Make improvements

- We now serialize the file descriptor table when spawning / executing
  processes on Windows. This means you can now inherit more stuff than
  just standard i/o. It's needed by bash, which duplicates the console
  to file descriptor #255. We also now do a better job serializing the
  environment variables, so you're less likely to encounter E2BIG when
  using your bash shell. We also no longer coerce environ to uppercase

- execve() on Windows now remotely controls its parent process to make
  them spawn a replacement for itself. Then it'll be able to terminate
  immediately once the spawn succeeds, without having to linger around
  for the lifetime as a shell process for proxying the exit code. When
  process worker thread running in the parent sees the child die, it's
  given a handle to the new child, to replace it in the process table.

- execve() and posix_spawn() on Windows will now provide CreateProcess
  an explicit handle list. This allows us to remove handle locks which
  enables better fork/spawn concurrency, with seriously correct thread
  safety. Other codebases like Go use the same technique. On the other
  hand fork() still favors the conventional WIN32 inheritence approach
  which can be a little bit messy, but is *controlled* by guaranteeing
  perfectly clean slates at both the spawning and execution boundaries

- sigset_t is now 64 bits. Having it be 128 bits was a mistake because
  there's no reason to use that and it's only supported by FreeBSD. By
  using the system word size, signal mask manipulation on Windows goes
  very fast. Furthermore @asyncsignalsafe funcs have been rewritten on
  Windows to take advantage of signal masking, now that it's much more
  pleasant to use.

- All the overlapped i/o code on Windows has been rewritten for pretty
  good signal and cancelation safety. We're now able to ensure overlap
  data structures are cleaned up so long as you don't longjmp() out of
  out of a signal handler that interrupted an i/o operation. Latencies
  are also improved thanks to the removal of lots of "busy wait" code.
  Waits should be optimal for everything except poll(), which shall be
  the last and final demon we slay in the win32 i/o horror show.

- getrusage() on Windows is now able to report RUSAGE_CHILDREN as well
  as RUSAGE_SELF, thanks to aggregation in the process manager thread.
This commit is contained in:
Justine Tunney 2023-10-08 05:36:18 -07:00
parent af7cb3c82f
commit 791f79fcb3
No known key found for this signature in database
GPG key ID: BE714B4575D6E328
382 changed files with 4008 additions and 4511 deletions

View file

@ -16,105 +16,46 @@
TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
PERFORMANCE OF THIS SOFTWARE.
*/
#include "libc/calls/calls.h"
#include "libc/assert.h"
#include "libc/calls/internal.h"
#include "libc/calls/struct/fd.internal.h"
#include "libc/calls/struct/sigaction.internal.h"
#include "libc/fmt/itoa.h"
#include "libc/intrin/dll.h"
#include "libc/intrin/handlock.internal.h"
#include "libc/intrin/weaken.h"
#include "libc/nt/accounting.h"
#include "libc/nt/console.h"
#include "libc/calls/struct/sigset.internal.h"
#include "libc/calls/syscall-nt.internal.h"
#include "libc/errno.h"
#include "libc/intrin/kprintf.h"
#include "libc/mem/mem.h"
#include "libc/nt/enum/processaccess.h"
#include "libc/nt/enum/startf.h"
#include "libc/nt/enum/status.h"
#include "libc/nt/enum/wait.h"
#include "libc/nt/errors.h"
#include "libc/nt/files.h"
#include "libc/nt/memory.h"
#include "libc/nt/process.h"
#include "libc/nt/runtime.h"
#include "libc/nt/struct/msg.h"
#include "libc/nt/struct/processinformation.h"
#include "libc/nt/struct/startupinfo.h"
#include "libc/nt/synchronization.h"
#include "libc/nt/thread.h"
#include "libc/nt/thunk/msabi.h"
#include "libc/proc/describefds.internal.h"
#include "libc/proc/ntspawn.h"
#include "libc/proc/proc.internal.h"
#include "libc/runtime/memtrack.internal.h"
#include "libc/runtime/runtime.h"
#include "libc/sock/sock.h"
#include "libc/str/str.h"
#include "libc/sysv/consts/o.h"
#include "libc/sysv/consts/sig.h"
#include "libc/sysv/errfuns.h"
#include "libc/thread/itimer.internal.h"
#include "libc/thread/posixthread.internal.h"
#include "libc/thread/tls.h"
#include "libc/thread/thread.h"
#ifdef __x86_64__
#define keywords textwindows dontinstrument
textwindows int sys_execve_nt(const char *program, char *const argv[],
char *const envp[]) {
// clang-format off
__msabi extern typeof(CloseHandle) *const __imp_CloseHandle;
__msabi extern typeof(DuplicateHandle) *const __imp_DuplicateHandle;
__msabi extern typeof(GenerateConsoleCtrlEvent) *const __imp_GenerateConsoleCtrlEvent;
__msabi extern typeof(GetCurrentThreadId) *const __imp_GetCurrentThreadId;
__msabi extern typeof(GetExitCodeProcess) *const __imp_GetExitCodeProcess;
__msabi extern typeof(GetLastError) *const __imp_GetLastError;
__msabi extern typeof(OpenThread) *const __imp_OpenThread;
__msabi extern typeof(SetConsoleCtrlHandler) *const __imp_SetConsoleCtrlHandler;
__msabi extern typeof(SetHandleInformation) *const __imp_SetHandleInformation;
__msabi extern typeof(TerminateThread) *const __imp_TerminateThread;
__msabi extern typeof(UnmapViewOfFile) *const __imp_UnmapViewOfFile;
__msabi extern typeof(WaitForSingleObject) *const __imp_WaitForSingleObject;
// clang-format on
// execve() needs to be @asyncsignalsafe
sigset_t m = __sig_block();
_pthread_lock();
extern long __klog_handle;
static void sys_execve_nt_relay(intptr_t, long, long, long);
void __stack_call(intptr_t, long, long, long,
void (*)(intptr_t, intptr_t, long, long),
intptr_t) wontreturn;
static keywords void PurgeHandle(intptr_t h) {
if (!h) return;
if (h == -1) return;
__imp_CloseHandle(h);
}
static keywords void PurgeThread(intptr_t h) {
if (h && h != -1) {
__imp_TerminateThread(h, SIGKILL);
__imp_CloseHandle(h);
}
}
static keywords void sys_execve_inherit(int64_t hands[3], bool32 bInherit) {
for (int i = 0; i < 3; ++i) {
if (hands[i] != -1) {
__imp_SetHandleInformation(hands[i], kNtHandleFlagInherit, bInherit);
}
}
}
keywords int sys_execve_nt(const char *program, char *const argv[],
char *const envp[]) {
size_t i;
__hand_lock();
pthread_spin_lock(&_pthread_lock);
// pass bitmask telling child which fds are sockets
int bits;
char buf[32], *v = 0;
if (_weaken(socket)) {
for (bits = i = 0; i < 3; ++i) {
if (g_fds.p[i].kind == kFdSocket) {
bits |= 1 << i;
}
}
FormatInt32(stpcpy(buf, "__STDIO_SOCKETS="), bits);
v = buf;
// new process should be a child of our parent
int64_t hParentProcess;
int ppid = sys_getppid_nt();
if (!(hParentProcess = OpenProcess(
kNtProcessDupHandle | kNtProcessCreateProcess, false, ppid))) {
_pthread_unlock();
__sig_unblock(m);
return -1;
}
// define stdio handles for the spawned subprocess
@ -122,96 +63,55 @@ keywords int sys_execve_nt(const char *program, char *const argv[],
.cb = sizeof(struct NtStartupInfo),
.dwFlags = kNtStartfUsestdhandles,
};
for (i = 0; i <= 2; ++i) {
if (g_fds.p[i].kind != kFdEmpty && //
!(g_fds.p[i].flags & O_CLOEXEC)) {
si.stdiofds[i] = g_fds.p[i].handle;
for (int fd = 0; fd < 3; ++fd) {
if (!__is_cloexec(g_fds.p + fd)) {
si.stdiofds[fd] = g_fds.p[fd].handle;
} else {
si.stdiofds[i] = -1;
si.stdiofds[fd] = -1;
}
}
// pass serialized file descriptor table in environment
char *fdspec;
int64_t *lpExplicitHandles;
uint32_t dwExplicitHandleCount;
if (!(fdspec = __describe_fds(g_fds.p, g_fds.n, &si, hParentProcess,
&lpExplicitHandles, &dwExplicitHandleCount))) {
CloseHandle(hParentProcess);
_pthread_unlock();
__sig_unblock(m);
return -1;
}
// launch the process
struct NtProcessInformation pi;
sys_execve_inherit(si.stdiofds, true);
int rc = ntspawn(program, argv, envp, v, 0, 0, true, 0, 0, &si, &pi);
int rc =
ntspawn(program, argv, envp, (char *[]){fdspec, 0}, 0, 0, hParentProcess,
lpExplicitHandles, dwExplicitHandleCount, &si, &pi);
__undescribe_fds(hParentProcess, lpExplicitHandles, dwExplicitHandleCount);
if (rc == -1) {
sys_execve_inherit(si.stdiofds, false);
__hand_unlock();
if (__imp_GetLastError() == kNtErrorSharingViolation) {
free(fdspec);
CloseHandle(hParentProcess);
__undescribe_fds(hParentProcess, lpExplicitHandles, dwExplicitHandleCount);
_pthread_unlock();
__sig_unblock(m);
if (GetLastError() == kNtErrorSharingViolation) {
return etxtbsy();
} else {
return -1;
}
}
PurgeHandle(pi.hThread);
// kill siblings
struct Dll *e;
for (e = dll_first(_pthread_list); e; e = dll_next(_pthread_list, e)) {
struct PosixThread *pt = POSIXTHREAD_CONTAINER(e);
if ((pthread_t)pt == __get_tls()->tib_pthread) continue;
PurgeThread(pt->tib->tib_syshand);
PurgeHandle(pt->semaphore);
// give child to libc/proc/proc.c worker thread in parent
int64_t handle;
if (!DuplicateHandle(GetCurrentProcess(), pi.hProcess, hParentProcess,
&handle, 0, false, kNtDuplicateSameAccess)) {
kprintf("failed to duplicate handle from %P into %d due to %s\n", ppid,
strerror(GetLastError()));
_Exit(1);
}
if (_weaken(__itimer)) {
PurgeThread(_weaken(__itimer)->thread);
}
if (_weaken(__proc)) {
PurgeThread(_weaken(__proc)->thread);
PurgeHandle(_weaken(__proc)->onstart);
}
// retreat to original win32-provided stack memory
__ftrace = 0;
__stack_call(pi.hProcess, 0, 0, 0, sys_execve_nt_relay, __oldstack);
}
// child is in same process group so wait for it to get killed by this
__msabi static keywords bool32 sys_execve_nt_event(uint32_t dwCtrlType) {
return true; // tell win32 we handled signal
}
// this function runs on the original tiny stack that windows gave us
// we need to keep the original process alive simply to pass an int32
// so we unmap all memory to avoid getting a double whammy after fork
static keywords void sys_execve_nt_relay(intptr_t h, long b, long c, long d) {
uint32_t i, dwExitCode;
// close more handles
__imp_SetConsoleCtrlHandler((void *)sys_execve_nt_event, 1);
for (i = 0; i < g_fds.n; ++i) {
if (g_fds.p[i].kind != kFdEmpty) {
PurgeHandle(g_fds.p[i].handle);
if (g_fds.p[i].kind == kFdConsole) {
PurgeHandle(g_fds.p[i].extra);
}
}
}
// free all the memory mmap created
for (i = 0; i < _mmi.i; ++i) {
__imp_UnmapViewOfFile((void *)((uintptr_t)_mmi.p[i].x << 16));
PurgeHandle(_mmi.p[i].h);
}
// wait for process to terminate
//
// WaitForSingleObject can return kNtWaitAbandoned which MSDN
// describes as a "sort of" successful status which indicates
// someone else didn't free a mutex and you should check that
// persistent resources haven't been left corrupted. not sure
// what those resources would be for process objects, however
// this status has actually been observed when waiting on 'em
do {
dwExitCode = 255;
if (__imp_WaitForSingleObject(h, -1) != kNtWaitFailed) {
__imp_GetExitCodeProcess(h, &dwExitCode);
}
} while (dwExitCode == kNtStillActive);
// propagate child exit status to parent
TerminateThisProcess(dwExitCode);
unassert(!(handle & 0xFFFFFFFFFF000000));
TerminateThisProcess(0x23000000u | handle);
}
#endif /* __x86_64__ */