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,
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 trace_ocfs2_dentry_revalidate(dentry, dentry->d_name.len,
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;
77 trace_ocfs2_dentry_revalidate_negative(dentry->d_name.len,
87 if (inode == osb->root_inode || is_bad_inode(inode))
90 spin_lock(&OCFS2_I(inode)->ip_lock);
91 /* did we or someone else delete this inode? */
92 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
93 spin_unlock(&OCFS2_I(inode)->ip_lock);
94 trace_ocfs2_dentry_revalidate_delete(
95 (unsigned long long)OCFS2_I(inode)->ip_blkno);
98 spin_unlock(&OCFS2_I(inode)->ip_lock);
101 * We don't need a cluster lock to test this because once an
102 * inode nlink hits zero, it never goes back.
104 if (inode->i_nlink == 0) {
105 trace_ocfs2_dentry_revalidate_orphaned(
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 trace_ocfs2_dentry_revalidate_nofsdata(
117 (unsigned long long)OCFS2_I(inode)->ip_blkno);
125 trace_ocfs2_dentry_revalidate_ret(ret);
129 static int ocfs2_match_dentry(struct dentry *dentry,
133 struct inode *parent;
136 * ocfs2_lookup() does a d_splice_alias() _before_ attaching
137 * to the lock data, so we skip those here, otherwise
138 * ocfs2_dentry_attach_lock() will get its original dentry
141 if (!dentry->d_fsdata)
144 if (!dentry->d_parent)
147 if (skip_unhashed && d_unhashed(dentry))
150 parent = dentry->d_parent->d_inode;
151 /* Negative parent dentry? */
155 /* Name is in a different directory. */
156 if (OCFS2_I(parent)->ip_blkno != parent_blkno)
163 * Walk the inode alias list, and find a dentry which has a given
164 * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
165 * is looking for a dentry_lock reference. The downconvert thread is
166 * looking to unhash aliases, so we allow it to skip any that already
167 * have that property.
169 struct dentry *ocfs2_find_local_alias(struct inode *inode,
173 struct hlist_node *p;
174 struct dentry *dentry;
176 spin_lock(&inode->i_lock);
177 hlist_for_each_entry(dentry, p, &inode->i_dentry, d_alias) {
178 spin_lock(&dentry->d_lock);
179 if (ocfs2_match_dentry(dentry, parent_blkno, skip_unhashed)) {
180 trace_ocfs2_find_local_alias(dentry->d_name.len,
181 dentry->d_name.name);
184 spin_unlock(&dentry->d_lock);
185 spin_unlock(&inode->i_lock);
188 spin_unlock(&dentry->d_lock);
190 spin_unlock(&inode->i_lock);
194 DEFINE_SPINLOCK(dentry_attach_lock);
197 * Attach this dentry to a cluster lock.
199 * Dentry locks cover all links in a given directory to a particular
200 * inode. We do this so that ocfs2 can build a lock name which all
201 * nodes in the cluster can agree on at all times. Shoving full names
202 * in the cluster lock won't work due to size restrictions. Covering
203 * links inside of a directory is a good compromise because it still
204 * allows us to use the parent directory lock to synchronize
207 * Call this function with the parent dir semaphore and the parent dir
210 * The dir semaphore will protect us from having to worry about
211 * concurrent processes on our node trying to attach a lock at the
214 * The dir cluster lock (held at either PR or EX mode) protects us
215 * from unlink and rename on other nodes.
217 * A dput() can happen asynchronously due to pruning, so we cover
218 * attaching and detaching the dentry lock with a
219 * dentry_attach_lock.
221 * A node which has done lookup on a name retains a protected read
222 * lock until final dput. If the user requests and unlink or rename,
223 * the protected read is upgraded to an exclusive lock. Other nodes
224 * who have seen the dentry will then be informed that they need to
225 * downgrade their lock, which will involve d_delete on the
226 * dentry. This happens in ocfs2_dentry_convert_worker().
228 int ocfs2_dentry_attach_lock(struct dentry *dentry,
233 struct dentry *alias;
234 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
236 trace_ocfs2_dentry_attach_lock(dentry->d_name.len, dentry->d_name.name,
237 (unsigned long long)parent_blkno, dl);
240 * Negative dentry. We ignore these for now.
242 * XXX: Could we can improve ocfs2_dentry_revalidate() by
248 if (!dentry->d_inode && dentry->d_fsdata) {
249 /* Converting a negative dentry to positive
250 Clear dentry->d_fsdata */
251 dentry->d_fsdata = dl = NULL;
255 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
256 " \"%.*s\": old parent: %llu, new: %llu\n",
257 dentry->d_name.len, dentry->d_name.name,
258 (unsigned long long)parent_blkno,
259 (unsigned long long)dl->dl_parent_blkno);
263 alias = ocfs2_find_local_alias(inode, parent_blkno, 0);
266 * Great, an alias exists, which means we must have a
267 * dentry lock already. We can just grab the lock off
268 * the alias and add it to the list.
270 * We're depending here on the fact that this dentry
271 * was found and exists in the dcache and so must have
272 * a reference to the dentry_lock because we can't
273 * race creates. Final dput() cannot happen on it
274 * since we have it pinned, so our reference is safe.
276 dl = alias->d_fsdata;
277 mlog_bug_on_msg(!dl, "parent %llu, ino %llu\n",
278 (unsigned long long)parent_blkno,
279 (unsigned long long)OCFS2_I(inode)->ip_blkno);
281 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
282 " \"%.*s\": old parent: %llu, new: %llu\n",
283 dentry->d_name.len, dentry->d_name.name,
284 (unsigned long long)parent_blkno,
285 (unsigned long long)dl->dl_parent_blkno);
287 trace_ocfs2_dentry_attach_lock_found(dl->dl_lockres.l_name,
288 (unsigned long long)parent_blkno,
289 (unsigned long long)OCFS2_I(inode)->ip_blkno);
295 * There are no other aliases
297 dl = kmalloc(sizeof(*dl), GFP_NOFS);
306 * Does this have to happen below, for all attaches, in case
307 * the struct inode gets blown away by the downconvert thread?
309 dl->dl_inode = igrab(inode);
310 dl->dl_parent_blkno = parent_blkno;
311 ocfs2_dentry_lock_res_init(dl, parent_blkno, inode);
314 spin_lock(&dentry_attach_lock);
315 dentry->d_fsdata = dl;
317 spin_unlock(&dentry_attach_lock);
320 * This actually gets us our PRMODE level lock. From now on,
321 * we'll have a notification if one of these names is
322 * destroyed on another node.
324 ret = ocfs2_dentry_lock(dentry, 0);
326 ocfs2_dentry_unlock(dentry, 0);
331 * In case of error, manually free the allocation and do the iput().
332 * We need to do this because error here means no d_instantiate(),
333 * which means iput() will not be called during dput(dentry).
335 if (ret < 0 && !alias) {
336 ocfs2_lock_res_free(&dl->dl_lockres);
337 BUG_ON(dl->dl_count != 1);
338 spin_lock(&dentry_attach_lock);
339 dentry->d_fsdata = NULL;
340 spin_unlock(&dentry_attach_lock);
350 DEFINE_SPINLOCK(dentry_list_lock);
352 /* We limit the number of dentry locks to drop in one go. We have
353 * this limit so that we don't starve other users of ocfs2_wq. */
354 #define DL_INODE_DROP_COUNT 64
356 /* Drop inode references from dentry locks */
357 static void __ocfs2_drop_dl_inodes(struct ocfs2_super *osb, int drop_count)
359 struct ocfs2_dentry_lock *dl;
361 spin_lock(&dentry_list_lock);
362 while (osb->dentry_lock_list && (drop_count < 0 || drop_count--)) {
363 dl = osb->dentry_lock_list;
364 osb->dentry_lock_list = dl->dl_next;
365 spin_unlock(&dentry_list_lock);
368 spin_lock(&dentry_list_lock);
370 spin_unlock(&dentry_list_lock);
373 void ocfs2_drop_dl_inodes(struct work_struct *work)
375 struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
378 __ocfs2_drop_dl_inodes(osb, DL_INODE_DROP_COUNT);
380 * Don't queue dropping if umount is in progress. We flush the
381 * list in ocfs2_dismount_volume
383 spin_lock(&dentry_list_lock);
384 if (osb->dentry_lock_list &&
385 !ocfs2_test_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED))
386 queue_work(ocfs2_wq, &osb->dentry_lock_work);
387 spin_unlock(&dentry_list_lock);
390 /* Flush the whole work queue */
391 void ocfs2_drop_all_dl_inodes(struct ocfs2_super *osb)
393 __ocfs2_drop_dl_inodes(osb, -1);
397 * ocfs2_dentry_iput() and friends.
399 * At this point, our particular dentry is detached from the inodes
400 * alias list, so there's no way that the locking code can find it.
402 * The interesting stuff happens when we determine that our lock needs
403 * to go away because this is the last subdir alias in the
404 * system. This function needs to handle a couple things:
406 * 1) Synchronizing lock shutdown with the downconvert threads. This
407 * is already handled for us via the lockres release drop function
408 * called in ocfs2_release_dentry_lock()
410 * 2) A race may occur when we're doing our lock shutdown and
411 * another process wants to create a new dentry lock. Right now we
412 * let them race, which means that for a very short while, this
413 * node might have two locks on a lock resource. This should be a
414 * problem though because one of them is in the process of being
417 static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
418 struct ocfs2_dentry_lock *dl)
420 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
421 ocfs2_lock_res_free(&dl->dl_lockres);
423 /* We leave dropping of inode reference to ocfs2_wq as that can
424 * possibly lead to inode deletion which gets tricky */
425 spin_lock(&dentry_list_lock);
426 if (!osb->dentry_lock_list &&
427 !ocfs2_test_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED))
428 queue_work(ocfs2_wq, &osb->dentry_lock_work);
429 dl->dl_next = osb->dentry_lock_list;
430 osb->dentry_lock_list = dl;
431 spin_unlock(&dentry_list_lock);
434 void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
435 struct ocfs2_dentry_lock *dl)
439 BUG_ON(dl->dl_count == 0);
441 spin_lock(&dentry_attach_lock);
443 unlock = !dl->dl_count;
444 spin_unlock(&dentry_attach_lock);
447 ocfs2_drop_dentry_lock(osb, dl);
450 static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
452 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
456 * No dentry lock is ok if we're disconnected or
459 if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
460 !d_unhashed(dentry)) {
461 unsigned long long ino = 0ULL;
463 ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
464 mlog(ML_ERROR, "Dentry is missing cluster lock. "
465 "inode: %llu, d_flags: 0x%x, d_name: %.*s\n",
466 ino, dentry->d_flags, dentry->d_name.len,
467 dentry->d_name.name);
473 mlog_bug_on_msg(dl->dl_count == 0, "dentry: %.*s, count: %u\n",
474 dentry->d_name.len, dentry->d_name.name,
477 ocfs2_dentry_lock_put(OCFS2_SB(dentry->d_sb), dl);
484 * d_move(), but keep the locks in sync.
486 * When we are done, "dentry" will have the parent dir and name of
487 * "target", which will be thrown away.
489 * We manually update the lock of "dentry" if need be.
491 * "target" doesn't have it's dentry lock touched - we allow the later
492 * dput() to handle this for us.
494 * This is called during ocfs2_rename(), while holding parent
495 * directory locks. The dentries have already been deleted on other
496 * nodes via ocfs2_remote_dentry_delete().
498 * Normally, the VFS handles the d_move() for the file system, after
499 * the ->rename() callback. OCFS2 wants to handle this internally, so
500 * the new lock can be created atomically with respect to the cluster.
502 void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
503 struct inode *old_dir, struct inode *new_dir)
506 struct ocfs2_super *osb = OCFS2_SB(old_dir->i_sb);
507 struct inode *inode = dentry->d_inode;
510 * Move within the same directory, so the actual lock info won't
513 * XXX: Is there any advantage to dropping the lock here?
515 if (old_dir == new_dir)
518 ocfs2_dentry_lock_put(osb, dentry->d_fsdata);
520 dentry->d_fsdata = NULL;
521 ret = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(new_dir)->ip_blkno);
526 d_move(dentry, target);
529 const struct dentry_operations ocfs2_dentry_ops = {
530 .d_revalidate = ocfs2_dentry_revalidate,
531 .d_iput = ocfs2_dentry_iput,