
if (TransmitFile(oh, ih, uptobytes, 0, &ov, 0, 0)) {
rc = uptobytes;
2020-06-15 14:18:57 +00:00
} else {
BEGIN_BLOCKING_OPERATION;
rc = SendfileBlock(oh, &ov);
END_BLOCKING_OPERATION;
}
if (rc != -1) {
if (opt_in_out_inoffset) {
*opt_in_out_inoffset = offset + rc;
npassert(SetFilePointerEx(ih, pos, 0, SEEK_SET));
} else {
npassert(SetFilePointerEx(ih, offset + rc, 0, SEEK_SET));
}
2020-06-15 14:18:57 +00:00
}
WSACloseEvent(ov.hEvent);
return rc;
2020-06-15 14:18:57 +00:00
}
static ssize_t sys_sendfile_bsd(int outfd, int infd,
int64_t *opt_in_out_inoffset,
size_t uptobytes) {
ssize_t rc;
2020-06-15 14:18:57 +00:00
int64_t offset, sbytes;
if (opt_in_out_inoffset) {
offset = *opt_in_out_inoffset;
2021-02-07 14:11:44 +00:00
} else if ((offset = lseek(infd, 0, SEEK_CUR)) == -1) {
2020-06-15 14:18:57 +00:00
return -1;
}
Undiamond Python headers This change gets the Python codebase into a state where it conforms to the conventions of this codebase. It's now possible to include headers from Python, without worrying about ordering. Python has traditionally solved that problem by "diamonding" everything in Python.h, but that's problematic since it means any change to any Python header invalidates all the build artifacts. Lastly it makes tooling not work. Since it is hard to explain to Emacs when I press C-c C-h to add an import line it shouldn't add the header that actually defines the symbol, and instead do follow the nonstandard Python convention. Progress has been made on letting Python load source code from the zip executable structure via the standard C library APIs. System calss now recognizes zip!FILENAME alternative URIs as equivalent to zip:FILENAME since Python uses colon as its delimiter. Some progress has been made on embedding the notice license terms into the Python object code. This is easier said than done since Python has an extremely complicated ownership story. - Some termios APIs have been added - Implement rewinddir() dirstream API - GetCpuCount() API added to Cosmopolitan Libc - More bugs in Cosmopolitan Libc have been fixed - zipobj.com now has flags for mangling the path - Fixed bug a priori with sendfile() on certain BSDs - Polyfill F_DUPFD and F_DUPFD_CLOEXEC across platforms - FIOCLEX / FIONCLEX now polyfilled for fast O_CLOEXEC changes - APE now supports a hybrid solution to no-self-modify for builds - Many BSD-only magnums added, e.g. O_SEARCH, O_SHLOCK, SF_NODISKIO
2021-08-12 07:42:14 +00:00
if (IsFreebsd()) {
rc = sys_sendfile_freebsd(infd, outfd, offset, uptobytes, 0, &sbytes, 0);
if (rc == -1 && errno == ENOBUFS) errno = ENOMEM;
Undiamond Python headers This change gets the Python codebase into a state where it conforms to the conventions of this codebase. It's now possible to include headers from Python, without worrying about ordering. Python has traditionally solved that problem by "diamonding" everything in Python.h, but that's problematic since it means any change to any Python header invalidates all the build artifacts. Lastly it makes tooling not work. Since it is hard to explain to Emacs when I press C-c C-h to add an import line it shouldn't add the header that actually defines the symbol, and instead do follow the nonstandard Python convention. Progress has been made on letting Python load source code from the zip executable structure via the standard C library APIs. System calss now recognizes zip!FILENAME alternative URIs as equivalent to zip:FILENAME since Python uses colon as its delimiter. Some progress has been made on embedding the notice license terms into the Python object code. This is easier said than done since Python has an extremely complicated ownership story. - Some termios APIs have been added - Implement rewinddir() dirstream API - GetCpuCount() API added to Cosmopolitan Libc - More bugs in Cosmopolitan Libc have been fixed - zipobj.com now has flags for mangling the path - Fixed bug a priori with sendfile() on certain BSDs - Polyfill F_DUPFD and F_DUPFD_CLOEXEC across platforms - FIOCLEX / FIONCLEX now polyfilled for fast O_CLOEXEC changes - APE now supports a hybrid solution to no-self-modify for builds - Many BSD-only magnums added, e.g. O_SEARCH, O_SHLOCK, SF_NODISKIO
2021-08-12 07:42:14 +00:00
} else {
2022-09-13 11:32:29 +00:00
sbytes = uptobytes;
Undiamond Python headers This change gets the Python codebase into a state where it conforms to the conventions of this codebase. It's now possible to include headers from Python, without worrying about ordering. Python has traditionally solved that problem by "diamonding" everything in Python.h, but that's problematic since it means any change to any Python header invalidates all the build artifacts. Lastly it makes tooling not work. Since it is hard to explain to Emacs when I press C-c C-h to add an import line it shouldn't add the header that actually defines the symbol, and instead do follow the nonstandard Python convention. Progress has been made on letting Python load source code from the zip executable structure via the standard C library APIs. System calss now recognizes zip!FILENAME alternative URIs as equivalent to zip:FILENAME since Python uses colon as its delimiter. Some progress has been made on embedding the notice license terms into the Python object code. This is easier said than done since Python has an extremely complicated ownership story. - Some termios APIs have been added - Implement rewinddir() dirstream API - GetCpuCount() API added to Cosmopolitan Libc - More bugs in Cosmopolitan Libc have been fixed - zipobj.com now has flags for mangling the path - Fixed bug a priori with sendfile() on certain BSDs - Polyfill F_DUPFD and F_DUPFD_CLOEXEC across platforms - FIOCLEX / FIONCLEX now polyfilled for fast O_CLOEXEC changes - APE now supports a hybrid solution to no-self-modify for builds - Many BSD-only magnums added, e.g. O_SEARCH, O_SHLOCK, SF_NODISKIO
2021-08-12 07:42:14 +00:00
rc = sys_sendfile_xnu(infd, outfd, offset, &sbytes, 0, 0);
}
if (rc == -1 && errno == ENOTSOCK) errno = EBADF;
Undiamond Python headers This change gets the Python codebase into a state where it conforms to the conventions of this codebase. It's now possible to include headers from Python, without worrying about ordering. Python has traditionally solved that problem by "diamonding" everything in Python.h, but that's problematic since it means any change to any Python header invalidates all the build artifacts. Lastly it makes tooling not work. Since it is hard to explain to Emacs when I press C-c C-h to add an import line it shouldn't add the header that actually defines the symbol, and instead do follow the nonstandard Python convention. Progress has been made on letting Python load source code from the zip executable structure via the standard C library APIs. System calss now recognizes zip!FILENAME alternative URIs as equivalent to zip:FILENAME since Python uses colon as its delimiter. Some progress has been made on embedding the notice license terms into the Python object code. This is easier said than done since Python has an extremely complicated ownership story. - Some termios APIs have been added - Implement rewinddir() dirstream API - GetCpuCount() API added to Cosmopolitan Libc - More bugs in Cosmopolitan Libc have been fixed - zipobj.com now has flags for mangling the path - Fixed bug a priori with sendfile() on certain BSDs - Polyfill F_DUPFD and F_DUPFD_CLOEXEC across platforms - FIOCLEX / FIONCLEX now polyfilled for fast O_CLOEXEC changes - APE now supports a hybrid solution to no-self-modify for builds - Many BSD-only magnums added, e.g. O_SEARCH, O_SHLOCK, SF_NODISKIO
2021-08-12 07:42:14 +00:00
if (rc != -1) {
if (opt_in_out_inoffset) {
*opt_in_out_inoffset += sbytes;
} else {
npassert(lseek(infd, offset + sbytes, SEEK_SET) == offset + sbytes);
2022-09-13 11:32:29 +00:00
}
2020-06-15 14:18:57 +00:00
return sbytes;
} else {
return -1;
}
}
/**
* Transfers data from file to network.
*
* @param outfd needs to be a socket
* @param infd needs to be a file
* @param opt_in_out_inoffset may be specified for pread()-like behavior
* in which case the file position won't be changed; otherwise, this
* shall read from the file pointer which is advanced accordingly
* @param uptobytes is the maximum number of bytes to send; some platforms
* block until everything's sent, whereas others won't; the behavior of
* zero is undefined; this value may overlap the end of file in which
* case what remains is sent; this is silently reduced to `0x7ffff000`
* @return number of bytes transmitted which may be fewer than requested in
* which case caller must be prepared to call sendfile() again
* @raise ESPIPE on Linux RHEL7+ if offset is used but `infd` isn't seekable,
* otherwise this could be EINVAL
* @raise EPIPE on most systems if socket has been shutdown for reading or
* the remote end closed the connection, otherwise this could be EINVAL
* @raise EBADF if `outfd` isn't a valid writeable stream sock descriptor
* @raise EAGAIN if `O_NONBLOCK` is in play and it would have blocked
* @raise EBADF if `infd` isn't a valid readable file descriptor
* @raise EFAULT if `opt_in_out_inoffset` is a bad pointer
* @raise EINVAL if `*opt_in_out_inoffset` is negative
* @raise EOVERFLOW is documented as possible on Linux
* @raise EIO if `infd` had a low-level i/o error
* @raise ENOMEM if we require more vespene gas
* @raise ENOTCONN if `outfd` isn't connected
* @raise ENOSYS on NetBSD and OpenBSD
2020-06-15 14:18:57 +00:00
* @see copy_file_range() for file file
* @see splice() for fd pipe
*/
ssize_t sendfile(int outfd, int infd, int64_t *opt_in_out_inoffset,
2020-06-15 14:18:57 +00:00
size_t uptobytes) {
ssize_t rc;
// We must reduce this due to the uint32_t type conversion on Windows
// which has a maximum of 0x7ffffffe. It also makes sendfile(..., -1)
// less error prone, since Linux may EINVAL if greater than INT64_MAX
uptobytes = MIN(uptobytes, 0x7ffff000);
if (IsAsan() && opt_in_out_inoffset &&
!__asan_is_valid(opt_in_out_inoffset, 8)) {
2022-09-13 11:32:29 +00:00
rc = efault();
} else if (IsLinux()) {
rc = sys_sendfile(outfd, infd, opt_in_out_inoffset, uptobytes);
2020-06-15 14:18:57 +00:00
} else if (IsFreebsd() || IsXnu()) {
rc = sys_sendfile_bsd(outfd, infd, opt_in_out_inoffset, uptobytes);
2020-06-15 14:18:57 +00:00
} else if (IsWindows()) {
rc = sys_sendfile_nt(outfd, infd, opt_in_out_inoffset, uptobytes);
2020-06-15 14:18:57 +00:00
} else {
rc = enosys();
2020-06-15 14:18:57 +00:00
}
STRACE("sendfile(%d, %d, %p, %'zu) → %ld% m", outfd, infd,
DescribeInOutInt64(rc, opt_in_out_inoffset), uptobytes, rc);
2022-09-13 11:32:29 +00:00
return rc;
2020-06-15 14:18:57 +00:00
}
__weak_reference(sendfile, sendfile64);