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/smp_lock.h>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
42 #include "extent_map.h"
44 #include "heartbeat.h"
55 #include "buffer_head_io.h"
57 struct ocfs2_find_inode_args
61 unsigned int fi_flags;
64 static int ocfs2_read_locked_inode(struct inode *inode,
65 struct ocfs2_find_inode_args *args);
66 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
67 static int ocfs2_find_actor(struct inode *inode, void *opaque);
68 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
70 struct buffer_head *fe_bh);
72 void ocfs2_set_inode_flags(struct inode *inode)
74 unsigned int flags = OCFS2_I(inode)->ip_attr;
76 inode->i_flags &= ~(S_IMMUTABLE |
77 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
79 if (flags & OCFS2_IMMUTABLE_FL)
80 inode->i_flags |= S_IMMUTABLE;
82 if (flags & OCFS2_SYNC_FL)
83 inode->i_flags |= S_SYNC;
84 if (flags & OCFS2_APPEND_FL)
85 inode->i_flags |= S_APPEND;
86 if (flags & OCFS2_NOATIME_FL)
87 inode->i_flags |= S_NOATIME;
88 if (flags & OCFS2_DIRSYNC_FL)
89 inode->i_flags |= S_DIRSYNC;
92 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
93 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
95 unsigned int flags = oi->vfs_inode.i_flags;
97 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
98 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
100 oi->ip_attr |= OCFS2_SYNC_FL;
101 if (flags & S_APPEND)
102 oi->ip_attr |= OCFS2_APPEND_FL;
103 if (flags & S_IMMUTABLE)
104 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
105 if (flags & S_NOATIME)
106 oi->ip_attr |= OCFS2_NOATIME_FL;
107 if (flags & S_DIRSYNC)
108 oi->ip_attr |= OCFS2_DIRSYNC_FL;
111 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
113 struct inode *inode = NULL;
114 struct super_block *sb = osb->sb;
115 struct ocfs2_find_inode_args args;
117 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
119 /* Ok. By now we've either got the offsets passed to us by the
120 * caller, or we just pulled them off the bh. Lets do some
121 * sanity checks to make sure they're OK. */
123 inode = ERR_PTR(-EINVAL);
124 mlog_errno(PTR_ERR(inode));
128 args.fi_blkno = blkno;
129 args.fi_flags = flags;
130 args.fi_ino = ino_from_blkno(sb, blkno);
132 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
133 ocfs2_init_locked_inode, &args);
134 /* inode was *not* in the inode cache. 2.6.x requires
135 * us to do our own read_inode call and unlock it
137 if (inode && inode->i_state & I_NEW) {
138 mlog(0, "Inode was not in inode cache, reading it.\n");
139 ocfs2_read_locked_inode(inode, &args);
140 unlock_new_inode(inode);
143 inode = ERR_PTR(-ENOMEM);
144 mlog_errno(PTR_ERR(inode));
147 if (is_bad_inode(inode)) {
149 inode = ERR_PTR(-ESTALE);
154 if (!IS_ERR(inode)) {
155 mlog(0, "returning inode with number %llu\n",
156 (unsigned long long)OCFS2_I(inode)->ip_blkno);
157 mlog_exit_ptr(inode);
165 * here's how inodes get read from disk:
166 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
167 * found? : return the in-memory inode
168 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
171 static int ocfs2_find_actor(struct inode *inode, void *opaque)
173 struct ocfs2_find_inode_args *args = NULL;
174 struct ocfs2_inode_info *oi = OCFS2_I(inode);
177 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
181 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
183 if (oi->ip_blkno != args->fi_blkno)
193 * initialize the new inode, but don't do anything that would cause
195 * return 0 on success, 1 on failure
197 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
199 struct ocfs2_find_inode_args *args = opaque;
201 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
203 inode->i_ino = args->fi_ino;
204 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
210 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
213 struct super_block *sb;
214 struct ocfs2_super *osb;
215 int status = -EINVAL;
217 mlog_entry("(0x%p, size:%llu)\n", inode,
218 (unsigned long long)le64_to_cpu(fe->i_size));
223 /* this means that read_inode cannot create a superblock inode
224 * today. change if needed. */
225 if (!OCFS2_IS_VALID_DINODE(fe) ||
226 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
227 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
228 "signature = %.*s, flags = 0x%x\n",
230 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
231 fe->i_signature, le32_to_cpu(fe->i_flags));
235 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
236 mlog(ML_ERROR, "file entry generation does not match "
237 "superblock! osb->fs_generation=%x, "
238 "fe->i_fs_generation=%x\n",
239 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
243 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
244 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
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(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
459 (unsigned long long)le64_to_cpu(fe->i_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;
517 if (fe->i_clusters) {
518 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
519 if (IS_ERR(handle)) {
520 status = PTR_ERR(handle);
525 status = ocfs2_journal_access(handle, inode, fe_bh,
526 OCFS2_JOURNAL_ACCESS_WRITE);
532 i_size_write(inode, 0);
534 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
540 ocfs2_commit_trans(osb, handle);
543 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
549 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
558 ocfs2_commit_trans(osb, handle);
563 static int ocfs2_remove_inode(struct inode *inode,
564 struct buffer_head *di_bh,
565 struct inode *orphan_dir_inode,
566 struct buffer_head *orphan_dir_bh)
569 struct inode *inode_alloc_inode = NULL;
570 struct buffer_head *inode_alloc_bh = NULL;
572 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
573 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
576 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
577 le16_to_cpu(di->i_suballoc_slot));
578 if (!inode_alloc_inode) {
584 mutex_lock(&inode_alloc_inode->i_mutex);
585 status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
587 mutex_unlock(&inode_alloc_inode->i_mutex);
593 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
594 if (IS_ERR(handle)) {
595 status = PTR_ERR(handle);
600 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
607 /* set the inodes dtime */
608 status = ocfs2_journal_access(handle, inode, di_bh,
609 OCFS2_JOURNAL_ACCESS_WRITE);
615 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
616 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
618 status = ocfs2_journal_dirty(handle, di_bh);
624 ocfs2_remove_from_cache(inode, di_bh);
626 status = ocfs2_free_dinode(handle, inode_alloc_inode,
632 ocfs2_commit_trans(osb, handle);
634 ocfs2_meta_unlock(inode_alloc_inode, 1);
635 mutex_unlock(&inode_alloc_inode->i_mutex);
636 brelse(inode_alloc_bh);
638 iput(inode_alloc_inode);
644 * Serialize with orphan dir recovery. If the process doing
645 * recovery on this orphan dir does an iget() with the dir
646 * i_mutex held, we'll deadlock here. Instead we detect this
647 * and exit early - recovery will wipe this inode for us.
649 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
654 spin_lock(&osb->osb_lock);
655 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
656 mlog(0, "Recovery is happening on orphan dir %d, will skip "
657 "this inode\n", slot);
661 /* This signals to the orphan recovery process that it should
662 * wait for us to handle the wipe. */
663 osb->osb_orphan_wipes[slot]++;
665 spin_unlock(&osb->osb_lock);
669 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
672 spin_lock(&osb->osb_lock);
673 osb->osb_orphan_wipes[slot]--;
674 spin_unlock(&osb->osb_lock);
676 wake_up(&osb->osb_wipe_event);
679 static int ocfs2_wipe_inode(struct inode *inode,
680 struct buffer_head *di_bh)
682 int status, orphaned_slot;
683 struct inode *orphan_dir_inode = NULL;
684 struct buffer_head *orphan_dir_bh = NULL;
685 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
686 struct ocfs2_dinode *di;
688 di = (struct ocfs2_dinode *) di_bh->b_data;
689 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
691 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
695 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
696 ORPHAN_DIR_SYSTEM_INODE,
698 if (!orphan_dir_inode) {
704 /* Lock the orphan dir. The lock will be held for the entire
705 * delete_inode operation. We do this now to avoid races with
706 * recovery completion on other nodes. */
707 mutex_lock(&orphan_dir_inode->i_mutex);
708 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
710 mutex_unlock(&orphan_dir_inode->i_mutex);
716 /* we do this while holding the orphan dir lock because we
717 * don't want recovery being run from another node to vote for
718 * an inode delete on us -- this will result in two nodes
719 * truncating the same file! */
720 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
723 goto bail_unlock_dir;
726 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
732 ocfs2_meta_unlock(orphan_dir_inode, 1);
733 mutex_unlock(&orphan_dir_inode->i_mutex);
734 brelse(orphan_dir_bh);
736 iput(orphan_dir_inode);
737 ocfs2_signal_wipe_completion(osb, orphaned_slot);
742 /* There is a series of simple checks that should be done before a
743 * vote is even considered. Encapsulate those in this function. */
744 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
747 struct ocfs2_inode_info *oi = OCFS2_I(inode);
748 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
750 /* We shouldn't be getting here for the root directory
752 if (inode == osb->root_inode) {
753 mlog(ML_ERROR, "Skipping delete of root inode.\n");
757 /* If we're coming from process_vote we can't go into our own
758 * voting [hello, deadlock city!], so unforuntately we just
759 * have to skip deleting this guy. That's OK though because
760 * the node who's doing the actual deleting should handle it
762 if (current == osb->vote_task) {
763 mlog(0, "Skipping delete of %lu because we're currently "
764 "in process_vote\n", inode->i_ino);
768 spin_lock(&oi->ip_lock);
769 /* OCFS2 *never* deletes system files. This should technically
770 * never get here as system file inodes should always have a
771 * positive link count. */
772 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
773 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
774 (unsigned long long)oi->ip_blkno);
778 /* If we have voted "yes" on the wipe of this inode for
779 * another node, it will be marked here so we can safely skip
780 * it. Recovery will cleanup any inodes we might inadvertantly
782 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
783 mlog(0, "Skipping delete of %lu because another node "
784 "has done this for us.\n", inode->i_ino);
790 spin_unlock(&oi->ip_lock);
795 /* Query the cluster to determine whether we should wipe an inode from
798 * Requires the inode to have the cluster lock. */
799 static int ocfs2_query_inode_wipe(struct inode *inode,
800 struct buffer_head *di_bh,
804 struct ocfs2_inode_info *oi = OCFS2_I(inode);
805 struct ocfs2_dinode *di;
809 /* While we were waiting for the cluster lock in
810 * ocfs2_delete_inode, another node might have asked to delete
811 * the inode. Recheck our flags to catch this. */
812 if (!ocfs2_inode_is_valid_to_delete(inode)) {
813 mlog(0, "Skipping delete of %llu because flags changed\n",
814 (unsigned long long)oi->ip_blkno);
818 /* Now that we have an up to date inode, we can double check
820 if (inode->i_nlink) {
821 mlog(0, "Skipping delete of %llu because nlink = %u\n",
822 (unsigned long long)oi->ip_blkno, inode->i_nlink);
826 /* Do some basic inode verification... */
827 di = (struct ocfs2_dinode *) di_bh->b_data;
828 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
829 /* for lack of a better error? */
832 "Inode %llu (on-disk %llu) not orphaned! "
833 "Disk flags 0x%x, inode flags 0x%x\n",
834 (unsigned long long)oi->ip_blkno,
835 (unsigned long long)le64_to_cpu(di->i_blkno),
836 le32_to_cpu(di->i_flags), oi->ip_flags);
840 /* has someone already deleted us?! baaad... */
848 * This is how ocfs2 determines whether an inode is still live
849 * within the cluster. Every node takes a shared read lock on
850 * the inode open lock in ocfs2_read_locked_inode(). When we
851 * get to ->delete_inode(), each node tries to convert it's
852 * lock to an exclusive. Trylocks are serialized by the inode
853 * meta data lock. If the upconvert suceeds, we know the inode
854 * is no longer live and can be deleted.
856 * Though we call this with the meta data lock held, the
857 * trylock keeps us from ABBA deadlock.
859 status = ocfs2_try_open_lock(inode, 1);
860 if (status == -EAGAIN) {
862 mlog(0, "Skipping delete of %llu because it is in use on"
863 "other nodes\n", (unsigned long long)oi->ip_blkno);
872 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
873 (unsigned long long)oi->ip_blkno,
874 le16_to_cpu(di->i_orphaned_slot));
880 /* Support function for ocfs2_delete_inode. Will help us keep the
881 * inode data in a consistent state for clear_inode. Always truncates
882 * pages, optionally sync's them first. */
883 static void ocfs2_cleanup_delete_inode(struct inode *inode,
886 mlog(0, "Cleanup inode %llu, sync = %d\n",
887 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
889 write_inode_now(inode, 1);
890 truncate_inode_pages(&inode->i_data, 0);
893 void ocfs2_delete_inode(struct inode *inode)
896 sigset_t blocked, oldset;
897 struct buffer_head *di_bh = NULL;
899 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
901 if (is_bad_inode(inode)) {
902 mlog(0, "Skipping delete of bad inode\n");
906 if (!ocfs2_inode_is_valid_to_delete(inode)) {
907 /* It's probably not necessary to truncate_inode_pages
908 * here but we do it for safety anyway (it will most
909 * likely be a no-op anyway) */
910 ocfs2_cleanup_delete_inode(inode, 0);
914 /* We want to block signals in delete_inode as the lock and
915 * messaging paths may return us -ERESTARTSYS. Which would
916 * cause us to exit early, resulting in inodes being orphaned
918 sigfillset(&blocked);
919 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
922 ocfs2_cleanup_delete_inode(inode, 1);
926 /* Lock down the inode. This gives us an up to date view of
927 * it's metadata (for verification), and allows us to
928 * serialize delete_inode votes.
930 * Even though we might be doing a truncate, we don't take the
931 * allocation lock here as it won't be needed - nobody will
932 * have the file open.
934 status = ocfs2_meta_lock(inode, &di_bh, 1);
936 if (status != -ENOENT)
938 ocfs2_cleanup_delete_inode(inode, 0);
942 /* Query the cluster. This will be the final decision made
943 * before we go ahead and wipe the inode. */
944 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
945 if (!wipe || status < 0) {
946 /* Error and inode busy vote both mean we won't be
947 * removing the inode, so they take almost the same
952 /* Someone in the cluster has voted to not wipe this
953 * inode, or it was never completely orphaned. Write
954 * out the pages and exit now. */
955 ocfs2_cleanup_delete_inode(inode, 1);
956 goto bail_unlock_inode;
959 ocfs2_cleanup_delete_inode(inode, 0);
961 status = ocfs2_wipe_inode(inode, di_bh);
963 if (status != -EDEADLK)
965 goto bail_unlock_inode;
969 * Mark the inode as successfully deleted.
971 * This is important for ocfs2_clear_inode() as it will check
972 * this flag and skip any checkpointing work
974 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
975 * the LVB for other nodes.
977 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
980 ocfs2_meta_unlock(inode, 1);
983 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
991 void ocfs2_clear_inode(struct inode *inode)
994 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1001 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1002 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1004 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1005 "Inode=%lu\n", inode->i_ino);
1007 /* For remove delete_inode vote, we hold open lock before,
1008 * now it is time to unlock PR and EX open locks. */
1009 ocfs2_open_unlock(inode);
1011 /* Do these before all the other work so that we don't bounce
1012 * the vote thread while waiting to destroy the locks. */
1013 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1014 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
1015 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
1016 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1018 /* We very well may get a clear_inode before all an inodes
1019 * metadata has hit disk. Of course, we can't drop any cluster
1020 * locks until the journal has finished with it. The only
1021 * exception here are successfully wiped inodes - their
1022 * metadata can now be considered to be part of the system
1023 * inodes from which it came. */
1024 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1025 ocfs2_checkpoint_inode(inode);
1027 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1028 "Clear inode of %llu, inode has io markers\n",
1029 (unsigned long long)oi->ip_blkno);
1031 ocfs2_extent_map_trunc(inode, 0);
1033 status = ocfs2_drop_inode_locks(inode);
1037 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1038 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1039 ocfs2_lock_res_free(&oi->ip_data_lockres);
1040 ocfs2_lock_res_free(&oi->ip_open_lockres);
1042 ocfs2_metadata_cache_purge(inode);
1044 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1045 "Clear inode of %llu, inode has %u cache items\n",
1046 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1048 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1049 "Clear inode of %llu, inode has a bad flag\n",
1050 (unsigned long long)oi->ip_blkno);
1052 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1053 "Clear inode of %llu, inode is locked\n",
1054 (unsigned long long)oi->ip_blkno);
1056 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1057 "Clear inode of %llu, io_mutex is locked\n",
1058 (unsigned long long)oi->ip_blkno);
1059 mutex_unlock(&oi->ip_io_mutex);
1062 * down_trylock() returns 0, down_write_trylock() returns 1
1065 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1066 "Clear inode of %llu, alloc_sem is locked\n",
1067 (unsigned long long)oi->ip_blkno);
1068 up_write(&oi->ip_alloc_sem);
1070 mlog_bug_on_msg(oi->ip_open_count,
1071 "Clear inode of %llu has open count %d\n",
1072 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1074 /* Clear all other flags. */
1075 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1076 oi->ip_created_trans = 0;
1077 oi->ip_last_trans = 0;
1078 oi->ip_dir_start_lookup = 0;
1079 oi->ip_blkno = 0ULL;
1085 /* Called under inode_lock, with no more references on the
1086 * struct inode, so it's safe here to check the flags field
1087 * and to manipulate i_nlink without any other locks. */
1088 void ocfs2_drop_inode(struct inode *inode)
1090 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1094 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1095 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1097 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1098 generic_delete_inode(inode);
1100 generic_drop_inode(inode);
1106 * TODO: this should probably be merged into ocfs2_get_block
1108 * However, you now need to pay attention to the cont_prepare_write()
1109 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1110 * expects never to extend).
1112 struct buffer_head *ocfs2_bread(struct inode *inode,
1113 int block, int *err, int reada)
1115 struct buffer_head *bh = NULL;
1118 int readflags = OCFS2_BH_CACHED;
1121 readflags |= OCFS2_BH_READAHEAD;
1123 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1124 i_size_read(inode)) {
1129 down_read(&OCFS2_I(inode)->ip_alloc_sem);
1130 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1132 up_read(&OCFS2_I(inode)->ip_alloc_sem);
1138 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1158 * This is called from our getattr.
1160 int ocfs2_inode_revalidate(struct dentry *dentry)
1162 struct inode *inode = dentry->d_inode;
1165 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1166 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1169 mlog(0, "eep, no inode!\n");
1174 spin_lock(&OCFS2_I(inode)->ip_lock);
1175 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1176 spin_unlock(&OCFS2_I(inode)->ip_lock);
1177 mlog(0, "inode deleted!\n");
1181 spin_unlock(&OCFS2_I(inode)->ip_lock);
1183 /* Let ocfs2_meta_lock do the work of updating our struct
1185 status = ocfs2_meta_lock(inode, NULL, 0);
1187 if (status != -ENOENT)
1191 ocfs2_meta_unlock(inode, 0);
1199 * Updates a disk inode from a
1201 * Only takes ip_lock.
1203 int ocfs2_mark_inode_dirty(handle_t *handle,
1204 struct inode *inode,
1205 struct buffer_head *bh)
1208 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1210 mlog_entry("(inode %llu)\n",
1211 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1213 status = ocfs2_journal_access(handle, inode, bh,
1214 OCFS2_JOURNAL_ACCESS_WRITE);
1220 spin_lock(&OCFS2_I(inode)->ip_lock);
1221 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1222 ocfs2_get_inode_flags(OCFS2_I(inode));
1223 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1224 spin_unlock(&OCFS2_I(inode)->ip_lock);
1226 fe->i_size = cpu_to_le64(i_size_read(inode));
1227 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1228 fe->i_uid = cpu_to_le32(inode->i_uid);
1229 fe->i_gid = cpu_to_le32(inode->i_gid);
1230 fe->i_mode = cpu_to_le16(inode->i_mode);
1231 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1232 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1233 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1234 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1235 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1236 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1238 status = ocfs2_journal_dirty(handle, bh);
1251 * Updates a struct inode from a disk inode.
1252 * does no i/o, only takes ip_lock.
1254 void ocfs2_refresh_inode(struct inode *inode,
1255 struct ocfs2_dinode *fe)
1257 spin_lock(&OCFS2_I(inode)->ip_lock);
1259 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1260 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1261 ocfs2_set_inode_flags(inode);
1262 i_size_write(inode, le64_to_cpu(fe->i_size));
1263 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1264 inode->i_uid = le32_to_cpu(fe->i_uid);
1265 inode->i_gid = le32_to_cpu(fe->i_gid);
1266 inode->i_mode = le16_to_cpu(fe->i_mode);
1267 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1268 inode->i_blocks = 0;
1270 inode->i_blocks = ocfs2_inode_sector_count(inode);
1271 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1272 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1273 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1274 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1275 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1276 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1278 spin_unlock(&OCFS2_I(inode)->ip_lock);