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>
32 #include <asm/byteorder.h>
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
41 #include "extent_map.h"
43 #include "heartbeat.h"
53 #include "buffer_head_io.h"
55 struct ocfs2_find_inode_args
59 unsigned int fi_flags;
60 unsigned int fi_sysfile_type;
63 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
65 static int ocfs2_read_locked_inode(struct inode *inode,
66 struct ocfs2_find_inode_args *args);
67 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
68 static int ocfs2_find_actor(struct inode *inode, void *opaque);
69 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
71 struct buffer_head *fe_bh);
73 void ocfs2_set_inode_flags(struct inode *inode)
75 unsigned int flags = OCFS2_I(inode)->ip_attr;
77 inode->i_flags &= ~(S_IMMUTABLE |
78 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
80 if (flags & OCFS2_IMMUTABLE_FL)
81 inode->i_flags |= S_IMMUTABLE;
83 if (flags & OCFS2_SYNC_FL)
84 inode->i_flags |= S_SYNC;
85 if (flags & OCFS2_APPEND_FL)
86 inode->i_flags |= S_APPEND;
87 if (flags & OCFS2_NOATIME_FL)
88 inode->i_flags |= S_NOATIME;
89 if (flags & OCFS2_DIRSYNC_FL)
90 inode->i_flags |= S_DIRSYNC;
93 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
94 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
96 unsigned int flags = oi->vfs_inode.i_flags;
98 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
99 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
101 oi->ip_attr |= OCFS2_SYNC_FL;
102 if (flags & S_APPEND)
103 oi->ip_attr |= OCFS2_APPEND_FL;
104 if (flags & S_IMMUTABLE)
105 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
106 if (flags & S_NOATIME)
107 oi->ip_attr |= OCFS2_NOATIME_FL;
108 if (flags & S_DIRSYNC)
109 oi->ip_attr |= OCFS2_DIRSYNC_FL;
112 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
115 struct inode *inode = NULL;
116 struct super_block *sb = osb->sb;
117 struct ocfs2_find_inode_args args;
119 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
121 /* Ok. By now we've either got the offsets passed to us by the
122 * caller, or we just pulled them off the bh. Lets do some
123 * sanity checks to make sure they're OK. */
125 inode = ERR_PTR(-EINVAL);
126 mlog_errno(PTR_ERR(inode));
130 args.fi_blkno = blkno;
131 args.fi_flags = flags;
132 args.fi_ino = ino_from_blkno(sb, blkno);
133 args.fi_sysfile_type = sysfile_type;
135 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
136 ocfs2_init_locked_inode, &args);
137 /* inode was *not* in the inode cache. 2.6.x requires
138 * us to do our own read_inode call and unlock it
140 if (inode && inode->i_state & I_NEW) {
141 mlog(0, "Inode was not in inode cache, reading it.\n");
142 ocfs2_read_locked_inode(inode, &args);
143 unlock_new_inode(inode);
146 inode = ERR_PTR(-ENOMEM);
147 mlog_errno(PTR_ERR(inode));
150 if (is_bad_inode(inode)) {
152 inode = ERR_PTR(-ESTALE);
157 if (!IS_ERR(inode)) {
158 mlog(0, "returning inode with number %llu\n",
159 (unsigned long long)OCFS2_I(inode)->ip_blkno);
160 mlog_exit_ptr(inode);
168 * here's how inodes get read from disk:
169 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
170 * found? : return the in-memory inode
171 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
174 static int ocfs2_find_actor(struct inode *inode, void *opaque)
176 struct ocfs2_find_inode_args *args = NULL;
177 struct ocfs2_inode_info *oi = OCFS2_I(inode);
180 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
184 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
186 if (oi->ip_blkno != args->fi_blkno)
196 * initialize the new inode, but don't do anything that would cause
198 * return 0 on success, 1 on failure
200 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
202 struct ocfs2_find_inode_args *args = opaque;
204 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
206 inode->i_ino = args->fi_ino;
207 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
208 if (args->fi_sysfile_type != 0)
209 lockdep_set_class(&inode->i_mutex,
210 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
216 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
219 struct super_block *sb;
220 struct ocfs2_super *osb;
221 int status = -EINVAL;
223 mlog_entry("(0x%p, size:%llu)\n", inode,
224 (unsigned long long)le64_to_cpu(fe->i_size));
229 /* this means that read_inode cannot create a superblock inode
230 * today. change if needed. */
231 if (!OCFS2_IS_VALID_DINODE(fe) ||
232 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
233 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
234 "signature = %.*s, flags = 0x%x\n",
236 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
237 fe->i_signature, le32_to_cpu(fe->i_flags));
241 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
242 mlog(ML_ERROR, "file entry generation does not match "
243 "superblock! osb->fs_generation=%x, "
244 "fe->i_fs_generation=%x\n",
245 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
249 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
250 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
251 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
253 inode->i_version = 1;
254 inode->i_generation = le32_to_cpu(fe->i_generation);
255 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
256 inode->i_mode = le16_to_cpu(fe->i_mode);
257 inode->i_uid = le32_to_cpu(fe->i_uid);
258 inode->i_gid = le32_to_cpu(fe->i_gid);
260 /* Fast symlinks will have i_size but no allocated clusters. */
261 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
264 inode->i_blocks = ocfs2_inode_sector_count(inode);
265 inode->i_mapping->a_ops = &ocfs2_aops;
266 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
267 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
268 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
269 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
270 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
271 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
273 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
275 "ip_blkno %llu != i_blkno %llu!\n",
276 (unsigned long long)OCFS2_I(inode)->ip_blkno,
277 (unsigned long long)le64_to_cpu(fe->i_blkno));
279 inode->i_nlink = le16_to_cpu(fe->i_links_count);
281 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
282 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
284 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
285 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
286 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
287 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
288 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
289 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
290 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
291 /* we can't actually hit this as read_inode can't
292 * handle superblocks today ;-) */
296 switch (inode->i_mode & S_IFMT) {
298 inode->i_fop = &ocfs2_fops;
299 inode->i_op = &ocfs2_file_iops;
300 i_size_write(inode, le64_to_cpu(fe->i_size));
303 inode->i_op = &ocfs2_dir_iops;
304 inode->i_fop = &ocfs2_dops;
305 i_size_write(inode, le64_to_cpu(fe->i_size));
308 if (ocfs2_inode_is_fast_symlink(inode))
309 inode->i_op = &ocfs2_fast_symlink_inode_operations;
311 inode->i_op = &ocfs2_symlink_inode_operations;
312 i_size_write(inode, le64_to_cpu(fe->i_size));
315 inode->i_op = &ocfs2_special_file_iops;
316 init_special_inode(inode, inode->i_mode,
322 inode->i_ino = ino_from_blkno(inode->i_sb,
323 le64_to_cpu(fe->i_blkno));
326 * If we ever want to create system files from kernel,
327 * the generation argument to
328 * ocfs2_inode_lock_res_init() will have to change.
330 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
332 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
333 OCFS2_LOCK_TYPE_META, 0, inode);
335 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
336 OCFS2_LOCK_TYPE_OPEN, 0, inode);
339 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
340 OCFS2_LOCK_TYPE_RW, inode->i_generation,
343 ocfs2_set_inode_flags(inode);
351 static int ocfs2_read_locked_inode(struct inode *inode,
352 struct ocfs2_find_inode_args *args)
354 struct super_block *sb;
355 struct ocfs2_super *osb;
356 struct ocfs2_dinode *fe;
357 struct buffer_head *bh = NULL;
358 int status, can_lock;
361 mlog_entry("(0x%p, 0x%p)\n", inode, args);
364 if (inode == NULL || inode->i_sb == NULL) {
365 mlog(ML_ERROR, "bad inode\n");
372 mlog(ML_ERROR, "bad inode args\n");
373 make_bad_inode(inode);
378 * To improve performance of cold-cache inode stats, we take
379 * the cluster lock here if possible.
381 * Generally, OCFS2 never trusts the contents of an inode
382 * unless it's holding a cluster lock, so taking it here isn't
383 * a correctness issue as much as it is a performance
386 * There are three times when taking the lock is not a good idea:
388 * 1) During startup, before we have initialized the DLM.
390 * 2) If we are reading certain system files which never get
391 * cluster locks (local alloc, truncate log).
393 * 3) If the process doing the iget() is responsible for
394 * orphan dir recovery. We're holding the orphan dir lock and
395 * can get into a deadlock with another process on another
396 * node in ->delete_inode().
398 * #1 and #2 can be simply solved by never taking the lock
399 * here for system files (which are the only type we read
400 * during mount). It's a heavier approach, but our main
401 * concern is user-accesible files anyway.
403 * #3 works itself out because we'll eventually take the
404 * cluster lock before trusting anything anyway.
406 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
407 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
408 && !ocfs2_mount_local(osb);
411 * To maintain backwards compatibility with older versions of
412 * ocfs2-tools, we still store the generation value for system
413 * files. The only ones that actually matter to userspace are
414 * the journals, but it's easier and inexpensive to just flag
415 * all system files similarly.
417 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
418 generation = osb->fs_generation;
420 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
421 OCFS2_LOCK_TYPE_META,
424 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
425 OCFS2_LOCK_TYPE_OPEN,
429 status = ocfs2_open_lock(inode);
431 make_bad_inode(inode);
435 status = ocfs2_inode_lock(inode, NULL, 0);
437 make_bad_inode(inode);
443 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
444 status = ocfs2_try_open_lock(inode, 0);
446 make_bad_inode(inode);
451 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
452 can_lock ? inode : NULL);
459 fe = (struct ocfs2_dinode *) bh->b_data;
460 if (!OCFS2_IS_VALID_DINODE(fe)) {
461 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
462 (unsigned long long)args->fi_blkno, 7,
468 * This is a code bug. Right now the caller needs to
469 * understand whether it is asking for a system file inode or
470 * not so the proper lock names can be built.
472 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
473 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
474 "Inode %llu: system file state is ambigous\n",
475 (unsigned long long)args->fi_blkno);
477 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
478 S_ISBLK(le16_to_cpu(fe->i_mode)))
479 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
481 if (ocfs2_populate_inode(inode, fe, 0) < 0)
484 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
490 ocfs2_inode_unlock(inode, 0);
493 make_bad_inode(inode);
502 void ocfs2_sync_blockdev(struct super_block *sb)
504 sync_blockdev(sb->s_bdev);
507 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
509 struct buffer_head *fe_bh)
512 struct ocfs2_truncate_context *tc = NULL;
513 struct ocfs2_dinode *fe;
514 handle_t *handle = NULL;
518 fe = (struct ocfs2_dinode *) fe_bh->b_data;
521 * This check will also skip truncate of inodes with inline
522 * data and fast symlinks.
524 if (fe->i_clusters) {
525 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
526 if (IS_ERR(handle)) {
527 status = PTR_ERR(handle);
532 status = ocfs2_journal_access(handle, inode, fe_bh,
533 OCFS2_JOURNAL_ACCESS_WRITE);
539 i_size_write(inode, 0);
541 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
547 ocfs2_commit_trans(osb, handle);
550 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
556 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
565 ocfs2_commit_trans(osb, handle);
570 static int ocfs2_remove_inode(struct inode *inode,
571 struct buffer_head *di_bh,
572 struct inode *orphan_dir_inode,
573 struct buffer_head *orphan_dir_bh)
576 struct inode *inode_alloc_inode = NULL;
577 struct buffer_head *inode_alloc_bh = NULL;
579 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
580 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
583 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
584 le16_to_cpu(di->i_suballoc_slot));
585 if (!inode_alloc_inode) {
591 mutex_lock(&inode_alloc_inode->i_mutex);
592 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
594 mutex_unlock(&inode_alloc_inode->i_mutex);
600 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
601 if (IS_ERR(handle)) {
602 status = PTR_ERR(handle);
607 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
614 /* set the inodes dtime */
615 status = ocfs2_journal_access(handle, inode, di_bh,
616 OCFS2_JOURNAL_ACCESS_WRITE);
622 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
623 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
625 status = ocfs2_journal_dirty(handle, di_bh);
631 ocfs2_remove_from_cache(inode, di_bh);
633 status = ocfs2_free_dinode(handle, inode_alloc_inode,
639 ocfs2_commit_trans(osb, handle);
641 ocfs2_inode_unlock(inode_alloc_inode, 1);
642 mutex_unlock(&inode_alloc_inode->i_mutex);
643 brelse(inode_alloc_bh);
645 iput(inode_alloc_inode);
651 * Serialize with orphan dir recovery. If the process doing
652 * recovery on this orphan dir does an iget() with the dir
653 * i_mutex held, we'll deadlock here. Instead we detect this
654 * and exit early - recovery will wipe this inode for us.
656 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
661 spin_lock(&osb->osb_lock);
662 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
663 mlog(0, "Recovery is happening on orphan dir %d, will skip "
664 "this inode\n", slot);
668 /* This signals to the orphan recovery process that it should
669 * wait for us to handle the wipe. */
670 osb->osb_orphan_wipes[slot]++;
672 spin_unlock(&osb->osb_lock);
676 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
679 spin_lock(&osb->osb_lock);
680 osb->osb_orphan_wipes[slot]--;
681 spin_unlock(&osb->osb_lock);
683 wake_up(&osb->osb_wipe_event);
686 static int ocfs2_wipe_inode(struct inode *inode,
687 struct buffer_head *di_bh)
689 int status, orphaned_slot;
690 struct inode *orphan_dir_inode = NULL;
691 struct buffer_head *orphan_dir_bh = NULL;
692 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
693 struct ocfs2_dinode *di;
695 di = (struct ocfs2_dinode *) di_bh->b_data;
696 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
698 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
702 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
703 ORPHAN_DIR_SYSTEM_INODE,
705 if (!orphan_dir_inode) {
711 /* Lock the orphan dir. The lock will be held for the entire
712 * delete_inode operation. We do this now to avoid races with
713 * recovery completion on other nodes. */
714 mutex_lock(&orphan_dir_inode->i_mutex);
715 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
717 mutex_unlock(&orphan_dir_inode->i_mutex);
723 /* we do this while holding the orphan dir lock because we
724 * don't want recovery being run from another node to try an
725 * inode delete underneath us -- this will result in two nodes
726 * truncating the same file! */
727 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
730 goto bail_unlock_dir;
733 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
739 ocfs2_inode_unlock(orphan_dir_inode, 1);
740 mutex_unlock(&orphan_dir_inode->i_mutex);
741 brelse(orphan_dir_bh);
743 iput(orphan_dir_inode);
744 ocfs2_signal_wipe_completion(osb, orphaned_slot);
749 /* There is a series of simple checks that should be done before a
750 * trylock is even considered. Encapsulate those in this function. */
751 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
754 struct ocfs2_inode_info *oi = OCFS2_I(inode);
755 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
757 /* We shouldn't be getting here for the root directory
759 if (inode == osb->root_inode) {
760 mlog(ML_ERROR, "Skipping delete of root inode.\n");
764 /* If we're coming from downconvert_thread we can't go into our own
765 * voting [hello, deadlock city!], so unforuntately we just
766 * have to skip deleting this guy. That's OK though because
767 * the node who's doing the actual deleting should handle it
769 if (current == osb->dc_task) {
770 mlog(0, "Skipping delete of %lu because we're currently "
771 "in downconvert\n", inode->i_ino);
775 spin_lock(&oi->ip_lock);
776 /* OCFS2 *never* deletes system files. This should technically
777 * never get here as system file inodes should always have a
778 * positive link count. */
779 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
780 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
781 (unsigned long long)oi->ip_blkno);
785 /* If we have allowd wipe of this inode for another node, it
786 * will be marked here so we can safely skip it. Recovery will
787 * cleanup any inodes we might inadvertantly skip here. */
788 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
789 mlog(0, "Skipping delete of %lu because another node "
790 "has done this for us.\n", inode->i_ino);
796 spin_unlock(&oi->ip_lock);
801 /* Query the cluster to determine whether we should wipe an inode from
804 * Requires the inode to have the cluster lock. */
805 static int ocfs2_query_inode_wipe(struct inode *inode,
806 struct buffer_head *di_bh,
810 struct ocfs2_inode_info *oi = OCFS2_I(inode);
811 struct ocfs2_dinode *di;
815 /* While we were waiting for the cluster lock in
816 * ocfs2_delete_inode, another node might have asked to delete
817 * the inode. Recheck our flags to catch this. */
818 if (!ocfs2_inode_is_valid_to_delete(inode)) {
819 mlog(0, "Skipping delete of %llu because flags changed\n",
820 (unsigned long long)oi->ip_blkno);
824 /* Now that we have an up to date inode, we can double check
826 if (inode->i_nlink) {
827 mlog(0, "Skipping delete of %llu because nlink = %u\n",
828 (unsigned long long)oi->ip_blkno, inode->i_nlink);
832 /* Do some basic inode verification... */
833 di = (struct ocfs2_dinode *) di_bh->b_data;
834 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
835 /* for lack of a better error? */
838 "Inode %llu (on-disk %llu) not orphaned! "
839 "Disk flags 0x%x, inode flags 0x%x\n",
840 (unsigned long long)oi->ip_blkno,
841 (unsigned long long)le64_to_cpu(di->i_blkno),
842 le32_to_cpu(di->i_flags), oi->ip_flags);
846 /* has someone already deleted us?! baaad... */
854 * This is how ocfs2 determines whether an inode is still live
855 * within the cluster. Every node takes a shared read lock on
856 * the inode open lock in ocfs2_read_locked_inode(). When we
857 * get to ->delete_inode(), each node tries to convert it's
858 * lock to an exclusive. Trylocks are serialized by the inode
859 * meta data lock. If the upconvert suceeds, we know the inode
860 * is no longer live and can be deleted.
862 * Though we call this with the meta data lock held, the
863 * trylock keeps us from ABBA deadlock.
865 status = ocfs2_try_open_lock(inode, 1);
866 if (status == -EAGAIN) {
868 mlog(0, "Skipping delete of %llu because it is in use on "
869 "other nodes\n", (unsigned long long)oi->ip_blkno);
878 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
879 (unsigned long long)oi->ip_blkno,
880 le16_to_cpu(di->i_orphaned_slot));
886 /* Support function for ocfs2_delete_inode. Will help us keep the
887 * inode data in a consistent state for clear_inode. Always truncates
888 * pages, optionally sync's them first. */
889 static void ocfs2_cleanup_delete_inode(struct inode *inode,
892 mlog(0, "Cleanup inode %llu, sync = %d\n",
893 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
895 write_inode_now(inode, 1);
896 truncate_inode_pages(&inode->i_data, 0);
899 void ocfs2_delete_inode(struct inode *inode)
902 sigset_t blocked, oldset;
903 struct buffer_head *di_bh = NULL;
905 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
907 if (is_bad_inode(inode)) {
908 mlog(0, "Skipping delete of bad inode\n");
912 if (!ocfs2_inode_is_valid_to_delete(inode)) {
913 /* It's probably not necessary to truncate_inode_pages
914 * here but we do it for safety anyway (it will most
915 * likely be a no-op anyway) */
916 ocfs2_cleanup_delete_inode(inode, 0);
920 /* We want to block signals in delete_inode as the lock and
921 * messaging paths may return us -ERESTARTSYS. Which would
922 * cause us to exit early, resulting in inodes being orphaned
924 sigfillset(&blocked);
925 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
928 ocfs2_cleanup_delete_inode(inode, 1);
932 /* Lock down the inode. This gives us an up to date view of
933 * it's metadata (for verification), and allows us to
934 * serialize delete_inode on multiple nodes.
936 * Even though we might be doing a truncate, we don't take the
937 * allocation lock here as it won't be needed - nobody will
938 * have the file open.
940 status = ocfs2_inode_lock(inode, &di_bh, 1);
942 if (status != -ENOENT)
944 ocfs2_cleanup_delete_inode(inode, 0);
948 /* Query the cluster. This will be the final decision made
949 * before we go ahead and wipe the inode. */
950 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
951 if (!wipe || status < 0) {
952 /* Error and remote inode busy both mean we won't be
953 * removing the inode, so they take almost the same
958 /* Someone in the cluster has disallowed a wipe of
959 * this inode, or it was never completely
960 * orphaned. Write out the pages and exit now. */
961 ocfs2_cleanup_delete_inode(inode, 1);
962 goto bail_unlock_inode;
965 ocfs2_cleanup_delete_inode(inode, 0);
967 status = ocfs2_wipe_inode(inode, di_bh);
969 if (status != -EDEADLK)
971 goto bail_unlock_inode;
975 * Mark the inode as successfully deleted.
977 * This is important for ocfs2_clear_inode() as it will check
978 * this flag and skip any checkpointing work
980 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
981 * the LVB for other nodes.
983 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
986 ocfs2_inode_unlock(inode, 1);
989 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
997 void ocfs2_clear_inode(struct inode *inode)
1000 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1007 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1008 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1010 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1011 "Inode=%lu\n", inode->i_ino);
1013 /* To preven remote deletes we hold open lock before, now it
1014 * is time to unlock PR and EX open locks. */
1015 ocfs2_open_unlock(inode);
1017 /* Do these before all the other work so that we don't bounce
1018 * the downconvert thread while waiting to destroy the locks. */
1019 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1020 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1021 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1023 /* We very well may get a clear_inode before all an inodes
1024 * metadata has hit disk. Of course, we can't drop any cluster
1025 * locks until the journal has finished with it. The only
1026 * exception here are successfully wiped inodes - their
1027 * metadata can now be considered to be part of the system
1028 * inodes from which it came. */
1029 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1030 ocfs2_checkpoint_inode(inode);
1032 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1033 "Clear inode of %llu, inode has io markers\n",
1034 (unsigned long long)oi->ip_blkno);
1036 ocfs2_extent_map_trunc(inode, 0);
1038 status = ocfs2_drop_inode_locks(inode);
1042 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1043 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1044 ocfs2_lock_res_free(&oi->ip_open_lockres);
1046 ocfs2_metadata_cache_purge(inode);
1048 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1049 "Clear inode of %llu, inode has %u cache items\n",
1050 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1052 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1053 "Clear inode of %llu, inode has a bad flag\n",
1054 (unsigned long long)oi->ip_blkno);
1056 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1057 "Clear inode of %llu, inode is locked\n",
1058 (unsigned long long)oi->ip_blkno);
1060 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1061 "Clear inode of %llu, io_mutex is locked\n",
1062 (unsigned long long)oi->ip_blkno);
1063 mutex_unlock(&oi->ip_io_mutex);
1066 * down_trylock() returns 0, down_write_trylock() returns 1
1069 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1070 "Clear inode of %llu, alloc_sem is locked\n",
1071 (unsigned long long)oi->ip_blkno);
1072 up_write(&oi->ip_alloc_sem);
1074 mlog_bug_on_msg(oi->ip_open_count,
1075 "Clear inode of %llu has open count %d\n",
1076 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1078 /* Clear all other flags. */
1079 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1080 oi->ip_created_trans = 0;
1081 oi->ip_last_trans = 0;
1082 oi->ip_dir_start_lookup = 0;
1083 oi->ip_blkno = 0ULL;
1089 /* Called under inode_lock, with no more references on the
1090 * struct inode, so it's safe here to check the flags field
1091 * and to manipulate i_nlink without any other locks. */
1092 void ocfs2_drop_inode(struct inode *inode)
1094 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1098 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1099 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1101 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1102 generic_delete_inode(inode);
1104 generic_drop_inode(inode);
1110 * TODO: this should probably be merged into ocfs2_get_block
1112 * However, you now need to pay attention to the cont_prepare_write()
1113 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1114 * expects never to extend).
1116 struct buffer_head *ocfs2_bread(struct inode *inode,
1117 int block, int *err, int reada)
1119 struct buffer_head *bh = NULL;
1122 int readflags = OCFS2_BH_CACHED;
1125 readflags |= OCFS2_BH_READAHEAD;
1127 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1128 i_size_read(inode)) {
1133 down_read(&OCFS2_I(inode)->ip_alloc_sem);
1134 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1136 up_read(&OCFS2_I(inode)->ip_alloc_sem);
1142 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1162 * This is called from our getattr.
1164 int ocfs2_inode_revalidate(struct dentry *dentry)
1166 struct inode *inode = dentry->d_inode;
1169 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1170 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1173 mlog(0, "eep, no inode!\n");
1178 spin_lock(&OCFS2_I(inode)->ip_lock);
1179 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1180 spin_unlock(&OCFS2_I(inode)->ip_lock);
1181 mlog(0, "inode deleted!\n");
1185 spin_unlock(&OCFS2_I(inode)->ip_lock);
1187 /* Let ocfs2_inode_lock do the work of updating our struct
1189 status = ocfs2_inode_lock(inode, NULL, 0);
1191 if (status != -ENOENT)
1195 ocfs2_inode_unlock(inode, 0);
1203 * Updates a disk inode from a
1205 * Only takes ip_lock.
1207 int ocfs2_mark_inode_dirty(handle_t *handle,
1208 struct inode *inode,
1209 struct buffer_head *bh)
1212 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1214 mlog_entry("(inode %llu)\n",
1215 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1217 status = ocfs2_journal_access(handle, inode, bh,
1218 OCFS2_JOURNAL_ACCESS_WRITE);
1224 spin_lock(&OCFS2_I(inode)->ip_lock);
1225 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1226 ocfs2_get_inode_flags(OCFS2_I(inode));
1227 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1228 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1229 spin_unlock(&OCFS2_I(inode)->ip_lock);
1231 fe->i_size = cpu_to_le64(i_size_read(inode));
1232 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1233 fe->i_uid = cpu_to_le32(inode->i_uid);
1234 fe->i_gid = cpu_to_le32(inode->i_gid);
1235 fe->i_mode = cpu_to_le16(inode->i_mode);
1236 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1237 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1238 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1239 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1240 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1241 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1243 status = ocfs2_journal_dirty(handle, bh);
1256 * Updates a struct inode from a disk inode.
1257 * does no i/o, only takes ip_lock.
1259 void ocfs2_refresh_inode(struct inode *inode,
1260 struct ocfs2_dinode *fe)
1262 spin_lock(&OCFS2_I(inode)->ip_lock);
1264 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1265 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1266 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1267 ocfs2_set_inode_flags(inode);
1268 i_size_write(inode, le64_to_cpu(fe->i_size));
1269 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1270 inode->i_uid = le32_to_cpu(fe->i_uid);
1271 inode->i_gid = le32_to_cpu(fe->i_gid);
1272 inode->i_mode = le16_to_cpu(fe->i_mode);
1273 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1274 inode->i_blocks = 0;
1276 inode->i_blocks = ocfs2_inode_sector_count(inode);
1277 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1278 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1279 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1280 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1281 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1282 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1284 spin_unlock(&OCFS2_I(inode)->ip_lock);