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"
54 #include "buffer_head_io.h"
56 struct ocfs2_find_inode_args
60 unsigned int fi_flags;
61 unsigned int fi_sysfile_type;
64 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
66 static int ocfs2_read_locked_inode(struct inode *inode,
67 struct ocfs2_find_inode_args *args);
68 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
69 static int ocfs2_find_actor(struct inode *inode, void *opaque);
70 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
72 struct buffer_head *fe_bh);
74 void ocfs2_set_inode_flags(struct inode *inode)
76 unsigned int flags = OCFS2_I(inode)->ip_attr;
78 inode->i_flags &= ~(S_IMMUTABLE |
79 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
81 if (flags & OCFS2_IMMUTABLE_FL)
82 inode->i_flags |= S_IMMUTABLE;
84 if (flags & OCFS2_SYNC_FL)
85 inode->i_flags |= S_SYNC;
86 if (flags & OCFS2_APPEND_FL)
87 inode->i_flags |= S_APPEND;
88 if (flags & OCFS2_NOATIME_FL)
89 inode->i_flags |= S_NOATIME;
90 if (flags & OCFS2_DIRSYNC_FL)
91 inode->i_flags |= S_DIRSYNC;
94 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
95 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
97 unsigned int flags = oi->vfs_inode.i_flags;
99 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
100 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
102 oi->ip_attr |= OCFS2_SYNC_FL;
103 if (flags & S_APPEND)
104 oi->ip_attr |= OCFS2_APPEND_FL;
105 if (flags & S_IMMUTABLE)
106 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
107 if (flags & S_NOATIME)
108 oi->ip_attr |= OCFS2_NOATIME_FL;
109 if (flags & S_DIRSYNC)
110 oi->ip_attr |= OCFS2_DIRSYNC_FL;
113 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
116 struct inode *inode = NULL;
117 struct super_block *sb = osb->sb;
118 struct ocfs2_find_inode_args args;
120 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
122 /* Ok. By now we've either got the offsets passed to us by the
123 * caller, or we just pulled them off the bh. Lets do some
124 * sanity checks to make sure they're OK. */
126 inode = ERR_PTR(-EINVAL);
127 mlog_errno(PTR_ERR(inode));
131 args.fi_blkno = blkno;
132 args.fi_flags = flags;
133 args.fi_ino = ino_from_blkno(sb, blkno);
134 args.fi_sysfile_type = sysfile_type;
136 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
137 ocfs2_init_locked_inode, &args);
138 /* inode was *not* in the inode cache. 2.6.x requires
139 * us to do our own read_inode call and unlock it
141 if (inode && inode->i_state & I_NEW) {
142 mlog(0, "Inode was not in inode cache, reading it.\n");
143 ocfs2_read_locked_inode(inode, &args);
144 unlock_new_inode(inode);
147 inode = ERR_PTR(-ENOMEM);
148 mlog_errno(PTR_ERR(inode));
151 if (is_bad_inode(inode)) {
153 inode = ERR_PTR(-ESTALE);
158 if (!IS_ERR(inode)) {
159 mlog(0, "returning inode with number %llu\n",
160 (unsigned long long)OCFS2_I(inode)->ip_blkno);
161 mlog_exit_ptr(inode);
169 * here's how inodes get read from disk:
170 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
171 * found? : return the in-memory inode
172 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
175 static int ocfs2_find_actor(struct inode *inode, void *opaque)
177 struct ocfs2_find_inode_args *args = NULL;
178 struct ocfs2_inode_info *oi = OCFS2_I(inode);
181 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
185 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
187 if (oi->ip_blkno != args->fi_blkno)
197 * initialize the new inode, but don't do anything that would cause
199 * return 0 on success, 1 on failure
201 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
203 struct ocfs2_find_inode_args *args = opaque;
205 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
207 inode->i_ino = args->fi_ino;
208 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
209 if (args->fi_sysfile_type != 0)
210 lockdep_set_class(&inode->i_mutex,
211 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
217 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
220 struct super_block *sb;
221 struct ocfs2_super *osb;
224 mlog_entry("(0x%p, size:%llu)\n", inode,
225 (unsigned long long)le64_to_cpu(fe->i_size));
230 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
231 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
235 * These have all been checked by ocfs2_read_inode_block() or set
236 * by ocfs2_mknod_locked(), so a failure is a code bug.
238 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
239 cannot create a superblock
240 inode today. change if
242 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
243 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
246 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
247 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
248 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
250 inode->i_version = 1;
251 inode->i_generation = le32_to_cpu(fe->i_generation);
252 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
253 inode->i_mode = le16_to_cpu(fe->i_mode);
254 inode->i_uid = le32_to_cpu(fe->i_uid);
255 inode->i_gid = le32_to_cpu(fe->i_gid);
257 /* Fast symlinks will have i_size but no allocated clusters. */
258 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
261 inode->i_blocks = ocfs2_inode_sector_count(inode);
262 inode->i_mapping->a_ops = &ocfs2_aops;
263 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
264 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
265 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
266 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
267 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
268 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
270 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
272 "ip_blkno %llu != i_blkno %llu!\n",
273 (unsigned long long)OCFS2_I(inode)->ip_blkno,
274 (unsigned long long)le64_to_cpu(fe->i_blkno));
276 inode->i_nlink = le16_to_cpu(fe->i_links_count);
278 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
279 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
280 inode->i_flags |= S_NOQUOTA;
283 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
284 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
285 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
286 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
287 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
288 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
289 inode->i_flags |= S_NOQUOTA;
290 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
291 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
292 /* we can't actually hit this as read_inode can't
293 * handle superblocks today ;-) */
297 switch (inode->i_mode & S_IFMT) {
300 inode->i_fop = &ocfs2_fops;
302 inode->i_fop = &ocfs2_fops_no_plocks;
303 inode->i_op = &ocfs2_file_iops;
304 i_size_write(inode, le64_to_cpu(fe->i_size));
307 inode->i_op = &ocfs2_dir_iops;
309 inode->i_fop = &ocfs2_dops;
311 inode->i_fop = &ocfs2_dops_no_plocks;
312 i_size_write(inode, le64_to_cpu(fe->i_size));
315 if (ocfs2_inode_is_fast_symlink(inode))
316 inode->i_op = &ocfs2_fast_symlink_inode_operations;
318 inode->i_op = &ocfs2_symlink_inode_operations;
319 i_size_write(inode, le64_to_cpu(fe->i_size));
322 inode->i_op = &ocfs2_special_file_iops;
323 init_special_inode(inode, inode->i_mode,
329 inode->i_ino = ino_from_blkno(inode->i_sb,
330 le64_to_cpu(fe->i_blkno));
333 * If we ever want to create system files from kernel,
334 * the generation argument to
335 * ocfs2_inode_lock_res_init() will have to change.
337 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
339 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
340 OCFS2_LOCK_TYPE_META, 0, inode);
342 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
343 OCFS2_LOCK_TYPE_OPEN, 0, inode);
346 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
347 OCFS2_LOCK_TYPE_RW, inode->i_generation,
350 ocfs2_set_inode_flags(inode);
355 static int ocfs2_read_locked_inode(struct inode *inode,
356 struct ocfs2_find_inode_args *args)
358 struct super_block *sb;
359 struct ocfs2_super *osb;
360 struct ocfs2_dinode *fe;
361 struct buffer_head *bh = NULL;
362 int status, can_lock;
365 mlog_entry("(0x%p, 0x%p)\n", inode, args);
368 if (inode == NULL || inode->i_sb == NULL) {
369 mlog(ML_ERROR, "bad inode\n");
376 mlog(ML_ERROR, "bad inode args\n");
377 make_bad_inode(inode);
382 * To improve performance of cold-cache inode stats, we take
383 * the cluster lock here if possible.
385 * Generally, OCFS2 never trusts the contents of an inode
386 * unless it's holding a cluster lock, so taking it here isn't
387 * a correctness issue as much as it is a performance
390 * There are three times when taking the lock is not a good idea:
392 * 1) During startup, before we have initialized the DLM.
394 * 2) If we are reading certain system files which never get
395 * cluster locks (local alloc, truncate log).
397 * 3) If the process doing the iget() is responsible for
398 * orphan dir recovery. We're holding the orphan dir lock and
399 * can get into a deadlock with another process on another
400 * node in ->delete_inode().
402 * #1 and #2 can be simply solved by never taking the lock
403 * here for system files (which are the only type we read
404 * during mount). It's a heavier approach, but our main
405 * concern is user-accesible files anyway.
407 * #3 works itself out because we'll eventually take the
408 * cluster lock before trusting anything anyway.
410 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
411 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
412 && !ocfs2_mount_local(osb);
415 * To maintain backwards compatibility with older versions of
416 * ocfs2-tools, we still store the generation value for system
417 * files. The only ones that actually matter to userspace are
418 * the journals, but it's easier and inexpensive to just flag
419 * all system files similarly.
421 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
422 generation = osb->fs_generation;
424 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
425 OCFS2_LOCK_TYPE_META,
428 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
429 OCFS2_LOCK_TYPE_OPEN,
433 status = ocfs2_open_lock(inode);
435 make_bad_inode(inode);
439 status = ocfs2_inode_lock(inode, NULL, 0);
441 make_bad_inode(inode);
447 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
448 status = ocfs2_try_open_lock(inode, 0);
450 make_bad_inode(inode);
456 status = ocfs2_read_inode_block_full(inode, &bh,
457 OCFS2_BH_IGNORE_CACHE);
459 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
461 status = ocfs2_validate_inode_block(osb->sb, bh);
469 fe = (struct ocfs2_dinode *) bh->b_data;
472 * This is a code bug. Right now the caller needs to
473 * understand whether it is asking for a system file inode or
474 * not so the proper lock names can be built.
476 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
477 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
478 "Inode %llu: system file state is ambigous\n",
479 (unsigned long long)args->fi_blkno);
481 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
482 S_ISBLK(le16_to_cpu(fe->i_mode)))
483 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
485 ocfs2_populate_inode(inode, fe, 0);
487 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
493 ocfs2_inode_unlock(inode, 0);
496 make_bad_inode(inode);
505 void ocfs2_sync_blockdev(struct super_block *sb)
507 sync_blockdev(sb->s_bdev);
510 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
512 struct buffer_head *fe_bh)
515 struct ocfs2_truncate_context *tc = NULL;
516 struct ocfs2_dinode *fe;
517 handle_t *handle = NULL;
521 fe = (struct ocfs2_dinode *) fe_bh->b_data;
524 * This check will also skip truncate of inodes with inline
525 * data and fast symlinks.
527 if (fe->i_clusters) {
528 if (ocfs2_should_order_data(inode))
529 ocfs2_begin_ordered_truncate(inode, 0);
531 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
532 if (IS_ERR(handle)) {
533 status = PTR_ERR(handle);
538 status = ocfs2_journal_access(handle, inode, fe_bh,
539 OCFS2_JOURNAL_ACCESS_WRITE);
545 i_size_write(inode, 0);
547 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
553 ocfs2_commit_trans(osb, handle);
556 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
562 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
571 ocfs2_commit_trans(osb, handle);
576 static int ocfs2_remove_inode(struct inode *inode,
577 struct buffer_head *di_bh,
578 struct inode *orphan_dir_inode,
579 struct buffer_head *orphan_dir_bh)
582 struct inode *inode_alloc_inode = NULL;
583 struct buffer_head *inode_alloc_bh = NULL;
585 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
586 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
589 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
590 le16_to_cpu(di->i_suballoc_slot));
591 if (!inode_alloc_inode) {
597 mutex_lock(&inode_alloc_inode->i_mutex);
598 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
600 mutex_unlock(&inode_alloc_inode->i_mutex);
606 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
607 if (IS_ERR(handle)) {
608 status = PTR_ERR(handle);
613 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
620 /* set the inodes dtime */
621 status = ocfs2_journal_access(handle, inode, di_bh,
622 OCFS2_JOURNAL_ACCESS_WRITE);
628 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
629 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
631 status = ocfs2_journal_dirty(handle, di_bh);
637 ocfs2_remove_from_cache(inode, di_bh);
639 status = ocfs2_free_dinode(handle, inode_alloc_inode,
645 ocfs2_commit_trans(osb, handle);
647 ocfs2_inode_unlock(inode_alloc_inode, 1);
648 mutex_unlock(&inode_alloc_inode->i_mutex);
649 brelse(inode_alloc_bh);
651 iput(inode_alloc_inode);
657 * Serialize with orphan dir recovery. If the process doing
658 * recovery on this orphan dir does an iget() with the dir
659 * i_mutex held, we'll deadlock here. Instead we detect this
660 * and exit early - recovery will wipe this inode for us.
662 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
667 spin_lock(&osb->osb_lock);
668 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
669 mlog(0, "Recovery is happening on orphan dir %d, will skip "
670 "this inode\n", slot);
674 /* This signals to the orphan recovery process that it should
675 * wait for us to handle the wipe. */
676 osb->osb_orphan_wipes[slot]++;
678 spin_unlock(&osb->osb_lock);
682 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
685 spin_lock(&osb->osb_lock);
686 osb->osb_orphan_wipes[slot]--;
687 spin_unlock(&osb->osb_lock);
689 wake_up(&osb->osb_wipe_event);
692 static int ocfs2_wipe_inode(struct inode *inode,
693 struct buffer_head *di_bh)
695 int status, orphaned_slot;
696 struct inode *orphan_dir_inode = NULL;
697 struct buffer_head *orphan_dir_bh = NULL;
698 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
699 struct ocfs2_dinode *di;
701 di = (struct ocfs2_dinode *) di_bh->b_data;
702 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
704 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
708 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
709 ORPHAN_DIR_SYSTEM_INODE,
711 if (!orphan_dir_inode) {
717 /* Lock the orphan dir. The lock will be held for the entire
718 * delete_inode operation. We do this now to avoid races with
719 * recovery completion on other nodes. */
720 mutex_lock(&orphan_dir_inode->i_mutex);
721 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
723 mutex_unlock(&orphan_dir_inode->i_mutex);
729 /* we do this while holding the orphan dir lock because we
730 * don't want recovery being run from another node to try an
731 * inode delete underneath us -- this will result in two nodes
732 * truncating the same file! */
733 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
736 goto bail_unlock_dir;
739 /*Free extended attribute resources associated with this inode.*/
740 status = ocfs2_xattr_remove(inode, di_bh);
743 goto bail_unlock_dir;
746 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
752 ocfs2_inode_unlock(orphan_dir_inode, 1);
753 mutex_unlock(&orphan_dir_inode->i_mutex);
754 brelse(orphan_dir_bh);
756 iput(orphan_dir_inode);
757 ocfs2_signal_wipe_completion(osb, orphaned_slot);
762 /* There is a series of simple checks that should be done before a
763 * trylock is even considered. Encapsulate those in this function. */
764 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
767 struct ocfs2_inode_info *oi = OCFS2_I(inode);
768 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
770 /* We shouldn't be getting here for the root directory
772 if (inode == osb->root_inode) {
773 mlog(ML_ERROR, "Skipping delete of root inode.\n");
777 /* If we're coming from downconvert_thread we can't go into our own
778 * voting [hello, deadlock city!], so unforuntately we just
779 * have to skip deleting this guy. That's OK though because
780 * the node who's doing the actual deleting should handle it
782 if (current == osb->dc_task) {
783 mlog(0, "Skipping delete of %lu because we're currently "
784 "in downconvert\n", inode->i_ino);
788 spin_lock(&oi->ip_lock);
789 /* OCFS2 *never* deletes system files. This should technically
790 * never get here as system file inodes should always have a
791 * positive link count. */
792 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
793 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
794 (unsigned long long)oi->ip_blkno);
798 /* If we have allowd wipe of this inode for another node, it
799 * will be marked here so we can safely skip it. Recovery will
800 * cleanup any inodes we might inadvertantly skip here. */
801 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
802 mlog(0, "Skipping delete of %lu because another node "
803 "has done this for us.\n", inode->i_ino);
809 spin_unlock(&oi->ip_lock);
814 /* Query the cluster to determine whether we should wipe an inode from
817 * Requires the inode to have the cluster lock. */
818 static int ocfs2_query_inode_wipe(struct inode *inode,
819 struct buffer_head *di_bh,
823 struct ocfs2_inode_info *oi = OCFS2_I(inode);
824 struct ocfs2_dinode *di;
828 /* While we were waiting for the cluster lock in
829 * ocfs2_delete_inode, another node might have asked to delete
830 * the inode. Recheck our flags to catch this. */
831 if (!ocfs2_inode_is_valid_to_delete(inode)) {
832 mlog(0, "Skipping delete of %llu because flags changed\n",
833 (unsigned long long)oi->ip_blkno);
837 /* Now that we have an up to date inode, we can double check
839 if (inode->i_nlink) {
840 mlog(0, "Skipping delete of %llu because nlink = %u\n",
841 (unsigned long long)oi->ip_blkno, inode->i_nlink);
845 /* Do some basic inode verification... */
846 di = (struct ocfs2_dinode *) di_bh->b_data;
847 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
848 /* for lack of a better error? */
851 "Inode %llu (on-disk %llu) not orphaned! "
852 "Disk flags 0x%x, inode flags 0x%x\n",
853 (unsigned long long)oi->ip_blkno,
854 (unsigned long long)le64_to_cpu(di->i_blkno),
855 le32_to_cpu(di->i_flags), oi->ip_flags);
859 /* has someone already deleted us?! baaad... */
867 * This is how ocfs2 determines whether an inode is still live
868 * within the cluster. Every node takes a shared read lock on
869 * the inode open lock in ocfs2_read_locked_inode(). When we
870 * get to ->delete_inode(), each node tries to convert it's
871 * lock to an exclusive. Trylocks are serialized by the inode
872 * meta data lock. If the upconvert suceeds, we know the inode
873 * is no longer live and can be deleted.
875 * Though we call this with the meta data lock held, the
876 * trylock keeps us from ABBA deadlock.
878 status = ocfs2_try_open_lock(inode, 1);
879 if (status == -EAGAIN) {
881 mlog(0, "Skipping delete of %llu because it is in use on "
882 "other nodes\n", (unsigned long long)oi->ip_blkno);
891 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
892 (unsigned long long)oi->ip_blkno,
893 le16_to_cpu(di->i_orphaned_slot));
899 /* Support function for ocfs2_delete_inode. Will help us keep the
900 * inode data in a consistent state for clear_inode. Always truncates
901 * pages, optionally sync's them first. */
902 static void ocfs2_cleanup_delete_inode(struct inode *inode,
905 mlog(0, "Cleanup inode %llu, sync = %d\n",
906 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
908 write_inode_now(inode, 1);
909 truncate_inode_pages(&inode->i_data, 0);
912 void ocfs2_delete_inode(struct inode *inode)
915 sigset_t blocked, oldset;
916 struct buffer_head *di_bh = NULL;
918 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
920 if (is_bad_inode(inode)) {
921 mlog(0, "Skipping delete of bad inode\n");
925 if (!ocfs2_inode_is_valid_to_delete(inode)) {
926 /* It's probably not necessary to truncate_inode_pages
927 * here but we do it for safety anyway (it will most
928 * likely be a no-op anyway) */
929 ocfs2_cleanup_delete_inode(inode, 0);
933 /* We want to block signals in delete_inode as the lock and
934 * messaging paths may return us -ERESTARTSYS. Which would
935 * cause us to exit early, resulting in inodes being orphaned
937 sigfillset(&blocked);
938 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
941 ocfs2_cleanup_delete_inode(inode, 1);
945 /* Lock down the inode. This gives us an up to date view of
946 * it's metadata (for verification), and allows us to
947 * serialize delete_inode on multiple nodes.
949 * Even though we might be doing a truncate, we don't take the
950 * allocation lock here as it won't be needed - nobody will
951 * have the file open.
953 status = ocfs2_inode_lock(inode, &di_bh, 1);
955 if (status != -ENOENT)
957 ocfs2_cleanup_delete_inode(inode, 0);
961 /* Query the cluster. This will be the final decision made
962 * before we go ahead and wipe the inode. */
963 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
964 if (!wipe || status < 0) {
965 /* Error and remote inode busy both mean we won't be
966 * removing the inode, so they take almost the same
971 /* Someone in the cluster has disallowed a wipe of
972 * this inode, or it was never completely
973 * orphaned. Write out the pages and exit now. */
974 ocfs2_cleanup_delete_inode(inode, 1);
975 goto bail_unlock_inode;
978 ocfs2_cleanup_delete_inode(inode, 0);
980 status = ocfs2_wipe_inode(inode, di_bh);
982 if (status != -EDEADLK)
984 goto bail_unlock_inode;
988 * Mark the inode as successfully deleted.
990 * This is important for ocfs2_clear_inode() as it will check
991 * this flag and skip any checkpointing work
993 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
994 * the LVB for other nodes.
996 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
999 ocfs2_inode_unlock(inode, 1);
1002 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1010 void ocfs2_clear_inode(struct inode *inode)
1013 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1020 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1021 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1023 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1024 "Inode=%lu\n", inode->i_ino);
1026 /* To preven remote deletes we hold open lock before, now it
1027 * is time to unlock PR and EX open locks. */
1028 ocfs2_open_unlock(inode);
1030 /* Do these before all the other work so that we don't bounce
1031 * the downconvert thread while waiting to destroy the locks. */
1032 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1033 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1034 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1036 /* We very well may get a clear_inode before all an inodes
1037 * metadata has hit disk. Of course, we can't drop any cluster
1038 * locks until the journal has finished with it. The only
1039 * exception here are successfully wiped inodes - their
1040 * metadata can now be considered to be part of the system
1041 * inodes from which it came. */
1042 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1043 ocfs2_checkpoint_inode(inode);
1045 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1046 "Clear inode of %llu, inode has io markers\n",
1047 (unsigned long long)oi->ip_blkno);
1049 ocfs2_extent_map_trunc(inode, 0);
1051 status = ocfs2_drop_inode_locks(inode);
1055 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1056 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1057 ocfs2_lock_res_free(&oi->ip_open_lockres);
1059 ocfs2_metadata_cache_purge(inode);
1061 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1062 "Clear inode of %llu, inode has %u cache items\n",
1063 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1065 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1066 "Clear inode of %llu, inode has a bad flag\n",
1067 (unsigned long long)oi->ip_blkno);
1069 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1070 "Clear inode of %llu, inode is locked\n",
1071 (unsigned long long)oi->ip_blkno);
1073 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1074 "Clear inode of %llu, io_mutex is locked\n",
1075 (unsigned long long)oi->ip_blkno);
1076 mutex_unlock(&oi->ip_io_mutex);
1079 * down_trylock() returns 0, down_write_trylock() returns 1
1082 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1083 "Clear inode of %llu, alloc_sem is locked\n",
1084 (unsigned long long)oi->ip_blkno);
1085 up_write(&oi->ip_alloc_sem);
1087 mlog_bug_on_msg(oi->ip_open_count,
1088 "Clear inode of %llu has open count %d\n",
1089 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1091 /* Clear all other flags. */
1092 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1093 oi->ip_created_trans = 0;
1094 oi->ip_last_trans = 0;
1095 oi->ip_dir_start_lookup = 0;
1096 oi->ip_blkno = 0ULL;
1099 * ip_jinode is used to track txns against this inode. We ensure that
1100 * the journal is flushed before journal shutdown. Thus it is safe to
1101 * have inodes get cleaned up after journal shutdown.
1103 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1110 /* Called under inode_lock, with no more references on the
1111 * struct inode, so it's safe here to check the flags field
1112 * and to manipulate i_nlink without any other locks. */
1113 void ocfs2_drop_inode(struct inode *inode)
1115 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1119 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1120 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1122 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1123 generic_delete_inode(inode);
1125 generic_drop_inode(inode);
1131 * This is called from our getattr.
1133 int ocfs2_inode_revalidate(struct dentry *dentry)
1135 struct inode *inode = dentry->d_inode;
1138 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1139 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1142 mlog(0, "eep, no inode!\n");
1147 spin_lock(&OCFS2_I(inode)->ip_lock);
1148 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1149 spin_unlock(&OCFS2_I(inode)->ip_lock);
1150 mlog(0, "inode deleted!\n");
1154 spin_unlock(&OCFS2_I(inode)->ip_lock);
1156 /* Let ocfs2_inode_lock do the work of updating our struct
1158 status = ocfs2_inode_lock(inode, NULL, 0);
1160 if (status != -ENOENT)
1164 ocfs2_inode_unlock(inode, 0);
1172 * Updates a disk inode from a
1174 * Only takes ip_lock.
1176 int ocfs2_mark_inode_dirty(handle_t *handle,
1177 struct inode *inode,
1178 struct buffer_head *bh)
1181 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1183 mlog_entry("(inode %llu)\n",
1184 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1186 status = ocfs2_journal_access(handle, inode, bh,
1187 OCFS2_JOURNAL_ACCESS_WRITE);
1193 spin_lock(&OCFS2_I(inode)->ip_lock);
1194 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1195 ocfs2_get_inode_flags(OCFS2_I(inode));
1196 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1197 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1198 spin_unlock(&OCFS2_I(inode)->ip_lock);
1200 fe->i_size = cpu_to_le64(i_size_read(inode));
1201 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1202 fe->i_uid = cpu_to_le32(inode->i_uid);
1203 fe->i_gid = cpu_to_le32(inode->i_gid);
1204 fe->i_mode = cpu_to_le16(inode->i_mode);
1205 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1206 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1207 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1208 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1209 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1210 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1212 status = ocfs2_journal_dirty(handle, bh);
1225 * Updates a struct inode from a disk inode.
1226 * does no i/o, only takes ip_lock.
1228 void ocfs2_refresh_inode(struct inode *inode,
1229 struct ocfs2_dinode *fe)
1231 spin_lock(&OCFS2_I(inode)->ip_lock);
1233 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1234 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1235 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1236 ocfs2_set_inode_flags(inode);
1237 i_size_write(inode, le64_to_cpu(fe->i_size));
1238 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1239 inode->i_uid = le32_to_cpu(fe->i_uid);
1240 inode->i_gid = le32_to_cpu(fe->i_gid);
1241 inode->i_mode = le16_to_cpu(fe->i_mode);
1242 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1243 inode->i_blocks = 0;
1245 inode->i_blocks = ocfs2_inode_sector_count(inode);
1246 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1247 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1248 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1249 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1250 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1251 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1253 spin_unlock(&OCFS2_I(inode)->ip_lock);
1256 int ocfs2_validate_inode_block(struct super_block *sb,
1257 struct buffer_head *bh)
1260 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1262 mlog(0, "Validating dinode %llu\n",
1263 (unsigned long long)bh->b_blocknr);
1265 BUG_ON(!buffer_uptodate(bh));
1267 if (!OCFS2_IS_VALID_DINODE(di)) {
1268 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1269 (unsigned long long)bh->b_blocknr, 7,
1274 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1275 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1276 (unsigned long long)bh->b_blocknr,
1277 (unsigned long long)le64_to_cpu(di->i_blkno));
1281 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1283 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1284 (unsigned long long)bh->b_blocknr);
1288 if (le32_to_cpu(di->i_fs_generation) !=
1289 OCFS2_SB(sb)->fs_generation) {
1291 "Invalid dinode #%llu: fs_generation is %u\n",
1292 (unsigned long long)bh->b_blocknr,
1293 le32_to_cpu(di->i_fs_generation));
1303 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1307 struct buffer_head *tmp = *bh;
1309 rc = ocfs2_read_blocks(inode, OCFS2_I(inode)->ip_blkno, 1, &tmp,
1310 flags, ocfs2_validate_inode_block);
1312 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1319 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1321 return ocfs2_read_inode_block_full(inode, bh, 0);