2025-01-26 10:04:02 +03:00 · 2012-04-26 15:48:12 +10:00 · 2012-04-26 15:48:12 +10:00 · c3c5dec52c
commit c3c5dec52c
parent ec72e30b8f
1 changed files with 783 additions and 0 deletions
--- a/ctdb/server/ctdb_lock.c
+++ b/ctdb/server/ctdb_lock.c
@ -0,0 +1,783 @@
+/* 
+   ctdb lock handling 
+   provide API to do non-blocking locks for single or all databases
+
+   Copyright (C) Amitay Isaacs  2012
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+   
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+   
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "includes.h"
+#include "include/ctdb_private.h"
+#include "include/ctdb_protocol.h"
+#include "tevent.h"
+#include "tdb.h"
+#include "db_wrap.h"
+#include "system/filesys.h"
+
+/* 
+ * Locking API
+ *
+ * 1. Try locking with non-blocking tdb calls
+ * 2. If it fails, create a child process to do blocking calls
+ * 3. If the child process cannot get locks within certain time, 
+ *    diagnose using /proc/locks and log warning message
+ * 4. Additionally terminate the blocking process
+ *
+ * ctdb_lock_record()
+ * ctdb_lock_db()
+ * ctdb_lock_alldb()
+ */
+
+enum lock_db_type {
+	LOCK_DB_RECORD,
+	LOCK_DB_ONE,
+	LOCK_DB_ALL_PRIO,
+};
+
+struct lock_db_handle {
+	enum lock_db_type type;
+	struct lock_db_handle *next, *prev;
+	struct ctdb_context *ctdb;
+	struct ctdb_db_context *ctdb_db;
+	TDB_DATA key;
+	uint32_t priority;
+	void (*lock_callback)(void *);
+	void *lock_private_data;
+	void (*timeout_callback)(void *);
+	void *timeout_private_data;
+	int fd[2];
+	pid_t child;
+	struct tevent_fd *tfd;
+	struct tevent_timer *ttimer;
+	struct timeval start_time;
+	bool lock_in_parent;
+};
+
+struct lock_record {
+	ino_t inode;
+	off_t start, end;
+};
+
+static bool ignore_db(struct ctdb_db_context *ctdb_db)
+{
+	const char *name = ctdb_db->db_name;
+	return (strstr(name, "notify") || strstr(name, "serverid"));
+}
+
+
+/*
+ * lock all databases
+ */
+static int ctdb_lockall(struct ctdb_context *ctdb, uint32_t priority)
+{
+	struct ctdb_db_context *ctdb_db;
+
+	for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
+		if (ctdb_db->priority != priority) {
+			continue;
+		}
+		if (ignore_db(ctdb_db)) {
+			continue;
+		}
+		if (tdb_lockall(ctdb_db->ltdb->tdb) != 0) {
+			DEBUG(DEBUG_ERR, ("Failed to lock database %s\n", ctdb_db->db_name));
+			return -1;
+		}
+	}
+
+	return 0;
+}
+
+
+/*
+ * unlock all databases
+ */
+static int ctdb_unlockall(struct ctdb_context *ctdb, uint32_t priority)
+{
+	struct ctdb_db_context *ctdb_db;
+
+	for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
+		if (ctdb_db->priority != priority) {
+			continue;
+		}
+		if (ignore_db(ctdb_db)) {
+			continue;
+		}
+		if (tdb_unlockall(ctdb_db->ltdb->tdb) != 0) {
+			DEBUG(DEBUG_ERR, ("Failed to unlock database %s\n", ctdb_db->db_name));
+			return -1;
+		}
+	}
+
+	return 0;
+}
+
+
+/*
+ * lock all databases - mark only
+ */
+static int ctdb_lockall_mark(struct ctdb_context *ctdb, uint32_t priority)
+{
+	struct ctdb_db_context *ctdb_db;
+
+	if (priority < 1 || priority > NUM_DB_PRIORITIES) {
+		DEBUG(DEBUG_ERR, ("Illegal priority %u in ctdb_lockall_mark", priority));
+		return -1;
+	}
+
+	/* TODO: What is this? */
+	if (ctdb->freeze_mode[priority] != CTDB_FREEZE_FROZEN) {
+		DEBUG(DEBUG_ERR, ("Attempt to mark all databases locked when not frozen"));
+		return -1;
+	}
+
+	for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
+		if (ctdb_db->priority != priority) {
+			continue;
+		}
+		if (ignore_db(ctdb_db)) {
+			continue;
+		}
+#if 0
+		/* FIXME: Why do we need this? Isn't tdb_lockall_mark sufficient? */
+		if (tdb_transaction_write_lock_mark(ctdb_db->ltdb->tdb) != 0) {
+			return -1;
+		}
+#endif
+		if (tdb_lockall_mark(ctdb_db->ltdb->tdb) != 0) {
+			/* FIXME: Shouldn't we unmark here? */
+			return -1;
+		}
+	}
+
+	return 0;
+}
+
+
+/*
+ * lock all databases - unmark only
+ */
+static int ctdb_lockall_unmark(struct ctdb_context *ctdb, uint32_t priority)
+{
+	struct ctdb_db_context *ctdb_db;
+
+	if (priority < 1 || priority > NUM_DB_PRIORITIES) {
+		DEBUG(DEBUG_ERR, ("Illegal priority %u in ctdb_lockall_unmark", priority));
+		return -1;
+	}
+
+	/* TODO: What is this? */
+	if (ctdb->freeze_mode[priority] != CTDB_FREEZE_FROZEN) {
+		DEBUG(DEBUG_ERR, ("Attempt to unmark all databases locked when not frozen"));
+		return -1;
+	}
+
+	for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
+		if (ctdb_db->priority != priority) {
+			continue;
+		}
+		if (ignore_db(ctdb_db)) {
+			continue;
+		}
+#if 0
+		if (tdb_transaction_write_lock_unmark(ctdb_db->ltdb->tdb) != 0) {
+			return -1;
+		}
+#endif
+		if (tdb_lockall_unmark(ctdb_db->ltdb->tdb) != 0) {
+			return -1;
+		}
+	}
+
+	return 0;
+}
+
+
+/*
+ * Lock record / db / all db depending on h->type
+ * Called from parent context.
+ */
+static bool ctdb_lock_item_nonblock(struct lock_db_handle *h)
+{
+	bool status = false;
+
+	switch (h->type) {
+	case LOCK_DB_RECORD:
+		if (tdb_chainlock_nonblock(h->ctdb_db->ltdb->tdb, h->key) == 0) {
+			status = true;
+		}
+		break;
+
+	case LOCK_DB_ONE:
+		if (tdb_lockall_nonblock(h->ctdb_db->ltdb->tdb) == 0) {
+			status = true;
+		}
+		break;
+
+	case LOCK_DB_ALL_PRIO:
+		/* Is there a chance we will ever succeed here? */
+		status = false;
+		break;
+	}
+
+	return status;
+}
+
+
+/*
+ * Lock record / db / all-db depending on h->type
+ * Called from child context.
+ */
+static bool ctdb_lock_item(struct lock_db_handle *h)
+{
+	bool status = false;
+
+	switch (h->type) {
+	case LOCK_DB_RECORD:
+		if (tdb_chainlock(h->ctdb_db->ltdb->tdb, h->key) == 0) {
+			status = true;
+		}
+		break;
+
+	case LOCK_DB_ONE:
+		if (tdb_lockall(h->ctdb_db->ltdb->tdb) == 0) {
+			status = true;
+		}
+		break;
+
+	case LOCK_DB_ALL_PRIO:
+		if (ctdb_lockall(h->ctdb, h->priority) == 0) {
+			status = true;
+		}
+		break;
+	}
+
+	return status;
+}
+
+
+/*
+ * Unlock record / db / all-db depending on h->type
+ */
+bool ctdb_unlock_item(struct lock_db_handle *h)
+{
+	bool status = true;
+
+	switch (h->type) {
+	case LOCK_DB_RECORD:
+		tdb_chainunlock(h->ctdb_db->ltdb->tdb, h->key);
+		break;
+
+	case LOCK_DB_ONE:
+		tdb_unlockall(h->ctdb_db->ltdb->tdb);
+		break;
+
+	case LOCK_DB_ALL_PRIO:
+		ctdb_unlockall(h->ctdb, h->priority);
+		break;
+	}
+
+	return status;
+}
+
+
+/*
+ * Parsing a line from /proc/locks, 
+ *
+ * FIXME: This should be moved to system_linux.c
+ */
+static bool parse_proc_locks_line(char *line, pid_t *pid,
+				  struct lock_record *curlock)
+{
+	char *ptr, *saveptr;
+
+	/* 57: POSIX  ADVISORY  WRITE 1301 00:11:15132 0 EOF */
+
+	/* Id: */
+	ptr = strtok_r(line, " ", &saveptr);
+	if (ptr == NULL) return false;
+
+	/* POSIX */
+	ptr = strtok_r(NULL, " ", &saveptr);
+	if (ptr == NULL || strcmp(ptr, "POSIX") != 0) {
+		return false;
+	}
+
+	/* ADVISORY */
+	ptr = strtok_r(NULL, " ", &saveptr);
+	if (ptr == NULL) return false;
+
+	/* WRITE */
+	ptr = strtok_r(NULL, " ", &saveptr);
+	if (ptr == NULL || strcmp(ptr, "WRITE") != 0) {
+		return false;
+	}
+
+	/* PID */
+	ptr = strtok_r(NULL, " ", &saveptr);
+	if (ptr == NULL) return false;
+	*pid = atoi(ptr);
+
+	/* MAJOR:MINOR:INODE */
+	ptr = strtok_r(NULL, " :", &saveptr);
+	if (ptr == NULL) return false;
+	ptr = strtok_r(NULL, " :", &saveptr);
+	if (ptr == NULL) return false;
+	ptr = strtok_r(NULL, " :", &saveptr);
+	if (ptr == NULL) return false;
+	curlock->inode = atol(ptr);
+
+	/* START OFFSET */
+	ptr = strtok_r(NULL, " ", &saveptr);
+	if (ptr == NULL) return false;
+	curlock->start = atol(ptr);
+
+	/* END OFFSET */
+	ptr = strtok_r(NULL, " ", &saveptr);
+	if (ptr == NULL) return false;
+	if (strncmp(ptr, "EOF", 3) == 0) {
+		curlock->end = (off_t)-1;
+	} else {
+		curlock->end = atol(ptr);
+	}
+
+	return true;
+}
+
+
+/*
+ * Find the process name from proceed id
+ *
+ * FIXME: This should be moved to system_linux.c
+ */
+static char *get_process_name(int pid)
+{
+	char path[32];
+	char buf[PATH_MAX];
+	char *ptr;
+	int n;
+
+	snprintf(path, sizeof(path), "/proc/%d/exe", pid);
+	n = readlink(path, buf, sizeof(buf));
+	if (n < 0) {
+		return NULL;
+	}
+
+	/* Remove any extra fields */
+	buf[n] = '\0';
+	ptr = strtok(buf, " ");
+	return strdup(ptr);
+}
+
+
+static bool get_tdb_lock_info(struct ctdb_db_context *ctdb_db, TDB_DATA key,
+			       struct lock_record *rec)
+{
+	struct stat statbuf;
+
+	if (stat(ctdb_db->db_path, &statbuf) < 0) {
+		DEBUG(DEBUG_ERR, ("Fail to get inode info for TDB %s", ctdb_db->db_path));
+		return false;
+	}
+	rec->inode = statbuf.st_ino;
+
+	/* TODO: find the required offset from key */
+	rec->start = 0;
+	rec->end = (off_t)-1;
+	return true;
+}
+
+
+static bool get_blocker_pid(struct lock_record *reqlock, pid_t *blocker_pid)
+{
+	FILE *fp;
+	char buf[1024];
+	char *ptr;
+	pid_t pid;
+	struct lock_record curlock;
+	bool status = false;
+
+	fp = fopen("/proc/locks", "r");
+	if (!fp) {
+		DEBUG(DEBUG_ERR, ("Failed to read locks information"));
+		return false;
+	}
+	while ((ptr = fgets(buf, sizeof(buf), fp)) != NULL) {
+		if (! parse_proc_locks_line(buf, &pid, &curlock)) {
+			continue;
+		}
+
+		if ((curlock.inode == reqlock->inode) && 
+		    (curlock.start > reqlock->end || 
+		     curlock.end < reqlock->start)) {
+			/* Outside the required range */
+			continue;
+		}
+		*blocker_pid = pid;
+		status = true;
+		break;
+	}
+	fclose(fp);
+
+	if (!status) {
+		DEBUG(DEBUG_WARNING, ("Failed to find blocking culprit"));
+	}
+
+	return status;
+}
+
+
+static void terminate_pid(pid_t blocker_pid, bool terminate)
+{
+	char *process_name;
+
+	process_name = get_process_name(blocker_pid);
+	if (process_name) {
+		DEBUG(DEBUG_INFO, ("Process %s (pid=%d) is blocking", process_name, blocker_pid));
+		free(process_name);
+	}
+	
+	/* Kill process if required */
+	if (terminate) {
+		kill(blocker_pid, SIGKILL);
+	}
+}
+
+
+/*
+ * Find processes that holds lock we are interested in
+ */
+void ctdb_lock_find_blocker(struct lock_db_handle *h, bool terminate)
+{
+	struct lock_record reqlock;
+	pid_t blocker_pid;
+	struct ctdb_db_context *ctdb_db;
+
+	switch (h->type) {
+	case LOCK_DB_RECORD:
+		if (!get_tdb_lock_info(h->ctdb_db, h->key, &reqlock)) {
+			return;
+		}
+		if (!get_blocker_pid(&reqlock, &blocker_pid)) {
+			return;
+		}
+		terminate_pid(blocker_pid, terminate);
+		break;
+
+	case LOCK_DB_ONE:
+		if (!get_tdb_lock_info(h->ctdb_db, tdb_null, &reqlock)) {
+			return;
+		}
+		if (!get_blocker_pid(&reqlock, &blocker_pid)) {
+			return;
+		}
+		terminate_pid(blocker_pid, terminate);
+		break;
+
+	case LOCK_DB_ALL_PRIO:
+		for (ctdb_db = h->ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
+			if (!get_tdb_lock_info(ctdb_db, tdb_null, &reqlock)) {
+				continue;
+			}
+			if (!get_blocker_pid(&reqlock, &blocker_pid)) {
+				continue;
+			}
+			terminate_pid(blocker_pid, terminate);
+		}
+		break;
+	}
+}
+
+
+/*
+ * Destructor to kill the child locking process
+ */
+static int lock_db_destructor(struct lock_db_handle *h)
+{
+	kill(h->child, SIGKILL);
+	return 0;
+}
+
+
+/*
+ * Callback routine when the required locks are obtained.
+ * Called from parent context
+ */
+static void lock_db_handler(struct tevent_context *ev, 
+			    struct tevent_fd *tfd,
+			    uint16_t flags,
+			    void *private_data)
+{
+	struct lock_db_handle *h;
+	TALLOC_CTX *tmp_ctx;
+
+	h = talloc_get_type_abort(private_data, struct lock_db_handle);
+
+	/* cancel the timeout event */
+	if (h->ttimer) {
+		talloc_free(h->ttimer);
+		h->ttimer = NULL;
+	}
+
+	tmp_ctx = talloc_new(ev);
+	talloc_steal(tmp_ctx, h);
+	
+	/* TODO: read the byte from the pipe and see if the lock succeeded */
+
+	switch (h->type) {
+	case LOCK_DB_RECORD:
+		tdb_chainlock_mark(h->ctdb_db->ltdb->tdb, h->key);
+		h->lock_callback(h->lock_private_data);
+		tdb_chainlock_unmark(h->ctdb_db->ltdb->tdb, h->key);
+		break;
+
+	case LOCK_DB_ONE:
+		tdb_lockall_mark(h->ctdb_db->ltdb->tdb);
+		h->lock_callback(h->lock_private_data);
+		tdb_lockall_unmark(h->ctdb_db->ltdb->tdb);
+		break;
+
+	case LOCK_DB_ALL_PRIO:
+		ctdb_lockall_mark(h->ctdb, h->priority);
+		h->lock_callback(h->lock_private_data);
+		ctdb_lockall_unmark(h->ctdb, h->priority);
+		break;
+	} 
+
+	talloc_free(tmp_ctx);
+}
+
+
+/*
+ * Callback routine when required locks are not obtained within timeout
+ * Called from parent context
+ */
+static void lock_db_timeout_handler(struct tevent_context *ev,
+				    struct tevent_timer *ttimer,
+				    struct timeval current_time,
+				    void *private_data)
+{
+	struct lock_db_handle *h;
+	TALLOC_CTX *tmp_ctx;
+
+	h = talloc_get_type_abort(private_data, struct lock_db_handle);
+
+	/* cancel the fd event */
+	if (h->tfd) {
+		talloc_free(h->tfd);
+		h->tfd = NULL;
+	}
+
+	tmp_ctx = talloc_new(ev);
+	talloc_steal(tmp_ctx, h);
+
+	/* TODO: Find out which process is holding required lock */
+	ctdb_lock_find_blocker(h, false);
+
+	if (h->timeout_callback) {
+		h->timeout_callback(h->timeout_private_data);
+	}
+
+	talloc_free(tmp_ctx);
+}
+
+
+/*
+ * Lock record / db / all dbs depending on type
+ * Set up callback handler and timeout handler
+ */
+static struct lock_db_handle *ctdb_lock_internal(struct ctdb_context *ctdb,
+						 struct ctdb_db_context *ctdb_db,
+						 TDB_DATA key,
+						 uint32_t priority,
+						 void (*lock_callback)(void *private_data),
+						 void *lock_private_data,
+						 void (*timeout_callback)(void *private_data),
+						 void *timeout_private_data,
+						 enum lock_db_type type)
+{
+	struct lock_db_handle *result;
+	int ret;
+	pid_t parent;
+
+	if (lock_callback == NULL) {
+		DEBUG(DEBUG_WARNING, ("No callback function specified, not locking"));
+		return NULL;
+	}
+
+	if ((result = talloc_zero(ctdb, struct lock_db_handle)) == NULL) {
+		return NULL;
+	}
+
+	result->type = type;
+	result->ctdb = ctdb;
+
+	switch (type) {
+	case LOCK_DB_RECORD:
+		result->ctdb_db = ctdb_db;
+		result->key = key;
+		break;
+
+	case LOCK_DB_ONE:
+		result->ctdb_db = ctdb_db;
+		break;
+
+	case LOCK_DB_ALL_PRIO:
+		result->priority = priority;
+		break;
+	}
+
+	result->lock_callback = lock_callback;
+	result->lock_private_data = lock_private_data;
+	result->timeout_callback = timeout_callback;
+	result->timeout_private_data = timeout_private_data;
+	result->lock_in_parent = false;
+	
+	result->child = -1;
+
+	if (ctdb_lock_item_nonblock(result) == 0) {
+		result->lock_callback(result->lock_private_data);
+		ctdb_unlock_item(result);
+		/* FIXME: How is result destroyed? */
+		return result;
+	}
+
+	ret = pipe(result->fd);
+	if (ret != 0) {
+		DEBUG(DEBUG_ERR, ("Failed to create pipe for ctdb_lock_internal\n"));
+		talloc_free(result);
+		return NULL;
+	}
+
+	parent = getpid();
+	result->child = ctdb_fork(ctdb);
+
+	if (result->child == (pid_t)-1) {
+		DEBUG(DEBUG_ERR, ("Failed to fork child for ctdb_lock_internal\n"));
+		close(result->fd[0]);
+		close(result->fd[1]);
+		talloc_free(result);
+		return NULL;
+	}
+
+	/* Child process - do locking here */
+	if (result->child == 0) {
+		char c;
+		close(result->fd[0]);
+		if (ctdb_lock_item(result)) {
+			c = 0;
+		} else {
+			c = -1;
+		}
+		write(result->fd[1], &c, 1);
+
+		/* Hang around, but if parent dies, terminate */
+		while (kill(parent, 0) == 0 || errno != ESRCH) {
+			sleep(5);
+		}
+		_exit(0);
+	}
+
+	/* Set up callback, when the child process gets lock */
+	close(result->fd[1]);
+	set_close_on_exec(result->fd[0]);
+
+	talloc_set_destructor(result, lock_db_destructor);
+
+	result->tfd = tevent_add_fd(ctdb->ev, result, result->fd[0],
+				EVENT_FD_READ, lock_db_handler, 
+				(void *)result);
+	if (result->tfd == NULL) {
+		talloc_free(result);
+		return NULL;
+	}
+	tevent_fd_set_auto_close(result->tfd);
+
+	/* Set up callback if we timeout waiting for child to get lock */
+	result->ttimer = tevent_add_timer(ctdb->ev, 
+					  result, 
+					  timeval_current_ofs(ctdb->tunable.control_timeout, 0),
+					  lock_db_timeout_handler,
+					  (void *)result);
+
+	result->start_time = timeval_current();
+	return result;
+}
+
+
+/*
+ * obtain a lock on a record in a database
+ */
+struct lock_db_handle *ctdb_lock_record(struct ctdb_context *ctdb,
+					struct ctdb_db_context *ctdb_db,
+					TDB_DATA key,
+					void (*lock_callback)(void *private_data),
+					void *lock_private_data,
+					void (*timeout_callback)(void *private_data),
+					void *timeout_private_data)
+{
+	return ctdb_lock_internal(ctdb, 
+				  ctdb_db, 
+				  key, 
+				  0,
+				  lock_callback, 
+				  lock_private_data,
+				  timeout_callback,
+				  timeout_private_data,
+				  LOCK_DB_RECORD);
+}
+
+
+/*
+ * obtain a lock on a database
+ */
+struct lock_db_handle *ctdb_lock_db(struct ctdb_context *ctdb,
+					struct ctdb_db_context *ctdb_db,
+					void (*lock_callback)(void *private_data),
+					void *lock_private_data,
+					void (*timeout_callback)(void *private_data),
+					void *timeout_private_data)
+{
+	return ctdb_lock_internal(ctdb, 
+				  ctdb_db, 
+				  tdb_null, 
+				  0,
+				  lock_callback, 
+				  lock_private_data,
+				  timeout_callback,
+				  timeout_private_data,
+				  LOCK_DB_ONE);
+}
+
+
+/*
+ * Obtain locks on all databases
+ */
+struct lock_db_handle *ctdb_lock_alldb(struct ctdb_context *ctdb,
+					uint32_t priority,
+					void (*lock_callback)(void *private_data),
+					void *lock_private_data,
+					void (*timeout_callback)(void *private_data),
+					void *timeout_private_data)
+{
+	return ctdb_lock_internal(ctdb, 
+				  NULL, 
+				  tdb_null, 
+				  priority,
+				  lock_callback, 
+				  lock_private_data,
+				  timeout_callback,
+				  timeout_private_data,
+				  LOCK_DB_ALL_PRIO);
+}