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 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
220 struct super_block *sb;
221 struct ocfs2_super *osb;
222 int status = -EINVAL;
225 mlog_entry("(0x%p, size:%llu)\n", inode,
226 (unsigned long long)le64_to_cpu(fe->i_size));
231 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
232 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
235 /* this means that read_inode cannot create a superblock inode
236 * today. change if needed. */
237 if (!OCFS2_IS_VALID_DINODE(fe) ||
238 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
239 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
240 "signature = %.*s, flags = 0x%x\n",
242 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
243 fe->i_signature, le32_to_cpu(fe->i_flags));
247 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
248 mlog(ML_ERROR, "file entry generation does not match "
249 "superblock! osb->fs_generation=%x, "
250 "fe->i_fs_generation=%x\n",
251 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
255 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
256 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
257 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
259 inode->i_version = 1;
260 inode->i_generation = le32_to_cpu(fe->i_generation);
261 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
262 inode->i_mode = le16_to_cpu(fe->i_mode);
263 inode->i_uid = le32_to_cpu(fe->i_uid);
264 inode->i_gid = le32_to_cpu(fe->i_gid);
266 /* Fast symlinks will have i_size but no allocated clusters. */
267 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
270 inode->i_blocks = ocfs2_inode_sector_count(inode);
271 inode->i_mapping->a_ops = &ocfs2_aops;
272 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
273 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
274 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
275 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
276 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
277 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
279 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
281 "ip_blkno %llu != i_blkno %llu!\n",
282 (unsigned long long)OCFS2_I(inode)->ip_blkno,
283 (unsigned long long)le64_to_cpu(fe->i_blkno));
285 inode->i_nlink = le16_to_cpu(fe->i_links_count);
287 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
288 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
290 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
291 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
292 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
293 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
294 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
295 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
296 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
297 /* we can't actually hit this as read_inode can't
298 * handle superblocks today ;-) */
302 switch (inode->i_mode & S_IFMT) {
305 inode->i_fop = &ocfs2_fops;
307 inode->i_fop = &ocfs2_fops_no_plocks;
308 inode->i_op = &ocfs2_file_iops;
309 i_size_write(inode, le64_to_cpu(fe->i_size));
312 inode->i_op = &ocfs2_dir_iops;
314 inode->i_fop = &ocfs2_dops;
316 inode->i_fop = &ocfs2_dops_no_plocks;
317 i_size_write(inode, le64_to_cpu(fe->i_size));
320 if (ocfs2_inode_is_fast_symlink(inode))
321 inode->i_op = &ocfs2_fast_symlink_inode_operations;
323 inode->i_op = &ocfs2_symlink_inode_operations;
324 i_size_write(inode, le64_to_cpu(fe->i_size));
327 inode->i_op = &ocfs2_special_file_iops;
328 init_special_inode(inode, inode->i_mode,
334 inode->i_ino = ino_from_blkno(inode->i_sb,
335 le64_to_cpu(fe->i_blkno));
338 * If we ever want to create system files from kernel,
339 * the generation argument to
340 * ocfs2_inode_lock_res_init() will have to change.
342 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
344 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
345 OCFS2_LOCK_TYPE_META, 0, inode);
347 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
348 OCFS2_LOCK_TYPE_OPEN, 0, inode);
351 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
352 OCFS2_LOCK_TYPE_RW, inode->i_generation,
355 ocfs2_set_inode_flags(inode);
363 static int ocfs2_read_locked_inode(struct inode *inode,
364 struct ocfs2_find_inode_args *args)
366 struct super_block *sb;
367 struct ocfs2_super *osb;
368 struct ocfs2_dinode *fe;
369 struct buffer_head *bh = NULL;
370 int status, can_lock;
373 mlog_entry("(0x%p, 0x%p)\n", inode, args);
376 if (inode == NULL || inode->i_sb == NULL) {
377 mlog(ML_ERROR, "bad inode\n");
384 mlog(ML_ERROR, "bad inode args\n");
385 make_bad_inode(inode);
390 * To improve performance of cold-cache inode stats, we take
391 * the cluster lock here if possible.
393 * Generally, OCFS2 never trusts the contents of an inode
394 * unless it's holding a cluster lock, so taking it here isn't
395 * a correctness issue as much as it is a performance
398 * There are three times when taking the lock is not a good idea:
400 * 1) During startup, before we have initialized the DLM.
402 * 2) If we are reading certain system files which never get
403 * cluster locks (local alloc, truncate log).
405 * 3) If the process doing the iget() is responsible for
406 * orphan dir recovery. We're holding the orphan dir lock and
407 * can get into a deadlock with another process on another
408 * node in ->delete_inode().
410 * #1 and #2 can be simply solved by never taking the lock
411 * here for system files (which are the only type we read
412 * during mount). It's a heavier approach, but our main
413 * concern is user-accesible files anyway.
415 * #3 works itself out because we'll eventually take the
416 * cluster lock before trusting anything anyway.
418 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
419 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
420 && !ocfs2_mount_local(osb);
423 * To maintain backwards compatibility with older versions of
424 * ocfs2-tools, we still store the generation value for system
425 * files. The only ones that actually matter to userspace are
426 * the journals, but it's easier and inexpensive to just flag
427 * all system files similarly.
429 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
430 generation = osb->fs_generation;
432 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
433 OCFS2_LOCK_TYPE_META,
436 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
437 OCFS2_LOCK_TYPE_OPEN,
441 status = ocfs2_open_lock(inode);
443 make_bad_inode(inode);
447 status = ocfs2_inode_lock(inode, NULL, 0);
449 make_bad_inode(inode);
455 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
456 status = ocfs2_try_open_lock(inode, 0);
458 make_bad_inode(inode);
463 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
464 can_lock ? inode : NULL);
471 fe = (struct ocfs2_dinode *) bh->b_data;
472 if (!OCFS2_IS_VALID_DINODE(fe)) {
473 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
474 (unsigned long long)args->fi_blkno, 7,
480 * This is a code bug. Right now the caller needs to
481 * understand whether it is asking for a system file inode or
482 * not so the proper lock names can be built.
484 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
485 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
486 "Inode %llu: system file state is ambigous\n",
487 (unsigned long long)args->fi_blkno);
489 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
490 S_ISBLK(le16_to_cpu(fe->i_mode)))
491 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
493 if (ocfs2_populate_inode(inode, fe, 0) < 0)
496 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
502 ocfs2_inode_unlock(inode, 0);
505 make_bad_inode(inode);
514 void ocfs2_sync_blockdev(struct super_block *sb)
516 sync_blockdev(sb->s_bdev);
519 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
521 struct buffer_head *fe_bh)
524 struct ocfs2_truncate_context *tc = NULL;
525 struct ocfs2_dinode *fe;
526 handle_t *handle = NULL;
530 fe = (struct ocfs2_dinode *) fe_bh->b_data;
533 * This check will also skip truncate of inodes with inline
534 * data and fast symlinks.
536 if (fe->i_clusters) {
537 if (ocfs2_should_order_data(inode))
538 ocfs2_begin_ordered_truncate(inode, 0);
540 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
541 if (IS_ERR(handle)) {
542 status = PTR_ERR(handle);
547 status = ocfs2_journal_access(handle, inode, fe_bh,
548 OCFS2_JOURNAL_ACCESS_WRITE);
554 i_size_write(inode, 0);
556 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
562 ocfs2_commit_trans(osb, handle);
565 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
571 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
580 ocfs2_commit_trans(osb, handle);
585 static int ocfs2_remove_inode(struct inode *inode,
586 struct buffer_head *di_bh,
587 struct inode *orphan_dir_inode,
588 struct buffer_head *orphan_dir_bh)
591 struct inode *inode_alloc_inode = NULL;
592 struct buffer_head *inode_alloc_bh = NULL;
594 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
595 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
598 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
599 le16_to_cpu(di->i_suballoc_slot));
600 if (!inode_alloc_inode) {
606 mutex_lock(&inode_alloc_inode->i_mutex);
607 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
609 mutex_unlock(&inode_alloc_inode->i_mutex);
615 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
616 if (IS_ERR(handle)) {
617 status = PTR_ERR(handle);
622 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
629 /* set the inodes dtime */
630 status = ocfs2_journal_access(handle, inode, di_bh,
631 OCFS2_JOURNAL_ACCESS_WRITE);
637 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
638 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
640 status = ocfs2_journal_dirty(handle, di_bh);
646 ocfs2_remove_from_cache(inode, di_bh);
648 status = ocfs2_free_dinode(handle, inode_alloc_inode,
654 ocfs2_commit_trans(osb, handle);
656 ocfs2_inode_unlock(inode_alloc_inode, 1);
657 mutex_unlock(&inode_alloc_inode->i_mutex);
658 brelse(inode_alloc_bh);
660 iput(inode_alloc_inode);
666 * Serialize with orphan dir recovery. If the process doing
667 * recovery on this orphan dir does an iget() with the dir
668 * i_mutex held, we'll deadlock here. Instead we detect this
669 * and exit early - recovery will wipe this inode for us.
671 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
676 spin_lock(&osb->osb_lock);
677 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
678 mlog(0, "Recovery is happening on orphan dir %d, will skip "
679 "this inode\n", slot);
683 /* This signals to the orphan recovery process that it should
684 * wait for us to handle the wipe. */
685 osb->osb_orphan_wipes[slot]++;
687 spin_unlock(&osb->osb_lock);
691 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
694 spin_lock(&osb->osb_lock);
695 osb->osb_orphan_wipes[slot]--;
696 spin_unlock(&osb->osb_lock);
698 wake_up(&osb->osb_wipe_event);
701 static int ocfs2_wipe_inode(struct inode *inode,
702 struct buffer_head *di_bh)
704 int status, orphaned_slot;
705 struct inode *orphan_dir_inode = NULL;
706 struct buffer_head *orphan_dir_bh = NULL;
707 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
708 struct ocfs2_dinode *di;
710 di = (struct ocfs2_dinode *) di_bh->b_data;
711 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
713 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
717 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
718 ORPHAN_DIR_SYSTEM_INODE,
720 if (!orphan_dir_inode) {
726 /* Lock the orphan dir. The lock will be held for the entire
727 * delete_inode operation. We do this now to avoid races with
728 * recovery completion on other nodes. */
729 mutex_lock(&orphan_dir_inode->i_mutex);
730 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
732 mutex_unlock(&orphan_dir_inode->i_mutex);
738 /* we do this while holding the orphan dir lock because we
739 * don't want recovery being run from another node to try an
740 * inode delete underneath us -- this will result in two nodes
741 * truncating the same file! */
742 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
745 goto bail_unlock_dir;
748 /*Free extended attribute resources associated with this inode.*/
749 status = ocfs2_xattr_remove(inode, di_bh);
752 goto bail_unlock_dir;
755 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
761 ocfs2_inode_unlock(orphan_dir_inode, 1);
762 mutex_unlock(&orphan_dir_inode->i_mutex);
763 brelse(orphan_dir_bh);
765 iput(orphan_dir_inode);
766 ocfs2_signal_wipe_completion(osb, orphaned_slot);
771 /* There is a series of simple checks that should be done before a
772 * trylock is even considered. Encapsulate those in this function. */
773 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
776 struct ocfs2_inode_info *oi = OCFS2_I(inode);
777 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
779 /* We shouldn't be getting here for the root directory
781 if (inode == osb->root_inode) {
782 mlog(ML_ERROR, "Skipping delete of root inode.\n");
786 /* If we're coming from downconvert_thread we can't go into our own
787 * voting [hello, deadlock city!], so unforuntately we just
788 * have to skip deleting this guy. That's OK though because
789 * the node who's doing the actual deleting should handle it
791 if (current == osb->dc_task) {
792 mlog(0, "Skipping delete of %lu because we're currently "
793 "in downconvert\n", inode->i_ino);
797 spin_lock(&oi->ip_lock);
798 /* OCFS2 *never* deletes system files. This should technically
799 * never get here as system file inodes should always have a
800 * positive link count. */
801 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
802 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
803 (unsigned long long)oi->ip_blkno);
807 /* If we have allowd wipe of this inode for another node, it
808 * will be marked here so we can safely skip it. Recovery will
809 * cleanup any inodes we might inadvertantly skip here. */
810 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
811 mlog(0, "Skipping delete of %lu because another node "
812 "has done this for us.\n", inode->i_ino);
818 spin_unlock(&oi->ip_lock);
823 /* Query the cluster to determine whether we should wipe an inode from
826 * Requires the inode to have the cluster lock. */
827 static int ocfs2_query_inode_wipe(struct inode *inode,
828 struct buffer_head *di_bh,
832 struct ocfs2_inode_info *oi = OCFS2_I(inode);
833 struct ocfs2_dinode *di;
837 /* While we were waiting for the cluster lock in
838 * ocfs2_delete_inode, another node might have asked to delete
839 * the inode. Recheck our flags to catch this. */
840 if (!ocfs2_inode_is_valid_to_delete(inode)) {
841 mlog(0, "Skipping delete of %llu because flags changed\n",
842 (unsigned long long)oi->ip_blkno);
846 /* Now that we have an up to date inode, we can double check
848 if (inode->i_nlink) {
849 mlog(0, "Skipping delete of %llu because nlink = %u\n",
850 (unsigned long long)oi->ip_blkno, inode->i_nlink);
854 /* Do some basic inode verification... */
855 di = (struct ocfs2_dinode *) di_bh->b_data;
856 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
857 /* for lack of a better error? */
860 "Inode %llu (on-disk %llu) not orphaned! "
861 "Disk flags 0x%x, inode flags 0x%x\n",
862 (unsigned long long)oi->ip_blkno,
863 (unsigned long long)le64_to_cpu(di->i_blkno),
864 le32_to_cpu(di->i_flags), oi->ip_flags);
868 /* has someone already deleted us?! baaad... */
876 * This is how ocfs2 determines whether an inode is still live
877 * within the cluster. Every node takes a shared read lock on
878 * the inode open lock in ocfs2_read_locked_inode(). When we
879 * get to ->delete_inode(), each node tries to convert it's
880 * lock to an exclusive. Trylocks are serialized by the inode
881 * meta data lock. If the upconvert suceeds, we know the inode
882 * is no longer live and can be deleted.
884 * Though we call this with the meta data lock held, the
885 * trylock keeps us from ABBA deadlock.
887 status = ocfs2_try_open_lock(inode, 1);
888 if (status == -EAGAIN) {
890 mlog(0, "Skipping delete of %llu because it is in use on "
891 "other nodes\n", (unsigned long long)oi->ip_blkno);
900 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
901 (unsigned long long)oi->ip_blkno,
902 le16_to_cpu(di->i_orphaned_slot));
908 /* Support function for ocfs2_delete_inode. Will help us keep the
909 * inode data in a consistent state for clear_inode. Always truncates
910 * pages, optionally sync's them first. */
911 static void ocfs2_cleanup_delete_inode(struct inode *inode,
914 mlog(0, "Cleanup inode %llu, sync = %d\n",
915 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
917 write_inode_now(inode, 1);
918 truncate_inode_pages(&inode->i_data, 0);
921 void ocfs2_delete_inode(struct inode *inode)
924 sigset_t blocked, oldset;
925 struct buffer_head *di_bh = NULL;
927 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
929 if (is_bad_inode(inode)) {
930 mlog(0, "Skipping delete of bad inode\n");
934 if (!ocfs2_inode_is_valid_to_delete(inode)) {
935 /* It's probably not necessary to truncate_inode_pages
936 * here but we do it for safety anyway (it will most
937 * likely be a no-op anyway) */
938 ocfs2_cleanup_delete_inode(inode, 0);
942 /* We want to block signals in delete_inode as the lock and
943 * messaging paths may return us -ERESTARTSYS. Which would
944 * cause us to exit early, resulting in inodes being orphaned
946 sigfillset(&blocked);
947 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
950 ocfs2_cleanup_delete_inode(inode, 1);
954 /* Lock down the inode. This gives us an up to date view of
955 * it's metadata (for verification), and allows us to
956 * serialize delete_inode on multiple nodes.
958 * Even though we might be doing a truncate, we don't take the
959 * allocation lock here as it won't be needed - nobody will
960 * have the file open.
962 status = ocfs2_inode_lock(inode, &di_bh, 1);
964 if (status != -ENOENT)
966 ocfs2_cleanup_delete_inode(inode, 0);
970 /* Query the cluster. This will be the final decision made
971 * before we go ahead and wipe the inode. */
972 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
973 if (!wipe || status < 0) {
974 /* Error and remote inode busy both mean we won't be
975 * removing the inode, so they take almost the same
980 /* Someone in the cluster has disallowed a wipe of
981 * this inode, or it was never completely
982 * orphaned. Write out the pages and exit now. */
983 ocfs2_cleanup_delete_inode(inode, 1);
984 goto bail_unlock_inode;
987 ocfs2_cleanup_delete_inode(inode, 0);
989 status = ocfs2_wipe_inode(inode, di_bh);
991 if (status != -EDEADLK)
993 goto bail_unlock_inode;
997 * Mark the inode as successfully deleted.
999 * This is important for ocfs2_clear_inode() as it will check
1000 * this flag and skip any checkpointing work
1002 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1003 * the LVB for other nodes.
1005 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1008 ocfs2_inode_unlock(inode, 1);
1011 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1019 void ocfs2_clear_inode(struct inode *inode)
1022 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1029 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1030 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1032 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1033 "Inode=%lu\n", inode->i_ino);
1035 /* To preven remote deletes we hold open lock before, now it
1036 * is time to unlock PR and EX open locks. */
1037 ocfs2_open_unlock(inode);
1039 /* Do these before all the other work so that we don't bounce
1040 * the downconvert thread while waiting to destroy the locks. */
1041 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1042 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1043 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1045 /* We very well may get a clear_inode before all an inodes
1046 * metadata has hit disk. Of course, we can't drop any cluster
1047 * locks until the journal has finished with it. The only
1048 * exception here are successfully wiped inodes - their
1049 * metadata can now be considered to be part of the system
1050 * inodes from which it came. */
1051 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1052 ocfs2_checkpoint_inode(inode);
1054 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1055 "Clear inode of %llu, inode has io markers\n",
1056 (unsigned long long)oi->ip_blkno);
1058 ocfs2_extent_map_trunc(inode, 0);
1060 status = ocfs2_drop_inode_locks(inode);
1064 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1065 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1066 ocfs2_lock_res_free(&oi->ip_open_lockres);
1068 ocfs2_metadata_cache_purge(inode);
1070 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1071 "Clear inode of %llu, inode has %u cache items\n",
1072 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1074 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1075 "Clear inode of %llu, inode has a bad flag\n",
1076 (unsigned long long)oi->ip_blkno);
1078 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1079 "Clear inode of %llu, inode is locked\n",
1080 (unsigned long long)oi->ip_blkno);
1082 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1083 "Clear inode of %llu, io_mutex is locked\n",
1084 (unsigned long long)oi->ip_blkno);
1085 mutex_unlock(&oi->ip_io_mutex);
1088 * down_trylock() returns 0, down_write_trylock() returns 1
1091 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1092 "Clear inode of %llu, alloc_sem is locked\n",
1093 (unsigned long long)oi->ip_blkno);
1094 up_write(&oi->ip_alloc_sem);
1096 mlog_bug_on_msg(oi->ip_open_count,
1097 "Clear inode of %llu has open count %d\n",
1098 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1100 /* Clear all other flags. */
1101 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1102 oi->ip_created_trans = 0;
1103 oi->ip_last_trans = 0;
1104 oi->ip_dir_start_lookup = 0;
1105 oi->ip_blkno = 0ULL;
1106 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1113 /* Called under inode_lock, with no more references on the
1114 * struct inode, so it's safe here to check the flags field
1115 * and to manipulate i_nlink without any other locks. */
1116 void ocfs2_drop_inode(struct inode *inode)
1118 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1122 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1123 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1125 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1126 generic_delete_inode(inode);
1128 generic_drop_inode(inode);
1134 * TODO: this should probably be merged into ocfs2_get_block
1136 * However, you now need to pay attention to the cont_prepare_write()
1137 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1138 * expects never to extend).
1140 struct buffer_head *ocfs2_bread(struct inode *inode,
1141 int block, int *err, int reada)
1143 struct buffer_head *bh = NULL;
1146 int readflags = OCFS2_BH_CACHED;
1149 readflags |= OCFS2_BH_READAHEAD;
1151 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1152 i_size_read(inode)) {
1157 down_read(&OCFS2_I(inode)->ip_alloc_sem);
1158 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1160 up_read(&OCFS2_I(inode)->ip_alloc_sem);
1166 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1186 * This is called from our getattr.
1188 int ocfs2_inode_revalidate(struct dentry *dentry)
1190 struct inode *inode = dentry->d_inode;
1193 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1194 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1197 mlog(0, "eep, no inode!\n");
1202 spin_lock(&OCFS2_I(inode)->ip_lock);
1203 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1204 spin_unlock(&OCFS2_I(inode)->ip_lock);
1205 mlog(0, "inode deleted!\n");
1209 spin_unlock(&OCFS2_I(inode)->ip_lock);
1211 /* Let ocfs2_inode_lock do the work of updating our struct
1213 status = ocfs2_inode_lock(inode, NULL, 0);
1215 if (status != -ENOENT)
1219 ocfs2_inode_unlock(inode, 0);
1227 * Updates a disk inode from a
1229 * Only takes ip_lock.
1231 int ocfs2_mark_inode_dirty(handle_t *handle,
1232 struct inode *inode,
1233 struct buffer_head *bh)
1236 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1238 mlog_entry("(inode %llu)\n",
1239 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1241 status = ocfs2_journal_access(handle, inode, bh,
1242 OCFS2_JOURNAL_ACCESS_WRITE);
1248 spin_lock(&OCFS2_I(inode)->ip_lock);
1249 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1250 ocfs2_get_inode_flags(OCFS2_I(inode));
1251 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1252 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1253 spin_unlock(&OCFS2_I(inode)->ip_lock);
1255 fe->i_size = cpu_to_le64(i_size_read(inode));
1256 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1257 fe->i_uid = cpu_to_le32(inode->i_uid);
1258 fe->i_gid = cpu_to_le32(inode->i_gid);
1259 fe->i_mode = cpu_to_le16(inode->i_mode);
1260 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1261 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1262 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1263 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1264 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1265 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1267 status = ocfs2_journal_dirty(handle, bh);
1280 * Updates a struct inode from a disk inode.
1281 * does no i/o, only takes ip_lock.
1283 void ocfs2_refresh_inode(struct inode *inode,
1284 struct ocfs2_dinode *fe)
1286 spin_lock(&OCFS2_I(inode)->ip_lock);
1288 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1289 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1290 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1291 ocfs2_set_inode_flags(inode);
1292 i_size_write(inode, le64_to_cpu(fe->i_size));
1293 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1294 inode->i_uid = le32_to_cpu(fe->i_uid);
1295 inode->i_gid = le32_to_cpu(fe->i_gid);
1296 inode->i_mode = le16_to_cpu(fe->i_mode);
1297 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1298 inode->i_blocks = 0;
1300 inode->i_blocks = ocfs2_inode_sector_count(inode);
1301 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1302 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1303 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1304 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1305 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1306 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1308 spin_unlock(&OCFS2_I(inode)->ip_lock);