overhaul cancellation to fix resource leaks and dangerous behavior with signals

this commit addresses two issues:

1. a race condition, whereby a cancellation request occurring after a
syscall returned from kernelspace but before the subsequent
CANCELPT_END would cause cancellable resource-allocating syscalls
(like open) to leak resources.

2. signal handlers invoked while the thread was blocked at a
cancellation point behaved as if asynchronous cancellation mode wer in
effect, resulting in potentially dangerous state corruption if a
cancellation request occurs.

the glibc/nptl implementation of threads shares both of these issues.

with this commit, both are fixed. however, cancellation points
encountered in a signal handler will not be acted upon if the signal
was received while the thread was already at a cancellation point.
they will of course be acted upon after the signal handler returns, so
in real-world usage where signal handlers quickly return, it should
not be a problem. it's possible to solve this problem too by having
sigaction() wrap all signal handlers with a function that uses a
pthread_cleanup handler to catch cancellation, patch up the saved
context, and return into the cancellable function that will catch and
act upon the cancellation. however that would be a lot of complexity
for minimal if any benefit...

arch/i386/pthread_arch.h

@@ -4,3 +4,6 @@ static inline struct pthread *__pthread_self()
 	__asm__ ("movl %%gs:0,%0" : "=r" (self) );
 	return self;
 }
+
+#define PC_AT_SYS(c) \
+	(*(uint16_t *)(((ucontext_t *)(c))->uc_mcontext.__gregs[14])==0x80cd)

arch/x86_64/pthread_arch.h

@@ -4,3 +4,6 @@ static inline struct pthread *__pthread_self()
 	__asm__ ("movq %%fs:0,%0" : "=r" (self) );
 	return self;
 }
+
+#define PC_AT_SYS(c) \
+	(*(uint16_t *)(((ucontext_t *)(c))->uc_mcontext.__gregs[16])==0x050f)

src/internal/libc.h

@@ -44,7 +44,8 @@ void __lockfile(FILE *);
 #define UNLOCK(x) (*(x)=0)
 #define CANCELPT(x) (libc.cancelpt ? libc.cancelpt((x)),0 : (void)(x),0)
 #define CANCELPT_BEGIN CANCELPT(1)
-#define CANCELPT_END CANCELPT(0)
+#define CANCELPT_TRY CANCELPT(0)
+#define CANCELPT_END CANCELPT(-1)
 
 extern char **__environ;
 #define environ __environ

src/process/waitid.c

 #include <sys/wait.h>
 #include "syscall.h"
+#include "libc.h"
 
 int waitid(idtype_t type, id_t id, siginfo_t *info, int options)
 {
-	return syscall(SYS_waitid, type, id, info, options, 0);
+	int r;
+	CANCELPT_BEGIN;
+	r = syscall(SYS_waitid, type, id, info, options, 0);
+	if (r<0) CANCELPT_TRY;
+	CANCELPT_END;
+	return r;
 }

src/process/waitpid.c

 #include <sys/wait.h>
 #include "syscall.h"
+#include "libc.h"
 
 pid_t waitpid(pid_t pid, int *status, int options)
 {
-	return syscall(SYS_wait4, pid, status, options, 0);
+	int r;
+	CANCELPT_BEGIN;
+	r = syscall(SYS_wait4, pid, status, options, 0);
+	if (r<0) CANCELPT_TRY;
+	CANCELPT_END;
+	return r;
 }

src/select/pselect.c

@@ -10,6 +10,7 @@ int pselect(int n, fd_set *rfds, fd_set *wfds, fd_set *efds, const struct timesp
 	if (ts) ts_tmp = *ts;
 	CANCELPT_BEGIN;
 	r = syscall(SYS_pselect6, n, rfds, wfds, efds, ts ? &ts_tmp : 0, data);
+	CANCELPT_TRY;
 	CANCELPT_END;
 	return r;
 }

src/select/select.c

@@ -7,6 +7,7 @@ int select(int n, fd_set *rfds, fd_set *wfds, fd_set *efds, struct timeval *tv)
 	int r;
 	CANCELPT_BEGIN;
 	r = syscall(SYS_select, n, rfds, wfds, efds, tv);
+	CANCELPT_TRY;
 	CANCELPT_END;
 	return r;
 }

src/signal/sigsuspend.c

@@ -7,6 +7,7 @@ int sigsuspend(const sigset_t *mask)
 	int ret;
 	CANCELPT_BEGIN;
 	ret = syscall(__NR_rt_sigsuspend, mask, 8);
+	if (ret<0) CANCELPT_TRY;
 	CANCELPT_END;
 	return ret;
 }

src/signal/sigtimedwait.c

@@ -9,6 +9,7 @@ int sigtimedwait(const sigset_t *mask, siginfo_t *si, const struct timespec *tim
 	CANCELPT_BEGIN;
 	do {
 		ret = syscall(__NR_rt_sigtimedwait, mask, si, timeout, 8);
+		if (ret<0) CANCELPT_TRY;
 	} while (ret<0 && errno==EINTR);
 	CANCELPT_END;
 	return ret;

src/termios/tcdrain.c

@@ -7,6 +7,7 @@ int tcdrain(int fd)
 	int ret;
 	CANCELPT_BEGIN;
 	ret = ioctl(fd, TCSBRK, 1);
+	CANCELPT_TRY;
 	CANCELPT_END;
 	return ret;
 }

src/thread/pthread_cond_timedwait.c

@@ -8,19 +8,21 @@ 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(0);
+	CANCELPT_BEGIN;
+	CANCELPT_END;
 
 	pthread_cleanup_push(relock, m);
 	c->_c_block = 1;
 	if ((r=pthread_mutex_unlock(m))) return r;
 
-	CANCELPT(1);
+	CANCELPT_BEGIN;
 	e = __timedwait(&c->_c_block, 1, c->_c_clock, ts, 0);
-	CANCELPT(0);
+	CANCELPT_END;
 
 	pthread_cleanup_pop(0);
 	if ((r=pthread_mutex_lock(m))) return r;
 
-	CANCELPT(0);
+	CANCELPT_BEGIN;
+	CANCELPT_END;
 	return e;
 }

src/thread/pthread_create.c

@@ -42,6 +42,12 @@ void __pthread_unwind_next(struct __ptcb *cb)
 static void docancel(struct pthread *self)
 {
 	struct __ptcb cb = { .__next = self->cancelbuf };
+	sigset_t set;
+	self->canceldisable = 1;
+	self->cancelasync = 0;
+	sigemptyset(&set);
+	sigaddset(&set, SIGCANCEL);
+	__libc_sigprocmask(SIG_UNBLOCK, &set, 0);
 	__pthread_unwind_next(&cb);
 }
 
@@ -50,17 +56,17 @@ static void cancel_handler(int sig, siginfo_t *si, void *ctx)
 	struct pthread *self = __pthread_self();
 	if (si->si_code > 0 || si->si_pid != self->pid) return;
 	self->cancel = 1;
-	if (self->canceldisable || (!self->cancelasync && !self->cancelpoint))
-		return;
-	docancel(self);
+	if (self->canceldisable) return;
+	if (self->cancelasync || (self->cancelpoint==1 && PC_AT_SYS(ctx)))
+		docancel(self);
 }
 
 static void cancelpt(int x)
 {
 	struct pthread *self = __pthread_self();
 	if (self->canceldisable) return;
-	self->cancelpoint = x;
-	if (self->cancel) docancel(self);
+	if ((self->cancelpoint+=x)==1 && x>=0 && self->cancel)
+		docancel(self);
 }
 
 /* "rsyscall" is a mechanism by which a thread can synchronously force all

src/thread/sem_timedwait.c

@@ -9,15 +9,18 @@ int sem_timedwait(sem_t *sem, const struct timespec *at)
 		if (a_fetch_add(sem->__val, -1) > 0) return 0;
 		val = a_fetch_add(sem->__val, 1)+1;
 		if (val==1) __wake(sem->__val, 1, 0);
-		CANCELPT_BEGIN;
 		if (at && at->tv_nsec >= 1000000000UL) {
 			errno = EINVAL;
 			return -1;
 		}
+		CANCELPT_BEGIN;
 		if (val <= 0 && __timedwait(sem->__val, val, CLOCK_REALTIME, at, 0) == ETIMEDOUT) {
 			errno = ETIMEDOUT;
+			CANCELPT_TRY;
+			CANCELPT_END;
 			return -1;
 		}
+		CANCELPT_TRY;
 		CANCELPT_END;
 	}
 }

src/time/nanosleep.c

@@ -8,6 +8,7 @@ int nanosleep(const struct timespec *req, struct timespec *rem)
 	int ret;
 	CANCELPT_BEGIN;
 	ret = syscall(SYS_nanosleep, req, rem);
+	CANCELPT_TRY;
 	CANCELPT_END;
 	return ret;
 }