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 #include <cluster/masklog.h>
41 #include "ocfs2_trace.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, unsigned int flags)
55 int ret = 0; /* if all else fails, just return false */
56 struct ocfs2_super *osb;
58 if (flags & LOOKUP_RCU)
61 inode = dentry->d_inode;
62 osb = OCFS2_SB(dentry->d_sb);
64 trace_ocfs2_dentry_revalidate(dentry, dentry->d_name.len,
67 /* For a negative dentry -
68 * check the generation number of the parent and compare with the
69 * one stored in the inode.
72 unsigned long gen = (unsigned long) dentry->d_fsdata;
74 OCFS2_I(dentry->d_parent->d_inode)->ip_dir_lock_gen;
76 trace_ocfs2_dentry_revalidate_negative(dentry->d_name.len,
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 trace_ocfs2_dentry_revalidate_delete(
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 trace_ocfs2_dentry_revalidate_orphaned(
105 (unsigned long long)OCFS2_I(inode)->ip_blkno,
106 S_ISDIR(inode->i_mode));
111 * If the last lookup failed to create dentry lock, let us
114 if (!dentry->d_fsdata) {
115 trace_ocfs2_dentry_revalidate_nofsdata(
116 (unsigned long long)OCFS2_I(inode)->ip_blkno);
124 trace_ocfs2_dentry_revalidate_ret(ret);
128 static int ocfs2_match_dentry(struct dentry *dentry,
132 struct inode *parent;
135 * ocfs2_lookup() does a d_splice_alias() _before_ attaching
136 * to the lock data, so we skip those here, otherwise
137 * ocfs2_dentry_attach_lock() will get its original dentry
140 if (!dentry->d_fsdata)
143 if (!dentry->d_parent)
146 if (skip_unhashed && d_unhashed(dentry))
149 parent = dentry->d_parent->d_inode;
150 /* Negative parent dentry? */
154 /* Name is in a different directory. */
155 if (OCFS2_I(parent)->ip_blkno != parent_blkno)
162 * Walk the inode alias list, and find a dentry which has a given
163 * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
164 * is looking for a dentry_lock reference. The downconvert thread is
165 * looking to unhash aliases, so we allow it to skip any that already
166 * have that property.
168 struct dentry *ocfs2_find_local_alias(struct inode *inode,
172 struct dentry *dentry;
174 spin_lock(&inode->i_lock);
175 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
176 spin_lock(&dentry->d_lock);
177 if (ocfs2_match_dentry(dentry, parent_blkno, skip_unhashed)) {
178 trace_ocfs2_find_local_alias(dentry->d_name.len,
179 dentry->d_name.name);
182 spin_unlock(&dentry->d_lock);
183 spin_unlock(&inode->i_lock);
186 spin_unlock(&dentry->d_lock);
188 spin_unlock(&inode->i_lock);
192 DEFINE_SPINLOCK(dentry_attach_lock);
195 * Attach this dentry to a cluster lock.
197 * Dentry locks cover all links in a given directory to a particular
198 * inode. We do this so that ocfs2 can build a lock name which all
199 * nodes in the cluster can agree on at all times. Shoving full names
200 * in the cluster lock won't work due to size restrictions. Covering
201 * links inside of a directory is a good compromise because it still
202 * allows us to use the parent directory lock to synchronize
205 * Call this function with the parent dir semaphore and the parent dir
208 * The dir semaphore will protect us from having to worry about
209 * concurrent processes on our node trying to attach a lock at the
212 * The dir cluster lock (held at either PR or EX mode) protects us
213 * from unlink and rename on other nodes.
215 * A dput() can happen asynchronously due to pruning, so we cover
216 * attaching and detaching the dentry lock with a
217 * dentry_attach_lock.
219 * A node which has done lookup on a name retains a protected read
220 * lock until final dput. If the user requests and unlink or rename,
221 * the protected read is upgraded to an exclusive lock. Other nodes
222 * who have seen the dentry will then be informed that they need to
223 * downgrade their lock, which will involve d_delete on the
224 * dentry. This happens in ocfs2_dentry_convert_worker().
226 int ocfs2_dentry_attach_lock(struct dentry *dentry,
231 struct dentry *alias;
232 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
234 trace_ocfs2_dentry_attach_lock(dentry->d_name.len, dentry->d_name.name,
235 (unsigned long long)parent_blkno, dl);
238 * Negative dentry. We ignore these for now.
240 * XXX: Could we can improve ocfs2_dentry_revalidate() by
246 if (!dentry->d_inode && dentry->d_fsdata) {
247 /* Converting a negative dentry to positive
248 Clear dentry->d_fsdata */
249 dentry->d_fsdata = dl = NULL;
253 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
254 " \"%.*s\": old parent: %llu, new: %llu\n",
255 dentry->d_name.len, dentry->d_name.name,
256 (unsigned long long)parent_blkno,
257 (unsigned long long)dl->dl_parent_blkno);
261 alias = ocfs2_find_local_alias(inode, parent_blkno, 0);
264 * Great, an alias exists, which means we must have a
265 * dentry lock already. We can just grab the lock off
266 * the alias and add it to the list.
268 * We're depending here on the fact that this dentry
269 * was found and exists in the dcache and so must have
270 * a reference to the dentry_lock because we can't
271 * race creates. Final dput() cannot happen on it
272 * since we have it pinned, so our reference is safe.
274 dl = alias->d_fsdata;
275 mlog_bug_on_msg(!dl, "parent %llu, ino %llu\n",
276 (unsigned long long)parent_blkno,
277 (unsigned long long)OCFS2_I(inode)->ip_blkno);
279 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
280 " \"%.*s\": old parent: %llu, new: %llu\n",
281 dentry->d_name.len, dentry->d_name.name,
282 (unsigned long long)parent_blkno,
283 (unsigned long long)dl->dl_parent_blkno);
285 trace_ocfs2_dentry_attach_lock_found(dl->dl_lockres.l_name,
286 (unsigned long long)parent_blkno,
287 (unsigned long long)OCFS2_I(inode)->ip_blkno);
293 * There are no other aliases
295 dl = kmalloc(sizeof(*dl), GFP_NOFS);
304 * Does this have to happen below, for all attaches, in case
305 * the struct inode gets blown away by the downconvert thread?
307 dl->dl_inode = igrab(inode);
308 dl->dl_parent_blkno = parent_blkno;
309 ocfs2_dentry_lock_res_init(dl, parent_blkno, inode);
312 spin_lock(&dentry_attach_lock);
313 dentry->d_fsdata = dl;
315 spin_unlock(&dentry_attach_lock);
318 * This actually gets us our PRMODE level lock. From now on,
319 * we'll have a notification if one of these names is
320 * destroyed on another node.
322 ret = ocfs2_dentry_lock(dentry, 0);
324 ocfs2_dentry_unlock(dentry, 0);
329 * In case of error, manually free the allocation and do the iput().
330 * We need to do this because error here means no d_instantiate(),
331 * which means iput() will not be called during dput(dentry).
333 if (ret < 0 && !alias) {
334 ocfs2_lock_res_free(&dl->dl_lockres);
335 BUG_ON(dl->dl_count != 1);
336 spin_lock(&dentry_attach_lock);
337 dentry->d_fsdata = NULL;
338 spin_unlock(&dentry_attach_lock);
348 DEFINE_SPINLOCK(dentry_list_lock);
350 /* We limit the number of dentry locks to drop in one go. We have
351 * this limit so that we don't starve other users of ocfs2_wq. */
352 #define DL_INODE_DROP_COUNT 64
354 /* Drop inode references from dentry locks */
355 static void __ocfs2_drop_dl_inodes(struct ocfs2_super *osb, int drop_count)
357 struct ocfs2_dentry_lock *dl;
359 spin_lock(&dentry_list_lock);
360 while (osb->dentry_lock_list && (drop_count < 0 || drop_count--)) {
361 dl = osb->dentry_lock_list;
362 osb->dentry_lock_list = dl->dl_next;
363 spin_unlock(&dentry_list_lock);
366 spin_lock(&dentry_list_lock);
368 spin_unlock(&dentry_list_lock);
371 void ocfs2_drop_dl_inodes(struct work_struct *work)
373 struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
376 __ocfs2_drop_dl_inodes(osb, DL_INODE_DROP_COUNT);
378 * Don't queue dropping if umount is in progress. We flush the
379 * list in ocfs2_dismount_volume
381 spin_lock(&dentry_list_lock);
382 if (osb->dentry_lock_list &&
383 !ocfs2_test_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED))
384 queue_work(ocfs2_wq, &osb->dentry_lock_work);
385 spin_unlock(&dentry_list_lock);
388 /* Flush the whole work queue */
389 void ocfs2_drop_all_dl_inodes(struct ocfs2_super *osb)
391 __ocfs2_drop_dl_inodes(osb, -1);
395 * ocfs2_dentry_iput() and friends.
397 * At this point, our particular dentry is detached from the inodes
398 * alias list, so there's no way that the locking code can find it.
400 * The interesting stuff happens when we determine that our lock needs
401 * to go away because this is the last subdir alias in the
402 * system. This function needs to handle a couple things:
404 * 1) Synchronizing lock shutdown with the downconvert threads. This
405 * is already handled for us via the lockres release drop function
406 * called in ocfs2_release_dentry_lock()
408 * 2) A race may occur when we're doing our lock shutdown and
409 * another process wants to create a new dentry lock. Right now we
410 * let them race, which means that for a very short while, this
411 * node might have two locks on a lock resource. This should be a
412 * problem though because one of them is in the process of being
415 static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
416 struct ocfs2_dentry_lock *dl)
418 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
419 ocfs2_lock_res_free(&dl->dl_lockres);
421 /* We leave dropping of inode reference to ocfs2_wq as that can
422 * possibly lead to inode deletion which gets tricky */
423 spin_lock(&dentry_list_lock);
424 if (!osb->dentry_lock_list &&
425 !ocfs2_test_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED))
426 queue_work(ocfs2_wq, &osb->dentry_lock_work);
427 dl->dl_next = osb->dentry_lock_list;
428 osb->dentry_lock_list = dl;
429 spin_unlock(&dentry_list_lock);
432 void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
433 struct ocfs2_dentry_lock *dl)
437 BUG_ON(dl->dl_count == 0);
439 spin_lock(&dentry_attach_lock);
441 unlock = !dl->dl_count;
442 spin_unlock(&dentry_attach_lock);
445 ocfs2_drop_dentry_lock(osb, dl);
448 static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
450 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
454 * No dentry lock is ok if we're disconnected or
457 if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
458 !d_unhashed(dentry)) {
459 unsigned long long ino = 0ULL;
461 ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
462 mlog(ML_ERROR, "Dentry is missing cluster lock. "
463 "inode: %llu, d_flags: 0x%x, d_name: %.*s\n",
464 ino, dentry->d_flags, dentry->d_name.len,
465 dentry->d_name.name);
471 mlog_bug_on_msg(dl->dl_count == 0, "dentry: %.*s, count: %u\n",
472 dentry->d_name.len, dentry->d_name.name,
475 ocfs2_dentry_lock_put(OCFS2_SB(dentry->d_sb), dl);
482 * d_move(), but keep the locks in sync.
484 * When we are done, "dentry" will have the parent dir and name of
485 * "target", which will be thrown away.
487 * We manually update the lock of "dentry" if need be.
489 * "target" doesn't have it's dentry lock touched - we allow the later
490 * dput() to handle this for us.
492 * This is called during ocfs2_rename(), while holding parent
493 * directory locks. The dentries have already been deleted on other
494 * nodes via ocfs2_remote_dentry_delete().
496 * Normally, the VFS handles the d_move() for the file system, after
497 * the ->rename() callback. OCFS2 wants to handle this internally, so
498 * the new lock can be created atomically with respect to the cluster.
500 void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
501 struct inode *old_dir, struct inode *new_dir)
504 struct ocfs2_super *osb = OCFS2_SB(old_dir->i_sb);
505 struct inode *inode = dentry->d_inode;
508 * Move within the same directory, so the actual lock info won't
511 * XXX: Is there any advantage to dropping the lock here?
513 if (old_dir == new_dir)
516 ocfs2_dentry_lock_put(osb, dentry->d_fsdata);
518 dentry->d_fsdata = NULL;
519 ret = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(new_dir)->ip_blkno);
524 d_move(dentry, target);
527 const struct dentry_operations ocfs2_dentry_ops = {
528 .d_revalidate = ocfs2_dentry_revalidate,
529 .d_iput = ocfs2_dentry_iput,