X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Flock.c;h=a71c95f6e5cc9479288deae43581361019361c38;hp=cad6480326897e27aa5d6c2e96a602ecc3a2b236;hb=614259f13c3e694fcd6b57fc05a329066e43c76d;hpb=f513c5701b184fd1c0a6f03431c7fdda3ab6d2cf

diff --git a/ccan/tdb2/lock.c b/ccan/tdb2/lock.c
index cad64803..a71c95f6 100644
--- a/ccan/tdb2/lock.c
+++ b/ccan/tdb2/lock.c
@@ -30,36 +30,93 @@
 #include <ccan/build_assert/build_assert.h>
 
 /* If we were threaded, we could wait for unlock, but we're not, so fail. */
-static enum TDB_ERROR owner_conflict(struct tdb_context *tdb, const char *call)
+enum TDB_ERROR owner_conflict(struct tdb_context *tdb, const char *call)
 {
 	return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
 			  "%s: lock owned by another tdb in this process.",
 			  call);
 }
 
-static int fcntl_lock(struct tdb_context *tdb,
-		      int rw, off_t off, off_t len, bool waitflag)
+/* If we fork, we no longer really own locks. */
+bool check_lock_pid(struct tdb_context *tdb, const char *call, bool log)
 {
-	struct flock fl;
+	/* No locks?  No problem! */
+	if (tdb->file->allrecord_lock.count == 0
+	    && tdb->file->num_lockrecs == 0) {
+		return true;
+	}
 
-	fl.l_type = rw;
-	fl.l_whence = SEEK_SET;
-	fl.l_start = off;
-	fl.l_len = len;
-	fl.l_pid = 0;
+	/* No fork?  No problem! */
+	if (tdb->file->locker == getpid()) {
+		return true;
+	}
 
-	add_stat(tdb, lock_lowlevel, 1);
-	if (waitflag)
-		return fcntl(tdb->file->fd, F_SETLKW, &fl);
-	else {
-		add_stat(tdb, lock_nonblock, 1);
-		return fcntl(tdb->file->fd, F_SETLK, &fl);
+	if (log) {
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+			   "%s: fork() detected after lock acquisition!"
+			   " (%u vs %u)", call, tdb->file->locker, getpid());
 	}
+	return false;
+}
+
+int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag,
+		   void *unused)
+{
+	struct flock fl;
+	int ret;
+
+	do {
+		fl.l_type = rw;
+		fl.l_whence = SEEK_SET;
+		fl.l_start = off;
+		fl.l_len = len;
+
+		if (waitflag)
+			ret = fcntl(fd, F_SETLKW, &fl);
+		else
+			ret = fcntl(fd, F_SETLK, &fl);
+	} while (ret != 0 && errno == EINTR);
+	return ret;
 }
 
-static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
+int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *unused)
 {
 	struct flock fl;
+	int ret;
+
+	do {
+		fl.l_type = F_UNLCK;
+		fl.l_whence = SEEK_SET;
+		fl.l_start = off;
+		fl.l_len = len;
+
+		ret = fcntl(fd, F_SETLKW, &fl);
+	} while (ret != 0 && errno == EINTR);
+	return ret;
+}
+
+static int lock(struct tdb_context *tdb,
+		      int rw, off_t off, off_t len, bool waitflag)
+{
+	int ret;
+	if (tdb->file->allrecord_lock.count == 0
+	    && tdb->file->num_lockrecs == 0) {
+		tdb->file->locker = getpid();
+	}
+
+	tdb->stats.lock_lowlevel++;
+	ret = tdb->lock_fn(tdb->file->fd, rw, off, len, waitflag,
+			   tdb->lock_data);
+	if (!waitflag) {
+		tdb->stats.lock_nonblock++;
+		if (ret != 0)
+			tdb->stats.lock_nonblock_fail++;
+	}
+	return ret;
+}
+
+static int unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
+{
 #if 0 /* Check they matched up locks and unlocks correctly. */
 	char line[80];
 	FILE *locks;
@@ -118,13 +175,7 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
 	fclose(locks);
 #endif
 
-	fl.l_type = F_UNLCK;
-	fl.l_whence = SEEK_SET;
-	fl.l_start = off;
-	fl.l_len = len;
-	fl.l_pid = 0;
-
-	return fcntl(tdb->file->fd, F_SETLKW, &fl);
+	return tdb->unlock_fn(tdb->file->fd, rw, off, len, tdb->lock_data);
 }
 
 /* a byte range locking function - return 0 on success
@@ -132,9 +183,9 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
 
    note that a len of zero means lock to end of file
 */
-static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
-				 int rw_type, tdb_off_t offset, tdb_off_t len,
-				 enum tdb_lock_flags flags)
+enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
+			  int rw_type, tdb_off_t offset, tdb_off_t len,
+			  enum tdb_lock_flags flags)
 {
 	int ret;
 
@@ -142,7 +193,7 @@ static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
 		return TDB_SUCCESS;
 	}
 
-	if (rw_type == F_WRLCK && tdb->read_only) {
+	if (rw_type == F_WRLCK && (tdb->flags & TDB_RDONLY)) {
 		return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
 				  "Write lock attempted on read-only database");
 	}
@@ -155,16 +206,13 @@ static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
 				  (long long)(offset + len));
 	}
 
-	do {
-		ret = fcntl_lock(tdb, rw_type, offset, len,
-				 flags & TDB_LOCK_WAIT);
-	} while (ret == -1 && errno == EINTR);
-
-	if (ret == -1) {
+	ret = lock(tdb, rw_type, offset, len, flags & TDB_LOCK_WAIT);
+	if (ret != 0) {
 		/* Generic lock error. errno set by fcntl.
 		 * EAGAIN is an expected return from non-blocking
 		 * locks. */
-		if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
+		if (!(flags & TDB_LOCK_PROBE)
+		    && (errno != EAGAIN && errno != EINTR)) {
 			tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				   "tdb_brlock failed (fd=%d) at"
 				   " offset %zu rw_type=%d flags=%d len=%zu:"
@@ -177,25 +225,22 @@ static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
 	return TDB_SUCCESS;
 }
 
-static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
-				   int rw_type, tdb_off_t offset, size_t len)
+enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
+			    int rw_type, tdb_off_t offset, size_t len)
 {
-	int ret;
-
 	if (tdb->flags & TDB_NOLOCK) {
 		return TDB_SUCCESS;
 	}
 
-	do {
-		ret = fcntl_unlock(tdb, rw_type, offset, len);
-	} while (ret == -1 && errno == EINTR);
+	if (!check_lock_pid(tdb, "tdb_brunlock", true))
+		return TDB_ERR_LOCK;
 
-	if (ret == -1) {
+	if (unlock(tdb, rw_type, offset, len) == -1) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_brunlock failed (fd=%d) at offset %zu"
-				  " rw_type=%d len=%zu",
+				  " rw_type=%d len=%zu: %s",
 				  tdb->file->fd, (size_t)offset, rw_type,
-				  (size_t)len);
+				  (size_t)len, strerror(errno));
 	}
 	return TDB_SUCCESS;
 }
@@ -206,10 +251,13 @@ static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
   deadlock detection and claim a deadlock when progress can be
   made. For those OSes we may loop for a while.
 */
-enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
+enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb, off_t start)
 {
 	int count = 1000;
 
+	if (!check_lock_pid(tdb, "tdb_transaction_prepare_commit", true))
+		return TDB_ERR_LOCK;
+
 	if (tdb->file->allrecord_lock.count != 1) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_allrecord_upgrade failed:"
@@ -229,8 +277,7 @@ enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
 
 	while (count--) {
 		struct timeval tv;
-		if (tdb_brlock(tdb, F_WRLCK,
-			       TDB_HASH_LOCK_START, 0,
+		if (tdb_brlock(tdb, F_WRLCK, start, 0,
 			       TDB_LOCK_WAIT|TDB_LOCK_PROBE) == TDB_SUCCESS) {
 			tdb->file->allrecord_lock.ltype = F_WRLCK;
 			tdb->file->allrecord_lock.off = 0;
@@ -244,8 +291,11 @@ enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
 		tv.tv_usec = 1;
 		select(0, NULL, NULL, NULL, &tv);
 	}
-	return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
-			  "tdb_allrecord_upgrade failed");
+
+	if (errno != EAGAIN && errno != EINTR)
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+			   "tdb_allrecord_upgrade failed");
+	return TDB_ERR_LOCK;
 }
 
 static struct tdb_lock *find_nestlock(struct tdb_context *tdb, tdb_off_t offset,
@@ -267,34 +317,38 @@ enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb)
 {
 	enum TDB_ERROR ecode;
 
+	if (!check_lock_pid(tdb, "tdb_transaction_prepare_commit", true))
+		return TDB_ERR_LOCK;
+
 	ecode = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
 				   false);
 	if (ecode != TDB_SUCCESS) {
 		return ecode;
 	}
 
-	ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
+	ecode = tdb_lock_open(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
 	if (ecode != TDB_SUCCESS) {
 		tdb_allrecord_unlock(tdb, F_WRLCK);
 		return ecode;
 	}
 	ecode = tdb_transaction_recover(tdb);
-	tdb_unlock_open(tdb);
+	tdb_unlock_open(tdb, F_WRLCK);
 	tdb_allrecord_unlock(tdb, F_WRLCK);
 
 	return ecode;
 }
 
 /* lock an offset in the database. */
-static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
-				    tdb_off_t offset, int ltype,
-				    enum tdb_lock_flags flags)
+enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
+			     tdb_off_t offset, int ltype,
+			     enum tdb_lock_flags flags)
 {
 	struct tdb_lock *new_lck;
 	enum TDB_ERROR ecode;
 
-	if (offset > (TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
-		      + tdb->file->map_size / 8)) {
+	if (!(tdb->flags & TDB_VERSION1)
+	    && offset > (TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
+			 + tdb->file->map_size / 8)) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_nest_lock: invalid offset %zu ltype=%d",
 				  (size_t)offset, ltype);
@@ -303,7 +357,11 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 	if (tdb->flags & TDB_NOLOCK)
 		return TDB_SUCCESS;
 
-	add_stat(tdb, locks, 1);
+	if (!check_lock_pid(tdb, "tdb_nest_lock", true)) {
+		return TDB_ERR_LOCK;
+	}
+
+	tdb->stats.locks++;
 
 	new_lck = find_nestlock(tdb, offset, NULL);
 	if (new_lck) {
@@ -322,12 +380,14 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 		return TDB_SUCCESS;
 	}
 
+#if 0
 	if (tdb->file->num_lockrecs
 	    && offset >= TDB_HASH_LOCK_START
 	    && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_nest_lock: already have a hash lock?");
 	}
+#endif
 
 	new_lck = (struct tdb_lock *)realloc(
 		tdb->file->lockrecs,
@@ -355,7 +415,7 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 			tdb_brunlock(tdb, ltype, offset, 1);
 
 			if (berr < 0)
-				return berr;
+				return TDB_OFF_TO_ERR(berr);
 			ecode = tdb_lock_and_recover(tdb);
 			if (ecode == TDB_SUCCESS) {
 				ecode = tdb_brlock(tdb, ltype, offset, 1,
@@ -376,8 +436,8 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 	return TDB_SUCCESS;
 }
 
-static enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
-				      tdb_off_t off, int ltype)
+enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
+			       tdb_off_t off, int ltype)
 {
 	struct tdb_lock *lck;
 	enum TDB_ERROR ecode;
@@ -432,9 +492,9 @@ void tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
 
 /* We only need to lock individual bytes, but Linux merges consecutive locks
  * so we lock in contiguous ranges. */
-static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
-				       int ltype, enum tdb_lock_flags flags,
-				       tdb_off_t off, tdb_off_t len)
+enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
+				int ltype, enum tdb_lock_flags flags,
+				tdb_off_t off, tdb_off_t len)
 {
 	enum TDB_ERROR ecode;
 	enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
@@ -447,8 +507,9 @@ static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
 	}
 
 	/* First we try non-blocking. */
-	if (tdb_brlock(tdb, ltype, off, len, nb_flags) == TDB_SUCCESS) {
-		return TDB_SUCCESS;
+	ecode = tdb_brlock(tdb, ltype, off, len, nb_flags);
+	if (ecode != TDB_ERR_LOCK) {
+		return ecode;
 	}
 
 	/* Try locking first half, then second. */
@@ -472,6 +533,19 @@ enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
 	enum TDB_ERROR ecode;
 	tdb_bool_err berr;
 
+	if (tdb->flags & TDB_VERSION1) {
+		if (tdb1_allrecord_lock(tdb, ltype, flags, upgradable) == -1)
+			return tdb->last_error;
+		return TDB_SUCCESS;
+	}
+
+	if (tdb->flags & TDB_NOLOCK)
+		return TDB_SUCCESS;
+
+	if (!check_lock_pid(tdb, "tdb_allrecord_lock", true)) {
+		return TDB_ERR_LOCK;
+	}
+
 	if (tdb->file->allrecord_lock.count) {
 		if (tdb->file->allrecord_lock.owner != tdb) {
 			return owner_conflict(tdb, "tdb_allrecord_lock");
@@ -504,28 +578,19 @@ enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
 				  " can't upgrade a write lock");
 	}
 
-	add_stat(tdb, locks, 1);
+	tdb->stats.locks++;
 again:
 	/* Lock hashes, gradually. */
 	ecode = tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
 				 TDB_HASH_LOCK_RANGE);
-	if (ecode != TDB_SUCCESS) {
-		if (!(flags & TDB_LOCK_PROBE)) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				   "tdb_allrecord_lock hashes failed");
-		}
+	if (ecode != TDB_SUCCESS)
 		return ecode;
-	}
 
 	/* Lock free tables: there to end of file. */
 	ecode = tdb_brlock(tdb, ltype,
 			   TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
 			   0, flags);
 	if (ecode != TDB_SUCCESS) {
-		if (!(flags & TDB_LOCK_PROBE)) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				 "tdb_allrecord_lock freetables failed");
-		}
 		tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
 			     TDB_HASH_LOCK_RANGE);
 		return ecode;
@@ -548,7 +613,7 @@ again:
 
 	tdb_allrecord_unlock(tdb, ltype);
 	if (berr < 0)
-		return berr;
+		return TDB_OFF_TO_ERR(berr);
 	ecode = tdb_lock_and_recover(tdb);
 	if (ecode != TDB_SUCCESS) {
 		return ecode;
@@ -556,14 +621,15 @@ again:
 	goto again;
 }
 
-enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
+enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
+			     int ltype, enum tdb_lock_flags flags)
 {
-	return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
+	return tdb_nest_lock(tdb, TDB_OPEN_LOCK, ltype, flags);
 }
 
-void tdb_unlock_open(struct tdb_context *tdb)
+void tdb_unlock_open(struct tdb_context *tdb, int ltype)
 {
-	tdb_nest_unlock(tdb, TDB_OPEN_LOCK, F_WRLCK);
+	tdb_nest_unlock(tdb, TDB_OPEN_LOCK, ltype);
 }
 
 bool tdb_has_open_lock(struct tdb_context *tdb)
@@ -587,6 +653,14 @@ void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
 /* unlock entire db */
 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
 {
+	if (tdb->flags & TDB_VERSION1) {
+		tdb1_allrecord_unlock(tdb, ltype);
+		return;
+	}
+
+	if (tdb->flags & TDB_NOLOCK)
+		return;
+
 	if (tdb->file->allrecord_lock.count == 0) {
 		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
 			   "tdb_allrecord_unlock: not locked!");
@@ -659,11 +733,14 @@ enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
 			       int ltype, enum tdb_lock_flags waitflag)
 {
 	/* FIXME: Do this properly, using hlock_range */
-	unsigned lock = TDB_HASH_LOCK_START
+	unsigned l = TDB_HASH_LOCK_START
 		+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
 
 	/* a allrecord lock allows us to avoid per chain locks */
 	if (tdb->file->allrecord_lock.count) {
+		if (!check_lock_pid(tdb, "tdb_lock_hashes", true))
+			return TDB_ERR_LOCK;
+
 		if (tdb->file->allrecord_lock.owner != tdb)
 			return owner_conflict(tdb, "tdb_lock_hashes");
 		if (ltype == tdb->file->allrecord_lock.ltype
@@ -689,14 +766,14 @@ enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
 				  " already have expansion lock");
 	}
 
-	return tdb_nest_lock(tdb, lock, ltype, waitflag);
+	return tdb_nest_lock(tdb, l, ltype, waitflag);
 }
 
 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
 				 tdb_off_t hash_lock,
 				 tdb_len_t hash_range, int ltype)
 {
-	unsigned lock = TDB_HASH_LOCK_START
+	unsigned l = TDB_HASH_LOCK_START
 		+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
 
 	if (tdb->flags & TDB_NOLOCK)
@@ -709,10 +786,15 @@ enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
 			return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 					  "tdb_unlock_hashes RO allrecord!");
 		}
+		if (tdb->file->allrecord_lock.owner != tdb) {
+			return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+					  "tdb_unlock_hashes:"
+					  " not locked by us!");
+		}
 		return TDB_SUCCESS;
 	}
 
-	return tdb_nest_unlock(tdb, lock, ltype);
+	return tdb_nest_unlock(tdb, l, ltype);
 }
 
 /* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
@@ -736,6 +818,13 @@ enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
 
 	/* a allrecord lock allows us to avoid per chain locks */
 	if (tdb->file->allrecord_lock.count) {
+		if (!check_lock_pid(tdb, "tdb_lock_free_bucket", true))
+			return TDB_ERR_LOCK;
+
+		if (tdb->file->allrecord_lock.owner != tdb) {
+			return owner_conflict(tdb, "tdb_lock_free_bucket");
+		}
+
 		if (tdb->file->allrecord_lock.ltype == F_WRLCK)
 			return 0;
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
@@ -786,6 +875,10 @@ void tdb_lock_cleanup(struct tdb_context *tdb)
 {
 	unsigned int i;
 
+	/* We don't want to warn: they're allowed to close tdb after fork. */
+	if (!check_lock_pid(tdb, "tdb_close", false))
+		return;
+
 	while (tdb->file->allrecord_lock.count
 	       && tdb->file->allrecord_lock.owner == tdb) {
 		tdb_allrecord_unlock(tdb, tdb->file->allrecord_lock.ltype);