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/quotaops.h>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
42 #include "blockcheck.h"
44 #include "extent_map.h"
46 #include "heartbeat.h"
57 #include "buffer_head_io.h"
59 struct ocfs2_find_inode_args
63 unsigned int fi_flags;
64 unsigned int fi_sysfile_type;
67 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
69 static int ocfs2_read_locked_inode(struct inode *inode,
70 struct ocfs2_find_inode_args *args);
71 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
72 static int ocfs2_find_actor(struct inode *inode, void *opaque);
73 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
75 struct buffer_head *fe_bh);
77 void ocfs2_set_inode_flags(struct inode *inode)
79 unsigned int flags = OCFS2_I(inode)->ip_attr;
81 inode->i_flags &= ~(S_IMMUTABLE |
82 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
84 if (flags & OCFS2_IMMUTABLE_FL)
85 inode->i_flags |= S_IMMUTABLE;
87 if (flags & OCFS2_SYNC_FL)
88 inode->i_flags |= S_SYNC;
89 if (flags & OCFS2_APPEND_FL)
90 inode->i_flags |= S_APPEND;
91 if (flags & OCFS2_NOATIME_FL)
92 inode->i_flags |= S_NOATIME;
93 if (flags & OCFS2_DIRSYNC_FL)
94 inode->i_flags |= S_DIRSYNC;
97 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
98 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
100 unsigned int flags = oi->vfs_inode.i_flags;
102 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
103 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
105 oi->ip_attr |= OCFS2_SYNC_FL;
106 if (flags & S_APPEND)
107 oi->ip_attr |= OCFS2_APPEND_FL;
108 if (flags & S_IMMUTABLE)
109 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
110 if (flags & S_NOATIME)
111 oi->ip_attr |= OCFS2_NOATIME_FL;
112 if (flags & S_DIRSYNC)
113 oi->ip_attr |= OCFS2_DIRSYNC_FL;
116 struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
118 struct ocfs2_find_inode_args args;
120 args.fi_blkno = blkno;
122 args.fi_ino = ino_from_blkno(sb, blkno);
123 args.fi_sysfile_type = 0;
125 return ilookup5(sb, blkno, ocfs2_find_actor, &args);
127 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
130 struct inode *inode = NULL;
131 struct super_block *sb = osb->sb;
132 struct ocfs2_find_inode_args args;
134 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
136 /* Ok. By now we've either got the offsets passed to us by the
137 * caller, or we just pulled them off the bh. Lets do some
138 * sanity checks to make sure they're OK. */
140 inode = ERR_PTR(-EINVAL);
141 mlog_errno(PTR_ERR(inode));
145 args.fi_blkno = blkno;
146 args.fi_flags = flags;
147 args.fi_ino = ino_from_blkno(sb, blkno);
148 args.fi_sysfile_type = sysfile_type;
150 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
151 ocfs2_init_locked_inode, &args);
152 /* inode was *not* in the inode cache. 2.6.x requires
153 * us to do our own read_inode call and unlock it
155 if (inode && inode->i_state & I_NEW) {
156 mlog(0, "Inode was not in inode cache, reading it.\n");
157 ocfs2_read_locked_inode(inode, &args);
158 unlock_new_inode(inode);
161 inode = ERR_PTR(-ENOMEM);
162 mlog_errno(PTR_ERR(inode));
165 if (is_bad_inode(inode)) {
167 inode = ERR_PTR(-ESTALE);
172 if (!IS_ERR(inode)) {
173 mlog(0, "returning inode with number %llu\n",
174 (unsigned long long)OCFS2_I(inode)->ip_blkno);
175 mlog_exit_ptr(inode);
183 * here's how inodes get read from disk:
184 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
185 * found? : return the in-memory inode
186 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
189 static int ocfs2_find_actor(struct inode *inode, void *opaque)
191 struct ocfs2_find_inode_args *args = NULL;
192 struct ocfs2_inode_info *oi = OCFS2_I(inode);
195 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
199 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
201 if (oi->ip_blkno != args->fi_blkno)
211 * initialize the new inode, but don't do anything that would cause
213 * return 0 on success, 1 on failure
215 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
217 struct ocfs2_find_inode_args *args = opaque;
219 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
221 inode->i_ino = args->fi_ino;
222 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
223 if (args->fi_sysfile_type != 0)
224 lockdep_set_class(&inode->i_mutex,
225 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
231 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
234 struct super_block *sb;
235 struct ocfs2_super *osb;
238 mlog_entry("(0x%p, size:%llu)\n", inode,
239 (unsigned long long)le64_to_cpu(fe->i_size));
244 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
245 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
249 * These have all been checked by ocfs2_read_inode_block() or set
250 * by ocfs2_mknod_locked(), so a failure is a code bug.
252 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
253 cannot create a superblock
254 inode today. change if
256 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
257 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
260 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
261 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
262 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
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)
275 inode->i_blocks = ocfs2_inode_sector_count(inode);
276 inode->i_mapping->a_ops = &ocfs2_aops;
277 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
278 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
279 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
280 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
281 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
282 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
284 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
286 "ip_blkno %llu != i_blkno %llu!\n",
287 (unsigned long long)OCFS2_I(inode)->ip_blkno,
288 (unsigned long long)le64_to_cpu(fe->i_blkno));
290 inode->i_nlink = ocfs2_read_links_count(fe);
292 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
293 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
294 inode->i_flags |= S_NOQUOTA;
297 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
298 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
299 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
300 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
301 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
302 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
303 inode->i_flags |= S_NOQUOTA;
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) {
314 inode->i_fop = &ocfs2_fops;
316 inode->i_fop = &ocfs2_fops_no_plocks;
317 inode->i_op = &ocfs2_file_iops;
318 i_size_write(inode, le64_to_cpu(fe->i_size));
321 inode->i_op = &ocfs2_dir_iops;
323 inode->i_fop = &ocfs2_dops;
325 inode->i_fop = &ocfs2_dops_no_plocks;
326 i_size_write(inode, le64_to_cpu(fe->i_size));
329 if (ocfs2_inode_is_fast_symlink(inode))
330 inode->i_op = &ocfs2_fast_symlink_inode_operations;
332 inode->i_op = &ocfs2_symlink_inode_operations;
333 i_size_write(inode, le64_to_cpu(fe->i_size));
336 inode->i_op = &ocfs2_special_file_iops;
337 init_special_inode(inode, inode->i_mode,
343 inode->i_ino = ino_from_blkno(inode->i_sb,
344 le64_to_cpu(fe->i_blkno));
347 * If we ever want to create system files from kernel,
348 * the generation argument to
349 * ocfs2_inode_lock_res_init() will have to change.
351 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
353 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
354 OCFS2_LOCK_TYPE_META, 0, inode);
356 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
357 OCFS2_LOCK_TYPE_OPEN, 0, inode);
360 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
361 OCFS2_LOCK_TYPE_RW, inode->i_generation,
364 ocfs2_set_inode_flags(inode);
366 OCFS2_I(inode)->ip_last_used_slot = 0;
367 OCFS2_I(inode)->ip_last_used_group = 0;
371 static int ocfs2_read_locked_inode(struct inode *inode,
372 struct ocfs2_find_inode_args *args)
374 struct super_block *sb;
375 struct ocfs2_super *osb;
376 struct ocfs2_dinode *fe;
377 struct buffer_head *bh = NULL;
378 int status, can_lock;
381 mlog_entry("(0x%p, 0x%p)\n", inode, args);
384 if (inode == NULL || inode->i_sb == NULL) {
385 mlog(ML_ERROR, "bad inode\n");
392 mlog(ML_ERROR, "bad inode args\n");
393 make_bad_inode(inode);
398 * To improve performance of cold-cache inode stats, we take
399 * the cluster lock here if possible.
401 * Generally, OCFS2 never trusts the contents of an inode
402 * unless it's holding a cluster lock, so taking it here isn't
403 * a correctness issue as much as it is a performance
406 * There are three times when taking the lock is not a good idea:
408 * 1) During startup, before we have initialized the DLM.
410 * 2) If we are reading certain system files which never get
411 * cluster locks (local alloc, truncate log).
413 * 3) If the process doing the iget() is responsible for
414 * orphan dir recovery. We're holding the orphan dir lock and
415 * can get into a deadlock with another process on another
416 * node in ->delete_inode().
418 * #1 and #2 can be simply solved by never taking the lock
419 * here for system files (which are the only type we read
420 * during mount). It's a heavier approach, but our main
421 * concern is user-accesible files anyway.
423 * #3 works itself out because we'll eventually take the
424 * cluster lock before trusting anything anyway.
426 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
427 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
428 && !ocfs2_mount_local(osb);
431 * To maintain backwards compatibility with older versions of
432 * ocfs2-tools, we still store the generation value for system
433 * files. The only ones that actually matter to userspace are
434 * the journals, but it's easier and inexpensive to just flag
435 * all system files similarly.
437 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
438 generation = osb->fs_generation;
440 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
441 OCFS2_LOCK_TYPE_META,
444 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
445 OCFS2_LOCK_TYPE_OPEN,
449 status = ocfs2_open_lock(inode);
451 make_bad_inode(inode);
455 status = ocfs2_inode_lock(inode, NULL, 0);
457 make_bad_inode(inode);
463 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
464 status = ocfs2_try_open_lock(inode, 0);
466 make_bad_inode(inode);
472 status = ocfs2_read_inode_block_full(inode, &bh,
473 OCFS2_BH_IGNORE_CACHE);
475 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
477 status = ocfs2_validate_inode_block(osb->sb, bh);
485 fe = (struct ocfs2_dinode *) bh->b_data;
488 * This is a code bug. Right now the caller needs to
489 * understand whether it is asking for a system file inode or
490 * not so the proper lock names can be built.
492 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
493 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
494 "Inode %llu: system file state is ambigous\n",
495 (unsigned long long)args->fi_blkno);
497 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
498 S_ISBLK(le16_to_cpu(fe->i_mode)))
499 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
501 ocfs2_populate_inode(inode, fe, 0);
503 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
509 ocfs2_inode_unlock(inode, 0);
512 make_bad_inode(inode);
521 void ocfs2_sync_blockdev(struct super_block *sb)
523 sync_blockdev(sb->s_bdev);
526 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
528 struct buffer_head *fe_bh)
531 struct ocfs2_truncate_context *tc = NULL;
532 struct ocfs2_dinode *fe;
533 handle_t *handle = NULL;
537 fe = (struct ocfs2_dinode *) fe_bh->b_data;
540 * This check will also skip truncate of inodes with inline
541 * data and fast symlinks.
543 if (fe->i_clusters) {
544 if (ocfs2_should_order_data(inode))
545 ocfs2_begin_ordered_truncate(inode, 0);
547 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
548 if (IS_ERR(handle)) {
549 status = PTR_ERR(handle);
554 status = ocfs2_journal_access_di(handle, inode, fe_bh,
555 OCFS2_JOURNAL_ACCESS_WRITE);
561 i_size_write(inode, 0);
563 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
569 ocfs2_commit_trans(osb, handle);
572 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
578 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
587 ocfs2_commit_trans(osb, handle);
592 static int ocfs2_remove_inode(struct inode *inode,
593 struct buffer_head *di_bh,
594 struct inode *orphan_dir_inode,
595 struct buffer_head *orphan_dir_bh)
598 struct inode *inode_alloc_inode = NULL;
599 struct buffer_head *inode_alloc_bh = NULL;
601 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
602 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
605 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
606 le16_to_cpu(di->i_suballoc_slot));
607 if (!inode_alloc_inode) {
613 mutex_lock(&inode_alloc_inode->i_mutex);
614 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
616 mutex_unlock(&inode_alloc_inode->i_mutex);
622 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
623 ocfs2_quota_trans_credits(inode->i_sb));
624 if (IS_ERR(handle)) {
625 status = PTR_ERR(handle);
630 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
637 /* set the inodes dtime */
638 status = ocfs2_journal_access_di(handle, inode, di_bh,
639 OCFS2_JOURNAL_ACCESS_WRITE);
645 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
646 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
648 status = ocfs2_journal_dirty(handle, di_bh);
654 ocfs2_remove_from_cache(inode, di_bh);
655 vfs_dq_free_inode(inode);
657 status = ocfs2_free_dinode(handle, inode_alloc_inode,
663 ocfs2_commit_trans(osb, handle);
665 ocfs2_inode_unlock(inode_alloc_inode, 1);
666 mutex_unlock(&inode_alloc_inode->i_mutex);
667 brelse(inode_alloc_bh);
669 iput(inode_alloc_inode);
675 * Serialize with orphan dir recovery. If the process doing
676 * recovery on this orphan dir does an iget() with the dir
677 * i_mutex held, we'll deadlock here. Instead we detect this
678 * and exit early - recovery will wipe this inode for us.
680 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
685 spin_lock(&osb->osb_lock);
686 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
687 mlog(0, "Recovery is happening on orphan dir %d, will skip "
688 "this inode\n", slot);
692 /* This signals to the orphan recovery process that it should
693 * wait for us to handle the wipe. */
694 osb->osb_orphan_wipes[slot]++;
696 spin_unlock(&osb->osb_lock);
700 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
703 spin_lock(&osb->osb_lock);
704 osb->osb_orphan_wipes[slot]--;
705 spin_unlock(&osb->osb_lock);
707 wake_up(&osb->osb_wipe_event);
710 static int ocfs2_wipe_inode(struct inode *inode,
711 struct buffer_head *di_bh)
713 int status, orphaned_slot;
714 struct inode *orphan_dir_inode = NULL;
715 struct buffer_head *orphan_dir_bh = NULL;
716 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
717 struct ocfs2_dinode *di;
719 di = (struct ocfs2_dinode *) di_bh->b_data;
720 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
722 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
726 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
727 ORPHAN_DIR_SYSTEM_INODE,
729 if (!orphan_dir_inode) {
735 /* Lock the orphan dir. The lock will be held for the entire
736 * delete_inode operation. We do this now to avoid races with
737 * recovery completion on other nodes. */
738 mutex_lock(&orphan_dir_inode->i_mutex);
739 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
741 mutex_unlock(&orphan_dir_inode->i_mutex);
747 /* we do this while holding the orphan dir lock because we
748 * don't want recovery being run from another node to try an
749 * inode delete underneath us -- this will result in two nodes
750 * truncating the same file! */
751 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
754 goto bail_unlock_dir;
757 /* Remove any dir index tree */
758 if (S_ISDIR(inode->i_mode)) {
759 status = ocfs2_dx_dir_truncate(inode, di_bh);
762 goto bail_unlock_dir;
766 /*Free extended attribute resources associated with this inode.*/
767 status = ocfs2_xattr_remove(inode, di_bh);
770 goto bail_unlock_dir;
773 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
779 ocfs2_inode_unlock(orphan_dir_inode, 1);
780 mutex_unlock(&orphan_dir_inode->i_mutex);
781 brelse(orphan_dir_bh);
783 iput(orphan_dir_inode);
784 ocfs2_signal_wipe_completion(osb, orphaned_slot);
789 /* There is a series of simple checks that should be done before a
790 * trylock is even considered. Encapsulate those in this function. */
791 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
794 struct ocfs2_inode_info *oi = OCFS2_I(inode);
795 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
797 /* We shouldn't be getting here for the root directory
799 if (inode == osb->root_inode) {
800 mlog(ML_ERROR, "Skipping delete of root inode.\n");
804 /* If we're coming from downconvert_thread we can't go into our own
805 * voting [hello, deadlock city!], so unforuntately we just
806 * have to skip deleting this guy. That's OK though because
807 * the node who's doing the actual deleting should handle it
809 if (current == osb->dc_task) {
810 mlog(0, "Skipping delete of %lu because we're currently "
811 "in downconvert\n", inode->i_ino);
815 spin_lock(&oi->ip_lock);
816 /* OCFS2 *never* deletes system files. This should technically
817 * never get here as system file inodes should always have a
818 * positive link count. */
819 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
820 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
821 (unsigned long long)oi->ip_blkno);
825 /* If we have allowd wipe of this inode for another node, it
826 * will be marked here so we can safely skip it. Recovery will
827 * cleanup any inodes we might inadvertantly skip here. */
828 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
829 mlog(0, "Skipping delete of %lu because another node "
830 "has done this for us.\n", inode->i_ino);
836 spin_unlock(&oi->ip_lock);
841 /* Query the cluster to determine whether we should wipe an inode from
844 * Requires the inode to have the cluster lock. */
845 static int ocfs2_query_inode_wipe(struct inode *inode,
846 struct buffer_head *di_bh,
850 struct ocfs2_inode_info *oi = OCFS2_I(inode);
851 struct ocfs2_dinode *di;
855 /* While we were waiting for the cluster lock in
856 * ocfs2_delete_inode, another node might have asked to delete
857 * the inode. Recheck our flags to catch this. */
858 if (!ocfs2_inode_is_valid_to_delete(inode)) {
859 mlog(0, "Skipping delete of %llu because flags changed\n",
860 (unsigned long long)oi->ip_blkno);
864 /* Now that we have an up to date inode, we can double check
866 if (inode->i_nlink) {
867 mlog(0, "Skipping delete of %llu because nlink = %u\n",
868 (unsigned long long)oi->ip_blkno, inode->i_nlink);
872 /* Do some basic inode verification... */
873 di = (struct ocfs2_dinode *) di_bh->b_data;
874 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
875 /* for lack of a better error? */
878 "Inode %llu (on-disk %llu) not orphaned! "
879 "Disk flags 0x%x, inode flags 0x%x\n",
880 (unsigned long long)oi->ip_blkno,
881 (unsigned long long)le64_to_cpu(di->i_blkno),
882 le32_to_cpu(di->i_flags), oi->ip_flags);
886 /* has someone already deleted us?! baaad... */
894 * This is how ocfs2 determines whether an inode is still live
895 * within the cluster. Every node takes a shared read lock on
896 * the inode open lock in ocfs2_read_locked_inode(). When we
897 * get to ->delete_inode(), each node tries to convert it's
898 * lock to an exclusive. Trylocks are serialized by the inode
899 * meta data lock. If the upconvert suceeds, we know the inode
900 * is no longer live and can be deleted.
902 * Though we call this with the meta data lock held, the
903 * trylock keeps us from ABBA deadlock.
905 status = ocfs2_try_open_lock(inode, 1);
906 if (status == -EAGAIN) {
908 mlog(0, "Skipping delete of %llu because it is in use on "
909 "other nodes\n", (unsigned long long)oi->ip_blkno);
918 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
919 (unsigned long long)oi->ip_blkno,
920 le16_to_cpu(di->i_orphaned_slot));
926 /* Support function for ocfs2_delete_inode. Will help us keep the
927 * inode data in a consistent state for clear_inode. Always truncates
928 * pages, optionally sync's them first. */
929 static void ocfs2_cleanup_delete_inode(struct inode *inode,
932 mlog(0, "Cleanup inode %llu, sync = %d\n",
933 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
935 write_inode_now(inode, 1);
936 truncate_inode_pages(&inode->i_data, 0);
939 void ocfs2_delete_inode(struct inode *inode)
942 sigset_t blocked, oldset;
943 struct buffer_head *di_bh = NULL;
945 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
947 /* When we fail in read_inode() we mark inode as bad. The second test
948 * catches the case when inode allocation fails before allocating
949 * a block for inode. */
950 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
951 mlog(0, "Skipping delete of bad inode\n");
955 if (!ocfs2_inode_is_valid_to_delete(inode)) {
956 /* It's probably not necessary to truncate_inode_pages
957 * here but we do it for safety anyway (it will most
958 * likely be a no-op anyway) */
959 ocfs2_cleanup_delete_inode(inode, 0);
963 /* We want to block signals in delete_inode as the lock and
964 * messaging paths may return us -ERESTARTSYS. Which would
965 * cause us to exit early, resulting in inodes being orphaned
967 sigfillset(&blocked);
968 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
971 ocfs2_cleanup_delete_inode(inode, 1);
976 * Synchronize us against ocfs2_get_dentry. We take this in
977 * shared mode so that all nodes can still concurrently
980 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
982 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
983 ocfs2_cleanup_delete_inode(inode, 0);
986 /* Lock down the inode. This gives us an up to date view of
987 * it's metadata (for verification), and allows us to
988 * serialize delete_inode on multiple nodes.
990 * Even though we might be doing a truncate, we don't take the
991 * allocation lock here as it won't be needed - nobody will
992 * have the file open.
994 status = ocfs2_inode_lock(inode, &di_bh, 1);
996 if (status != -ENOENT)
998 ocfs2_cleanup_delete_inode(inode, 0);
999 goto bail_unlock_nfs_sync;
1002 /* Query the cluster. This will be the final decision made
1003 * before we go ahead and wipe the inode. */
1004 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1005 if (!wipe || status < 0) {
1006 /* Error and remote inode busy both mean we won't be
1007 * removing the inode, so they take almost the same
1012 /* Someone in the cluster has disallowed a wipe of
1013 * this inode, or it was never completely
1014 * orphaned. Write out the pages and exit now. */
1015 ocfs2_cleanup_delete_inode(inode, 1);
1016 goto bail_unlock_inode;
1019 ocfs2_cleanup_delete_inode(inode, 0);
1021 status = ocfs2_wipe_inode(inode, di_bh);
1023 if (status != -EDEADLK)
1025 goto bail_unlock_inode;
1029 * Mark the inode as successfully deleted.
1031 * This is important for ocfs2_clear_inode() as it will check
1032 * this flag and skip any checkpointing work
1034 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1035 * the LVB for other nodes.
1037 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1040 ocfs2_inode_unlock(inode, 1);
1043 bail_unlock_nfs_sync:
1044 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1047 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1055 void ocfs2_clear_inode(struct inode *inode)
1058 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1065 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1066 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1068 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1069 "Inode=%lu\n", inode->i_ino);
1071 /* To preven remote deletes we hold open lock before, now it
1072 * is time to unlock PR and EX open locks. */
1073 ocfs2_open_unlock(inode);
1075 /* Do these before all the other work so that we don't bounce
1076 * the downconvert thread while waiting to destroy the locks. */
1077 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1078 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1079 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1081 /* We very well may get a clear_inode before all an inodes
1082 * metadata has hit disk. Of course, we can't drop any cluster
1083 * locks until the journal has finished with it. The only
1084 * exception here are successfully wiped inodes - their
1085 * metadata can now be considered to be part of the system
1086 * inodes from which it came. */
1087 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1088 ocfs2_checkpoint_inode(inode);
1090 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1091 "Clear inode of %llu, inode has io markers\n",
1092 (unsigned long long)oi->ip_blkno);
1094 ocfs2_extent_map_trunc(inode, 0);
1096 status = ocfs2_drop_inode_locks(inode);
1100 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1101 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1102 ocfs2_lock_res_free(&oi->ip_open_lockres);
1104 ocfs2_metadata_cache_purge(inode);
1106 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1107 "Clear inode of %llu, inode has %u cache items\n",
1108 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1110 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1111 "Clear inode of %llu, inode has a bad flag\n",
1112 (unsigned long long)oi->ip_blkno);
1114 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1115 "Clear inode of %llu, inode is locked\n",
1116 (unsigned long long)oi->ip_blkno);
1118 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1119 "Clear inode of %llu, io_mutex is locked\n",
1120 (unsigned long long)oi->ip_blkno);
1121 mutex_unlock(&oi->ip_io_mutex);
1124 * down_trylock() returns 0, down_write_trylock() returns 1
1127 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1128 "Clear inode of %llu, alloc_sem is locked\n",
1129 (unsigned long long)oi->ip_blkno);
1130 up_write(&oi->ip_alloc_sem);
1132 mlog_bug_on_msg(oi->ip_open_count,
1133 "Clear inode of %llu has open count %d\n",
1134 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1136 /* Clear all other flags. */
1137 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1138 oi->ip_created_trans = 0;
1139 oi->ip_last_trans = 0;
1140 oi->ip_dir_start_lookup = 0;
1141 oi->ip_blkno = 0ULL;
1144 * ip_jinode is used to track txns against this inode. We ensure that
1145 * the journal is flushed before journal shutdown. Thus it is safe to
1146 * have inodes get cleaned up after journal shutdown.
1148 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1155 /* Called under inode_lock, with no more references on the
1156 * struct inode, so it's safe here to check the flags field
1157 * and to manipulate i_nlink without any other locks. */
1158 void ocfs2_drop_inode(struct inode *inode)
1160 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1164 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1165 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1167 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1168 generic_delete_inode(inode);
1170 generic_drop_inode(inode);
1176 * This is called from our getattr.
1178 int ocfs2_inode_revalidate(struct dentry *dentry)
1180 struct inode *inode = dentry->d_inode;
1183 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1184 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1187 mlog(0, "eep, no inode!\n");
1192 spin_lock(&OCFS2_I(inode)->ip_lock);
1193 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1194 spin_unlock(&OCFS2_I(inode)->ip_lock);
1195 mlog(0, "inode deleted!\n");
1199 spin_unlock(&OCFS2_I(inode)->ip_lock);
1201 /* Let ocfs2_inode_lock do the work of updating our struct
1203 status = ocfs2_inode_lock(inode, NULL, 0);
1205 if (status != -ENOENT)
1209 ocfs2_inode_unlock(inode, 0);
1217 * Updates a disk inode from a
1219 * Only takes ip_lock.
1221 int ocfs2_mark_inode_dirty(handle_t *handle,
1222 struct inode *inode,
1223 struct buffer_head *bh)
1226 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1228 mlog_entry("(inode %llu)\n",
1229 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1231 status = ocfs2_journal_access_di(handle, inode, bh,
1232 OCFS2_JOURNAL_ACCESS_WRITE);
1238 spin_lock(&OCFS2_I(inode)->ip_lock);
1239 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1240 ocfs2_get_inode_flags(OCFS2_I(inode));
1241 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1242 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1243 spin_unlock(&OCFS2_I(inode)->ip_lock);
1245 fe->i_size = cpu_to_le64(i_size_read(inode));
1246 ocfs2_set_links_count(fe, inode->i_nlink);
1247 fe->i_uid = cpu_to_le32(inode->i_uid);
1248 fe->i_gid = cpu_to_le32(inode->i_gid);
1249 fe->i_mode = cpu_to_le16(inode->i_mode);
1250 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1251 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1252 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1253 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1254 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1255 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1257 status = ocfs2_journal_dirty(handle, bh);
1270 * Updates a struct inode from a disk inode.
1271 * does no i/o, only takes ip_lock.
1273 void ocfs2_refresh_inode(struct inode *inode,
1274 struct ocfs2_dinode *fe)
1276 spin_lock(&OCFS2_I(inode)->ip_lock);
1278 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1279 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1280 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1281 ocfs2_set_inode_flags(inode);
1282 i_size_write(inode, le64_to_cpu(fe->i_size));
1283 inode->i_nlink = ocfs2_read_links_count(fe);
1284 inode->i_uid = le32_to_cpu(fe->i_uid);
1285 inode->i_gid = le32_to_cpu(fe->i_gid);
1286 inode->i_mode = le16_to_cpu(fe->i_mode);
1287 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1288 inode->i_blocks = 0;
1290 inode->i_blocks = ocfs2_inode_sector_count(inode);
1291 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1292 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1293 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1294 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1295 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1296 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1298 spin_unlock(&OCFS2_I(inode)->ip_lock);
1301 int ocfs2_validate_inode_block(struct super_block *sb,
1302 struct buffer_head *bh)
1305 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1307 mlog(0, "Validating dinode %llu\n",
1308 (unsigned long long)bh->b_blocknr);
1310 BUG_ON(!buffer_uptodate(bh));
1313 * If the ecc fails, we return the error but otherwise
1314 * leave the filesystem running. We know any error is
1315 * local to this block.
1317 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1319 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1320 (unsigned long long)bh->b_blocknr);
1325 * Errors after here are fatal.
1330 if (!OCFS2_IS_VALID_DINODE(di)) {
1331 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1332 (unsigned long long)bh->b_blocknr, 7,
1337 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1338 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1339 (unsigned long long)bh->b_blocknr,
1340 (unsigned long long)le64_to_cpu(di->i_blkno));
1344 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1346 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1347 (unsigned long long)bh->b_blocknr);
1351 if (le32_to_cpu(di->i_fs_generation) !=
1352 OCFS2_SB(sb)->fs_generation) {
1354 "Invalid dinode #%llu: fs_generation is %u\n",
1355 (unsigned long long)bh->b_blocknr,
1356 le32_to_cpu(di->i_fs_generation));
1366 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1370 struct buffer_head *tmp = *bh;
1372 rc = ocfs2_read_blocks(inode, OCFS2_I(inode)->ip_blkno, 1, &tmp,
1373 flags, ocfs2_validate_inode_block);
1375 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1382 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1384 return ocfs2_read_inode_block_full(inode, bh, 0);