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 = %"MLFu64")\n", 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 %"MLFu64"\n",
138 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:%"MLFu64")\n", inode, fe->i_size);
227 /* this means that read_inode cannot create a superblock inode
228 * today. change if needed. */
229 if (!OCFS2_IS_VALID_DINODE(fe) ||
230 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
231 mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%"MLFu64", "
232 "signature = %.*s, flags = 0x%x\n",
233 inode->i_ino, le64_to_cpu(fe->i_blkno), 7,
234 fe->i_signature, le32_to_cpu(fe->i_flags));
238 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
239 mlog(ML_ERROR, "file entry generation does not match "
240 "superblock! osb->fs_generation=%x, "
241 "fe->i_fs_generation=%x\n",
242 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
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);
252 inode->i_blksize = (u32)osb->s_clustersize;
254 /* Fast symlinks will have i_size but no allocated clusters. */
255 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
259 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
260 inode->i_mapping->a_ops = &ocfs2_aops;
261 inode->i_flags |= S_NOATIME;
262 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
263 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
264 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
265 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
266 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
267 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
269 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
271 "ip_blkno %"MLFu64" != i_blkno %"MLFu64"!\n",
272 OCFS2_I(inode)->ip_blkno, fe->i_blkno);
274 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
275 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
278 inode->i_ino = ino_from_blkno(inode->i_sb,
279 le64_to_cpu(fe->i_blkno));
281 mlog(0, "blkno = %"MLFu64", ino = %lu, create_ino = %s\n",
282 fe->i_blkno, inode->i_ino, create_ino ? "true" : "false");
284 inode->i_nlink = le16_to_cpu(fe->i_links_count);
286 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
287 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
288 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
289 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
290 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
291 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
292 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
293 /* we can't actually hit this as read_inode can't
294 * handle superblocks today ;-) */
298 switch (inode->i_mode & S_IFMT) {
300 inode->i_fop = &ocfs2_fops;
301 inode->i_op = &ocfs2_file_iops;
302 i_size_write(inode, le64_to_cpu(fe->i_size));
305 inode->i_op = &ocfs2_dir_iops;
306 inode->i_fop = &ocfs2_dops;
307 i_size_write(inode, le64_to_cpu(fe->i_size));
310 if (ocfs2_inode_is_fast_symlink(inode))
311 inode->i_op = &ocfs2_fast_symlink_inode_operations;
313 inode->i_op = &ocfs2_symlink_inode_operations;
314 i_size_write(inode, le64_to_cpu(fe->i_size));
317 inode->i_op = &ocfs2_special_file_iops;
318 init_special_inode(inode, inode->i_mode,
323 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
324 OCFS2_LOCK_TYPE_RW, inode);
325 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
326 OCFS2_LOCK_TYPE_META, inode);
327 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
328 OCFS2_LOCK_TYPE_DATA, inode);
336 static int ocfs2_read_locked_inode(struct inode *inode,
337 struct ocfs2_find_inode_args *args)
339 struct super_block *sb;
340 struct ocfs2_super *osb;
341 struct ocfs2_dinode *fe;
342 struct buffer_head *bh = NULL;
346 mlog_entry("(0x%p, 0x%p)\n", inode, args);
349 if (inode == NULL || inode->i_sb == NULL) {
350 mlog(ML_ERROR, "bad inode\n");
357 mlog(ML_ERROR, "bad inode args\n");
358 make_bad_inode(inode);
362 /* Read the FE off disk. This is safe because the kernel only
363 * does one read_inode2 for a new inode, and if it doesn't
364 * exist yet then nobody can be working on it! */
365 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL);
368 make_bad_inode(inode);
372 fe = (struct ocfs2_dinode *) bh->b_data;
373 if (!OCFS2_IS_VALID_DINODE(fe)) {
374 mlog(ML_ERROR, "Invalid dinode #%"MLFu64": signature = %.*s\n",
375 fe->i_blkno, 7, fe->i_signature);
376 make_bad_inode(inode);
380 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
383 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
384 S_ISBLK(le16_to_cpu(fe->i_mode)))
385 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
388 if (ocfs2_populate_inode(inode, fe, 0) < 0) {
389 mlog(ML_ERROR, "populate inode failed! i_blkno=%"MLFu64", "
390 "i_ino=%lu\n", fe->i_blkno, inode->i_ino);
391 make_bad_inode(inode);
395 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
398 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
410 void ocfs2_sync_blockdev(struct super_block *sb)
412 sync_blockdev(sb->s_bdev);
415 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
417 struct buffer_head *fe_bh)
420 struct ocfs2_journal_handle *handle = NULL;
421 struct ocfs2_truncate_context *tc = NULL;
422 struct ocfs2_dinode *fe;
426 fe = (struct ocfs2_dinode *) fe_bh->b_data;
428 /* zero allocation, zero truncate :) */
432 handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS);
433 if (IS_ERR(handle)) {
434 status = PTR_ERR(handle);
440 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
446 ocfs2_commit_trans(handle);
449 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
455 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
462 ocfs2_commit_trans(handle);
468 static int ocfs2_remove_inode(struct inode *inode,
469 struct buffer_head *di_bh,
470 struct inode *orphan_dir_inode,
471 struct buffer_head *orphan_dir_bh)
474 struct inode *inode_alloc_inode = NULL;
475 struct buffer_head *inode_alloc_bh = NULL;
476 struct ocfs2_journal_handle *handle;
477 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
478 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
481 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
482 le16_to_cpu(di->i_suballoc_slot));
483 if (!inode_alloc_inode) {
489 mutex_lock(&inode_alloc_inode->i_mutex);
490 status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1);
492 mutex_unlock(&inode_alloc_inode->i_mutex);
498 handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS);
499 if (IS_ERR(handle)) {
500 status = PTR_ERR(handle);
505 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
512 /* set the inodes dtime */
513 status = ocfs2_journal_access(handle, inode, di_bh,
514 OCFS2_JOURNAL_ACCESS_WRITE);
520 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
521 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
523 status = ocfs2_journal_dirty(handle, di_bh);
529 ocfs2_remove_from_cache(inode, di_bh);
531 status = ocfs2_free_dinode(handle, inode_alloc_inode,
537 ocfs2_commit_trans(handle);
539 ocfs2_meta_unlock(inode_alloc_inode, 1);
540 mutex_unlock(&inode_alloc_inode->i_mutex);
541 brelse(inode_alloc_bh);
543 iput(inode_alloc_inode);
549 * Serialize with orphan dir recovery. If the process doing
550 * recovery on this orphan dir does an iget() with the dir
551 * i_mutex held, we'll deadlock here. Instead we detect this
552 * and exit early - recovery will wipe this inode for us.
554 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
559 spin_lock(&osb->osb_lock);
560 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
561 mlog(0, "Recovery is happening on orphan dir %d, will skip "
562 "this inode\n", slot);
566 /* This signals to the orphan recovery process that it should
567 * wait for us to handle the wipe. */
568 osb->osb_orphan_wipes[slot]++;
570 spin_unlock(&osb->osb_lock);
574 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
577 spin_lock(&osb->osb_lock);
578 osb->osb_orphan_wipes[slot]--;
579 spin_unlock(&osb->osb_lock);
581 wake_up(&osb->osb_wipe_event);
584 static int ocfs2_wipe_inode(struct inode *inode,
585 struct buffer_head *di_bh)
587 int status, orphaned_slot;
588 struct inode *orphan_dir_inode = NULL;
589 struct buffer_head *orphan_dir_bh = NULL;
590 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
592 /* We've already voted on this so it should be readonly - no
593 * spinlock needed. */
594 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
596 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
600 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
601 ORPHAN_DIR_SYSTEM_INODE,
603 if (!orphan_dir_inode) {
609 /* Lock the orphan dir. The lock will be held for the entire
610 * delete_inode operation. We do this now to avoid races with
611 * recovery completion on other nodes. */
612 mutex_lock(&orphan_dir_inode->i_mutex);
613 status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
615 mutex_unlock(&orphan_dir_inode->i_mutex);
621 /* we do this while holding the orphan dir lock because we
622 * don't want recovery being run from another node to vote for
623 * an inode delete on us -- this will result in two nodes
624 * truncating the same file! */
625 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
628 goto bail_unlock_dir;
631 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
637 ocfs2_meta_unlock(orphan_dir_inode, 1);
638 mutex_unlock(&orphan_dir_inode->i_mutex);
639 brelse(orphan_dir_bh);
641 iput(orphan_dir_inode);
642 ocfs2_signal_wipe_completion(osb, orphaned_slot);
647 /* There is a series of simple checks that should be done before a
648 * vote is even considered. Encapsulate those in this function. */
649 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
652 struct ocfs2_inode_info *oi = OCFS2_I(inode);
653 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
655 /* We shouldn't be getting here for the root directory
657 if (inode == osb->root_inode) {
658 mlog(ML_ERROR, "Skipping delete of root inode.\n");
662 /* If we're coming from process_vote we can't go into our own
663 * voting [hello, deadlock city!], so unforuntately we just
664 * have to skip deleting this guy. That's OK though because
665 * the node who's doing the actual deleting should handle it
667 if (current == osb->vote_task) {
668 mlog(0, "Skipping delete of %lu because we're currently "
669 "in process_vote\n", inode->i_ino);
673 spin_lock(&oi->ip_lock);
674 /* OCFS2 *never* deletes system files. This should technically
675 * never get here as system file inodes should always have a
676 * positive link count. */
677 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
678 mlog(ML_ERROR, "Skipping delete of system file %"MLFu64".\n",
683 /* If we have voted "yes" on the wipe of this inode for
684 * another node, it will be marked here so we can safely skip
685 * it. Recovery will cleanup any inodes we might inadvertantly
687 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
688 mlog(0, "Skipping delete of %lu because another node "
689 "has done this for us.\n", inode->i_ino);
695 spin_unlock(&oi->ip_lock);
700 /* Query the cluster to determine whether we should wipe an inode from
703 * Requires the inode to have the cluster lock. */
704 static int ocfs2_query_inode_wipe(struct inode *inode,
705 struct buffer_head *di_bh,
709 struct ocfs2_inode_info *oi = OCFS2_I(inode);
710 struct ocfs2_dinode *di;
714 /* While we were waiting for the cluster lock in
715 * ocfs2_delete_inode, another node might have asked to delete
716 * the inode. Recheck our flags to catch this. */
717 if (!ocfs2_inode_is_valid_to_delete(inode)) {
718 mlog(0, "Skipping delete of %"MLFu64" because flags changed\n",
723 /* Now that we have an up to date inode, we can double check
725 if (inode->i_nlink) {
726 mlog(0, "Skipping delete of %"MLFu64" because nlink = %u\n",
727 oi->ip_blkno, inode->i_nlink);
731 /* Do some basic inode verification... */
732 di = (struct ocfs2_dinode *) di_bh->b_data;
733 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
734 /* for lack of a better error? */
737 "Inode %"MLFu64" (on-disk %"MLFu64") not orphaned! "
738 "Disk flags 0x%x, inode flags 0x%x\n",
739 oi->ip_blkno, di->i_blkno, di->i_flags, oi->ip_flags);
743 /* has someone already deleted us?! baaad... */
750 status = ocfs2_request_delete_vote(inode);
751 /* -EBUSY means that other nodes are still using the
752 * inode. We're done here though, so avoid doing anything on
753 * disk and let them worry about deleting it. */
754 if (status == -EBUSY) {
756 mlog(0, "Skipping delete of %"MLFu64" because it is in use on"
757 "other nodes\n", oi->ip_blkno);
765 spin_lock(&oi->ip_lock);
766 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
767 /* Nobody knew which slot this inode was orphaned
768 * into. This may happen during node death and
769 * recovery knows how to clean it up so we can safely
770 * ignore this inode for now on. */
771 mlog(0, "Nobody knew where inode %"MLFu64" was orphaned!\n",
776 mlog(0, "Inode %"MLFu64" is ok to wipe from orphan dir %d\n",
777 oi->ip_blkno, oi->ip_orphaned_slot);
779 spin_unlock(&oi->ip_lock);
785 /* Support function for ocfs2_delete_inode. Will help us keep the
786 * inode data in a consistent state for clear_inode. Always truncates
787 * pages, optionally sync's them first. */
788 static void ocfs2_cleanup_delete_inode(struct inode *inode,
791 mlog(0, "Cleanup inode %"MLFu64", sync = %d\n",
792 OCFS2_I(inode)->ip_blkno, sync_data);
794 write_inode_now(inode, 1);
795 truncate_inode_pages(&inode->i_data, 0);
798 void ocfs2_delete_inode(struct inode *inode)
801 sigset_t blocked, oldset;
802 struct buffer_head *di_bh = NULL;
804 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
806 if (is_bad_inode(inode)) {
807 mlog(0, "Skipping delete of bad inode\n");
811 if (!ocfs2_inode_is_valid_to_delete(inode)) {
812 /* It's probably not necessary to truncate_inode_pages
813 * here but we do it for safety anyway (it will most
814 * likely be a no-op anyway) */
815 ocfs2_cleanup_delete_inode(inode, 0);
819 /* We want to block signals in delete_inode as the lock and
820 * messaging paths may return us -ERESTARTSYS. Which would
821 * cause us to exit early, resulting in inodes being orphaned
823 sigfillset(&blocked);
824 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
827 ocfs2_cleanup_delete_inode(inode, 1);
831 /* Lock down the inode. This gives us an up to date view of
832 * it's metadata (for verification), and allows us to
833 * serialize delete_inode votes.
835 * Even though we might be doing a truncate, we don't take the
836 * allocation lock here as it won't be needed - nobody will
837 * have the file open.
839 status = ocfs2_meta_lock(inode, NULL, &di_bh, 1);
841 if (status != -ENOENT)
843 ocfs2_cleanup_delete_inode(inode, 0);
847 /* Query the cluster. This will be the final decision made
848 * before we go ahead and wipe the inode. */
849 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
850 if (!wipe || status < 0) {
851 /* Error and inode busy vote both mean we won't be
852 * removing the inode, so they take almost the same
857 /* Someone in the cluster has voted to not wipe this
858 * inode, or it was never completely orphaned. Write
859 * out the pages and exit now. */
860 ocfs2_cleanup_delete_inode(inode, 1);
861 goto bail_unlock_inode;
864 ocfs2_cleanup_delete_inode(inode, 0);
866 status = ocfs2_wipe_inode(inode, di_bh);
868 if (status != -EDEADLK)
870 goto bail_unlock_inode;
873 /* Mark the inode as successfully deleted. This is important
874 * for ocfs2_clear_inode as it will check this flag and skip
875 * any checkpointing work */
876 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
879 ocfs2_meta_unlock(inode, 1);
882 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
890 void ocfs2_clear_inode(struct inode *inode)
893 struct ocfs2_inode_info *oi = OCFS2_I(inode);
900 mlog(0, "Clearing inode: %"MLFu64", nlink = %u\n",
901 OCFS2_I(inode)->ip_blkno, inode->i_nlink);
903 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
904 "Inode=%lu\n", inode->i_ino);
906 /* Do these before all the other work so that we don't bounce
907 * the vote thread while waiting to destroy the locks. */
908 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
909 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
910 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
912 /* We very well may get a clear_inode before all an inodes
913 * metadata has hit disk. Of course, we can't drop any cluster
914 * locks until the journal has finished with it. The only
915 * exception here are successfully wiped inodes - their
916 * metadata can now be considered to be part of the system
917 * inodes from which it came. */
918 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
919 ocfs2_checkpoint_inode(inode);
921 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
922 "Clear inode of %"MLFu64", inode has io markers\n",
925 ocfs2_extent_map_drop(inode, 0);
926 ocfs2_extent_map_init(inode);
928 status = ocfs2_drop_inode_locks(inode);
932 ocfs2_lock_res_free(&oi->ip_rw_lockres);
933 ocfs2_lock_res_free(&oi->ip_meta_lockres);
934 ocfs2_lock_res_free(&oi->ip_data_lockres);
936 ocfs2_metadata_cache_purge(inode);
938 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
939 "Clear inode of %"MLFu64", inode has %u cache items\n",
940 oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
942 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
943 "Clear inode of %"MLFu64", inode has a bad flag\n",
946 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
947 "Clear inode of %"MLFu64", inode is locked\n",
950 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
951 "Clear inode of %"MLFu64", io_mutex is locked\n",
953 mutex_unlock(&oi->ip_io_mutex);
956 * down_trylock() returns 0, down_write_trylock() returns 1
959 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
960 "Clear inode of %"MLFu64", alloc_sem is locked\n",
962 up_write(&oi->ip_alloc_sem);
964 mlog_bug_on_msg(oi->ip_open_count,
965 "Clear inode of %"MLFu64" has open count %d\n",
966 oi->ip_blkno, oi->ip_open_count);
967 mlog_bug_on_msg(!list_empty(&oi->ip_handle_list),
968 "Clear inode of %"MLFu64" has non empty handle list\n",
970 mlog_bug_on_msg(oi->ip_handle,
971 "Clear inode of %"MLFu64" has non empty handle pointer\n",
974 /* Clear all other flags. */
975 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
976 oi->ip_created_trans = 0;
977 oi->ip_last_trans = 0;
978 oi->ip_dir_start_lookup = 0;
985 /* Called under inode_lock, with no more references on the
986 * struct inode, so it's safe here to check the flags field
987 * and to manipulate i_nlink without any other locks. */
988 void ocfs2_drop_inode(struct inode *inode)
990 struct ocfs2_inode_info *oi = OCFS2_I(inode);
994 mlog(0, "Drop inode %"MLFu64", nlink = %u, ip_flags = 0x%x\n",
995 oi->ip_blkno, inode->i_nlink, oi->ip_flags);
997 /* Testing ip_orphaned_slot here wouldn't work because we may
998 * not have gotten a delete_inode vote from any other nodes
1000 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
1001 mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
1005 generic_drop_inode(inode);
1011 * TODO: this should probably be merged into ocfs2_get_block
1013 * However, you now need to pay attention to the cont_prepare_write()
1014 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1015 * expects never to extend).
1017 struct buffer_head *ocfs2_bread(struct inode *inode,
1018 int block, int *err, int reada)
1020 struct buffer_head *bh = NULL;
1023 int readflags = OCFS2_BH_CACHED;
1026 /* only turn this on if we know we can deal with read_block
1027 * returning nothing */
1029 readflags |= OCFS2_BH_READAHEAD;
1032 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1033 i_size_read(inode)) {
1038 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1045 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1065 * This is called from our getattr.
1067 int ocfs2_inode_revalidate(struct dentry *dentry)
1069 struct inode *inode = dentry->d_inode;
1072 mlog_entry("(inode = 0x%p, ino = %"MLFu64")\n", inode,
1073 inode ? OCFS2_I(inode)->ip_blkno : 0ULL);
1076 mlog(0, "eep, no inode!\n");
1081 spin_lock(&OCFS2_I(inode)->ip_lock);
1082 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1083 spin_unlock(&OCFS2_I(inode)->ip_lock);
1084 mlog(0, "inode deleted!\n");
1088 spin_unlock(&OCFS2_I(inode)->ip_lock);
1090 /* Let ocfs2_meta_lock do the work of updating our struct
1092 status = ocfs2_meta_lock(inode, NULL, NULL, 0);
1094 if (status != -ENOENT)
1098 ocfs2_meta_unlock(inode, 0);
1106 * Updates a disk inode from a
1108 * Only takes ip_lock.
1110 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle,
1111 struct inode *inode,
1112 struct buffer_head *bh)
1115 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1117 mlog_entry("(inode %"MLFu64")\n", OCFS2_I(inode)->ip_blkno);
1119 status = ocfs2_journal_access(handle, inode, bh,
1120 OCFS2_JOURNAL_ACCESS_WRITE);
1126 spin_lock(&OCFS2_I(inode)->ip_lock);
1127 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1128 spin_unlock(&OCFS2_I(inode)->ip_lock);
1130 fe->i_size = cpu_to_le64(i_size_read(inode));
1131 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1132 fe->i_uid = cpu_to_le32(inode->i_uid);
1133 fe->i_gid = cpu_to_le32(inode->i_gid);
1134 fe->i_mode = cpu_to_le16(inode->i_mode);
1135 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1136 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1137 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1138 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1139 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1140 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1142 status = ocfs2_journal_dirty(handle, bh);
1155 * Updates a struct inode from a disk inode.
1156 * does no i/o, only takes ip_lock.
1158 void ocfs2_refresh_inode(struct inode *inode,
1159 struct ocfs2_dinode *fe)
1161 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1163 spin_lock(&OCFS2_I(inode)->ip_lock);
1165 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1166 i_size_write(inode, le64_to_cpu(fe->i_size));
1167 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1168 inode->i_uid = le32_to_cpu(fe->i_uid);
1169 inode->i_gid = le32_to_cpu(fe->i_gid);
1170 inode->i_mode = le16_to_cpu(fe->i_mode);
1171 inode->i_blksize = (u32) osb->s_clustersize;
1172 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1173 inode->i_blocks = 0;
1175 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1176 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1177 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1178 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1179 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1180 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1181 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1183 spin_unlock(&OCFS2_I(inode)->ip_lock);