overhaul pthread cancellation

this patch improves the correctness, simplicity, and size of
cancellation-related code. modulo any small errors, it should now be
completely conformant, safe, and resource-leak free.

the notion of entering and exiting cancellation-point context has been
completely eliminated and replaced with alternative syscall assembly
code for cancellable syscalls. the assembly is responsible for setting
up execution context information (stack pointer and address of the
syscall instruction) which the cancellation signal handler can use to
determine whether the interrupted code was in a cancellable state.

these changes eliminate race conditions in the previous generation of
cancellation handling code (whereby a cancellation request received
just prior to the syscall would not be processed, leaving the syscall
to block, potentially indefinitely), and remedy an issue where
non-cancellable syscalls made from signal handlers became cancellable
if the signal handler interrupted a cancellation point.

x86_64 asm is untested and may need a second try to get it right.

src/thread/__timedwait.c

@@ -2,7 +2,7 @@
 #include <errno.h>
 #include "futex.h"
 #include "syscall.h"
-#include <stdio.h>
+
 int __timedwait(volatile int *addr, int val, clockid_t clk, const struct timespec *at, int priv)
 {
 	int r;

src/thread/__timedwait_cp.c

@@ -0,0 +1,23 @@
+#include <time.h>
+#include <errno.h>
+#include "futex.h"
+#include "syscall.h"
+
+int __timedwait_cp(volatile int *addr, int val, clockid_t clk, const struct timespec *at, int priv)
+{
+	int r;
+	struct timespec to;
+	if (at) {
+		clock_gettime(clk, &to);
+		to.tv_sec = at->tv_sec - to.tv_sec;
+		if ((to.tv_nsec = at->tv_nsec - to.tv_nsec) < 0) {
+			to.tv_sec--;
+			to.tv_nsec += 1000000000;
+		}
+		if (to.tv_sec < 0) return ETIMEDOUT;
+	}
+	if (priv) priv = 128; priv=0;
+	r = -__syscall_cp(SYS_futex, (long)addr, FUTEX_WAIT | priv, val, at ? (long)&to : 0);
+	if (r == ETIMEDOUT || r == EINTR) return r;
+	return 0;
+}

src/thread/cancel_dummy.c

@@ -0,0 +1,8 @@
+#include "pthread_impl.h"
+
+static long sccp(long nr, long u, long v, long w, long x, long y, long z)
+{
+	return (__syscall)(nr, u, v, w, x, y, z);
+}
+
+weak_alias(sccp, __syscall_cp);

src/thread/cancel_impl.c

@@ -0,0 +1,70 @@
+#include "pthread_impl.h"
+
+long __syscall_cp_asm(volatile void *, long, long, long, long, long, long, long);
+
+long (__syscall_cp)(long nr, long u, long v, long w, long x, long y, long z)
+{
+	pthread_t self;
+	uintptr_t old_sp, old_ip;
+	long r;
+
+	if (!libc.lock || (self = __pthread_self())->canceldisable)
+		return __syscall(nr, u, v, w, x, y, z);
+
+	old_sp = self->cp_sp;
+	old_ip = self->cp_ip;
+	self->cp_sp = 0;
+	self->cp_ip = 0;
+	r = __syscall_cp_asm(&self->cp_sp, nr, u, v, w, x, y, z);
+	self->cp_sp = old_sp;
+	self->cp_ip = old_ip;
+	if (r == -EINTR && self->cancel) pthread_exit(PTHREAD_CANCELED);
+	return r;
+}
+
+static void cancel_handler(int sig, siginfo_t *si, void *ctx)
+{
+	pthread_t self = __pthread_self();
+	ucontext_t *uc = ctx;
+	uintptr_t sp = ((uintptr_t *)&uc->uc_mcontext)[CANCEL_REG_SP];
+	uintptr_t ip = ((uintptr_t *)&uc->uc_mcontext)[CANCEL_REG_IP];
+
+	if (!self->cancel || self->canceldisable) return;
+
+	if (self->cancelasync) pthread_exit(PTHREAD_CANCELED);
+
+	if (sp != self->cp_sp) {
+		if (!sp) return;
+		sigaddset(&uc->uc_sigmask, SIGCANCEL);
+		__syscall(SYS_tgkill, self->pid, self->tid, SIGCANCEL);
+		return;
+	}
+
+	if (ip <= self->cp_ip) pthread_exit(PTHREAD_CANCELED);
+}
+
+static void testcancel()
+{
+	pthread_t self = __pthread_self();
+	if (self->cancel && !self->canceldisable)
+		pthread_exit(PTHREAD_CANCELED);
+}
+
+static void init_cancellation()
+{
+	struct sigaction sa = {
+		.sa_flags = SA_SIGINFO | SA_RESTART,
+		.sa_sigaction = cancel_handler
+	};
+	sigfillset(&sa.sa_mask);
+	__libc_sigaction(SIGCANCEL, &sa, 0);
+	libc.testcancel = testcancel;
+}
+
+int pthread_cancel(pthread_t t)
+{
+	static pthread_once_t once;
+	pthread_once(&once, init_cancellation);
+	a_store(&t->cancel, 1);
+	return pthread_kill(t, SIGCANCEL);
+}

src/thread/i386/syscall_cp.s

@@ -0,0 +1,36 @@
+.text
+.global __syscall_cp_asm
+.type   __syscall_cp_asm,%function
+__syscall_cp_asm:
+	pushl %ebx
+	pushl %esi
+	pushl %edi
+	pushl %ebp
+	leal 20(%esp),%ebp
+	call 1f
+1:	popl %eax
+	movl (%ebp),%ecx
+	addl $[1f-1b],%eax
+	movl %eax,4(%ecx)
+	movl %esp,(%ecx)
+	movl 8(%ecx),%eax
+	testl %eax,%eax
+	jnz 2f
+	movl 4(%ebp),%eax
+	movl 8(%ebp),%ebx
+	movl 12(%ebp),%ecx
+	movl 16(%ebp),%edx
+	movl 20(%ebp),%esi
+	movl 24(%ebp),%edi
+	movl 28(%ebp),%ebp
+1:	int $128
+	popl %ebp
+	popl %edi
+	popl %esi
+	popl %ebx
+	ret
+2:	xorl %eax,%eax
+	movl %eax,4(%ecx)
+	movl %eax,(%ecx)
+	pushl $-1
+	call pthread_exit

src/thread/pthread_cancel.c

@@ -1,7 +0,0 @@
-#include "pthread_impl.h"
-
-int pthread_cancel(pthread_t t)
-{
-	a_store(&t->cancel, 1);
-	return pthread_kill(t, SIGCANCEL);
-}

src/thread/pthread_cond_timedwait.c

@@ -8,22 +8,19 @@ static void relock(void *m)
 int pthread_cond_timedwait(pthread_cond_t *c, pthread_mutex_t *m, const struct timespec *ts)
 {
 	int r, e=0;
-	CANCELPT_BEGIN;
-	CANCELPT_END;
+
+	pthread_testcancel();
 
 	pthread_cleanup_push(relock, m);
 	c->_c_block = 1;
 	if ((r=pthread_mutex_unlock(m))) return r;
 
-	CANCELPT_BEGIN;
-	do e = __timedwait(&c->_c_block, 1, c->_c_clock, ts, 0);
+	do e = __timedwait_cp(&c->_c_block, 1, c->_c_clock, ts, 0);
 	while (e == EINTR);
-	CANCELPT_END;
 
 	pthread_cleanup_pop(0);
 	if ((r=pthread_mutex_lock(m))) return r;
 
-	CANCELPT_BEGIN;
-	CANCELPT_END;
+	pthread_testcancel();
 	return e;
 }

src/thread/pthread_create.c

@@ -19,6 +19,7 @@ weak_alias(dummy_1, __pthread_tsd_run_dtors);
 void __pthread_unwind_next(struct __ptcb *cb)
 {
 	pthread_t self;
+	int n;
 
 	if (cb->__next) longjmp((void *)cb->__next->__jb, 1);
 
@@ -31,8 +32,9 @@ void __pthread_unwind_next(struct __ptcb *cb)
 	/* Mark this thread dead before decrementing count */
 	self->dead = 1;
 
-	if (!a_fetch_add(&libc.threads_minus_1, -1))
-		exit(0);
+	do n = libc.threads_minus_1;
+	while (n && a_cas(&libc.threads_minus_1, n, n-1)!=n);
+	if (!n) exit(0);
 
 	if (self->detached && self->map_base) {
 		__syscall(SYS_rt_sigprocmask, SIG_BLOCK, (long)(uint64_t[1]){-1},0,8);
@@ -42,43 +44,16 @@ void __pthread_unwind_next(struct __ptcb *cb)
 	__syscall(SYS_exit, 0);
 }
 
-static void docancel(struct pthread *self)
-{
-	struct __ptcb cb = { .__next = self->cancelbuf };
-	self->canceldisable = 1;
-	self->cancelasync = 0;
-	__pthread_unwind_next(&cb);
-}
-
-static void cancel_handler(int sig, siginfo_t *si, void *ctx)
-{
-	struct pthread *self = __pthread_self();
-	if (self->cancel && !self->canceldisable &&
-	    (self->cancelasync || (self->cancelpoint==1 && PC_AT_SYS(ctx))))
-		docancel(self);
-}
-
-static void cancelpt(int x)
-{
-	struct pthread *self = __pthread_self();
-	if ((self->cancelpoint+=x)==1 && self->cancel
-		&& x<2U && !self->canceldisable) docancel(self);
-}
-
 static void init_threads()
 {
-	struct sigaction sa = { .sa_flags = SA_SIGINFO | SA_RESTART };
+	sigset_t set;
 	libc.lock = __lock;
 	libc.lockfile = __lockfile;
-	libc.cancelpt = cancelpt;
 
-	sigemptyset(&sa.sa_mask);
-	sa.sa_sigaction = cancel_handler;
-	__libc_sigaction(SIGCANCEL, &sa, 0);
-
-	sigaddset(&sa.sa_mask, SIGSYSCALL);
-	sigaddset(&sa.sa_mask, SIGCANCEL);
-	__libc_sigprocmask(SIG_UNBLOCK, &sa.sa_mask, 0);
+	sigemptyset(&set);
+	sigaddset(&set, SIGSYSCALL);
+	sigaddset(&set, SIGCANCEL);
+	__libc_sigprocmask(SIG_UNBLOCK, &set, 0);
 }
 
 static int start(void *p)
@@ -159,6 +134,9 @@ int pthread_create(pthread_t *res, const pthread_attr_t *attr, void *(*entry)(vo
 void pthread_exit(void *result)
 {
 	struct pthread *self = pthread_self();
+	struct __ptcb cb = { .__next = self->cancelbuf };
 	self->result = result;
-	docancel(self);
+	self->canceldisable = 1;
+	self->cancelasync = 0;
+	__pthread_unwind_next(&cb);
 }

src/thread/pthread_join.c

@@ -3,9 +3,7 @@
 int pthread_join(pthread_t t, void **res)
 {
 	int tmp = t->tid;
-	CANCELPT_BEGIN;
-	if (tmp) __wait(&t->tid, 0, tmp, 1);
-	CANCELPT_END;
+	if (tmp) __timedwait_cp(&t->tid, tmp, 0, 0, 1);
 	if (res) *res = t->result;
 	if (t->map_base) munmap(t->map_base, t->map_size);
 	return 0;

src/thread/pthread_testcancel.c

@@ -2,6 +2,5 @@
 
 void pthread_testcancel()
 {
-	CANCELPT_BEGIN;
-	CANCELPT_END;
+	if (libc.testcancel) libc.testcancel();
 }

src/thread/sem_timedwait.c

@@ -21,19 +21,16 @@ int sem_timedwait(sem_t *sem, const struct timespec *at)
 	a_inc(sem->__val+1);
 	pthread_cleanup_push(cleanup, sem->__val+1)
 
-	CANCELPT_BEGIN;
 	for (;;) {
 		r = 0;
 		if (!sem_trywait(sem)) break;
-		r = __timedwait(sem->__val, 0, CLOCK_REALTIME, at, 0);
+		r = __timedwait_cp(sem->__val, 0, CLOCK_REALTIME, at, 0);
 		if (r) {
 			errno = r;
 			r = -1;
 			break;
 		}
-		CANCELPT_TRY;
 	}
-	CANCELPT_END;
 
 	pthread_cleanup_pop(1);
 

src/thread/x86_64/syscall_cp.s

@@ -0,0 +1,24 @@
+.text
+.global __syscall_cp_asm
+.type   __syscall_cp_asm,%function
+__syscall_cp_asm:
+	lea 1f(%rip),%rax
+	mov %rax,8(%rdi)
+	mov %rsp,(%rdi)
+	mov 16(%rdi),%eax
+	test %eax,%eax
+	jnz 2f
+	mov %rsi,%rax
+	mov %rdx,%rdi
+	mov %rcx,%rsi
+	mov %r8,%rdx
+	mov %r9,%r10
+	mov 8(%rsp),%r8
+	mov 16(%rsp),%r9
+1:	syscall
+	ret
+2:	xor %edi,%edi
+	mov %rdi,8(%r10)
+	mov %rdi,(%r10)
+	dec %rdi
+	jmp pthread_exit