1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * dentry cache handling code
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/namei.h>
31 #define MLOG_MASK_PREFIX ML_DCACHE
32 #include <cluster/masklog.h>
43 void ocfs2_dentry_attach_gen(struct dentry
*dentry
)
46 OCFS2_I(dentry
->d_parent
->d_inode
)->ip_dir_lock_gen
;
47 BUG_ON(dentry
->d_inode
);
48 dentry
->d_fsdata
= (void *)gen
;
52 static int ocfs2_dentry_revalidate(struct dentry
*dentry
,
56 int ret
= 0; /* if all else fails, just return false */
57 struct ocfs2_super
*osb
;
59 if (nd
&& nd
->flags
& LOOKUP_RCU
)
62 inode
= dentry
->d_inode
;
63 osb
= OCFS2_SB(dentry
->d_sb
);
65 mlog_entry("(0x%p, '%.*s')\n", dentry
,
66 dentry
->d_name
.len
, dentry
->d_name
.name
);
68 /* For a negative dentry -
69 * check the generation number of the parent and compare with the
70 * one stored in the inode.
73 unsigned long gen
= (unsigned long) dentry
->d_fsdata
;
75 OCFS2_I(dentry
->d_parent
->d_inode
)->ip_dir_lock_gen
;
76 mlog(0, "negative dentry: %.*s parent gen: %lu "
78 dentry
->d_name
.len
, dentry
->d_name
.name
, pgen
, gen
);
86 if (inode
== osb
->root_inode
|| is_bad_inode(inode
))
89 spin_lock(&OCFS2_I(inode
)->ip_lock
);
90 /* did we or someone else delete this inode? */
91 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
92 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
93 mlog(0, "inode (%llu) deleted, returning false\n",
94 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
97 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
100 * We don't need a cluster lock to test this because once an
101 * inode nlink hits zero, it never goes back.
103 if (inode
->i_nlink
== 0) {
104 mlog(0, "Inode %llu orphaned, returning false "
106 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
107 S_ISDIR(inode
->i_mode
));
112 * If the last lookup failed to create dentry lock, let us
115 if (!dentry
->d_fsdata
) {
116 mlog(0, "Inode %llu doesn't have dentry lock, "
118 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
131 static int ocfs2_match_dentry(struct dentry
*dentry
,
135 struct inode
*parent
;
138 * ocfs2_lookup() does a d_splice_alias() _before_ attaching
139 * to the lock data, so we skip those here, otherwise
140 * ocfs2_dentry_attach_lock() will get its original dentry
143 if (!dentry
->d_fsdata
)
146 if (!dentry
->d_parent
)
149 if (skip_unhashed
&& d_unhashed(dentry
))
152 parent
= dentry
->d_parent
->d_inode
;
153 /* Negative parent dentry? */
157 /* Name is in a different directory. */
158 if (OCFS2_I(parent
)->ip_blkno
!= parent_blkno
)
165 * Walk the inode alias list, and find a dentry which has a given
166 * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
167 * is looking for a dentry_lock reference. The downconvert thread is
168 * looking to unhash aliases, so we allow it to skip any that already
169 * have that property.
171 struct dentry
*ocfs2_find_local_alias(struct inode
*inode
,
176 struct dentry
*dentry
= NULL
;
178 spin_lock(&inode
->i_lock
);
179 list_for_each(p
, &inode
->i_dentry
) {
180 dentry
= list_entry(p
, struct dentry
, d_alias
);
182 spin_lock(&dentry
->d_lock
);
183 if (ocfs2_match_dentry(dentry
, parent_blkno
, skip_unhashed
)) {
184 mlog(0, "dentry found: %.*s\n",
185 dentry
->d_name
.len
, dentry
->d_name
.name
);
188 spin_unlock(&dentry
->d_lock
);
191 spin_unlock(&dentry
->d_lock
);
196 spin_unlock(&inode
->i_lock
);
201 DEFINE_SPINLOCK(dentry_attach_lock
);
204 * Attach this dentry to a cluster lock.
206 * Dentry locks cover all links in a given directory to a particular
207 * inode. We do this so that ocfs2 can build a lock name which all
208 * nodes in the cluster can agree on at all times. Shoving full names
209 * in the cluster lock won't work due to size restrictions. Covering
210 * links inside of a directory is a good compromise because it still
211 * allows us to use the parent directory lock to synchronize
214 * Call this function with the parent dir semaphore and the parent dir
217 * The dir semaphore will protect us from having to worry about
218 * concurrent processes on our node trying to attach a lock at the
221 * The dir cluster lock (held at either PR or EX mode) protects us
222 * from unlink and rename on other nodes.
224 * A dput() can happen asynchronously due to pruning, so we cover
225 * attaching and detaching the dentry lock with a
226 * dentry_attach_lock.
228 * A node which has done lookup on a name retains a protected read
229 * lock until final dput. If the user requests and unlink or rename,
230 * the protected read is upgraded to an exclusive lock. Other nodes
231 * who have seen the dentry will then be informed that they need to
232 * downgrade their lock, which will involve d_delete on the
233 * dentry. This happens in ocfs2_dentry_convert_worker().
235 int ocfs2_dentry_attach_lock(struct dentry
*dentry
,
240 struct dentry
*alias
;
241 struct ocfs2_dentry_lock
*dl
= dentry
->d_fsdata
;
243 mlog(0, "Attach \"%.*s\", parent %llu, fsdata: %p\n",
244 dentry
->d_name
.len
, dentry
->d_name
.name
,
245 (unsigned long long)parent_blkno
, dl
);
248 * Negative dentry. We ignore these for now.
250 * XXX: Could we can improve ocfs2_dentry_revalidate() by
256 if (!dentry
->d_inode
&& dentry
->d_fsdata
) {
257 /* Converting a negative dentry to positive
258 Clear dentry->d_fsdata */
259 dentry
->d_fsdata
= dl
= NULL
;
263 mlog_bug_on_msg(dl
->dl_parent_blkno
!= parent_blkno
,
264 " \"%.*s\": old parent: %llu, new: %llu\n",
265 dentry
->d_name
.len
, dentry
->d_name
.name
,
266 (unsigned long long)parent_blkno
,
267 (unsigned long long)dl
->dl_parent_blkno
);
271 alias
= ocfs2_find_local_alias(inode
, parent_blkno
, 0);
274 * Great, an alias exists, which means we must have a
275 * dentry lock already. We can just grab the lock off
276 * the alias and add it to the list.
278 * We're depending here on the fact that this dentry
279 * was found and exists in the dcache and so must have
280 * a reference to the dentry_lock because we can't
281 * race creates. Final dput() cannot happen on it
282 * since we have it pinned, so our reference is safe.
284 dl
= alias
->d_fsdata
;
285 mlog_bug_on_msg(!dl
, "parent %llu, ino %llu\n",
286 (unsigned long long)parent_blkno
,
287 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
289 mlog_bug_on_msg(dl
->dl_parent_blkno
!= parent_blkno
,
290 " \"%.*s\": old parent: %llu, new: %llu\n",
291 dentry
->d_name
.len
, dentry
->d_name
.name
,
292 (unsigned long long)parent_blkno
,
293 (unsigned long long)dl
->dl_parent_blkno
);
295 mlog(0, "Found: %s\n", dl
->dl_lockres
.l_name
);
301 * There are no other aliases
303 dl
= kmalloc(sizeof(*dl
), GFP_NOFS
);
312 * Does this have to happen below, for all attaches, in case
313 * the struct inode gets blown away by the downconvert thread?
315 dl
->dl_inode
= igrab(inode
);
316 dl
->dl_parent_blkno
= parent_blkno
;
317 ocfs2_dentry_lock_res_init(dl
, parent_blkno
, inode
);
320 spin_lock(&dentry_attach_lock
);
321 dentry
->d_fsdata
= dl
;
323 spin_unlock(&dentry_attach_lock
);
326 * This actually gets us our PRMODE level lock. From now on,
327 * we'll have a notification if one of these names is
328 * destroyed on another node.
330 ret
= ocfs2_dentry_lock(dentry
, 0);
332 ocfs2_dentry_unlock(dentry
, 0);
337 * In case of error, manually free the allocation and do the iput().
338 * We need to do this because error here means no d_instantiate(),
339 * which means iput() will not be called during dput(dentry).
341 if (ret
< 0 && !alias
) {
342 ocfs2_lock_res_free(&dl
->dl_lockres
);
343 BUG_ON(dl
->dl_count
!= 1);
344 spin_lock(&dentry_attach_lock
);
345 dentry
->d_fsdata
= NULL
;
346 spin_unlock(&dentry_attach_lock
);
356 DEFINE_SPINLOCK(dentry_list_lock
);
358 /* We limit the number of dentry locks to drop in one go. We have
359 * this limit so that we don't starve other users of ocfs2_wq. */
360 #define DL_INODE_DROP_COUNT 64
362 /* Drop inode references from dentry locks */
363 static void __ocfs2_drop_dl_inodes(struct ocfs2_super
*osb
, int drop_count
)
365 struct ocfs2_dentry_lock
*dl
;
367 spin_lock(&dentry_list_lock
);
368 while (osb
->dentry_lock_list
&& (drop_count
< 0 || drop_count
--)) {
369 dl
= osb
->dentry_lock_list
;
370 osb
->dentry_lock_list
= dl
->dl_next
;
371 spin_unlock(&dentry_list_lock
);
374 spin_lock(&dentry_list_lock
);
376 spin_unlock(&dentry_list_lock
);
379 void ocfs2_drop_dl_inodes(struct work_struct
*work
)
381 struct ocfs2_super
*osb
= container_of(work
, struct ocfs2_super
,
384 __ocfs2_drop_dl_inodes(osb
, DL_INODE_DROP_COUNT
);
386 * Don't queue dropping if umount is in progress. We flush the
387 * list in ocfs2_dismount_volume
389 spin_lock(&dentry_list_lock
);
390 if (osb
->dentry_lock_list
&&
391 !ocfs2_test_osb_flag(osb
, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED
))
392 queue_work(ocfs2_wq
, &osb
->dentry_lock_work
);
393 spin_unlock(&dentry_list_lock
);
396 /* Flush the whole work queue */
397 void ocfs2_drop_all_dl_inodes(struct ocfs2_super
*osb
)
399 __ocfs2_drop_dl_inodes(osb
, -1);
403 * ocfs2_dentry_iput() and friends.
405 * At this point, our particular dentry is detached from the inodes
406 * alias list, so there's no way that the locking code can find it.
408 * The interesting stuff happens when we determine that our lock needs
409 * to go away because this is the last subdir alias in the
410 * system. This function needs to handle a couple things:
412 * 1) Synchronizing lock shutdown with the downconvert threads. This
413 * is already handled for us via the lockres release drop function
414 * called in ocfs2_release_dentry_lock()
416 * 2) A race may occur when we're doing our lock shutdown and
417 * another process wants to create a new dentry lock. Right now we
418 * let them race, which means that for a very short while, this
419 * node might have two locks on a lock resource. This should be a
420 * problem though because one of them is in the process of being
423 static void ocfs2_drop_dentry_lock(struct ocfs2_super
*osb
,
424 struct ocfs2_dentry_lock
*dl
)
426 ocfs2_simple_drop_lockres(osb
, &dl
->dl_lockres
);
427 ocfs2_lock_res_free(&dl
->dl_lockres
);
429 /* We leave dropping of inode reference to ocfs2_wq as that can
430 * possibly lead to inode deletion which gets tricky */
431 spin_lock(&dentry_list_lock
);
432 if (!osb
->dentry_lock_list
&&
433 !ocfs2_test_osb_flag(osb
, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED
))
434 queue_work(ocfs2_wq
, &osb
->dentry_lock_work
);
435 dl
->dl_next
= osb
->dentry_lock_list
;
436 osb
->dentry_lock_list
= dl
;
437 spin_unlock(&dentry_list_lock
);
440 void ocfs2_dentry_lock_put(struct ocfs2_super
*osb
,
441 struct ocfs2_dentry_lock
*dl
)
445 BUG_ON(dl
->dl_count
== 0);
447 spin_lock(&dentry_attach_lock
);
449 unlock
= !dl
->dl_count
;
450 spin_unlock(&dentry_attach_lock
);
453 ocfs2_drop_dentry_lock(osb
, dl
);
456 static void ocfs2_dentry_iput(struct dentry
*dentry
, struct inode
*inode
)
458 struct ocfs2_dentry_lock
*dl
= dentry
->d_fsdata
;
462 * No dentry lock is ok if we're disconnected or
465 if (!(dentry
->d_flags
& DCACHE_DISCONNECTED
) &&
466 !d_unhashed(dentry
)) {
467 unsigned long long ino
= 0ULL;
469 ino
= (unsigned long long)OCFS2_I(inode
)->ip_blkno
;
470 mlog(ML_ERROR
, "Dentry is missing cluster lock. "
471 "inode: %llu, d_flags: 0x%x, d_name: %.*s\n",
472 ino
, dentry
->d_flags
, dentry
->d_name
.len
,
473 dentry
->d_name
.name
);
479 mlog_bug_on_msg(dl
->dl_count
== 0, "dentry: %.*s, count: %u\n",
480 dentry
->d_name
.len
, dentry
->d_name
.name
,
483 ocfs2_dentry_lock_put(OCFS2_SB(dentry
->d_sb
), dl
);
490 * d_move(), but keep the locks in sync.
492 * When we are done, "dentry" will have the parent dir and name of
493 * "target", which will be thrown away.
495 * We manually update the lock of "dentry" if need be.
497 * "target" doesn't have it's dentry lock touched - we allow the later
498 * dput() to handle this for us.
500 * This is called during ocfs2_rename(), while holding parent
501 * directory locks. The dentries have already been deleted on other
502 * nodes via ocfs2_remote_dentry_delete().
504 * Normally, the VFS handles the d_move() for the file system, after
505 * the ->rename() callback. OCFS2 wants to handle this internally, so
506 * the new lock can be created atomically with respect to the cluster.
508 void ocfs2_dentry_move(struct dentry
*dentry
, struct dentry
*target
,
509 struct inode
*old_dir
, struct inode
*new_dir
)
512 struct ocfs2_super
*osb
= OCFS2_SB(old_dir
->i_sb
);
513 struct inode
*inode
= dentry
->d_inode
;
516 * Move within the same directory, so the actual lock info won't
519 * XXX: Is there any advantage to dropping the lock here?
521 if (old_dir
== new_dir
)
524 ocfs2_dentry_lock_put(osb
, dentry
->d_fsdata
);
526 dentry
->d_fsdata
= NULL
;
527 ret
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(new_dir
)->ip_blkno
);
532 d_move(dentry
, target
);
535 const struct dentry_operations ocfs2_dentry_ops
= {
536 .d_revalidate
= ocfs2_dentry_revalidate
,
537 .d_iput
= ocfs2_dentry_iput
,