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;
63 static int ocfs2_read_locked_inode(struct inode *inode,
64 struct ocfs2_find_inode_args *args);
65 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
66 static int ocfs2_find_actor(struct inode *inode, void *opaque);
67 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
69 struct buffer_head *fe_bh);
71 void ocfs2_set_inode_flags(struct inode *inode)
73 unsigned int flags = OCFS2_I(inode)->ip_attr;
75 inode->i_flags &= ~(S_IMMUTABLE |
76 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
78 if (flags & OCFS2_IMMUTABLE_FL)
79 inode->i_flags |= S_IMMUTABLE;
81 if (flags & OCFS2_SYNC_FL)
82 inode->i_flags |= S_SYNC;
83 if (flags & OCFS2_APPEND_FL)
84 inode->i_flags |= S_APPEND;
85 if (flags & OCFS2_NOATIME_FL)
86 inode->i_flags |= S_NOATIME;
87 if (flags & OCFS2_DIRSYNC_FL)
88 inode->i_flags |= S_DIRSYNC;
91 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
92 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
94 unsigned int flags = oi->vfs_inode.i_flags;
96 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
97 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
99 oi->ip_attr |= OCFS2_SYNC_FL;
100 if (flags & S_APPEND)
101 oi->ip_attr |= OCFS2_APPEND_FL;
102 if (flags & S_IMMUTABLE)
103 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
104 if (flags & S_NOATIME)
105 oi->ip_attr |= OCFS2_NOATIME_FL;
106 if (flags & S_DIRSYNC)
107 oi->ip_attr |= OCFS2_DIRSYNC_FL;
110 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
112 struct inode *inode = NULL;
113 struct super_block *sb = osb->sb;
114 struct ocfs2_find_inode_args args;
116 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
118 /* Ok. By now we've either got the offsets passed to us by the
119 * caller, or we just pulled them off the bh. Lets do some
120 * sanity checks to make sure they're OK. */
122 inode = ERR_PTR(-EINVAL);
123 mlog_errno(PTR_ERR(inode));
127 args.fi_blkno = blkno;
128 args.fi_flags = flags;
129 args.fi_ino = ino_from_blkno(sb, blkno);
131 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
132 ocfs2_init_locked_inode, &args);
133 /* inode was *not* in the inode cache. 2.6.x requires
134 * us to do our own read_inode call and unlock it
136 if (inode && inode->i_state & I_NEW) {
137 mlog(0, "Inode was not in inode cache, reading it.\n");
138 ocfs2_read_locked_inode(inode, &args);
139 unlock_new_inode(inode);
142 inode = ERR_PTR(-ENOMEM);
143 mlog_errno(PTR_ERR(inode));
146 if (is_bad_inode(inode)) {
148 inode = ERR_PTR(-ESTALE);
153 if (!IS_ERR(inode)) {
154 mlog(0, "returning inode with number %llu\n",
155 (unsigned long long)OCFS2_I(inode)->ip_blkno);
156 mlog_exit_ptr(inode);
164 * here's how inodes get read from disk:
165 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
166 * found? : return the in-memory inode
167 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
170 static int ocfs2_find_actor(struct inode *inode, void *opaque)
172 struct ocfs2_find_inode_args *args = NULL;
173 struct ocfs2_inode_info *oi = OCFS2_I(inode);
176 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
180 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
182 if (oi->ip_blkno != args->fi_blkno)
192 * initialize the new inode, but don't do anything that would cause
194 * return 0 on success, 1 on failure
196 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
198 struct ocfs2_find_inode_args *args = opaque;
200 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
202 inode->i_ino = args->fi_ino;
203 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
209 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
212 struct super_block *sb;
213 struct ocfs2_super *osb;
214 int status = -EINVAL;
216 mlog_entry("(0x%p, size:%llu)\n", inode,
217 (unsigned long long)le64_to_cpu(fe->i_size));
222 /* this means that read_inode cannot create a superblock inode
223 * today. change if needed. */
224 if (!OCFS2_IS_VALID_DINODE(fe) ||
225 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
226 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
227 "signature = %.*s, flags = 0x%x\n",
229 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
230 fe->i_signature, le32_to_cpu(fe->i_flags));
234 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
235 mlog(ML_ERROR, "file entry generation does not match "
236 "superblock! osb->fs_generation=%x, "
237 "fe->i_fs_generation=%x\n",
238 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
242 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
243 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
244 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
246 inode->i_version = 1;
247 inode->i_generation = le32_to_cpu(fe->i_generation);
248 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
249 inode->i_mode = le16_to_cpu(fe->i_mode);
250 inode->i_uid = le32_to_cpu(fe->i_uid);
251 inode->i_gid = le32_to_cpu(fe->i_gid);
253 /* Fast symlinks will have i_size but no allocated clusters. */
254 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
257 inode->i_blocks = ocfs2_inode_sector_count(inode);
258 inode->i_mapping->a_ops = &ocfs2_aops;
259 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
260 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
261 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
262 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
263 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
264 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
266 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
268 "ip_blkno %llu != i_blkno %llu!\n",
269 (unsigned long long)OCFS2_I(inode)->ip_blkno,
270 (unsigned long long)le64_to_cpu(fe->i_blkno));
272 inode->i_nlink = le16_to_cpu(fe->i_links_count);
274 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
275 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
277 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
278 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
279 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
280 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
281 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
282 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
283 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
284 /* we can't actually hit this as read_inode can't
285 * handle superblocks today ;-) */
289 switch (inode->i_mode & S_IFMT) {
291 inode->i_fop = &ocfs2_fops;
292 inode->i_op = &ocfs2_file_iops;
293 i_size_write(inode, le64_to_cpu(fe->i_size));
296 inode->i_op = &ocfs2_dir_iops;
297 inode->i_fop = &ocfs2_dops;
298 i_size_write(inode, le64_to_cpu(fe->i_size));
301 if (ocfs2_inode_is_fast_symlink(inode))
302 inode->i_op = &ocfs2_fast_symlink_inode_operations;
304 inode->i_op = &ocfs2_symlink_inode_operations;
305 i_size_write(inode, le64_to_cpu(fe->i_size));
308 inode->i_op = &ocfs2_special_file_iops;
309 init_special_inode(inode, inode->i_mode,
315 inode->i_ino = ino_from_blkno(inode->i_sb,
316 le64_to_cpu(fe->i_blkno));
319 * If we ever want to create system files from kernel,
320 * the generation argument to
321 * ocfs2_inode_lock_res_init() will have to change.
323 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
325 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
326 OCFS2_LOCK_TYPE_META, 0, inode);
328 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
329 OCFS2_LOCK_TYPE_OPEN, 0, inode);
332 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
333 OCFS2_LOCK_TYPE_RW, inode->i_generation,
336 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
337 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
340 ocfs2_set_inode_flags(inode);
348 static int ocfs2_read_locked_inode(struct inode *inode,
349 struct ocfs2_find_inode_args *args)
351 struct super_block *sb;
352 struct ocfs2_super *osb;
353 struct ocfs2_dinode *fe;
354 struct buffer_head *bh = NULL;
355 int status, can_lock;
358 mlog_entry("(0x%p, 0x%p)\n", inode, args);
361 if (inode == NULL || inode->i_sb == NULL) {
362 mlog(ML_ERROR, "bad inode\n");
369 mlog(ML_ERROR, "bad inode args\n");
370 make_bad_inode(inode);
375 * To improve performance of cold-cache inode stats, we take
376 * the cluster lock here if possible.
378 * Generally, OCFS2 never trusts the contents of an inode
379 * unless it's holding a cluster lock, so taking it here isn't
380 * a correctness issue as much as it is a performance
383 * There are three times when taking the lock is not a good idea:
385 * 1) During startup, before we have initialized the DLM.
387 * 2) If we are reading certain system files which never get
388 * cluster locks (local alloc, truncate log).
390 * 3) If the process doing the iget() is responsible for
391 * orphan dir recovery. We're holding the orphan dir lock and
392 * can get into a deadlock with another process on another
393 * node in ->delete_inode().
395 * #1 and #2 can be simply solved by never taking the lock
396 * here for system files (which are the only type we read
397 * during mount). It's a heavier approach, but our main
398 * concern is user-accesible files anyway.
400 * #3 works itself out because we'll eventually take the
401 * cluster lock before trusting anything anyway.
403 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
404 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
405 && !ocfs2_mount_local(osb);
408 * To maintain backwards compatibility with older versions of
409 * ocfs2-tools, we still store the generation value for system
410 * files. The only ones that actually matter to userspace are
411 * the journals, but it's easier and inexpensive to just flag
412 * all system files similarly.
414 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
415 generation = osb->fs_generation;
417 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
418 OCFS2_LOCK_TYPE_META,
421 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
422 OCFS2_LOCK_TYPE_OPEN,
426 status = ocfs2_open_lock(inode);
428 make_bad_inode(inode);
432 status = ocfs2_meta_lock(inode, NULL, 0);
434 make_bad_inode(inode);
440 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
441 status = ocfs2_try_open_lock(inode, 0);
443 make_bad_inode(inode);
448 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
449 can_lock ? inode : NULL);
456 fe = (struct ocfs2_dinode *) bh->b_data;
457 if (!OCFS2_IS_VALID_DINODE(fe)) {
458 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
459 (unsigned long long)args->fi_blkno, 7,
465 * This is a code bug. Right now the caller needs to
466 * understand whether it is asking for a system file inode or
467 * not so the proper lock names can be built.
469 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
470 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
471 "Inode %llu: system file state is ambigous\n",
472 (unsigned long long)args->fi_blkno);
474 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
475 S_ISBLK(le16_to_cpu(fe->i_mode)))
476 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
478 if (ocfs2_populate_inode(inode, fe, 0) < 0)
481 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
487 ocfs2_meta_unlock(inode, 0);
490 make_bad_inode(inode);
499 void ocfs2_sync_blockdev(struct super_block *sb)
501 sync_blockdev(sb->s_bdev);
504 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
506 struct buffer_head *fe_bh)
509 struct ocfs2_truncate_context *tc = NULL;
510 struct ocfs2_dinode *fe;
511 handle_t *handle = NULL;
515 fe = (struct ocfs2_dinode *) fe_bh->b_data;
518 * This check will also skip truncate of inodes with inline
519 * data and fast symlinks.
521 if (fe->i_clusters) {
522 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
523 if (IS_ERR(handle)) {
524 status = PTR_ERR(handle);
529 status = ocfs2_journal_access(handle, inode, fe_bh,
530 OCFS2_JOURNAL_ACCESS_WRITE);
536 i_size_write(inode, 0);
538 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
544 ocfs2_commit_trans(osb, handle);
547 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
553 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
562 ocfs2_commit_trans(osb, handle);
567 static int ocfs2_remove_inode(struct inode *inode,
568 struct buffer_head *di_bh,
569 struct inode *orphan_dir_inode,
570 struct buffer_head *orphan_dir_bh)
573 struct inode *inode_alloc_inode = NULL;
574 struct buffer_head *inode_alloc_bh = NULL;
576 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
577 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
580 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
581 le16_to_cpu(di->i_suballoc_slot));
582 if (!inode_alloc_inode) {
588 mutex_lock(&inode_alloc_inode->i_mutex);
589 status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
591 mutex_unlock(&inode_alloc_inode->i_mutex);
597 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
598 if (IS_ERR(handle)) {
599 status = PTR_ERR(handle);
604 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
611 /* set the inodes dtime */
612 status = ocfs2_journal_access(handle, inode, di_bh,
613 OCFS2_JOURNAL_ACCESS_WRITE);
619 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
620 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
622 status = ocfs2_journal_dirty(handle, di_bh);
628 ocfs2_remove_from_cache(inode, di_bh);
630 status = ocfs2_free_dinode(handle, inode_alloc_inode,
636 ocfs2_commit_trans(osb, handle);
638 ocfs2_meta_unlock(inode_alloc_inode, 1);
639 mutex_unlock(&inode_alloc_inode->i_mutex);
640 brelse(inode_alloc_bh);
642 iput(inode_alloc_inode);
648 * Serialize with orphan dir recovery. If the process doing
649 * recovery on this orphan dir does an iget() with the dir
650 * i_mutex held, we'll deadlock here. Instead we detect this
651 * and exit early - recovery will wipe this inode for us.
653 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
658 spin_lock(&osb->osb_lock);
659 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
660 mlog(0, "Recovery is happening on orphan dir %d, will skip "
661 "this inode\n", slot);
665 /* This signals to the orphan recovery process that it should
666 * wait for us to handle the wipe. */
667 osb->osb_orphan_wipes[slot]++;
669 spin_unlock(&osb->osb_lock);
673 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
676 spin_lock(&osb->osb_lock);
677 osb->osb_orphan_wipes[slot]--;
678 spin_unlock(&osb->osb_lock);
680 wake_up(&osb->osb_wipe_event);
683 static int ocfs2_wipe_inode(struct inode *inode,
684 struct buffer_head *di_bh)
686 int status, orphaned_slot;
687 struct inode *orphan_dir_inode = NULL;
688 struct buffer_head *orphan_dir_bh = NULL;
689 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
690 struct ocfs2_dinode *di;
692 di = (struct ocfs2_dinode *) di_bh->b_data;
693 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
695 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
699 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
700 ORPHAN_DIR_SYSTEM_INODE,
702 if (!orphan_dir_inode) {
708 /* Lock the orphan dir. The lock will be held for the entire
709 * delete_inode operation. We do this now to avoid races with
710 * recovery completion on other nodes. */
711 mutex_lock(&orphan_dir_inode->i_mutex);
712 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
714 mutex_unlock(&orphan_dir_inode->i_mutex);
720 /* we do this while holding the orphan dir lock because we
721 * don't want recovery being run from another node to vote for
722 * an inode delete on us -- this will result in two nodes
723 * truncating the same file! */
724 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
727 goto bail_unlock_dir;
730 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
736 ocfs2_meta_unlock(orphan_dir_inode, 1);
737 mutex_unlock(&orphan_dir_inode->i_mutex);
738 brelse(orphan_dir_bh);
740 iput(orphan_dir_inode);
741 ocfs2_signal_wipe_completion(osb, orphaned_slot);
746 /* There is a series of simple checks that should be done before a
747 * vote is even considered. Encapsulate those in this function. */
748 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
751 struct ocfs2_inode_info *oi = OCFS2_I(inode);
752 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
754 /* We shouldn't be getting here for the root directory
756 if (inode == osb->root_inode) {
757 mlog(ML_ERROR, "Skipping delete of root inode.\n");
761 /* If we're coming from process_vote we can't go into our own
762 * voting [hello, deadlock city!], so unforuntately we just
763 * have to skip deleting this guy. That's OK though because
764 * the node who's doing the actual deleting should handle it
766 if (current == osb->vote_task) {
767 mlog(0, "Skipping delete of %lu because we're currently "
768 "in process_vote\n", inode->i_ino);
772 spin_lock(&oi->ip_lock);
773 /* OCFS2 *never* deletes system files. This should technically
774 * never get here as system file inodes should always have a
775 * positive link count. */
776 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
777 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
778 (unsigned long long)oi->ip_blkno);
782 /* If we have voted "yes" on the wipe of this inode for
783 * another node, it will be marked here so we can safely skip
784 * it. Recovery will cleanup any inodes we might inadvertantly
786 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
787 mlog(0, "Skipping delete of %lu because another node "
788 "has done this for us.\n", inode->i_ino);
794 spin_unlock(&oi->ip_lock);
799 /* Query the cluster to determine whether we should wipe an inode from
802 * Requires the inode to have the cluster lock. */
803 static int ocfs2_query_inode_wipe(struct inode *inode,
804 struct buffer_head *di_bh,
808 struct ocfs2_inode_info *oi = OCFS2_I(inode);
809 struct ocfs2_dinode *di;
813 /* While we were waiting for the cluster lock in
814 * ocfs2_delete_inode, another node might have asked to delete
815 * the inode. Recheck our flags to catch this. */
816 if (!ocfs2_inode_is_valid_to_delete(inode)) {
817 mlog(0, "Skipping delete of %llu because flags changed\n",
818 (unsigned long long)oi->ip_blkno);
822 /* Now that we have an up to date inode, we can double check
824 if (inode->i_nlink) {
825 mlog(0, "Skipping delete of %llu because nlink = %u\n",
826 (unsigned long long)oi->ip_blkno, inode->i_nlink);
830 /* Do some basic inode verification... */
831 di = (struct ocfs2_dinode *) di_bh->b_data;
832 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
833 /* for lack of a better error? */
836 "Inode %llu (on-disk %llu) not orphaned! "
837 "Disk flags 0x%x, inode flags 0x%x\n",
838 (unsigned long long)oi->ip_blkno,
839 (unsigned long long)le64_to_cpu(di->i_blkno),
840 le32_to_cpu(di->i_flags), oi->ip_flags);
844 /* has someone already deleted us?! baaad... */
852 * This is how ocfs2 determines whether an inode is still live
853 * within the cluster. Every node takes a shared read lock on
854 * the inode open lock in ocfs2_read_locked_inode(). When we
855 * get to ->delete_inode(), each node tries to convert it's
856 * lock to an exclusive. Trylocks are serialized by the inode
857 * meta data lock. If the upconvert suceeds, we know the inode
858 * is no longer live and can be deleted.
860 * Though we call this with the meta data lock held, the
861 * trylock keeps us from ABBA deadlock.
863 status = ocfs2_try_open_lock(inode, 1);
864 if (status == -EAGAIN) {
866 mlog(0, "Skipping delete of %llu because it is in use on "
867 "other nodes\n", (unsigned long long)oi->ip_blkno);
876 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
877 (unsigned long long)oi->ip_blkno,
878 le16_to_cpu(di->i_orphaned_slot));
884 /* Support function for ocfs2_delete_inode. Will help us keep the
885 * inode data in a consistent state for clear_inode. Always truncates
886 * pages, optionally sync's them first. */
887 static void ocfs2_cleanup_delete_inode(struct inode *inode,
890 mlog(0, "Cleanup inode %llu, sync = %d\n",
891 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
893 write_inode_now(inode, 1);
894 truncate_inode_pages(&inode->i_data, 0);
897 void ocfs2_delete_inode(struct inode *inode)
900 sigset_t blocked, oldset;
901 struct buffer_head *di_bh = NULL;
903 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
905 if (is_bad_inode(inode)) {
906 mlog(0, "Skipping delete of bad inode\n");
910 if (!ocfs2_inode_is_valid_to_delete(inode)) {
911 /* It's probably not necessary to truncate_inode_pages
912 * here but we do it for safety anyway (it will most
913 * likely be a no-op anyway) */
914 ocfs2_cleanup_delete_inode(inode, 0);
918 /* We want to block signals in delete_inode as the lock and
919 * messaging paths may return us -ERESTARTSYS. Which would
920 * cause us to exit early, resulting in inodes being orphaned
922 sigfillset(&blocked);
923 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
926 ocfs2_cleanup_delete_inode(inode, 1);
930 /* Lock down the inode. This gives us an up to date view of
931 * it's metadata (for verification), and allows us to
932 * serialize delete_inode votes.
934 * Even though we might be doing a truncate, we don't take the
935 * allocation lock here as it won't be needed - nobody will
936 * have the file open.
938 status = ocfs2_meta_lock(inode, &di_bh, 1);
940 if (status != -ENOENT)
942 ocfs2_cleanup_delete_inode(inode, 0);
946 /* Query the cluster. This will be the final decision made
947 * before we go ahead and wipe the inode. */
948 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
949 if (!wipe || status < 0) {
950 /* Error and inode busy vote both mean we won't be
951 * removing the inode, so they take almost the same
956 /* Someone in the cluster has voted to not wipe this
957 * inode, or it was never completely orphaned. Write
958 * out the pages and exit now. */
959 ocfs2_cleanup_delete_inode(inode, 1);
960 goto bail_unlock_inode;
963 ocfs2_cleanup_delete_inode(inode, 0);
965 status = ocfs2_wipe_inode(inode, di_bh);
967 if (status != -EDEADLK)
969 goto bail_unlock_inode;
973 * Mark the inode as successfully deleted.
975 * This is important for ocfs2_clear_inode() as it will check
976 * this flag and skip any checkpointing work
978 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
979 * the LVB for other nodes.
981 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
984 ocfs2_meta_unlock(inode, 1);
987 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
995 void ocfs2_clear_inode(struct inode *inode)
998 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1005 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1006 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1008 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1009 "Inode=%lu\n", inode->i_ino);
1011 /* For remove delete_inode vote, we hold open lock before,
1012 * now it is time to unlock PR and EX open locks. */
1013 ocfs2_open_unlock(inode);
1015 /* Do these before all the other work so that we don't bounce
1016 * the vote thread while waiting to destroy the locks. */
1017 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1018 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
1019 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
1020 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1022 /* We very well may get a clear_inode before all an inodes
1023 * metadata has hit disk. Of course, we can't drop any cluster
1024 * locks until the journal has finished with it. The only
1025 * exception here are successfully wiped inodes - their
1026 * metadata can now be considered to be part of the system
1027 * inodes from which it came. */
1028 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1029 ocfs2_checkpoint_inode(inode);
1031 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1032 "Clear inode of %llu, inode has io markers\n",
1033 (unsigned long long)oi->ip_blkno);
1035 ocfs2_extent_map_trunc(inode, 0);
1037 status = ocfs2_drop_inode_locks(inode);
1041 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1042 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1043 ocfs2_lock_res_free(&oi->ip_data_lockres);
1044 ocfs2_lock_res_free(&oi->ip_open_lockres);
1046 ocfs2_metadata_cache_purge(inode);
1048 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1049 "Clear inode of %llu, inode has %u cache items\n",
1050 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1052 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1053 "Clear inode of %llu, inode has a bad flag\n",
1054 (unsigned long long)oi->ip_blkno);
1056 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1057 "Clear inode of %llu, inode is locked\n",
1058 (unsigned long long)oi->ip_blkno);
1060 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1061 "Clear inode of %llu, io_mutex is locked\n",
1062 (unsigned long long)oi->ip_blkno);
1063 mutex_unlock(&oi->ip_io_mutex);
1066 * down_trylock() returns 0, down_write_trylock() returns 1
1069 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1070 "Clear inode of %llu, alloc_sem is locked\n",
1071 (unsigned long long)oi->ip_blkno);
1072 up_write(&oi->ip_alloc_sem);
1074 mlog_bug_on_msg(oi->ip_open_count,
1075 "Clear inode of %llu has open count %d\n",
1076 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1078 /* Clear all other flags. */
1079 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1080 oi->ip_created_trans = 0;
1081 oi->ip_last_trans = 0;
1082 oi->ip_dir_start_lookup = 0;
1083 oi->ip_blkno = 0ULL;
1089 /* Called under inode_lock, with no more references on the
1090 * struct inode, so it's safe here to check the flags field
1091 * and to manipulate i_nlink without any other locks. */
1092 void ocfs2_drop_inode(struct inode *inode)
1094 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1098 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1099 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1101 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1102 generic_delete_inode(inode);
1104 generic_drop_inode(inode);
1110 * TODO: this should probably be merged into ocfs2_get_block
1112 * However, you now need to pay attention to the cont_prepare_write()
1113 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1114 * expects never to extend).
1116 struct buffer_head *ocfs2_bread(struct inode *inode,
1117 int block, int *err, int reada)
1119 struct buffer_head *bh = NULL;
1122 int readflags = OCFS2_BH_CACHED;
1125 readflags |= OCFS2_BH_READAHEAD;
1127 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1128 i_size_read(inode)) {
1133 down_read(&OCFS2_I(inode)->ip_alloc_sem);
1134 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1136 up_read(&OCFS2_I(inode)->ip_alloc_sem);
1142 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1162 * This is called from our getattr.
1164 int ocfs2_inode_revalidate(struct dentry *dentry)
1166 struct inode *inode = dentry->d_inode;
1169 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1170 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1173 mlog(0, "eep, no inode!\n");
1178 spin_lock(&OCFS2_I(inode)->ip_lock);
1179 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1180 spin_unlock(&OCFS2_I(inode)->ip_lock);
1181 mlog(0, "inode deleted!\n");
1185 spin_unlock(&OCFS2_I(inode)->ip_lock);
1187 /* Let ocfs2_meta_lock do the work of updating our struct
1189 status = ocfs2_meta_lock(inode, NULL, 0);
1191 if (status != -ENOENT)
1195 ocfs2_meta_unlock(inode, 0);
1203 * Updates a disk inode from a
1205 * Only takes ip_lock.
1207 int ocfs2_mark_inode_dirty(handle_t *handle,
1208 struct inode *inode,
1209 struct buffer_head *bh)
1212 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1214 mlog_entry("(inode %llu)\n",
1215 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1217 status = ocfs2_journal_access(handle, inode, bh,
1218 OCFS2_JOURNAL_ACCESS_WRITE);
1224 spin_lock(&OCFS2_I(inode)->ip_lock);
1225 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1226 ocfs2_get_inode_flags(OCFS2_I(inode));
1227 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1228 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1229 spin_unlock(&OCFS2_I(inode)->ip_lock);
1231 fe->i_size = cpu_to_le64(i_size_read(inode));
1232 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1233 fe->i_uid = cpu_to_le32(inode->i_uid);
1234 fe->i_gid = cpu_to_le32(inode->i_gid);
1235 fe->i_mode = cpu_to_le16(inode->i_mode);
1236 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1237 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1238 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1239 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1240 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1241 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1243 status = ocfs2_journal_dirty(handle, bh);
1256 * Updates a struct inode from a disk inode.
1257 * does no i/o, only takes ip_lock.
1259 void ocfs2_refresh_inode(struct inode *inode,
1260 struct ocfs2_dinode *fe)
1262 spin_lock(&OCFS2_I(inode)->ip_lock);
1264 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1265 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1266 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1267 ocfs2_set_inode_flags(inode);
1268 i_size_write(inode, le64_to_cpu(fe->i_size));
1269 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1270 inode->i_uid = le32_to_cpu(fe->i_uid);
1271 inode->i_gid = le32_to_cpu(fe->i_gid);
1272 inode->i_mode = le16_to_cpu(fe->i_mode);
1273 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1274 inode->i_blocks = 0;
1276 inode->i_blocks = ocfs2_inode_sector_count(inode);
1277 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1278 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1279 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1280 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1281 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1282 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1284 spin_unlock(&OCFS2_I(inode)->ip_lock);