blob: e9d7c2038c0f55c1f0cf48386a2f6eed12a2d805 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dcache.c
5 *
6 * dentry cache handling code
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
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.
14 *
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.
19 *
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.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/namei.h>
30
31#define MLOG_MASK_PREFIX ML_DCACHE
32#include <cluster/masklog.h>
33
34#include "ocfs2.h"
35
36#include "alloc.h"
37#include "dcache.h"
Mark Fasheh80c05842006-09-08 14:43:18 -070038#include "dlmglue.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080039#include "file.h"
40#include "inode.h"
Jan Karaea455f82009-01-12 23:20:31 +010041#include "super.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080042
Mark Fasheh80c05842006-09-08 14:43:18 -070043
Mark Fashehccd979b2005-12-15 14:31:24 -080044static int ocfs2_dentry_revalidate(struct dentry *dentry,
45 struct nameidata *nd)
46{
47 struct inode *inode = dentry->d_inode;
48 int ret = 0; /* if all else fails, just return false */
Mark Fasheh80c05842006-09-08 14:43:18 -070049 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -080050
51 mlog_entry("(0x%p, '%.*s')\n", dentry,
52 dentry->d_name.len, dentry->d_name.name);
53
54 /* Never trust a negative dentry - force a new lookup. */
55 if (inode == NULL) {
56 mlog(0, "negative dentry: %.*s\n", dentry->d_name.len,
57 dentry->d_name.name);
58 goto bail;
59 }
60
Mark Fashehccd979b2005-12-15 14:31:24 -080061 BUG_ON(!osb);
62
Mark Fasheh80c05842006-09-08 14:43:18 -070063 if (inode == osb->root_inode || is_bad_inode(inode))
64 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -080065
Mark Fasheh80c05842006-09-08 14:43:18 -070066 spin_lock(&OCFS2_I(inode)->ip_lock);
67 /* did we or someone else delete this inode? */
68 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
69 spin_unlock(&OCFS2_I(inode)->ip_lock);
70 mlog(0, "inode (%llu) deleted, returning false\n",
71 (unsigned long long)OCFS2_I(inode)->ip_blkno);
72 goto bail;
73 }
74 spin_unlock(&OCFS2_I(inode)->ip_lock);
75
76 /*
77 * We don't need a cluster lock to test this because once an
78 * inode nlink hits zero, it never goes back.
79 */
80 if (inode->i_nlink == 0) {
81 mlog(0, "Inode %llu orphaned, returning false "
82 "dir = %d\n",
83 (unsigned long long)OCFS2_I(inode)->ip_blkno,
84 S_ISDIR(inode->i_mode));
85 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -080086 }
87
88 ret = 1;
89
90bail:
91 mlog_exit(ret);
92
93 return ret;
94}
95
Mark Fasheh80c05842006-09-08 14:43:18 -070096static int ocfs2_match_dentry(struct dentry *dentry,
97 u64 parent_blkno,
98 int skip_unhashed)
99{
100 struct inode *parent;
101
102 /*
103 * ocfs2_lookup() does a d_splice_alias() _before_ attaching
104 * to the lock data, so we skip those here, otherwise
105 * ocfs2_dentry_attach_lock() will get its original dentry
106 * back.
107 */
108 if (!dentry->d_fsdata)
109 return 0;
110
111 if (!dentry->d_parent)
112 return 0;
113
114 if (skip_unhashed && d_unhashed(dentry))
115 return 0;
116
117 parent = dentry->d_parent->d_inode;
118 /* Negative parent dentry? */
119 if (!parent)
120 return 0;
121
122 /* Name is in a different directory. */
123 if (OCFS2_I(parent)->ip_blkno != parent_blkno)
124 return 0;
125
126 return 1;
127}
128
129/*
130 * Walk the inode alias list, and find a dentry which has a given
131 * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
Mark Fasheh34d024f2007-09-24 15:56:19 -0700132 * is looking for a dentry_lock reference. The downconvert thread is
133 * looking to unhash aliases, so we allow it to skip any that already
134 * have that property.
Mark Fasheh80c05842006-09-08 14:43:18 -0700135 */
136struct dentry *ocfs2_find_local_alias(struct inode *inode,
137 u64 parent_blkno,
138 int skip_unhashed)
139{
140 struct list_head *p;
141 struct dentry *dentry = NULL;
142
143 spin_lock(&dcache_lock);
144
145 list_for_each(p, &inode->i_dentry) {
146 dentry = list_entry(p, struct dentry, d_alias);
147
148 if (ocfs2_match_dentry(dentry, parent_blkno, skip_unhashed)) {
149 mlog(0, "dentry found: %.*s\n",
150 dentry->d_name.len, dentry->d_name.name);
151
152 dget_locked(dentry);
153 break;
154 }
155
156 dentry = NULL;
157 }
158
159 spin_unlock(&dcache_lock);
160
161 return dentry;
162}
163
Mark Fashehd680efe2006-09-08 14:14:34 -0700164DEFINE_SPINLOCK(dentry_attach_lock);
165
Mark Fasheh80c05842006-09-08 14:43:18 -0700166/*
167 * Attach this dentry to a cluster lock.
168 *
169 * Dentry locks cover all links in a given directory to a particular
170 * inode. We do this so that ocfs2 can build a lock name which all
171 * nodes in the cluster can agree on at all times. Shoving full names
172 * in the cluster lock won't work due to size restrictions. Covering
173 * links inside of a directory is a good compromise because it still
174 * allows us to use the parent directory lock to synchronize
175 * operations.
176 *
177 * Call this function with the parent dir semaphore and the parent dir
178 * cluster lock held.
179 *
180 * The dir semaphore will protect us from having to worry about
181 * concurrent processes on our node trying to attach a lock at the
182 * same time.
183 *
184 * The dir cluster lock (held at either PR or EX mode) protects us
185 * from unlink and rename on other nodes.
186 *
Mark Fasheh80c05842006-09-08 14:43:18 -0700187 * A dput() can happen asynchronously due to pruning, so we cover
188 * attaching and detaching the dentry lock with a
189 * dentry_attach_lock.
190 *
191 * A node which has done lookup on a name retains a protected read
192 * lock until final dput. If the user requests and unlink or rename,
193 * the protected read is upgraded to an exclusive lock. Other nodes
194 * who have seen the dentry will then be informed that they need to
195 * downgrade their lock, which will involve d_delete on the
196 * dentry. This happens in ocfs2_dentry_convert_worker().
197 */
198int ocfs2_dentry_attach_lock(struct dentry *dentry,
199 struct inode *inode,
Mark Fasheh0027dd52006-09-21 16:51:28 -0700200 u64 parent_blkno)
Mark Fasheh80c05842006-09-08 14:43:18 -0700201{
202 int ret;
203 struct dentry *alias;
204 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
205
Mark Fasheh0027dd52006-09-21 16:51:28 -0700206 mlog(0, "Attach \"%.*s\", parent %llu, fsdata: %p\n",
Mark Fasheh80c05842006-09-08 14:43:18 -0700207 dentry->d_name.len, dentry->d_name.name,
Mark Fasheh0027dd52006-09-21 16:51:28 -0700208 (unsigned long long)parent_blkno, dl);
Mark Fasheh80c05842006-09-08 14:43:18 -0700209
210 /*
211 * Negative dentry. We ignore these for now.
212 *
213 * XXX: Could we can improve ocfs2_dentry_revalidate() by
214 * tracking these?
215 */
216 if (!inode)
217 return 0;
218
219 if (dl) {
220 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
221 " \"%.*s\": old parent: %llu, new: %llu\n",
222 dentry->d_name.len, dentry->d_name.name,
223 (unsigned long long)parent_blkno,
224 (unsigned long long)dl->dl_parent_blkno);
225 return 0;
226 }
227
228 alias = ocfs2_find_local_alias(inode, parent_blkno, 0);
229 if (alias) {
230 /*
231 * Great, an alias exists, which means we must have a
232 * dentry lock already. We can just grab the lock off
233 * the alias and add it to the list.
234 *
235 * We're depending here on the fact that this dentry
236 * was found and exists in the dcache and so must have
237 * a reference to the dentry_lock because we can't
238 * race creates. Final dput() cannot happen on it
239 * since we have it pinned, so our reference is safe.
240 */
241 dl = alias->d_fsdata;
Mark Fasheh0027dd52006-09-21 16:51:28 -0700242 mlog_bug_on_msg(!dl, "parent %llu, ino %llu\n",
Mark Fasheh80c05842006-09-08 14:43:18 -0700243 (unsigned long long)parent_blkno,
Mark Fasheh0027dd52006-09-21 16:51:28 -0700244 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fasheh80c05842006-09-08 14:43:18 -0700245
246 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
247 " \"%.*s\": old parent: %llu, new: %llu\n",
248 dentry->d_name.len, dentry->d_name.name,
249 (unsigned long long)parent_blkno,
250 (unsigned long long)dl->dl_parent_blkno);
251
252 mlog(0, "Found: %s\n", dl->dl_lockres.l_name);
253
254 goto out_attach;
255 }
256
257 /*
258 * There are no other aliases
259 */
260 dl = kmalloc(sizeof(*dl), GFP_NOFS);
261 if (!dl) {
262 ret = -ENOMEM;
263 mlog_errno(ret);
264 return ret;
265 }
266
267 dl->dl_count = 0;
268 /*
269 * Does this have to happen below, for all attaches, in case
Mark Fasheh34d024f2007-09-24 15:56:19 -0700270 * the struct inode gets blown away by the downconvert thread?
Mark Fasheh80c05842006-09-08 14:43:18 -0700271 */
272 dl->dl_inode = igrab(inode);
273 dl->dl_parent_blkno = parent_blkno;
274 ocfs2_dentry_lock_res_init(dl, parent_blkno, inode);
275
276out_attach:
277 spin_lock(&dentry_attach_lock);
278 dentry->d_fsdata = dl;
279 dl->dl_count++;
280 spin_unlock(&dentry_attach_lock);
281
282 /*
Mark Fasheh80c05842006-09-08 14:43:18 -0700283 * This actually gets us our PRMODE level lock. From now on,
284 * we'll have a notification if one of these names is
285 * destroyed on another node.
286 */
287 ret = ocfs2_dentry_lock(dentry, 0);
Mark Fasheh0027dd52006-09-21 16:51:28 -0700288 if (!ret)
289 ocfs2_dentry_unlock(dentry, 0);
290 else
Mark Fasheh80c05842006-09-08 14:43:18 -0700291 mlog_errno(ret);
Mark Fasheh80c05842006-09-08 14:43:18 -0700292
Mark Fasheh80c05842006-09-08 14:43:18 -0700293 dput(alias);
294
295 return ret;
296}
297
Jan Karaea455f82009-01-12 23:20:31 +0100298static DEFINE_SPINLOCK(dentry_list_lock);
299
300/* We limit the number of dentry locks to drop in one go. We have
301 * this limit so that we don't starve other users of ocfs2_wq. */
302#define DL_INODE_DROP_COUNT 64
303
304/* Drop inode references from dentry locks */
305void ocfs2_drop_dl_inodes(struct work_struct *work)
306{
307 struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
308 dentry_lock_work);
309 struct ocfs2_dentry_lock *dl;
310 int drop_count = DL_INODE_DROP_COUNT;
311
312 spin_lock(&dentry_list_lock);
313 while (osb->dentry_lock_list && drop_count--) {
314 dl = osb->dentry_lock_list;
315 osb->dentry_lock_list = dl->dl_next;
316 spin_unlock(&dentry_list_lock);
317 iput(dl->dl_inode);
318 kfree(dl);
319 spin_lock(&dentry_list_lock);
320 }
321 if (osb->dentry_lock_list)
322 queue_work(ocfs2_wq, &osb->dentry_lock_work);
323 spin_unlock(&dentry_list_lock);
324}
325
Mark Fasheh80c05842006-09-08 14:43:18 -0700326/*
327 * ocfs2_dentry_iput() and friends.
328 *
329 * At this point, our particular dentry is detached from the inodes
330 * alias list, so there's no way that the locking code can find it.
331 *
332 * The interesting stuff happens when we determine that our lock needs
333 * to go away because this is the last subdir alias in the
334 * system. This function needs to handle a couple things:
335 *
336 * 1) Synchronizing lock shutdown with the downconvert threads. This
337 * is already handled for us via the lockres release drop function
338 * called in ocfs2_release_dentry_lock()
339 *
340 * 2) A race may occur when we're doing our lock shutdown and
341 * another process wants to create a new dentry lock. Right now we
342 * let them race, which means that for a very short while, this
343 * node might have two locks on a lock resource. This should be a
344 * problem though because one of them is in the process of being
345 * thrown out.
346 */
347static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
348 struct ocfs2_dentry_lock *dl)
349{
350 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
351 ocfs2_lock_res_free(&dl->dl_lockres);
Jan Karaea455f82009-01-12 23:20:31 +0100352
353 /* We leave dropping of inode reference to ocfs2_wq as that can
354 * possibly lead to inode deletion which gets tricky */
355 spin_lock(&dentry_list_lock);
356 if (!osb->dentry_lock_list)
357 queue_work(ocfs2_wq, &osb->dentry_lock_work);
358 dl->dl_next = osb->dentry_lock_list;
359 osb->dentry_lock_list = dl;
360 spin_unlock(&dentry_list_lock);
Mark Fasheh80c05842006-09-08 14:43:18 -0700361}
362
363void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
364 struct ocfs2_dentry_lock *dl)
365{
Jan Karaea455f82009-01-12 23:20:31 +0100366 int unlock;
Mark Fasheh80c05842006-09-08 14:43:18 -0700367
368 BUG_ON(dl->dl_count == 0);
369
370 spin_lock(&dentry_attach_lock);
371 dl->dl_count--;
372 unlock = !dl->dl_count;
373 spin_unlock(&dentry_attach_lock);
374
375 if (unlock)
376 ocfs2_drop_dentry_lock(osb, dl);
377}
378
379static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
380{
381 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
382
Mark Fashehbccb9da2007-11-07 16:35:14 -0800383 if (!dl) {
384 /*
385 * No dentry lock is ok if we're disconnected or
386 * unhashed.
387 */
388 if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
389 !d_unhashed(dentry)) {
390 unsigned long long ino = 0ULL;
391 if (inode)
392 ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
393 mlog(ML_ERROR, "Dentry is missing cluster lock. "
394 "inode: %llu, d_flags: 0x%x, d_name: %.*s\n",
395 ino, dentry->d_flags, dentry->d_name.len,
396 dentry->d_name.name);
397 }
Mark Fasheh80c05842006-09-08 14:43:18 -0700398
Mark Fasheh80c05842006-09-08 14:43:18 -0700399 goto out;
Mark Fashehbccb9da2007-11-07 16:35:14 -0800400 }
Mark Fasheh80c05842006-09-08 14:43:18 -0700401
402 mlog_bug_on_msg(dl->dl_count == 0, "dentry: %.*s, count: %u\n",
403 dentry->d_name.len, dentry->d_name.name,
404 dl->dl_count);
405
406 ocfs2_dentry_lock_put(OCFS2_SB(dentry->d_sb), dl);
407
408out:
409 iput(inode);
410}
411
412/*
413 * d_move(), but keep the locks in sync.
414 *
415 * When we are done, "dentry" will have the parent dir and name of
416 * "target", which will be thrown away.
417 *
418 * We manually update the lock of "dentry" if need be.
419 *
420 * "target" doesn't have it's dentry lock touched - we allow the later
421 * dput() to handle this for us.
422 *
423 * This is called during ocfs2_rename(), while holding parent
424 * directory locks. The dentries have already been deleted on other
425 * nodes via ocfs2_remote_dentry_delete().
426 *
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +0200427 * Normally, the VFS handles the d_move() for the file system, after
Mark Fasheh80c05842006-09-08 14:43:18 -0700428 * the ->rename() callback. OCFS2 wants to handle this internally, so
429 * the new lock can be created atomically with respect to the cluster.
430 */
431void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
432 struct inode *old_dir, struct inode *new_dir)
433{
434 int ret;
435 struct ocfs2_super *osb = OCFS2_SB(old_dir->i_sb);
436 struct inode *inode = dentry->d_inode;
437
438 /*
439 * Move within the same directory, so the actual lock info won't
440 * change.
441 *
442 * XXX: Is there any advantage to dropping the lock here?
443 */
444 if (old_dir == new_dir)
Mark Fasheh1ba9da22006-09-08 14:22:54 -0700445 goto out_move;
Mark Fasheh80c05842006-09-08 14:43:18 -0700446
447 ocfs2_dentry_lock_put(osb, dentry->d_fsdata);
448
449 dentry->d_fsdata = NULL;
Mark Fasheh0027dd52006-09-21 16:51:28 -0700450 ret = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(new_dir)->ip_blkno);
Mark Fasheh80c05842006-09-08 14:43:18 -0700451 if (ret)
452 mlog_errno(ret);
Mark Fasheh1ba9da22006-09-08 14:22:54 -0700453
454out_move:
455 d_move(dentry, target);
Mark Fasheh80c05842006-09-08 14:43:18 -0700456}
457
Mark Fashehccd979b2005-12-15 14:31:24 -0800458struct dentry_operations ocfs2_dentry_ops = {
459 .d_revalidate = ocfs2_dentry_revalidate,
Mark Fasheh80c05842006-09-08 14:43:18 -0700460 .d_iput = ocfs2_dentry_iput,
Mark Fashehccd979b2005-12-15 14:31:24 -0800461};