Use umtx to implement process sharable semaphore, to make this work,

now type sema_t is a structure which can be put in a shared memory area,
and multiple processes can operate it concurrently.
User can either use mmap(MAP_SHARED) + sem_init(pshared=1) or use sem_open()
to initialize a shared semaphore.
Named semaphore uses file system and is located in /tmp directory, and its
file name is prefixed with 'SEMD', so now it is chroot or jail friendly.
In simplist cases, both for named and un-named semaphore, userland code
does not have to enter kernel to reduce/increase semaphore's count.
The semaphore is designed to be crash-safe, it means even if an application
is crashed in the middle of operating semaphore, the semaphore state is
still safely recovered by later use, there is no waiter counter maintained
by userland code.
The main semaphore code is in libc and libthr only has some necessary stubs,
this makes it possible that a non-threaded application can use semaphore
without linking to thread library.
Old semaphore implementation is kept libc to maintain binary compatibility.
The kernel ksem API is no longer used in the new implemenation.

Discussed on: threads@

1 files changed, 174 insertions, 58 deletions

diff --git a/lib/libc/gen/sem.c b/lib/libc/gen/sem.c
index cd31d31..4ec7890 100644
--- a/lib/libc/gen/sem.c
+++ b/lib/libc/gen/sem.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (C) 2010 David Xu <davidxu@freebsd.org>.
  * Copyright (C) 2000 Jason Evans <jasone@freebsd.org>.
  * All rights reserved.
  * 
@@ -59,34 +60,71 @@
 #include "namespace.h"
 #include <sys/types.h>
 #include <sys/queue.h>
+#include <machine/atomic.h>
 #include <errno.h>
+#include <sys/umtx.h>
+#include <sys/_semaphore.h>
+#include <limits.h>
 #include <fcntl.h>
 #include <pthread.h>
-#include <semaphore.h>
 #include <stdarg.h>
 #include <stdlib.h>
 #include <time.h>
-#include <_semaphore.h>
 #include "un-namespace.h"
 #include "libc_private.h"
 
+/*
+ * Old semaphore definitions.
+ */
+struct sem {
+#define SEM_MAGIC       ((u_int32_t) 0x09fa4012)
+        u_int32_t       magic;
+        pthread_mutex_t lock;
+        pthread_cond_t  gtzero;
+        u_int32_t       count;
+        u_int32_t       nwaiters;
+#define SEM_USER        (NULL)
+        semid_t         semid;  /* semaphore id if kernel (shared) semaphore */
+        int             syssem; /* 1 if kernel (shared) semaphore */
+        LIST_ENTRY(sem) entry;
+        struct sem      **backpointer;
+};
+
+typedef struct sem* sem_t;
+
+#define SEM_FAILED     ((sem_t *)0)
+#define SEM_VALUE_MAX  __INT_MAX
+
+#define SYM_FB10(sym)                   __CONCAT(sym, _fb10)
+#define SYM_FBP10(sym)                  __CONCAT(sym, _fbp10)
+#define WEAK_REF(sym, alias)            __weak_reference(sym, alias)
+#define SYM_COMPAT(sym, impl, ver)      __sym_compat(sym, impl, ver)
+#define SYM_DEFAULT(sym, impl, ver)     __sym_default(sym, impl, ver)
+ 
+#define FB10_COMPAT(func, sym)                          \
+        WEAK_REF(func, SYM_FB10(sym));                  \
+        SYM_COMPAT(sym, SYM_FB10(sym), FBSD_1.0)
+
+#define FB10_COMPAT_PRIVATE(func, sym)                  \
+        WEAK_REF(func, SYM_FBP10(sym));                 \
+        SYM_DEFAULT(sym, SYM_FBP10(sym), FBSDprivate_1.0)
+
 static sem_t sem_alloc(unsigned int value, semid_t semid, int system_sem);
 static void  sem_free(sem_t sem);
 
-static LIST_HEAD(, sem) named_sems = LIST_HEAD_INITIALIZER(named_sems);
+static LIST_HEAD(, sem) named_sems = LIST_HEAD_INITIALIZER(&named_sems);
 static pthread_mutex_t named_sems_mtx = PTHREAD_MUTEX_INITIALIZER;
 
-__weak_reference(__sem_init, sem_init);
-__weak_reference(__sem_destroy, sem_destroy);
-__weak_reference(__sem_open, sem_open);
-__weak_reference(__sem_close, sem_close);
-__weak_reference(__sem_unlink, sem_unlink);
-__weak_reference(__sem_wait, sem_wait);
-__weak_reference(__sem_trywait, sem_trywait);
-__weak_reference(__sem_timedwait, sem_timedwait);
-__weak_reference(__sem_post, sem_post);
-__weak_reference(__sem_getvalue, sem_getvalue);
-
+FB10_COMPAT(_libc_sem_init_compat, sem_init);
+FB10_COMPAT(_libc_sem_destroy_compat, sem_destroy);
+FB10_COMPAT(_libc_sem_open_compat, sem_open);
+FB10_COMPAT(_libc_sem_close_compat, sem_close);
+FB10_COMPAT(_libc_sem_unlink_compat, sem_unlink);
+FB10_COMPAT(_libc_sem_wait_compat, sem_wait);
+FB10_COMPAT(_libc_sem_trywait_compat, sem_trywait);
+FB10_COMPAT(_libc_sem_timedwait_compat, sem_timedwait);
+FB10_COMPAT(_libc_sem_post_compat, sem_post);
+FB10_COMPAT(_libc_sem_getvalue_compat, sem_getvalue);
 
 static inline int
 sem_check_validity(sem_t *sem)
@@ -104,8 +142,6 @@ static void
 sem_free(sem_t sem)
 {
 
-	_pthread_mutex_destroy(&sem->lock);
-	_pthread_cond_destroy(&sem->gtzero);
 	sem->magic = 0;
 	free(sem);
 }
@@ -131,13 +167,11 @@ sem_alloc(unsigned int value, semid_t semid, int system_sem)
 	sem->magic = SEM_MAGIC;
 	sem->semid = semid;
 	sem->syssem = system_sem;
-	sem->lock = PTHREAD_MUTEX_INITIALIZER;
-	sem->gtzero = PTHREAD_COND_INITIALIZER;
 	return (sem);
 }
 
 int
-__sem_init(sem_t *sem, int pshared, unsigned int value)
+_libc_sem_init_compat(sem_t *sem, int pshared, unsigned int value)
 {
 	semid_t semid;
 
@@ -147,26 +181,27 @@ __sem_init(sem_t *sem, int pshared, unsigned int value)
 	 * pthread_cond_wait() is just a stub that doesn't really
 	 * wait.
 	 */
-	if (ksem_init(&semid, value) != 0)
+	semid = (semid_t)SEM_USER;
+	if ((pshared != 0) && ksem_init(&semid, value) != 0)
 		return (-1);
 
-	(*sem) = sem_alloc(value, semid, 1);
+	*sem = sem_alloc(value, semid, pshared);
 	if ((*sem) == NULL) {
-		ksem_destroy(semid);
+		if (pshared != 0)
+			ksem_destroy(semid);
 		return (-1);
 	}
 	return (0);
 }
 
 int
-__sem_destroy(sem_t *sem)
+_libc_sem_destroy_compat(sem_t *sem)
 {
 	int retval;
 
 	if (sem_check_validity(sem) != 0)
 		return (-1);
 
-	_pthread_mutex_lock(&(*sem)->lock);
 	/*
 	 * If this is a system semaphore let the kernel track it otherwise
 	 * make sure there are no waiters.
@@ -181,18 +216,14 @@ __sem_destroy(sem_t *sem)
 		retval = 0;
 		(*sem)->magic = 0;
 	}
-	_pthread_mutex_unlock(&(*sem)->lock);
 
-	if (retval == 0) {
-		_pthread_mutex_destroy(&(*sem)->lock);
-		_pthread_cond_destroy(&(*sem)->gtzero);
+	if (retval == 0)
 		sem_free(*sem);
-	}
 	return (retval);
 }
 
 sem_t *
-__sem_open(const char *name, int oflag, ...)
+_libc_sem_open_compat(const char *name, int oflag, ...)
 {
 	sem_t *sem;
 	sem_t s;
@@ -255,7 +286,7 @@ err:
 }
 
 int
-__sem_close(sem_t *sem)
+_libc_sem_close_compat(sem_t *sem)
 {
 
 	if (sem_check_validity(sem) != 0)
@@ -280,68 +311,156 @@ __sem_close(sem_t *sem)
 }
 
 int
-__sem_unlink(const char *name)
+_libc_sem_unlink_compat(const char *name)
 {
 
 	return (ksem_unlink(name));
 }
 
-int
-__sem_wait(sem_t *sem)
+static int
+enable_async_cancel(void)
 {
+	int old;
 
-	if (sem_check_validity(sem) != 0)
+	_pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
+	return (old);
+}
+
+static void
+restore_async_cancel(int val)
+{
+	_pthread_setcanceltype(val, NULL);
+}
+
+static int
+_umtx_wait_uint(volatile unsigned *mtx, unsigned id, const struct timespec *timeout)
+{
+	if (timeout && (timeout->tv_sec < 0 || (timeout->tv_sec == 0 &&
+	    timeout->tv_nsec <= 0))) {
+		errno = ETIMEDOUT;
 		return (-1);
+	}
+	return _umtx_op(__DEVOLATILE(void *, mtx),
+		UMTX_OP_WAIT_UINT_PRIVATE, id, NULL, __DECONST(void*, timeout));
+}
 
-	return (ksem_wait((*sem)->semid));
+static int
+_umtx_wake(volatile void *mtx)
+{
+	return _umtx_op(__DEVOLATILE(void *, mtx), UMTX_OP_WAKE_PRIVATE,
+			1, NULL, NULL);
+}
+
+#define TIMESPEC_SUB(dst, src, val)                             \
+        do {                                                    \
+                (dst)->tv_sec = (src)->tv_sec - (val)->tv_sec;  \
+                (dst)->tv_nsec = (src)->tv_nsec - (val)->tv_nsec; \
+                if ((dst)->tv_nsec < 0) {                       \
+                        (dst)->tv_sec--;                        \
+                        (dst)->tv_nsec += 1000000000;           \
+                }                                               \
+        } while (0)
+
+
+static void
+sem_cancel_handler(void *arg)
+{
+	sem_t *sem = arg;
+
+	atomic_add_int(&(*sem)->nwaiters, -1);
+	if ((*sem)->nwaiters && (*sem)->count)
+		_umtx_wake(&(*sem)->count);
 }
 
 int
-__sem_trywait(sem_t *sem)
+_libc_sem_timedwait_compat(sem_t * __restrict sem,
+	const struct timespec * __restrict abstime)
 {
-	int retval;
+	struct timespec ts, ts2;
+	int val, retval, saved_cancel;
 
 	if (sem_check_validity(sem) != 0)
 		return (-1);
 
-	if ((*sem)->syssem != 0)
- 		retval = ksem_trywait((*sem)->semid);
-	else {
-		_pthread_mutex_lock(&(*sem)->lock);
-		if ((*sem)->count > 0) {
-			(*sem)->count--;
-			retval = 0;
-		} else {
-			errno = EAGAIN;
-			retval = -1;
+	if ((*sem)->syssem != 0) {
+		saved_cancel = enable_async_cancel();
+		retval = ksem_wait((*sem)->semid);
+		restore_async_cancel(saved_cancel);
+		return (retval);
+	}
+
+	retval = 0;
+	_pthread_testcancel();
+	for (;;) {
+		while ((val = (*sem)->count) > 0) {
+			if (atomic_cmpset_acq_int(&(*sem)->count, val, val - 1))
+				return (0);
+		}
+		if (retval)
+			break;
+		if (abstime) {
+			if (abstime->tv_nsec >= 1000000000 || abstime->tv_nsec < 0) {
+				errno = EINVAL;
+				return (-1);
+			}
+			clock_gettime(CLOCK_REALTIME, &ts);
+	                TIMESPEC_SUB(&ts2, abstime, &ts);
 		}
-		_pthread_mutex_unlock(&(*sem)->lock);
+		atomic_add_int(&(*sem)->nwaiters, 1);
+		pthread_cleanup_push(sem_cancel_handler, sem);
+		saved_cancel = enable_async_cancel();
+		retval = _umtx_wait_uint(&(*sem)->count, 0, abstime ? &ts2 : NULL);
+		restore_async_cancel(saved_cancel);
+		pthread_cleanup_pop(0);
+		atomic_add_int(&(*sem)->nwaiters, -1);
 	}
 	return (retval);
 }
 
 int
-__sem_timedwait(sem_t * __restrict sem,
-    const struct timespec * __restrict abs_timeout)
+_libc_sem_wait_compat(sem_t *sem)
 {
+	return _libc_sem_timedwait_compat(sem, NULL);
+}
+
+int
+_libc_sem_trywait_compat(sem_t *sem)
+{
+	int val;
+
 	if (sem_check_validity(sem) != 0)
 		return (-1);
 
-	return (ksem_timedwait((*sem)->semid, abs_timeout));
+	if ((*sem)->syssem != 0)
+ 		return ksem_trywait((*sem)->semid);
+
+	while ((val = (*sem)->count) > 0) {
+		if (atomic_cmpset_acq_int(&(*sem)->count, val, val - 1))
+			return (0);
+	}
+	errno = EAGAIN;
+	return (-1);
 }
 
 int
-__sem_post(sem_t *sem)
+_libc_sem_post_compat(sem_t *sem)
 {
 
 	if (sem_check_validity(sem) != 0)
 		return (-1);
 
-	return (ksem_post((*sem)->semid));
+	if ((*sem)->syssem != 0)
+		return ksem_post((*sem)->semid);
+
+	atomic_add_rel_int(&(*sem)->count, 1);
+
+	if ((*sem)->nwaiters)
+		return _umtx_wake(&(*sem)->count);
+	return (0);
 }
 
 int
-__sem_getvalue(sem_t * __restrict sem, int * __restrict sval)
+_libc_sem_getvalue_compat(sem_t * __restrict sem, int * __restrict sval)
 {
 	int retval;
 
@@ -351,10 +470,7 @@ __sem_getvalue(sem_t * __restrict sem, int * __restrict sval)
 	if ((*sem)->syssem != 0)
 		retval = ksem_getvalue((*sem)->semid, sval);
 	else {
-		_pthread_mutex_lock(&(*sem)->lock);
 		*sval = (int)(*sem)->count;
-		_pthread_mutex_unlock(&(*sem)->lock);
-
 		retval = 0;
 	}
 	return (retval);