Implement support for POSIX thread cancellations

This change makes some miracle modifications to the System Five system
call support, which lets us have safe, correct, and atomic handling of
thread cancellations. It all turned out to be cheaper than anticipated
because it wasn't necessary to modify the system call veneers. We were
able to encode the cancellability of each system call into the magnums
found in libc/sysv/syscalls.sh. Since cancellations are so waq, we are
also supporting a lovely Musl Libc mask feature for raising ECANCELED.

Makefile

@@ -61,7 +61,7 @@
 
 SHELL      = build/bootstrap/cocmd.com
 HOSTS     ?= freebsd openbsd netbsd rhel7 rhel5 xnu win10 win10:31336
-MAKEFLAGS += -j --no-builtin-rules
+MAKEFLAGS += --no-builtin-rules
 
 .SUFFIXES:
 .DELETE_ON_ERROR:

libc/calls/__sig2.c

@@ -204,6 +204,7 @@ textwindows int __sig_raise(int sig, int si_code) {
   if (1 <= sig && sig <= 64) {
     if (!__sig_is_masked(sig)) {
       ++__sig_count;
+      // TODO(jart): ucontext_t support
       __sig_handle(false, sig, si_code, 0);
       return 0;
     } else {

libc/calls/clock_nanosleep-nt.c

@@ -34,7 +34,7 @@ textwindows int sys_clock_nanosleep_nt(int clock, int flags,
     for (;;) {
       if (sys_clock_gettime_nt(clock, &now)) return -1;
       if (_timespec_gte(now, abs)) return 0;
-      if (_check_interrupts(false, g_fds.p)) return eintr();
+      if (_check_interrupts(false, g_fds.p)) return -1;
       SleepEx(MIN(__SIG_POLLING_INTERVAL_MS,
                   _timespec_tomillis(_timespec_sub(abs, now))),
               false);
@@ -47,7 +47,7 @@ textwindows int sys_clock_nanosleep_nt(int clock, int flags,
       if (_timespec_gte(now, abs)) return 0;
       if (_check_interrupts(false, g_fds.p)) {
         if (rem) *rem = _timespec_sub(abs, now);
-        return eintr();
+        return -1;
       }
       SleepEx(MIN(__SIG_POLLING_INTERVAL_MS,
                   _timespec_tomillis(_timespec_sub(abs, now))),

libc/calls/clock_nanosleep.c

@@ -80,6 +80,7 @@
  * @raise EINVAL if `flags` has an unrecognized value
  * @raise EINVAL if `req->tv_nsec ∉ [0,1000000000)`
  * @raise ENOSYS on bare metal
+ * @cancellationpoint
  * @returnserrno
  * @norestart
  */

libc/calls/fcntl-sysv.c

@@ -26,7 +26,7 @@
 #include "libc/sysv/consts/o.h"
 #include "libc/sysv/errfuns.h"
 
-int sys_fcntl(int fd, int cmd, uintptr_t arg) {
+int sys_fcntl(int fd, int cmd, uintptr_t arg, int impl(int, int, ...)) {
   int e, rc;
   bool islock;
   if ((islock = cmd == F_GETLK ||  //
@@ -40,12 +40,12 @@ int sys_fcntl(int fd, int cmd, uintptr_t arg) {
     cosmo2flock(arg);
   }
   e = errno;
-  rc = __sys_fcntl(fd, cmd, arg);
+  rc = impl(fd, cmd, arg);
   if (islock) {
     flock2cosmo(arg);
   } else if (rc == -1 && cmd == F_DUPFD_CLOEXEC && errno == EINVAL) {
     errno = e;
-    rc = __fixupnewfd(__sys_fcntl(fd, F_DUPFD, arg), O_CLOEXEC);
+    rc = __fixupnewfd(impl(fd, F_DUPFD, arg), O_CLOEXEC);
   }
   return rc;
 }

libc/calls/fcntl.c

@@ -94,6 +94,7 @@
  * @raise EDEADLK if `cmd` was `F_SETLKW` and waiting would deadlock
  * @raise EMFILE if `cmd` is `F_DUPFD` or `F_DUPFD_CLOEXEC` and
  *     `RLIMIT_NOFILE` would be exceeded
+ * @cancellationpoint when `cmd` is `F_SETLKW`
  * @asyncsignalsafe
  * @restartable
  */
@@ -111,7 +112,11 @@ int fcntl(int fd, int cmd, ...) {
       if (fd < g_fds.n && g_fds.p[fd].kind == kFdZip) {
         rc = _weaken(__zipos_fcntl)(fd, cmd, arg);
       } else if (!IsWindows()) {
-        rc = sys_fcntl(fd, cmd, arg);
+        if (cmd == F_SETLKW) {
+          rc = sys_fcntl(fd, cmd, arg, __sys_fcntl_cp);
+        } else {
+          rc = sys_fcntl(fd, cmd, arg, __sys_fcntl);
+        }
       } else {
         rc = sys_fcntl_nt(fd, cmd, arg);
       }

libc/calls/fdatasync.c

@@ -30,6 +30,7 @@
  * @return 0 on success, or -1 w/ errno
  * @see sync(), fsync(), sync_file_range()
  * @see __nosync to secretly disable
+ * @cancellationpoint
  * @asyncsignalsafe
  */
 int fdatasync(int fd) {

libc/calls/flock.c

@@ -17,10 +17,10 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/calls.h"
-#include "libc/intrin/strace.internal.h"
 #include "libc/calls/syscall-nt.internal.h"
 #include "libc/calls/syscall-sysv.internal.h"
 #include "libc/dce.h"
+#include "libc/intrin/strace.internal.h"
 
 /**
  * Acquires lock on file.
@@ -30,6 +30,7 @@
  * @param op can have LOCK_{SH,EX,NB,UN} for shared, exclusive,
  *     non-blocking, and unlocking
  * @return 0 on success, or -1 w/ errno
+ * @cancellationpoint
  * @restartable
  */
 int flock(int fd, int op) {

libc/calls/fstatfs.c

@@ -18,16 +18,17 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/calls.h"
 #include "libc/calls/internal.h"
-#include "libc/intrin/strace.internal.h"
 #include "libc/calls/struct/statfs-meta.internal.h"
 #include "libc/calls/struct/statfs.internal.h"
 #include "libc/dce.h"
+#include "libc/intrin/strace.internal.h"
 #include "libc/runtime/stack.h"
 #include "libc/sysv/errfuns.h"
 
 /**
  * Returns information about filesystem.
  * @return 0 on success, or -1 w/ errno
+ * @cancellationpoint
  */
 int fstatfs(int fd, struct statfs *sf) {
   int rc;

libc/calls/fsync.c

@@ -30,6 +30,7 @@
  * @return 0 on success, or -1 w/ errno
  * @see fdatasync(), sync_file_range()
  * @see __nosync to secretly disable
+ * @cancellationpoint
  * @asyncsignalsafe
  */
 int fsync(int fd) {

libc/calls/ftruncate.c

@@ -56,7 +56,7 @@
  * @raise EINVAL if `fd` is a non-file, e.g. pipe, socket
  * @raise EINVAL if `fd` wasn't opened in a writeable mode
  * @raise ENOSYS on bare metal
- * @see truncate()
+ * @cancellationpoint
  * @asyncsignalsafe
  * @threadsafe
  */

libc/calls/internal.h

@@ -43,7 +43,7 @@ forceinline size_t _clampio(size_t size) {
 }
 
 int sys_close_nt(struct Fd *, int) hidden;
-bool _check_interrupts(bool, struct Fd *) hidden;
+int _check_interrupts(bool, struct Fd *) hidden;
 int sys_openat_metal(int, const char *, int, unsigned);
 
 COSMOPOLITAN_C_END_

libc/calls/interrupts-nt.c

@@ -18,24 +18,33 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/assert.h"
 #include "libc/calls/calls.h"
+#include "libc/calls/internal.h"
 #include "libc/calls/sig.internal.h"
 #include "libc/calls/state.internal.h"
 #include "libc/calls/struct/fd.internal.h"
 #include "libc/calls/struct/sigaction.h"
 #include "libc/calls/syscall_support-nt.internal.h"
 #include "libc/dce.h"
+#include "libc/errno.h"
 #include "libc/intrin/lockcmpxchgp.h"
 #include "libc/intrin/strace.internal.h"
 #include "libc/intrin/weaken.h"
+#include "libc/sysv/errfuns.h"
+#include "libc/thread/thread.h"
 #include "libc/thread/tls.h"
 
-textwindows bool _check_interrupts(bool restartable, struct Fd *fd) {
-  bool res;
+textwindows int _check_interrupts(bool restartable, struct Fd *fd) {
+  int rc;
+  if (_weaken(pthread_testcancel_np) &&
+      (rc = _weaken(pthread_testcancel_np)())) {
+    errno = rc;
+    return -1;
+  }
   if (_weaken(_check_sigalrm)) _weaken(_check_sigalrm)();
   if (!__tls_enabled || !(__get_tls()->tib_flags & TIB_FLAG_TIME_CRITICAL)) {
     if (_weaken(_check_sigchld)) _weaken(_check_sigchld)();
     if (fd && _weaken(_check_sigwinch)) _weaken(_check_sigwinch)(fd);
   }
-  res = _weaken(__sig_check) && _weaken(__sig_check)(restartable);
-  return res;
+  if (_weaken(__sig_check) && _weaken(__sig_check)(restartable)) return eintr();
+  return 0;
 }

libc/calls/nanosleep.c

@@ -32,6 +32,7 @@
  * @raise EFAULT if `req` is NULL or `req` / `rem` is a bad pointer
  * @raise ENOSYS on bare metal
  * @see clock_nanosleep()
+ * @cancellationpoint
  * @norestart
  */
 int nanosleep(const struct timespec *req, struct timespec *rem) {

libc/calls/ntcontext2linux.c

@@ -21,6 +21,8 @@
 #include "libc/nt/struct/context.h"
 #include "libc/str/str.h"
 
+// TODO(jart): uc_sigmask support
+
 privileged void _ntcontext2linux(ucontext_t *ctx, const struct NtContext *cr) {
   if (!cr) return;
   ctx->uc_mcontext.eflags = cr->EFlags;

libc/calls/open.c

@@ -29,6 +29,7 @@
  * @param file specifies filesystem path to open
  * @return file descriptor, or -1 w/ errno
  * @see openat() for further documentation
+ * @cancellationpoint
  * @asyncsignalsafe
  * @restartable
  * @threadsafe

libc/calls/openat.c

@@ -141,6 +141,7 @@
  * @raise ELOOP if `flags` had `O_NOFOLLOW` and `file` is a symbolic link
  * @raise ELOOP if a loop was detected resolving components of `file`
  * @raise EISDIR if writing is requested and `file` names a directory
+ * @cancellationpoint
  * @asyncsignalsafe
  * @restartable
  * @threadsafe

libc/calls/pause-nt.c

@@ -18,8 +18,9 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/internal.h"
 #include "libc/calls/sig.internal.h"
-#include "libc/intrin/strace.internal.h"
 #include "libc/calls/syscall_support-nt.internal.h"
+#include "libc/errno.h"
+#include "libc/intrin/strace.internal.h"
 #include "libc/nt/errors.h"
 #include "libc/nt/synchronization.h"
 #include "libc/sysv/errfuns.h"
@@ -31,7 +32,7 @@ textwindows int sys_pause_nt(void) {
   for (;;) {
 
     if (_check_interrupts(false, g_fds.p)) {
-      return eintr();
+      return -1;
     }
 
     if (SleepEx(__SIG_POLLING_INTERVAL_MS, true) == kNtWaitIoCompletion) {

libc/calls/pause.c

@@ -17,9 +17,9 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/calls.h"
-#include "libc/intrin/strace.internal.h"
 #include "libc/calls/syscall_support-nt.internal.h"
 #include "libc/dce.h"
+#include "libc/intrin/strace.internal.h"
 #include "libc/sock/internal.h"
 
 /**
@@ -37,6 +37,7 @@
  * However this has a tinier footprint and better logging.
  *
  * @return -1 w/ errno set to EINTR
+ * @cancellationpoint
  * @see sigsuspend()
  * @norestart
  */

libc/calls/poll-nt.c

@@ -21,7 +21,6 @@
 #include "libc/calls/sig.internal.h"
 #include "libc/calls/state.internal.h"
 #include "libc/calls/struct/sigaction.h"
-#include "libc/errno.h"
 #include "libc/intrin/bits.h"
 #include "libc/intrin/strace.internal.h"
 #include "libc/macros.internal.h"
@@ -64,8 +63,7 @@ textwindows int sys_poll_nt(struct pollfd *fds, uint64_t nfds, uint64_t *ms,
   if (sigmask) {
     __sig_mask(SIG_SETMASK, sigmask, &oldmask);
   }
-  if (_check_interrupts(false, g_fds.p)) {
-    rc = eintr();
+  if ((rc = _check_interrupts(false, g_fds.p))) {
     goto ReturnPath;
   }
 
@@ -190,8 +188,7 @@ textwindows int sys_poll_nt(struct pollfd *fds, uint64_t nfds, uint64_t *ms,
     }
     // otherwise loop limitlessly for timeout to elapse while
     // checking for signal delivery interrupts, along the way
-    if (_check_interrupts(false, g_fds.p)) {
-      rc = eintr();
+    if ((rc = _check_interrupts(false, g_fds.p))) {
       goto ReturnPath;
     }
   }

libc/calls/poll.c

@@ -55,6 +55,7 @@
  * @return fds[𝑖].revents is always zero initializaed and then will
  *     be populated with POLL{IN,OUT,PRI,HUP,ERR,NVAL} if something
  *     was determined about the file descriptor
+ * @cancellationpoint
  * @asyncsignalsafe
  * @threadsafe
  * @norestart

libc/calls/ppoll.c

@@ -50,6 +50,7 @@
  *
  * @param timeout if null will block indefinitely
  * @param sigmask may be null in which case no mask change happens
+ * @cancellationpoint
  * @asyncsignalsafe
  * @threadsafe
  * @norestart

libc/calls/pread.c

@@ -48,6 +48,7 @@
  * @raise EIO if a complicated i/o error happened
  * @raise EINTR if signal was delivered instead
  * @see pwrite(), write()
+ * @cancellationpoint
  * @asyncsignalsafe
  * @threadsafe
  * @vforksafe

libc/calls/preadv.c

@@ -108,6 +108,7 @@ static ssize_t Preadv(int fd, struct iovec *iov, int iovlen, int64_t off) {
  * Reads with maximum generality.
  *
  * @return number of bytes actually read, or -1 w/ errno
+ * @cancellationpoint
  * @asyncsignalsafe
  * @vforksafe
  */

libc/calls/pwrite.c

@@ -42,6 +42,7 @@
  * @return [1..size] bytes on success, or -1 w/ errno; noting zero is
  *     impossible unless size was passed as zero to do an error check
  * @see pread(), write()
+ * @cancellationpoint
  * @asyncsignalsafe
  * @threadsafe
  * @vforksafe

libc/calls/pwritev.c

@@ -114,6 +114,7 @@ static ssize_t Pwritev(int fd, const struct iovec *iov, int iovlen,
  * call using pwrite().
  *
  * @return number of bytes actually sent, or -1 w/ errno
+ * @cancellationpoint
  * @asyncsignalsafe
  * @vforksafe
  */

libc/calls/read-nt.c

@@ -53,7 +53,7 @@ static textwindows ssize_t sys_read_nt_impl(struct Fd *fd, void *data,
       }
       if (_check_interrupts(true, g_fds.p)) {
         POLLTRACE("sys_read_nt interrupted");
-        return eintr();
+        return -1;
       }
     }
     POLLTRACE("sys_read_nt ready to read");
@@ -94,7 +94,7 @@ textwindows ssize_t sys_read_nt(struct Fd *fd, const struct iovec *iov,
   uint32_t size;
   size_t i, total;
   if (opt_offset < -1) return einval();
-  if (_check_interrupts(true, fd)) return eintr();
+  if (_check_interrupts(true, fd)) return -1;
   while (iovlen && !iov[0].iov_len) iov++, iovlen--;
   if (iovlen) {
     for (total = i = 0; i < iovlen; ++i) {

libc/calls/read.c

@@ -56,6 +56,7 @@
  *     or `SO_RCVTIMEO` is in play and the time interval elapsed
  * @raise ENOBUFS is specified by POSIX
  * @raise ENXIO is specified by POSIX
+ * @cancellationpoint
  * @asyncsignalsafe
  * @restartable
  * @vforksafe

libc/calls/readv.c

@@ -43,6 +43,7 @@
  * performance boost in the case of a single small iovec.
  *
  * @return number of bytes actually read, or -1 w/ errno
+ * @cancellationpoint
  * @restartable
  */
 ssize_t readv(int fd, const struct iovec *iov, int iovlen) {

libc/calls/sigblockall.c

@@ -17,15 +17,10 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/struct/sigset.h"
+#include "libc/str/str.h"
 
-/**
- * Blocks all signals without strace logging.
- *
- * @param neu is new signal mask for process
- * @return old signal mask
- */
 sigset_t _sigblockall(void) {
   sigset_t ss;
-  sigfillset(&ss);
+  memset(&ss, -1, sizeof(ss));
   return _sigsetmask(ss);
 }