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 #define OCFS2_FI_FLAG_NOWAIT 0x1
58 #define OCFS2_FI_FLAG_DELETE 0x2
59 struct ocfs2_find_inode_args
63 unsigned int fi_flags;
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 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
78 struct ocfs2_find_inode_args args;
80 /* ocfs2_ilookup_for_vote should *only* be called from the
82 BUG_ON(current != osb->vote_task);
84 args.fi_blkno = blkno;
85 args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
87 args.fi_flags |= OCFS2_FI_FLAG_DELETE;
88 args.fi_ino = ino_from_blkno(osb->sb, blkno);
89 return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
92 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno)
94 struct inode *inode = NULL;
95 struct super_block *sb = osb->sb;
96 struct ocfs2_find_inode_args args;
98 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
100 /* Ok. By now we've either got the offsets passed to us by the
101 * caller, or we just pulled them off the bh. Lets do some
102 * sanity checks to make sure they're OK. */
104 inode = ERR_PTR(-EINVAL);
105 mlog_errno(PTR_ERR(inode));
109 args.fi_blkno = blkno;
111 args.fi_ino = ino_from_blkno(sb, blkno);
113 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
114 ocfs2_init_locked_inode, &args);
115 /* inode was *not* in the inode cache. 2.6.x requires
116 * us to do our own read_inode call and unlock it
118 if (inode && inode->i_state & I_NEW) {
119 mlog(0, "Inode was not in inode cache, reading it.\n");
120 ocfs2_read_locked_inode(inode, &args);
121 unlock_new_inode(inode);
124 inode = ERR_PTR(-ENOMEM);
125 mlog_errno(PTR_ERR(inode));
128 if (is_bad_inode(inode)) {
130 inode = ERR_PTR(-ESTALE);
131 mlog_errno(PTR_ERR(inode));
136 if (!IS_ERR(inode)) {
137 mlog(0, "returning inode with number %llu\n",
138 (unsigned long long)OCFS2_I(inode)->ip_blkno);
139 mlog_exit_ptr(inode);
141 mlog_errno(PTR_ERR(inode));
148 * here's how inodes get read from disk:
149 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
150 * found? : return the in-memory inode
151 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
154 static int ocfs2_find_actor(struct inode *inode, void *opaque)
156 struct ocfs2_find_inode_args *args = NULL;
157 struct ocfs2_inode_info *oi = OCFS2_I(inode);
160 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
164 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
166 if (oi->ip_blkno != args->fi_blkno)
169 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
170 * ocfs2_ilookup_for_vote which won't create an inode for one
171 * that isn't found. The vote thread which doesn't want to get
172 * an inode which is in the process of going away - otherwise
173 * the call to __wait_on_freeing_inode in find_inode_fast will
174 * cause it to deadlock on an inode which may be waiting on a
175 * vote (or lock release) in delete_inode */
176 if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
177 (inode->i_state & (I_FREEING|I_CLEAR))) {
178 /* As stated above, we're not going to return an
179 * inode. In the case of a delete vote, the voting
180 * code is going to signal the other node to go
181 * ahead. Mark that state here, so this freeing inode
182 * has the state when it gets to delete_inode. */
183 if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
184 spin_lock(&oi->ip_lock);
185 ocfs2_mark_inode_remotely_deleted(inode);
186 spin_unlock(&oi->ip_lock);
198 * initialize the new inode, but don't do anything that would cause
200 * return 0 on success, 1 on failure
202 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
204 struct ocfs2_find_inode_args *args = opaque;
206 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
208 inode->i_ino = args->fi_ino;
209 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
215 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
218 struct super_block *sb;
219 struct ocfs2_super *osb;
220 int status = -EINVAL;
222 mlog_entry("(0x%p, size:%llu)\n", inode,
223 (unsigned long long)fe->i_size);
228 /* this means that read_inode cannot create a superblock inode
229 * today. change if needed. */
230 if (!OCFS2_IS_VALID_DINODE(fe) ||
231 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
232 mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
233 "signature = %.*s, flags = 0x%x\n",
235 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
236 fe->i_signature, le32_to_cpu(fe->i_flags));
240 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
241 mlog(ML_ERROR, "file entry generation does not match "
242 "superblock! osb->fs_generation=%x, "
243 "fe->i_fs_generation=%x\n",
244 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
248 inode->i_version = 1;
249 inode->i_generation = le32_to_cpu(fe->i_generation);
250 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
251 inode->i_mode = le16_to_cpu(fe->i_mode);
252 inode->i_uid = le32_to_cpu(fe->i_uid);
253 inode->i_gid = le32_to_cpu(fe->i_gid);
255 /* Fast symlinks will have i_size but no allocated clusters. */
256 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
260 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
261 inode->i_mapping->a_ops = &ocfs2_aops;
262 inode->i_flags |= S_NOATIME;
263 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
264 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
265 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
266 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
267 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
268 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
270 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
272 "ip_blkno %llu != i_blkno %llu!\n",
273 (unsigned long long)OCFS2_I(inode)->ip_blkno,
274 (unsigned long long)fe->i_blkno);
276 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
277 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
280 inode->i_ino = ino_from_blkno(inode->i_sb,
281 le64_to_cpu(fe->i_blkno));
283 mlog(0, "blkno = %llu, ino = %lu, create_ino = %s\n",
284 (unsigned long long)fe->i_blkno, inode->i_ino, create_ino ? "true" : "false");
286 inode->i_nlink = le16_to_cpu(fe->i_links_count);
288 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
289 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
290 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
291 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
292 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
293 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
294 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
295 /* we can't actually hit this as read_inode can't
296 * handle superblocks today ;-) */
300 switch (inode->i_mode & S_IFMT) {
302 inode->i_fop = &ocfs2_fops;
303 inode->i_op = &ocfs2_file_iops;
304 i_size_write(inode, le64_to_cpu(fe->i_size));
307 inode->i_op = &ocfs2_dir_iops;
308 inode->i_fop = &ocfs2_dops;
309 i_size_write(inode, le64_to_cpu(fe->i_size));
312 if (ocfs2_inode_is_fast_symlink(inode))
313 inode->i_op = &ocfs2_fast_symlink_inode_operations;
315 inode->i_op = &ocfs2_symlink_inode_operations;
316 i_size_write(inode, le64_to_cpu(fe->i_size));
319 inode->i_op = &ocfs2_special_file_iops;
320 init_special_inode(inode, inode->i_mode,
325 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
326 OCFS2_LOCK_TYPE_RW, inode);
327 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
328 OCFS2_LOCK_TYPE_META, inode);
329 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
330 OCFS2_LOCK_TYPE_DATA, inode);
338 static int ocfs2_read_locked_inode(struct inode *inode,
339 struct ocfs2_find_inode_args *args)
341 struct super_block *sb;
342 struct ocfs2_super *osb;
343 struct ocfs2_dinode *fe;
344 struct buffer_head *bh = NULL;
348 mlog_entry("(0x%p, 0x%p)\n", inode, args);
351 if (inode == NULL || inode->i_sb == NULL) {
352 mlog(ML_ERROR, "bad inode\n");
359 mlog(ML_ERROR, "bad inode args\n");
360 make_bad_inode(inode);
364 /* Read the FE off disk. This is safe because the kernel only
365 * does one read_inode2 for a new inode, and if it doesn't
366 * exist yet then nobody can be working on it! */
367 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL);
370 make_bad_inode(inode);
374 fe = (struct ocfs2_dinode *) bh->b_data;
375 if (!OCFS2_IS_VALID_DINODE(fe)) {
376 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
377 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
378 make_bad_inode(inode);
382 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
385 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
386 S_ISBLK(le16_to_cpu(fe->i_mode)))
387 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
390 if (ocfs2_populate_inode(inode, fe, 0) < 0) {
391 mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
392 (unsigned long long)fe->i_blkno, inode->i_ino);
393 make_bad_inode(inode);
397 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
400 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
412 void ocfs2_sync_blockdev(struct super_block *sb)
414 sync_blockdev(sb->s_bdev);
417 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
419 struct buffer_head *fe_bh)
422 struct ocfs2_journal_handle *handle = NULL;
423 struct ocfs2_truncate_context *tc = NULL;
424 struct ocfs2_dinode *fe;
428 fe = (struct ocfs2_dinode *) fe_bh->b_data;
430 /* zero allocation, zero truncate :) */
434 handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS);
435 if (IS_ERR(handle)) {
436 status = PTR_ERR(handle);
442 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
448 ocfs2_commit_trans(handle);
451 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
457 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
464 ocfs2_commit_trans(handle);
470 static int ocfs2_remove_inode(struct inode *inode,
471 struct buffer_head *di_bh,
472 struct inode *orphan_dir_inode,
473 struct buffer_head *orphan_dir_bh)
476 struct inode *inode_alloc_inode = NULL;
477 struct buffer_head *inode_alloc_bh = NULL;
478 struct ocfs2_journal_handle *handle;
479 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
480 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
483 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
484 le16_to_cpu(di->i_suballoc_slot));
485 if (!inode_alloc_inode) {
491 mutex_lock(&inode_alloc_inode->i_mutex);
492 status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1);
494 mutex_unlock(&inode_alloc_inode->i_mutex);
500 handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS);
501 if (IS_ERR(handle)) {
502 status = PTR_ERR(handle);
507 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
514 /* set the inodes dtime */
515 status = ocfs2_journal_access(handle, inode, di_bh,
516 OCFS2_JOURNAL_ACCESS_WRITE);
522 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
523 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
525 status = ocfs2_journal_dirty(handle, di_bh);
531 ocfs2_remove_from_cache(inode, di_bh);
533 status = ocfs2_free_dinode(handle, inode_alloc_inode,
539 ocfs2_commit_trans(handle);
541 ocfs2_meta_unlock(inode_alloc_inode, 1);
542 mutex_unlock(&inode_alloc_inode->i_mutex);
543 brelse(inode_alloc_bh);
545 iput(inode_alloc_inode);
551 * Serialize with orphan dir recovery. If the process doing
552 * recovery on this orphan dir does an iget() with the dir
553 * i_mutex held, we'll deadlock here. Instead we detect this
554 * and exit early - recovery will wipe this inode for us.
556 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
561 spin_lock(&osb->osb_lock);
562 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
563 mlog(0, "Recovery is happening on orphan dir %d, will skip "
564 "this inode\n", slot);
568 /* This signals to the orphan recovery process that it should
569 * wait for us to handle the wipe. */
570 osb->osb_orphan_wipes[slot]++;
572 spin_unlock(&osb->osb_lock);
576 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
579 spin_lock(&osb->osb_lock);
580 osb->osb_orphan_wipes[slot]--;
581 spin_unlock(&osb->osb_lock);
583 wake_up(&osb->osb_wipe_event);
586 static int ocfs2_wipe_inode(struct inode *inode,
587 struct buffer_head *di_bh)
589 int status, orphaned_slot;
590 struct inode *orphan_dir_inode = NULL;
591 struct buffer_head *orphan_dir_bh = NULL;
592 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
594 /* We've already voted on this so it should be readonly - no
595 * spinlock needed. */
596 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
598 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
602 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
603 ORPHAN_DIR_SYSTEM_INODE,
605 if (!orphan_dir_inode) {
611 /* Lock the orphan dir. The lock will be held for the entire
612 * delete_inode operation. We do this now to avoid races with
613 * recovery completion on other nodes. */
614 mutex_lock(&orphan_dir_inode->i_mutex);
615 status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
617 mutex_unlock(&orphan_dir_inode->i_mutex);
623 /* we do this while holding the orphan dir lock because we
624 * don't want recovery being run from another node to vote for
625 * an inode delete on us -- this will result in two nodes
626 * truncating the same file! */
627 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
630 goto bail_unlock_dir;
633 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
639 ocfs2_meta_unlock(orphan_dir_inode, 1);
640 mutex_unlock(&orphan_dir_inode->i_mutex);
641 brelse(orphan_dir_bh);
643 iput(orphan_dir_inode);
644 ocfs2_signal_wipe_completion(osb, orphaned_slot);
649 /* There is a series of simple checks that should be done before a
650 * vote is even considered. Encapsulate those in this function. */
651 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
654 struct ocfs2_inode_info *oi = OCFS2_I(inode);
655 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
657 /* We shouldn't be getting here for the root directory
659 if (inode == osb->root_inode) {
660 mlog(ML_ERROR, "Skipping delete of root inode.\n");
664 /* If we're coming from process_vote we can't go into our own
665 * voting [hello, deadlock city!], so unforuntately we just
666 * have to skip deleting this guy. That's OK though because
667 * the node who's doing the actual deleting should handle it
669 if (current == osb->vote_task) {
670 mlog(0, "Skipping delete of %lu because we're currently "
671 "in process_vote\n", inode->i_ino);
675 spin_lock(&oi->ip_lock);
676 /* OCFS2 *never* deletes system files. This should technically
677 * never get here as system file inodes should always have a
678 * positive link count. */
679 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
680 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
681 (unsigned long long)oi->ip_blkno);
685 /* If we have voted "yes" on the wipe of this inode for
686 * another node, it will be marked here so we can safely skip
687 * it. Recovery will cleanup any inodes we might inadvertantly
689 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
690 mlog(0, "Skipping delete of %lu because another node "
691 "has done this for us.\n", inode->i_ino);
697 spin_unlock(&oi->ip_lock);
702 /* Query the cluster to determine whether we should wipe an inode from
705 * Requires the inode to have the cluster lock. */
706 static int ocfs2_query_inode_wipe(struct inode *inode,
707 struct buffer_head *di_bh,
711 struct ocfs2_inode_info *oi = OCFS2_I(inode);
712 struct ocfs2_dinode *di;
716 /* While we were waiting for the cluster lock in
717 * ocfs2_delete_inode, another node might have asked to delete
718 * the inode. Recheck our flags to catch this. */
719 if (!ocfs2_inode_is_valid_to_delete(inode)) {
720 mlog(0, "Skipping delete of %llu because flags changed\n",
721 (unsigned long long)oi->ip_blkno);
725 /* Now that we have an up to date inode, we can double check
727 if (inode->i_nlink) {
728 mlog(0, "Skipping delete of %llu because nlink = %u\n",
729 (unsigned long long)oi->ip_blkno, inode->i_nlink);
733 /* Do some basic inode verification... */
734 di = (struct ocfs2_dinode *) di_bh->b_data;
735 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
736 /* for lack of a better error? */
739 "Inode %llu (on-disk %llu) not orphaned! "
740 "Disk flags 0x%x, inode flags 0x%x\n",
741 (unsigned long long)oi->ip_blkno,
742 (unsigned long long)di->i_blkno, di->i_flags,
747 /* has someone already deleted us?! baaad... */
754 status = ocfs2_request_delete_vote(inode);
755 /* -EBUSY means that other nodes are still using the
756 * inode. We're done here though, so avoid doing anything on
757 * disk and let them worry about deleting it. */
758 if (status == -EBUSY) {
760 mlog(0, "Skipping delete of %llu because it is in use on"
761 "other nodes\n", (unsigned long long)oi->ip_blkno);
769 spin_lock(&oi->ip_lock);
770 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
771 /* Nobody knew which slot this inode was orphaned
772 * into. This may happen during node death and
773 * recovery knows how to clean it up so we can safely
774 * ignore this inode for now on. */
775 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
776 (unsigned long long)oi->ip_blkno);
780 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
781 (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
783 spin_unlock(&oi->ip_lock);
789 /* Support function for ocfs2_delete_inode. Will help us keep the
790 * inode data in a consistent state for clear_inode. Always truncates
791 * pages, optionally sync's them first. */
792 static void ocfs2_cleanup_delete_inode(struct inode *inode,
795 mlog(0, "Cleanup inode %llu, sync = %d\n",
796 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
798 write_inode_now(inode, 1);
799 truncate_inode_pages(&inode->i_data, 0);
802 void ocfs2_delete_inode(struct inode *inode)
805 sigset_t blocked, oldset;
806 struct buffer_head *di_bh = NULL;
808 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
810 if (is_bad_inode(inode)) {
811 mlog(0, "Skipping delete of bad inode\n");
815 if (!ocfs2_inode_is_valid_to_delete(inode)) {
816 /* It's probably not necessary to truncate_inode_pages
817 * here but we do it for safety anyway (it will most
818 * likely be a no-op anyway) */
819 ocfs2_cleanup_delete_inode(inode, 0);
823 /* We want to block signals in delete_inode as the lock and
824 * messaging paths may return us -ERESTARTSYS. Which would
825 * cause us to exit early, resulting in inodes being orphaned
827 sigfillset(&blocked);
828 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
831 ocfs2_cleanup_delete_inode(inode, 1);
835 /* Lock down the inode. This gives us an up to date view of
836 * it's metadata (for verification), and allows us to
837 * serialize delete_inode votes.
839 * Even though we might be doing a truncate, we don't take the
840 * allocation lock here as it won't be needed - nobody will
841 * have the file open.
843 status = ocfs2_meta_lock(inode, NULL, &di_bh, 1);
845 if (status != -ENOENT)
847 ocfs2_cleanup_delete_inode(inode, 0);
851 /* Query the cluster. This will be the final decision made
852 * before we go ahead and wipe the inode. */
853 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
854 if (!wipe || status < 0) {
855 /* Error and inode busy vote both mean we won't be
856 * removing the inode, so they take almost the same
861 /* Someone in the cluster has voted to not wipe this
862 * inode, or it was never completely orphaned. Write
863 * out the pages and exit now. */
864 ocfs2_cleanup_delete_inode(inode, 1);
865 goto bail_unlock_inode;
868 ocfs2_cleanup_delete_inode(inode, 0);
870 status = ocfs2_wipe_inode(inode, di_bh);
872 if (status != -EDEADLK)
874 goto bail_unlock_inode;
877 /* Mark the inode as successfully deleted. This is important
878 * for ocfs2_clear_inode as it will check this flag and skip
879 * any checkpointing work */
880 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
883 ocfs2_meta_unlock(inode, 1);
886 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
894 void ocfs2_clear_inode(struct inode *inode)
897 struct ocfs2_inode_info *oi = OCFS2_I(inode);
904 mlog(0, "Clearing inode: %llu, nlink = %u\n",
905 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
907 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
908 "Inode=%lu\n", inode->i_ino);
910 /* Do these before all the other work so that we don't bounce
911 * the vote thread while waiting to destroy the locks. */
912 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
913 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
914 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
916 /* We very well may get a clear_inode before all an inodes
917 * metadata has hit disk. Of course, we can't drop any cluster
918 * locks until the journal has finished with it. The only
919 * exception here are successfully wiped inodes - their
920 * metadata can now be considered to be part of the system
921 * inodes from which it came. */
922 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
923 ocfs2_checkpoint_inode(inode);
925 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
926 "Clear inode of %llu, inode has io markers\n",
927 (unsigned long long)oi->ip_blkno);
929 ocfs2_extent_map_drop(inode, 0);
930 ocfs2_extent_map_init(inode);
932 status = ocfs2_drop_inode_locks(inode);
936 ocfs2_lock_res_free(&oi->ip_rw_lockres);
937 ocfs2_lock_res_free(&oi->ip_meta_lockres);
938 ocfs2_lock_res_free(&oi->ip_data_lockres);
940 ocfs2_metadata_cache_purge(inode);
942 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
943 "Clear inode of %llu, inode has %u cache items\n",
944 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
946 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
947 "Clear inode of %llu, inode has a bad flag\n",
948 (unsigned long long)oi->ip_blkno);
950 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
951 "Clear inode of %llu, inode is locked\n",
952 (unsigned long long)oi->ip_blkno);
954 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
955 "Clear inode of %llu, io_mutex is locked\n",
956 (unsigned long long)oi->ip_blkno);
957 mutex_unlock(&oi->ip_io_mutex);
960 * down_trylock() returns 0, down_write_trylock() returns 1
963 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
964 "Clear inode of %llu, alloc_sem is locked\n",
965 (unsigned long long)oi->ip_blkno);
966 up_write(&oi->ip_alloc_sem);
968 mlog_bug_on_msg(oi->ip_open_count,
969 "Clear inode of %llu has open count %d\n",
970 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
971 mlog_bug_on_msg(!list_empty(&oi->ip_handle_list),
972 "Clear inode of %llu has non empty handle list\n",
973 (unsigned long long)oi->ip_blkno);
974 mlog_bug_on_msg(oi->ip_handle,
975 "Clear inode of %llu has non empty handle pointer\n",
976 (unsigned long long)oi->ip_blkno);
978 /* Clear all other flags. */
979 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
980 oi->ip_created_trans = 0;
981 oi->ip_last_trans = 0;
982 oi->ip_dir_start_lookup = 0;
989 /* Called under inode_lock, with no more references on the
990 * struct inode, so it's safe here to check the flags field
991 * and to manipulate i_nlink without any other locks. */
992 void ocfs2_drop_inode(struct inode *inode)
994 struct ocfs2_inode_info *oi = OCFS2_I(inode);
998 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
999 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1001 /* Testing ip_orphaned_slot here wouldn't work because we may
1002 * not have gotten a delete_inode vote from any other nodes
1004 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
1005 mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
1009 generic_drop_inode(inode);
1015 * TODO: this should probably be merged into ocfs2_get_block
1017 * However, you now need to pay attention to the cont_prepare_write()
1018 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1019 * expects never to extend).
1021 struct buffer_head *ocfs2_bread(struct inode *inode,
1022 int block, int *err, int reada)
1024 struct buffer_head *bh = NULL;
1027 int readflags = OCFS2_BH_CACHED;
1030 /* only turn this on if we know we can deal with read_block
1031 * returning nothing */
1033 readflags |= OCFS2_BH_READAHEAD;
1036 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1037 i_size_read(inode)) {
1042 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1049 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1069 * This is called from our getattr.
1071 int ocfs2_inode_revalidate(struct dentry *dentry)
1073 struct inode *inode = dentry->d_inode;
1076 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1077 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1080 mlog(0, "eep, no inode!\n");
1085 spin_lock(&OCFS2_I(inode)->ip_lock);
1086 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1087 spin_unlock(&OCFS2_I(inode)->ip_lock);
1088 mlog(0, "inode deleted!\n");
1092 spin_unlock(&OCFS2_I(inode)->ip_lock);
1094 /* Let ocfs2_meta_lock do the work of updating our struct
1096 status = ocfs2_meta_lock(inode, NULL, NULL, 0);
1098 if (status != -ENOENT)
1102 ocfs2_meta_unlock(inode, 0);
1110 * Updates a disk inode from a
1112 * Only takes ip_lock.
1114 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle,
1115 struct inode *inode,
1116 struct buffer_head *bh)
1119 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1121 mlog_entry("(inode %llu)\n",
1122 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1124 status = ocfs2_journal_access(handle, inode, bh,
1125 OCFS2_JOURNAL_ACCESS_WRITE);
1131 spin_lock(&OCFS2_I(inode)->ip_lock);
1132 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1133 spin_unlock(&OCFS2_I(inode)->ip_lock);
1135 fe->i_size = cpu_to_le64(i_size_read(inode));
1136 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1137 fe->i_uid = cpu_to_le32(inode->i_uid);
1138 fe->i_gid = cpu_to_le32(inode->i_gid);
1139 fe->i_mode = cpu_to_le16(inode->i_mode);
1140 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1141 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1142 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1143 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1144 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1145 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1147 status = ocfs2_journal_dirty(handle, bh);
1160 * Updates a struct inode from a disk inode.
1161 * does no i/o, only takes ip_lock.
1163 void ocfs2_refresh_inode(struct inode *inode,
1164 struct ocfs2_dinode *fe)
1166 spin_lock(&OCFS2_I(inode)->ip_lock);
1168 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1169 i_size_write(inode, le64_to_cpu(fe->i_size));
1170 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1171 inode->i_uid = le32_to_cpu(fe->i_uid);
1172 inode->i_gid = le32_to_cpu(fe->i_gid);
1173 inode->i_mode = le16_to_cpu(fe->i_mode);
1174 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1175 inode->i_blocks = 0;
1177 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1178 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1179 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1180 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1181 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1182 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1183 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1185 spin_unlock(&OCFS2_I(inode)->ip_lock);