2 * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License version 2.
10 #include <linux/miscdevice.h>
11 #include <linux/poll.h>
12 #include <linux/dlm.h>
13 #include <linux/dlm_plock.h>
14 #include <linux/slab.h>
16 #include "dlm_internal.h"
17 #include "lockspace.h"
19 static spinlock_t ops_lock;
20 static struct list_head send_list;
21 static struct list_head recv_list;
22 static wait_queue_head_t send_wq;
23 static wait_queue_head_t recv_wq;
26 struct list_head list;
28 struct dlm_plock_info info;
29 int (*callback)(struct file_lock *fl, int result);
40 static inline void set_version(struct dlm_plock_info *info)
42 info->version[0] = DLM_PLOCK_VERSION_MAJOR;
43 info->version[1] = DLM_PLOCK_VERSION_MINOR;
44 info->version[2] = DLM_PLOCK_VERSION_PATCH;
47 static int check_version(struct dlm_plock_info *info)
49 if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
50 (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
51 log_print("plock device version mismatch: "
52 "kernel (%u.%u.%u), user (%u.%u.%u)",
53 DLM_PLOCK_VERSION_MAJOR,
54 DLM_PLOCK_VERSION_MINOR,
55 DLM_PLOCK_VERSION_PATCH,
64 static void send_op(struct plock_op *op)
66 set_version(&op->info);
67 INIT_LIST_HEAD(&op->list);
69 list_add_tail(&op->list, &send_list);
70 spin_unlock(&ops_lock);
74 /* If a process was killed while waiting for the only plock on a file,
75 locks_remove_posix will not see any lock on the file so it won't
76 send an unlock-close to us to pass on to userspace to clean up the
77 abandoned waiter. So, we have to insert the unlock-close when the
78 lock call is interrupted. */
80 static void do_unlock_close(struct dlm_ls *ls, u64 number,
81 struct file *file, struct file_lock *fl)
85 op = kzalloc(sizeof(*op), GFP_NOFS);
89 op->info.optype = DLM_PLOCK_OP_UNLOCK;
90 op->info.pid = fl->fl_pid;
91 op->info.fsid = ls->ls_global_id;
92 op->info.number = number;
94 op->info.end = OFFSET_MAX;
95 if (fl->fl_lmops && fl->fl_lmops->lm_grant)
96 op->info.owner = (__u64) fl->fl_pid;
98 op->info.owner = (__u64)(long) fl->fl_owner;
100 op->info.flags |= DLM_PLOCK_FL_CLOSE;
104 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
105 int cmd, struct file_lock *fl)
109 struct plock_xop *xop;
112 ls = dlm_find_lockspace_local(lockspace);
116 xop = kzalloc(sizeof(*xop), GFP_NOFS);
123 op->info.optype = DLM_PLOCK_OP_LOCK;
124 op->info.pid = fl->fl_pid;
125 op->info.ex = (fl->fl_type == F_WRLCK);
126 op->info.wait = IS_SETLKW(cmd);
127 op->info.fsid = ls->ls_global_id;
128 op->info.number = number;
129 op->info.start = fl->fl_start;
130 op->info.end = fl->fl_end;
131 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
132 /* fl_owner is lockd which doesn't distinguish
133 processes on the nfs client */
134 op->info.owner = (__u64) fl->fl_pid;
135 op->callback = fl->fl_lmops->lm_grant;
136 locks_init_lock(&xop->flc);
137 locks_copy_lock(&xop->flc, fl);
141 op->info.owner = (__u64)(long) fl->fl_owner;
147 rv = wait_event_interruptible(recv_wq, (op->done != 0));
148 if (rv == -ERESTARTSYS) {
149 log_debug(ls, "dlm_posix_lock: wait killed %llx",
150 (unsigned long long)number);
151 spin_lock(&ops_lock);
153 spin_unlock(&ops_lock);
155 do_unlock_close(ls, number, file, fl);
159 rv = FILE_LOCK_DEFERRED;
163 spin_lock(&ops_lock);
164 if (!list_empty(&op->list)) {
165 log_error(ls, "dlm_posix_lock: op on list %llx",
166 (unsigned long long)number);
169 spin_unlock(&ops_lock);
174 if (locks_lock_file_wait(file, fl) < 0)
175 log_error(ls, "dlm_posix_lock: vfs lock error %llx",
176 (unsigned long long)number);
181 dlm_put_lockspace(ls);
184 EXPORT_SYMBOL_GPL(dlm_posix_lock);
186 /* Returns failure iff a successful lock operation should be canceled */
187 static int dlm_plock_callback(struct plock_op *op)
190 struct file_lock *fl;
191 struct file_lock *flc;
192 int (*notify)(struct file_lock *fl, int result) = NULL;
193 struct plock_xop *xop = (struct plock_xop *)op;
196 spin_lock(&ops_lock);
197 if (!list_empty(&op->list)) {
198 log_print("dlm_plock_callback: op on list %llx",
199 (unsigned long long)op->info.number);
202 spin_unlock(&ops_lock);
204 /* check if the following 2 are still valid or make a copy */
208 notify = op->callback;
211 notify(fl, op->info.rv);
215 /* got fs lock; bookkeep locally as well: */
216 flc->fl_flags &= ~FL_SLEEP;
217 if (posix_lock_file(file, flc, NULL)) {
219 * This can only happen in the case of kmalloc() failure.
220 * The filesystem's own lock is the authoritative lock,
221 * so a failure to get the lock locally is not a disaster.
222 * As long as the fs cannot reliably cancel locks (especially
223 * in a low-memory situation), we're better off ignoring
224 * this failure than trying to recover.
226 log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
227 (unsigned long long)op->info.number, file, fl);
232 /* XXX: We need to cancel the fs lock here: */
233 log_print("dlm_plock_callback: lock granted after lock request "
234 "failed; dangling lock!\n");
243 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
244 struct file_lock *fl)
249 unsigned char fl_flags = fl->fl_flags;
251 ls = dlm_find_lockspace_local(lockspace);
255 op = kzalloc(sizeof(*op), GFP_NOFS);
261 /* cause the vfs unlock to return ENOENT if lock is not found */
262 fl->fl_flags |= FL_EXISTS;
264 rv = locks_lock_file_wait(file, fl);
270 log_error(ls, "dlm_posix_unlock: vfs unlock error %d %llx",
271 rv, (unsigned long long)number);
274 op->info.optype = DLM_PLOCK_OP_UNLOCK;
275 op->info.pid = fl->fl_pid;
276 op->info.fsid = ls->ls_global_id;
277 op->info.number = number;
278 op->info.start = fl->fl_start;
279 op->info.end = fl->fl_end;
280 if (fl->fl_lmops && fl->fl_lmops->lm_grant)
281 op->info.owner = (__u64) fl->fl_pid;
283 op->info.owner = (__u64)(long) fl->fl_owner;
285 if (fl->fl_flags & FL_CLOSE) {
286 op->info.flags |= DLM_PLOCK_FL_CLOSE;
293 wait_event(recv_wq, (op->done != 0));
295 spin_lock(&ops_lock);
296 if (!list_empty(&op->list)) {
297 log_error(ls, "dlm_posix_unlock: op on list %llx",
298 (unsigned long long)number);
301 spin_unlock(&ops_lock);
311 dlm_put_lockspace(ls);
312 fl->fl_flags = fl_flags;
315 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
317 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
318 struct file_lock *fl)
324 ls = dlm_find_lockspace_local(lockspace);
328 op = kzalloc(sizeof(*op), GFP_NOFS);
334 op->info.optype = DLM_PLOCK_OP_GET;
335 op->info.pid = fl->fl_pid;
336 op->info.ex = (fl->fl_type == F_WRLCK);
337 op->info.fsid = ls->ls_global_id;
338 op->info.number = number;
339 op->info.start = fl->fl_start;
340 op->info.end = fl->fl_end;
341 if (fl->fl_lmops && fl->fl_lmops->lm_grant)
342 op->info.owner = (__u64) fl->fl_pid;
344 op->info.owner = (__u64)(long) fl->fl_owner;
347 wait_event(recv_wq, (op->done != 0));
349 spin_lock(&ops_lock);
350 if (!list_empty(&op->list)) {
351 log_error(ls, "dlm_posix_get: op on list %llx",
352 (unsigned long long)number);
355 spin_unlock(&ops_lock);
357 /* info.rv from userspace is 1 for conflict, 0 for no-conflict,
358 -ENOENT if there are no locks on the file */
362 fl->fl_type = F_UNLCK;
367 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
368 fl->fl_flags = FL_POSIX;
369 fl->fl_pid = op->info.pid;
370 fl->fl_start = op->info.start;
371 fl->fl_end = op->info.end;
377 dlm_put_lockspace(ls);
380 EXPORT_SYMBOL_GPL(dlm_posix_get);
382 /* a read copies out one plock request from the send list */
383 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
386 struct dlm_plock_info info;
387 struct plock_op *op = NULL;
389 if (count < sizeof(info))
392 spin_lock(&ops_lock);
393 if (!list_empty(&send_list)) {
394 op = list_entry(send_list.next, struct plock_op, list);
395 if (op->info.flags & DLM_PLOCK_FL_CLOSE)
398 list_move(&op->list, &recv_list);
399 memcpy(&info, &op->info, sizeof(info));
401 spin_unlock(&ops_lock);
406 /* there is no need to get a reply from userspace for unlocks
407 that were generated by the vfs cleaning up for a close
408 (the process did not make an unlock call). */
410 if (op->info.flags & DLM_PLOCK_FL_CLOSE)
413 if (copy_to_user(u, &info, sizeof(info)))
418 /* a write copies in one plock result that should match a plock_op
420 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
423 struct dlm_plock_info info;
425 int found = 0, do_callback = 0;
427 if (count != sizeof(info))
430 if (copy_from_user(&info, u, sizeof(info)))
433 if (check_version(&info))
436 spin_lock(&ops_lock);
437 list_for_each_entry(op, &recv_list, list) {
438 if (op->info.fsid == info.fsid &&
439 op->info.number == info.number &&
440 op->info.owner == info.owner) {
441 list_del_init(&op->list);
442 memcpy(&op->info, &info, sizeof(info));
451 spin_unlock(&ops_lock);
455 dlm_plock_callback(op);
459 log_print("dev_write no op %x %llx", info.fsid,
460 (unsigned long long)info.number);
464 static unsigned int dev_poll(struct file *file, poll_table *wait)
466 unsigned int mask = 0;
468 poll_wait(file, &send_wq, wait);
470 spin_lock(&ops_lock);
471 if (!list_empty(&send_list))
472 mask = POLLIN | POLLRDNORM;
473 spin_unlock(&ops_lock);
478 static const struct file_operations dev_fops = {
482 .owner = THIS_MODULE,
483 .llseek = noop_llseek,
486 static struct miscdevice plock_dev_misc = {
487 .minor = MISC_DYNAMIC_MINOR,
488 .name = DLM_PLOCK_MISC_NAME,
492 int dlm_plock_init(void)
496 spin_lock_init(&ops_lock);
497 INIT_LIST_HEAD(&send_list);
498 INIT_LIST_HEAD(&recv_list);
499 init_waitqueue_head(&send_wq);
500 init_waitqueue_head(&recv_wq);
502 rv = misc_register(&plock_dev_misc);
504 log_print("dlm_plock_init: misc_register failed %d", rv);
508 void dlm_plock_exit(void)
510 misc_deregister(&plock_dev_misc);