1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * vfs' aops, fops, dops and iops
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/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/smp_lock.h>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
42 #include "extent_map.h"
44 #include "heartbeat.h"
55 #include "buffer_head_io.h"
57 struct ocfs2_find_inode_args
61 unsigned int fi_flags;
64 static int ocfs2_read_locked_inode(struct inode *inode,
65 struct ocfs2_find_inode_args *args);
66 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
67 static int ocfs2_find_actor(struct inode *inode, void *opaque);
68 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
70 struct buffer_head *fe_bh);
72 void ocfs2_set_inode_flags(struct inode *inode)
74 unsigned int flags = OCFS2_I(inode)->ip_attr;
76 inode->i_flags &= ~(S_IMMUTABLE |
77 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
79 if (flags & OCFS2_IMMUTABLE_FL)
80 inode->i_flags |= S_IMMUTABLE;
82 if (flags & OCFS2_SYNC_FL)
83 inode->i_flags |= S_SYNC;
84 if (flags & OCFS2_APPEND_FL)
85 inode->i_flags |= S_APPEND;
86 if (flags & OCFS2_NOATIME_FL)
87 inode->i_flags |= S_NOATIME;
88 if (flags & OCFS2_DIRSYNC_FL)
89 inode->i_flags |= S_DIRSYNC;
92 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
96 struct ocfs2_find_inode_args args;
98 /* ocfs2_ilookup_for_vote should *only* be called from the
100 BUG_ON(current != osb->vote_task);
102 args.fi_blkno = blkno;
103 args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
105 args.fi_flags |= OCFS2_FI_FLAG_DELETE;
106 args.fi_ino = ino_from_blkno(osb->sb, blkno);
107 return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
110 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
112 struct inode *inode = NULL;
113 struct super_block *sb = osb->sb;
114 struct ocfs2_find_inode_args args;
116 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
118 /* Ok. By now we've either got the offsets passed to us by the
119 * caller, or we just pulled them off the bh. Lets do some
120 * sanity checks to make sure they're OK. */
122 inode = ERR_PTR(-EINVAL);
123 mlog_errno(PTR_ERR(inode));
127 args.fi_blkno = blkno;
128 args.fi_flags = flags;
129 args.fi_ino = ino_from_blkno(sb, blkno);
131 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
132 ocfs2_init_locked_inode, &args);
133 /* inode was *not* in the inode cache. 2.6.x requires
134 * us to do our own read_inode call and unlock it
136 if (inode && inode->i_state & I_NEW) {
137 mlog(0, "Inode was not in inode cache, reading it.\n");
138 ocfs2_read_locked_inode(inode, &args);
139 unlock_new_inode(inode);
142 inode = ERR_PTR(-ENOMEM);
143 mlog_errno(PTR_ERR(inode));
146 if (is_bad_inode(inode)) {
148 inode = ERR_PTR(-ESTALE);
153 if (!IS_ERR(inode)) {
154 mlog(0, "returning inode with number %llu\n",
155 (unsigned long long)OCFS2_I(inode)->ip_blkno);
156 mlog_exit_ptr(inode);
164 * here's how inodes get read from disk:
165 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
166 * found? : return the in-memory inode
167 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
170 static int ocfs2_find_actor(struct inode *inode, void *opaque)
172 struct ocfs2_find_inode_args *args = NULL;
173 struct ocfs2_inode_info *oi = OCFS2_I(inode);
176 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
180 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
182 if (oi->ip_blkno != args->fi_blkno)
185 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
186 * ocfs2_ilookup_for_vote which won't create an inode for one
187 * that isn't found. The vote thread which doesn't want to get
188 * an inode which is in the process of going away - otherwise
189 * the call to __wait_on_freeing_inode in find_inode_fast will
190 * cause it to deadlock on an inode which may be waiting on a
191 * vote (or lock release) in delete_inode */
192 if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
193 (inode->i_state & (I_FREEING|I_CLEAR))) {
194 /* As stated above, we're not going to return an
195 * inode. In the case of a delete vote, the voting
196 * code is going to signal the other node to go
197 * ahead. Mark that state here, so this freeing inode
198 * has the state when it gets to delete_inode. */
199 if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
200 spin_lock(&oi->ip_lock);
201 ocfs2_mark_inode_remotely_deleted(inode);
202 spin_unlock(&oi->ip_lock);
214 * initialize the new inode, but don't do anything that would cause
216 * return 0 on success, 1 on failure
218 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
220 struct ocfs2_find_inode_args *args = opaque;
222 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
224 inode->i_ino = args->fi_ino;
225 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
231 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
234 struct super_block *sb;
235 struct ocfs2_super *osb;
236 int status = -EINVAL;
238 mlog_entry("(0x%p, size:%llu)\n", inode,
239 (unsigned long long)fe->i_size);
244 /* this means that read_inode cannot create a superblock inode
245 * today. change if needed. */
246 if (!OCFS2_IS_VALID_DINODE(fe) ||
247 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
248 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
249 "signature = %.*s, flags = 0x%x\n",
251 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
252 fe->i_signature, le32_to_cpu(fe->i_flags));
256 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
257 mlog(ML_ERROR, "file entry generation does not match "
258 "superblock! osb->fs_generation=%x, "
259 "fe->i_fs_generation=%x\n",
260 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
264 inode->i_version = 1;
265 inode->i_generation = le32_to_cpu(fe->i_generation);
266 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
267 inode->i_mode = le16_to_cpu(fe->i_mode);
268 inode->i_uid = le32_to_cpu(fe->i_uid);
269 inode->i_gid = le32_to_cpu(fe->i_gid);
271 /* Fast symlinks will have i_size but no allocated clusters. */
272 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
276 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
277 inode->i_mapping->a_ops = &ocfs2_aops;
278 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
279 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
280 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
281 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
282 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
283 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
285 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
287 "ip_blkno %llu != i_blkno %llu!\n",
288 (unsigned long long)OCFS2_I(inode)->ip_blkno,
289 (unsigned long long)fe->i_blkno);
291 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
292 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
294 inode->i_nlink = le16_to_cpu(fe->i_links_count);
296 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
297 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
299 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
300 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
301 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
302 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
303 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
304 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
305 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
306 /* we can't actually hit this as read_inode can't
307 * handle superblocks today ;-) */
311 switch (inode->i_mode & S_IFMT) {
313 inode->i_fop = &ocfs2_fops;
314 inode->i_op = &ocfs2_file_iops;
315 i_size_write(inode, le64_to_cpu(fe->i_size));
318 inode->i_op = &ocfs2_dir_iops;
319 inode->i_fop = &ocfs2_dops;
320 i_size_write(inode, le64_to_cpu(fe->i_size));
323 if (ocfs2_inode_is_fast_symlink(inode))
324 inode->i_op = &ocfs2_fast_symlink_inode_operations;
326 inode->i_op = &ocfs2_symlink_inode_operations;
327 i_size_write(inode, le64_to_cpu(fe->i_size));
330 inode->i_op = &ocfs2_special_file_iops;
331 init_special_inode(inode, inode->i_mode,
337 inode->i_ino = ino_from_blkno(inode->i_sb,
338 le64_to_cpu(fe->i_blkno));
341 * If we ever want to create system files from kernel,
342 * the generation argument to
343 * ocfs2_inode_lock_res_init() will have to change.
345 BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
347 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
348 OCFS2_LOCK_TYPE_META, 0, inode);
350 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
351 OCFS2_LOCK_TYPE_OPEN, 0, inode);
354 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
355 OCFS2_LOCK_TYPE_RW, inode->i_generation,
358 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
359 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
362 ocfs2_set_inode_flags(inode);
370 static int ocfs2_read_locked_inode(struct inode *inode,
371 struct ocfs2_find_inode_args *args)
373 struct super_block *sb;
374 struct ocfs2_super *osb;
375 struct ocfs2_dinode *fe;
376 struct buffer_head *bh = NULL;
377 int status, can_lock;
380 mlog_entry("(0x%p, 0x%p)\n", inode, args);
383 if (inode == NULL || inode->i_sb == NULL) {
384 mlog(ML_ERROR, "bad inode\n");
391 mlog(ML_ERROR, "bad inode args\n");
392 make_bad_inode(inode);
397 * To improve performance of cold-cache inode stats, we take
398 * the cluster lock here if possible.
400 * Generally, OCFS2 never trusts the contents of an inode
401 * unless it's holding a cluster lock, so taking it here isn't
402 * a correctness issue as much as it is a performance
405 * There are three times when taking the lock is not a good idea:
407 * 1) During startup, before we have initialized the DLM.
409 * 2) If we are reading certain system files which never get
410 * cluster locks (local alloc, truncate log).
412 * 3) If the process doing the iget() is responsible for
413 * orphan dir recovery. We're holding the orphan dir lock and
414 * can get into a deadlock with another process on another
415 * node in ->delete_inode().
417 * #1 and #2 can be simply solved by never taking the lock
418 * here for system files (which are the only type we read
419 * during mount). It's a heavier approach, but our main
420 * concern is user-accesible files anyway.
422 * #3 works itself out because we'll eventually take the
423 * cluster lock before trusting anything anyway.
425 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
426 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
427 && !ocfs2_mount_local(osb);
430 * To maintain backwards compatibility with older versions of
431 * ocfs2-tools, we still store the generation value for system
432 * files. The only ones that actually matter to userspace are
433 * the journals, but it's easier and inexpensive to just flag
434 * all system files similarly.
436 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
437 generation = osb->fs_generation;
439 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
440 OCFS2_LOCK_TYPE_META,
443 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
444 OCFS2_LOCK_TYPE_OPEN,
448 status = ocfs2_open_lock(inode);
450 make_bad_inode(inode);
454 status = ocfs2_meta_lock(inode, NULL, 0);
456 make_bad_inode(inode);
462 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
463 status = ocfs2_try_open_lock(inode, 0);
465 make_bad_inode(inode);
470 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
471 can_lock ? inode : NULL);
478 fe = (struct ocfs2_dinode *) bh->b_data;
479 if (!OCFS2_IS_VALID_DINODE(fe)) {
480 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
481 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
486 * This is a code bug. Right now the caller needs to
487 * understand whether it is asking for a system file inode or
488 * not so the proper lock names can be built.
490 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
491 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
492 "Inode %llu: system file state is ambigous\n",
493 (unsigned long long)args->fi_blkno);
495 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
496 S_ISBLK(le16_to_cpu(fe->i_mode)))
497 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
499 if (ocfs2_populate_inode(inode, fe, 0) < 0)
502 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
508 ocfs2_meta_unlock(inode, 0);
511 make_bad_inode(inode);
520 void ocfs2_sync_blockdev(struct super_block *sb)
522 sync_blockdev(sb->s_bdev);
525 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
527 struct buffer_head *fe_bh)
530 handle_t *handle = NULL;
531 struct ocfs2_truncate_context *tc = NULL;
532 struct ocfs2_dinode *fe;
536 fe = (struct ocfs2_dinode *) fe_bh->b_data;
538 /* zero allocation, zero truncate :) */
542 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
543 if (IS_ERR(handle)) {
544 status = PTR_ERR(handle);
550 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
556 ocfs2_commit_trans(osb, handle);
559 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
565 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
572 ocfs2_commit_trans(osb, handle);
578 static int ocfs2_remove_inode(struct inode *inode,
579 struct buffer_head *di_bh,
580 struct inode *orphan_dir_inode,
581 struct buffer_head *orphan_dir_bh)
584 struct inode *inode_alloc_inode = NULL;
585 struct buffer_head *inode_alloc_bh = NULL;
587 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
588 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
591 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
592 le16_to_cpu(di->i_suballoc_slot));
593 if (!inode_alloc_inode) {
599 mutex_lock(&inode_alloc_inode->i_mutex);
600 status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
602 mutex_unlock(&inode_alloc_inode->i_mutex);
608 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
609 if (IS_ERR(handle)) {
610 status = PTR_ERR(handle);
615 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
622 /* set the inodes dtime */
623 status = ocfs2_journal_access(handle, inode, di_bh,
624 OCFS2_JOURNAL_ACCESS_WRITE);
630 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
631 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
633 status = ocfs2_journal_dirty(handle, di_bh);
639 ocfs2_remove_from_cache(inode, di_bh);
641 status = ocfs2_free_dinode(handle, inode_alloc_inode,
647 ocfs2_commit_trans(osb, handle);
649 ocfs2_meta_unlock(inode_alloc_inode, 1);
650 mutex_unlock(&inode_alloc_inode->i_mutex);
651 brelse(inode_alloc_bh);
653 iput(inode_alloc_inode);
659 * Serialize with orphan dir recovery. If the process doing
660 * recovery on this orphan dir does an iget() with the dir
661 * i_mutex held, we'll deadlock here. Instead we detect this
662 * and exit early - recovery will wipe this inode for us.
664 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
669 spin_lock(&osb->osb_lock);
670 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
671 mlog(0, "Recovery is happening on orphan dir %d, will skip "
672 "this inode\n", slot);
676 /* This signals to the orphan recovery process that it should
677 * wait for us to handle the wipe. */
678 osb->osb_orphan_wipes[slot]++;
680 spin_unlock(&osb->osb_lock);
684 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
687 spin_lock(&osb->osb_lock);
688 osb->osb_orphan_wipes[slot]--;
689 spin_unlock(&osb->osb_lock);
691 wake_up(&osb->osb_wipe_event);
694 static int ocfs2_wipe_inode(struct inode *inode,
695 struct buffer_head *di_bh)
697 int status, orphaned_slot;
698 struct inode *orphan_dir_inode = NULL;
699 struct buffer_head *orphan_dir_bh = NULL;
700 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
701 struct ocfs2_dinode *di;
703 di = (struct ocfs2_dinode *) di_bh->b_data;
704 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
706 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
710 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
711 ORPHAN_DIR_SYSTEM_INODE,
713 if (!orphan_dir_inode) {
719 /* Lock the orphan dir. The lock will be held for the entire
720 * delete_inode operation. We do this now to avoid races with
721 * recovery completion on other nodes. */
722 mutex_lock(&orphan_dir_inode->i_mutex);
723 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
725 mutex_unlock(&orphan_dir_inode->i_mutex);
731 /* we do this while holding the orphan dir lock because we
732 * don't want recovery being run from another node to vote for
733 * an inode delete on us -- this will result in two nodes
734 * truncating the same file! */
735 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
738 goto bail_unlock_dir;
741 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
747 ocfs2_meta_unlock(orphan_dir_inode, 1);
748 mutex_unlock(&orphan_dir_inode->i_mutex);
749 brelse(orphan_dir_bh);
751 iput(orphan_dir_inode);
752 ocfs2_signal_wipe_completion(osb, orphaned_slot);
757 /* There is a series of simple checks that should be done before a
758 * vote is even considered. Encapsulate those in this function. */
759 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
762 struct ocfs2_inode_info *oi = OCFS2_I(inode);
763 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
765 /* We shouldn't be getting here for the root directory
767 if (inode == osb->root_inode) {
768 mlog(ML_ERROR, "Skipping delete of root inode.\n");
772 /* If we're coming from process_vote we can't go into our own
773 * voting [hello, deadlock city!], so unforuntately we just
774 * have to skip deleting this guy. That's OK though because
775 * the node who's doing the actual deleting should handle it
777 if (current == osb->vote_task) {
778 mlog(0, "Skipping delete of %lu because we're currently "
779 "in process_vote\n", inode->i_ino);
783 spin_lock(&oi->ip_lock);
784 /* OCFS2 *never* deletes system files. This should technically
785 * never get here as system file inodes should always have a
786 * positive link count. */
787 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
788 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
789 (unsigned long long)oi->ip_blkno);
793 /* If we have voted "yes" on the wipe of this inode for
794 * another node, it will be marked here so we can safely skip
795 * it. Recovery will cleanup any inodes we might inadvertantly
797 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
798 mlog(0, "Skipping delete of %lu because another node "
799 "has done this for us.\n", inode->i_ino);
805 spin_unlock(&oi->ip_lock);
810 static int ocfs2_request_delete(struct inode *inode)
813 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
815 if (ocfs2_inode_is_new(inode))
818 if (ocfs2_node_map_is_only(osb, &osb->mounted_map,
822 * This is how ocfs2 determines whether an inode is still live
823 * within the cluster. Every node takes a shared read lock on
824 * the inode open lock in ocfs2_read_locked_inode(). When we
825 * get to ->delete_inode(), each node tries to convert it's
826 * lock to an exclusive. Trylocks are serialized by the inode
827 * meta data lock. If the upconvert suceeds, we know the inode
828 * is no longer live and can be deleted.
830 * Though we call this with the meta data lock held, the
831 * trylock keeps us from ABBA deadlock.
833 status = ocfs2_try_open_lock(inode, 1);
834 if (status < 0 && status != -EAGAIN)
839 /* Query the cluster to determine whether we should wipe an inode from
842 * Requires the inode to have the cluster lock. */
843 static int ocfs2_query_inode_wipe(struct inode *inode,
844 struct buffer_head *di_bh,
848 struct ocfs2_inode_info *oi = OCFS2_I(inode);
849 struct ocfs2_dinode *di;
853 /* While we were waiting for the cluster lock in
854 * ocfs2_delete_inode, another node might have asked to delete
855 * the inode. Recheck our flags to catch this. */
856 if (!ocfs2_inode_is_valid_to_delete(inode)) {
857 mlog(0, "Skipping delete of %llu because flags changed\n",
858 (unsigned long long)oi->ip_blkno);
862 /* Now that we have an up to date inode, we can double check
864 if (inode->i_nlink) {
865 mlog(0, "Skipping delete of %llu because nlink = %u\n",
866 (unsigned long long)oi->ip_blkno, inode->i_nlink);
870 /* Do some basic inode verification... */
871 di = (struct ocfs2_dinode *) di_bh->b_data;
872 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
873 /* for lack of a better error? */
876 "Inode %llu (on-disk %llu) not orphaned! "
877 "Disk flags 0x%x, inode flags 0x%x\n",
878 (unsigned long long)oi->ip_blkno,
879 (unsigned long long)di->i_blkno, di->i_flags,
884 /* has someone already deleted us?! baaad... */
891 status = ocfs2_request_delete(inode);
892 /* -EAGAIN means that other nodes are still using the
893 * inode. We're done here though, so avoid doing anything on
894 * disk and let them worry about deleting it. */
895 if (status == -EAGAIN) {
897 mlog(0, "Skipping delete of %llu because it is in use on"
898 "other nodes\n", (unsigned long long)oi->ip_blkno);
907 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
908 (unsigned long long)oi->ip_blkno,
909 le16_to_cpu(di->i_orphaned_slot));
915 /* Support function for ocfs2_delete_inode. Will help us keep the
916 * inode data in a consistent state for clear_inode. Always truncates
917 * pages, optionally sync's them first. */
918 static void ocfs2_cleanup_delete_inode(struct inode *inode,
921 mlog(0, "Cleanup inode %llu, sync = %d\n",
922 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
924 write_inode_now(inode, 1);
925 truncate_inode_pages(&inode->i_data, 0);
928 void ocfs2_delete_inode(struct inode *inode)
931 sigset_t blocked, oldset;
932 struct buffer_head *di_bh = NULL;
934 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
936 if (is_bad_inode(inode)) {
937 mlog(0, "Skipping delete of bad inode\n");
941 if (!ocfs2_inode_is_valid_to_delete(inode)) {
942 /* It's probably not necessary to truncate_inode_pages
943 * here but we do it for safety anyway (it will most
944 * likely be a no-op anyway) */
945 ocfs2_cleanup_delete_inode(inode, 0);
949 /* We want to block signals in delete_inode as the lock and
950 * messaging paths may return us -ERESTARTSYS. Which would
951 * cause us to exit early, resulting in inodes being orphaned
953 sigfillset(&blocked);
954 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
957 ocfs2_cleanup_delete_inode(inode, 1);
961 /* Lock down the inode. This gives us an up to date view of
962 * it's metadata (for verification), and allows us to
963 * serialize delete_inode votes.
965 * Even though we might be doing a truncate, we don't take the
966 * allocation lock here as it won't be needed - nobody will
967 * have the file open.
969 status = ocfs2_meta_lock(inode, &di_bh, 1);
971 if (status != -ENOENT)
973 ocfs2_cleanup_delete_inode(inode, 0);
977 /* Query the cluster. This will be the final decision made
978 * before we go ahead and wipe the inode. */
979 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
980 if (!wipe || status < 0) {
981 /* Error and inode busy vote both mean we won't be
982 * removing the inode, so they take almost the same
987 /* Someone in the cluster has voted to not wipe this
988 * inode, or it was never completely orphaned. Write
989 * out the pages and exit now. */
990 ocfs2_cleanup_delete_inode(inode, 1);
991 goto bail_unlock_inode;
994 ocfs2_cleanup_delete_inode(inode, 0);
996 status = ocfs2_wipe_inode(inode, di_bh);
998 if (status != -EDEADLK)
1000 goto bail_unlock_inode;
1004 * Mark the inode as successfully deleted.
1006 * This is important for ocfs2_clear_inode() as it will check
1007 * this flag and skip any checkpointing work
1009 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1010 * the LVB for other nodes.
1012 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1015 ocfs2_meta_unlock(inode, 1);
1018 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1026 void ocfs2_clear_inode(struct inode *inode)
1029 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1036 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1037 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1039 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1040 "Inode=%lu\n", inode->i_ino);
1042 /* For remove delete_inode vote, we hold open lock before,
1043 * now it is time to unlock PR and EX open locks. */
1044 ocfs2_open_unlock(inode);
1046 /* Do these before all the other work so that we don't bounce
1047 * the vote thread while waiting to destroy the locks. */
1048 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1049 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
1050 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
1051 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1053 /* We very well may get a clear_inode before all an inodes
1054 * metadata has hit disk. Of course, we can't drop any cluster
1055 * locks until the journal has finished with it. The only
1056 * exception here are successfully wiped inodes - their
1057 * metadata can now be considered to be part of the system
1058 * inodes from which it came. */
1059 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1060 ocfs2_checkpoint_inode(inode);
1062 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1063 "Clear inode of %llu, inode has io markers\n",
1064 (unsigned long long)oi->ip_blkno);
1066 ocfs2_extent_map_drop(inode, 0);
1067 ocfs2_extent_map_init(inode);
1069 status = ocfs2_drop_inode_locks(inode);
1073 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1074 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1075 ocfs2_lock_res_free(&oi->ip_data_lockres);
1076 ocfs2_lock_res_free(&oi->ip_open_lockres);
1078 ocfs2_metadata_cache_purge(inode);
1080 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1081 "Clear inode of %llu, inode has %u cache items\n",
1082 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1084 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1085 "Clear inode of %llu, inode has a bad flag\n",
1086 (unsigned long long)oi->ip_blkno);
1088 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1089 "Clear inode of %llu, inode is locked\n",
1090 (unsigned long long)oi->ip_blkno);
1092 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1093 "Clear inode of %llu, io_mutex is locked\n",
1094 (unsigned long long)oi->ip_blkno);
1095 mutex_unlock(&oi->ip_io_mutex);
1098 * down_trylock() returns 0, down_write_trylock() returns 1
1101 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1102 "Clear inode of %llu, alloc_sem is locked\n",
1103 (unsigned long long)oi->ip_blkno);
1104 up_write(&oi->ip_alloc_sem);
1106 mlog_bug_on_msg(oi->ip_open_count,
1107 "Clear inode of %llu has open count %d\n",
1108 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1110 /* Clear all other flags. */
1111 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1112 oi->ip_created_trans = 0;
1113 oi->ip_last_trans = 0;
1114 oi->ip_dir_start_lookup = 0;
1115 oi->ip_blkno = 0ULL;
1121 /* Called under inode_lock, with no more references on the
1122 * struct inode, so it's safe here to check the flags field
1123 * and to manipulate i_nlink without any other locks. */
1124 void ocfs2_drop_inode(struct inode *inode)
1126 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1130 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1131 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1133 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1134 generic_delete_inode(inode);
1136 generic_drop_inode(inode);
1142 * TODO: this should probably be merged into ocfs2_get_block
1144 * However, you now need to pay attention to the cont_prepare_write()
1145 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1146 * expects never to extend).
1148 struct buffer_head *ocfs2_bread(struct inode *inode,
1149 int block, int *err, int reada)
1151 struct buffer_head *bh = NULL;
1154 int readflags = OCFS2_BH_CACHED;
1157 readflags |= OCFS2_BH_READAHEAD;
1159 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1160 i_size_read(inode)) {
1165 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1172 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1192 * This is called from our getattr.
1194 int ocfs2_inode_revalidate(struct dentry *dentry)
1196 struct inode *inode = dentry->d_inode;
1199 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1200 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1203 mlog(0, "eep, no inode!\n");
1208 spin_lock(&OCFS2_I(inode)->ip_lock);
1209 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1210 spin_unlock(&OCFS2_I(inode)->ip_lock);
1211 mlog(0, "inode deleted!\n");
1215 spin_unlock(&OCFS2_I(inode)->ip_lock);
1217 /* Let ocfs2_meta_lock do the work of updating our struct
1219 status = ocfs2_meta_lock(inode, NULL, 0);
1221 if (status != -ENOENT)
1225 ocfs2_meta_unlock(inode, 0);
1233 * Updates a disk inode from a
1235 * Only takes ip_lock.
1237 int ocfs2_mark_inode_dirty(handle_t *handle,
1238 struct inode *inode,
1239 struct buffer_head *bh)
1242 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1244 mlog_entry("(inode %llu)\n",
1245 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1247 status = ocfs2_journal_access(handle, inode, bh,
1248 OCFS2_JOURNAL_ACCESS_WRITE);
1254 spin_lock(&OCFS2_I(inode)->ip_lock);
1255 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1256 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1257 spin_unlock(&OCFS2_I(inode)->ip_lock);
1259 fe->i_size = cpu_to_le64(i_size_read(inode));
1260 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1261 fe->i_uid = cpu_to_le32(inode->i_uid);
1262 fe->i_gid = cpu_to_le32(inode->i_gid);
1263 fe->i_mode = cpu_to_le16(inode->i_mode);
1264 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1265 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1266 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1267 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1268 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1269 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1271 status = ocfs2_journal_dirty(handle, bh);
1284 * Updates a struct inode from a disk inode.
1285 * does no i/o, only takes ip_lock.
1287 void ocfs2_refresh_inode(struct inode *inode,
1288 struct ocfs2_dinode *fe)
1290 spin_lock(&OCFS2_I(inode)->ip_lock);
1292 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1293 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1294 ocfs2_set_inode_flags(inode);
1295 i_size_write(inode, le64_to_cpu(fe->i_size));
1296 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1297 inode->i_uid = le32_to_cpu(fe->i_uid);
1298 inode->i_gid = le32_to_cpu(fe->i_gid);
1299 inode->i_mode = le16_to_cpu(fe->i_mode);
1300 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1301 inode->i_blocks = 0;
1303 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1304 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1305 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1306 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1307 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1308 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1309 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1311 spin_unlock(&OCFS2_I(inode)->ip_lock);