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);
254 inode->i_blksize = (u32)osb->s_clustersize;
256 /* Fast symlinks will have i_size but no allocated clusters. */
257 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
261 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
262 inode->i_mapping->a_ops = &ocfs2_aops;
263 inode->i_flags |= S_NOATIME;
264 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
265 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
266 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
267 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
268 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
269 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
271 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
273 "ip_blkno %llu != i_blkno %llu!\n",
274 (unsigned long long)OCFS2_I(inode)->ip_blkno,
275 (unsigned long long)fe->i_blkno);
277 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
278 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
281 inode->i_ino = ino_from_blkno(inode->i_sb,
282 le64_to_cpu(fe->i_blkno));
284 mlog(0, "blkno = %llu, ino = %lu, create_ino = %s\n",
285 (unsigned long long)fe->i_blkno, inode->i_ino, create_ino ? "true" : "false");
287 inode->i_nlink = le16_to_cpu(fe->i_links_count);
289 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
290 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
291 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
292 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
293 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
294 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
295 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
296 /* we can't actually hit this as read_inode can't
297 * handle superblocks today ;-) */
301 switch (inode->i_mode & S_IFMT) {
303 inode->i_fop = &ocfs2_fops;
304 inode->i_op = &ocfs2_file_iops;
305 i_size_write(inode, le64_to_cpu(fe->i_size));
308 inode->i_op = &ocfs2_dir_iops;
309 inode->i_fop = &ocfs2_dops;
310 i_size_write(inode, le64_to_cpu(fe->i_size));
313 if (ocfs2_inode_is_fast_symlink(inode))
314 inode->i_op = &ocfs2_fast_symlink_inode_operations;
316 inode->i_op = &ocfs2_symlink_inode_operations;
317 i_size_write(inode, le64_to_cpu(fe->i_size));
320 inode->i_op = &ocfs2_special_file_iops;
321 init_special_inode(inode, inode->i_mode,
326 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
327 OCFS2_LOCK_TYPE_RW, inode);
328 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
329 OCFS2_LOCK_TYPE_META, inode);
330 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
331 OCFS2_LOCK_TYPE_DATA, inode);
340 static int ocfs2_read_locked_inode(struct inode *inode,
341 struct ocfs2_find_inode_args *args)
343 struct super_block *sb;
344 struct ocfs2_super *osb;
345 struct ocfs2_dinode *fe;
346 struct buffer_head *bh = NULL;
350 mlog_entry("(0x%p, 0x%p)\n", inode, args);
353 if (inode == NULL || inode->i_sb == NULL) {
354 mlog(ML_ERROR, "bad inode\n");
361 mlog(ML_ERROR, "bad inode args\n");
362 make_bad_inode(inode);
366 /* Read the FE off disk. This is safe because the kernel only
367 * does one read_inode2 for a new inode, and if it doesn't
368 * exist yet then nobody can be working on it! */
369 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL);
372 make_bad_inode(inode);
376 fe = (struct ocfs2_dinode *) bh->b_data;
377 if (!OCFS2_IS_VALID_DINODE(fe)) {
378 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
379 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
380 make_bad_inode(inode);
384 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
387 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
388 S_ISBLK(le16_to_cpu(fe->i_mode)))
389 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
392 if (ocfs2_populate_inode(inode, fe, 0) < 0) {
393 mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
394 (unsigned long long)fe->i_blkno, inode->i_ino);
395 make_bad_inode(inode);
399 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
402 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
414 void ocfs2_sync_blockdev(struct super_block *sb)
416 sync_blockdev(sb->s_bdev);
419 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
421 struct buffer_head *fe_bh)
424 struct ocfs2_journal_handle *handle = NULL;
425 struct ocfs2_truncate_context *tc = NULL;
426 struct ocfs2_dinode *fe;
430 fe = (struct ocfs2_dinode *) fe_bh->b_data;
432 /* zero allocation, zero truncate :) */
436 handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS);
437 if (IS_ERR(handle)) {
438 status = PTR_ERR(handle);
444 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
450 ocfs2_commit_trans(handle);
453 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
459 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
466 ocfs2_commit_trans(handle);
472 static int ocfs2_remove_inode(struct inode *inode,
473 struct buffer_head *di_bh,
474 struct inode *orphan_dir_inode,
475 struct buffer_head *orphan_dir_bh)
478 struct inode *inode_alloc_inode = NULL;
479 struct buffer_head *inode_alloc_bh = NULL;
480 struct ocfs2_journal_handle *handle;
481 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
482 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
485 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
486 le16_to_cpu(di->i_suballoc_slot));
487 if (!inode_alloc_inode) {
493 mutex_lock(&inode_alloc_inode->i_mutex);
494 status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1);
496 mutex_unlock(&inode_alloc_inode->i_mutex);
502 handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS);
503 if (IS_ERR(handle)) {
504 status = PTR_ERR(handle);
509 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
516 /* set the inodes dtime */
517 status = ocfs2_journal_access(handle, inode, di_bh,
518 OCFS2_JOURNAL_ACCESS_WRITE);
524 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
525 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
527 status = ocfs2_journal_dirty(handle, di_bh);
533 ocfs2_remove_from_cache(inode, di_bh);
535 status = ocfs2_free_dinode(handle, inode_alloc_inode,
541 ocfs2_commit_trans(handle);
543 ocfs2_meta_unlock(inode_alloc_inode, 1);
544 mutex_unlock(&inode_alloc_inode->i_mutex);
545 brelse(inode_alloc_bh);
547 iput(inode_alloc_inode);
553 * Serialize with orphan dir recovery. If the process doing
554 * recovery on this orphan dir does an iget() with the dir
555 * i_mutex held, we'll deadlock here. Instead we detect this
556 * and exit early - recovery will wipe this inode for us.
558 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
563 spin_lock(&osb->osb_lock);
564 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
565 mlog(0, "Recovery is happening on orphan dir %d, will skip "
566 "this inode\n", slot);
570 /* This signals to the orphan recovery process that it should
571 * wait for us to handle the wipe. */
572 osb->osb_orphan_wipes[slot]++;
574 spin_unlock(&osb->osb_lock);
578 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
581 spin_lock(&osb->osb_lock);
582 osb->osb_orphan_wipes[slot]--;
583 spin_unlock(&osb->osb_lock);
585 wake_up(&osb->osb_wipe_event);
588 static int ocfs2_wipe_inode(struct inode *inode,
589 struct buffer_head *di_bh)
591 int status, orphaned_slot;
592 struct inode *orphan_dir_inode = NULL;
593 struct buffer_head *orphan_dir_bh = NULL;
594 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
596 /* We've already voted on this so it should be readonly - no
597 * spinlock needed. */
598 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
600 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
604 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
605 ORPHAN_DIR_SYSTEM_INODE,
607 if (!orphan_dir_inode) {
613 /* Lock the orphan dir. The lock will be held for the entire
614 * delete_inode operation. We do this now to avoid races with
615 * recovery completion on other nodes. */
616 mutex_lock(&orphan_dir_inode->i_mutex);
617 status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
619 mutex_unlock(&orphan_dir_inode->i_mutex);
625 /* we do this while holding the orphan dir lock because we
626 * don't want recovery being run from another node to vote for
627 * an inode delete on us -- this will result in two nodes
628 * truncating the same file! */
629 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
632 goto bail_unlock_dir;
635 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
641 ocfs2_meta_unlock(orphan_dir_inode, 1);
642 mutex_unlock(&orphan_dir_inode->i_mutex);
643 brelse(orphan_dir_bh);
645 iput(orphan_dir_inode);
646 ocfs2_signal_wipe_completion(osb, orphaned_slot);
651 /* There is a series of simple checks that should be done before a
652 * vote is even considered. Encapsulate those in this function. */
653 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
656 struct ocfs2_inode_info *oi = OCFS2_I(inode);
657 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
659 /* We shouldn't be getting here for the root directory
661 if (inode == osb->root_inode) {
662 mlog(ML_ERROR, "Skipping delete of root inode.\n");
666 /* If we're coming from process_vote we can't go into our own
667 * voting [hello, deadlock city!], so unforuntately we just
668 * have to skip deleting this guy. That's OK though because
669 * the node who's doing the actual deleting should handle it
671 if (current == osb->vote_task) {
672 mlog(0, "Skipping delete of %lu because we're currently "
673 "in process_vote\n", inode->i_ino);
677 spin_lock(&oi->ip_lock);
678 /* OCFS2 *never* deletes system files. This should technically
679 * never get here as system file inodes should always have a
680 * positive link count. */
681 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
682 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
683 (unsigned long long)oi->ip_blkno);
687 /* If we have voted "yes" on the wipe of this inode for
688 * another node, it will be marked here so we can safely skip
689 * it. Recovery will cleanup any inodes we might inadvertantly
691 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
692 mlog(0, "Skipping delete of %lu because another node "
693 "has done this for us.\n", inode->i_ino);
699 spin_unlock(&oi->ip_lock);
704 /* Query the cluster to determine whether we should wipe an inode from
707 * Requires the inode to have the cluster lock. */
708 static int ocfs2_query_inode_wipe(struct inode *inode,
709 struct buffer_head *di_bh,
713 struct ocfs2_inode_info *oi = OCFS2_I(inode);
714 struct ocfs2_dinode *di;
718 /* While we were waiting for the cluster lock in
719 * ocfs2_delete_inode, another node might have asked to delete
720 * the inode. Recheck our flags to catch this. */
721 if (!ocfs2_inode_is_valid_to_delete(inode)) {
722 mlog(0, "Skipping delete of %llu because flags changed\n",
723 (unsigned long long)oi->ip_blkno);
727 /* Now that we have an up to date inode, we can double check
729 if (inode->i_nlink) {
730 mlog(0, "Skipping delete of %llu because nlink = %u\n",
731 (unsigned long long)oi->ip_blkno, inode->i_nlink);
735 /* Do some basic inode verification... */
736 di = (struct ocfs2_dinode *) di_bh->b_data;
737 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
738 /* for lack of a better error? */
741 "Inode %llu (on-disk %llu) not orphaned! "
742 "Disk flags 0x%x, inode flags 0x%x\n",
743 (unsigned long long)oi->ip_blkno,
744 (unsigned long long)di->i_blkno, di->i_flags,
749 /* has someone already deleted us?! baaad... */
756 status = ocfs2_request_delete_vote(inode);
757 /* -EBUSY means that other nodes are still using the
758 * inode. We're done here though, so avoid doing anything on
759 * disk and let them worry about deleting it. */
760 if (status == -EBUSY) {
762 mlog(0, "Skipping delete of %llu because it is in use on"
763 "other nodes\n", (unsigned long long)oi->ip_blkno);
771 spin_lock(&oi->ip_lock);
772 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
773 /* Nobody knew which slot this inode was orphaned
774 * into. This may happen during node death and
775 * recovery knows how to clean it up so we can safely
776 * ignore this inode for now on. */
777 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
778 (unsigned long long)oi->ip_blkno);
782 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
783 (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
785 spin_unlock(&oi->ip_lock);
791 /* Support function for ocfs2_delete_inode. Will help us keep the
792 * inode data in a consistent state for clear_inode. Always truncates
793 * pages, optionally sync's them first. */
794 static void ocfs2_cleanup_delete_inode(struct inode *inode,
797 mlog(0, "Cleanup inode %llu, sync = %d\n",
798 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
800 write_inode_now(inode, 1);
801 truncate_inode_pages(&inode->i_data, 0);
804 void ocfs2_delete_inode(struct inode *inode)
807 sigset_t blocked, oldset;
808 struct buffer_head *di_bh = NULL;
810 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
812 if (is_bad_inode(inode)) {
813 mlog(0, "Skipping delete of bad inode\n");
817 if (!ocfs2_inode_is_valid_to_delete(inode)) {
818 /* It's probably not necessary to truncate_inode_pages
819 * here but we do it for safety anyway (it will most
820 * likely be a no-op anyway) */
821 ocfs2_cleanup_delete_inode(inode, 0);
825 /* We want to block signals in delete_inode as the lock and
826 * messaging paths may return us -ERESTARTSYS. Which would
827 * cause us to exit early, resulting in inodes being orphaned
829 sigfillset(&blocked);
830 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
833 ocfs2_cleanup_delete_inode(inode, 1);
837 /* Lock down the inode. This gives us an up to date view of
838 * it's metadata (for verification), and allows us to
839 * serialize delete_inode votes.
841 * Even though we might be doing a truncate, we don't take the
842 * allocation lock here as it won't be needed - nobody will
843 * have the file open.
845 status = ocfs2_meta_lock(inode, NULL, &di_bh, 1);
847 if (status != -ENOENT)
849 ocfs2_cleanup_delete_inode(inode, 0);
853 /* Query the cluster. This will be the final decision made
854 * before we go ahead and wipe the inode. */
855 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
856 if (!wipe || status < 0) {
857 /* Error and inode busy vote both mean we won't be
858 * removing the inode, so they take almost the same
863 /* Someone in the cluster has voted to not wipe this
864 * inode, or it was never completely orphaned. Write
865 * out the pages and exit now. */
866 ocfs2_cleanup_delete_inode(inode, 1);
867 goto bail_unlock_inode;
870 ocfs2_cleanup_delete_inode(inode, 0);
872 status = ocfs2_wipe_inode(inode, di_bh);
874 if (status != -EDEADLK)
876 goto bail_unlock_inode;
879 /* Mark the inode as successfully deleted. This is important
880 * for ocfs2_clear_inode as it will check this flag and skip
881 * any checkpointing work */
882 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
885 ocfs2_meta_unlock(inode, 1);
888 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
896 void ocfs2_clear_inode(struct inode *inode)
899 struct ocfs2_inode_info *oi = OCFS2_I(inode);
906 mlog(0, "Clearing inode: %llu, nlink = %u\n",
907 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
909 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
910 "Inode=%lu\n", inode->i_ino);
912 /* Do these before all the other work so that we don't bounce
913 * the vote thread while waiting to destroy the locks. */
914 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
915 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
916 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
918 /* We very well may get a clear_inode before all an inodes
919 * metadata has hit disk. Of course, we can't drop any cluster
920 * locks until the journal has finished with it. The only
921 * exception here are successfully wiped inodes - their
922 * metadata can now be considered to be part of the system
923 * inodes from which it came. */
924 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
925 ocfs2_checkpoint_inode(inode);
927 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
928 "Clear inode of %llu, inode has io markers\n",
929 (unsigned long long)oi->ip_blkno);
931 ocfs2_extent_map_drop(inode, 0);
932 ocfs2_extent_map_init(inode);
934 status = ocfs2_drop_inode_locks(inode);
938 ocfs2_lock_res_free(&oi->ip_rw_lockres);
939 ocfs2_lock_res_free(&oi->ip_meta_lockres);
940 ocfs2_lock_res_free(&oi->ip_data_lockres);
942 ocfs2_metadata_cache_purge(inode);
944 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
945 "Clear inode of %llu, inode has %u cache items\n",
946 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
948 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
949 "Clear inode of %llu, inode has a bad flag\n",
950 (unsigned long long)oi->ip_blkno);
952 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
953 "Clear inode of %llu, inode is locked\n",
954 (unsigned long long)oi->ip_blkno);
956 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
957 "Clear inode of %llu, io_mutex is locked\n",
958 (unsigned long long)oi->ip_blkno);
959 mutex_unlock(&oi->ip_io_mutex);
962 * down_trylock() returns 0, down_write_trylock() returns 1
965 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
966 "Clear inode of %llu, alloc_sem is locked\n",
967 (unsigned long long)oi->ip_blkno);
968 up_write(&oi->ip_alloc_sem);
970 mlog_bug_on_msg(oi->ip_open_count,
971 "Clear inode of %llu has open count %d\n",
972 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
973 mlog_bug_on_msg(!list_empty(&oi->ip_handle_list),
974 "Clear inode of %llu has non empty handle list\n",
975 (unsigned long long)oi->ip_blkno);
976 mlog_bug_on_msg(oi->ip_handle,
977 "Clear inode of %llu has non empty handle pointer\n",
978 (unsigned long long)oi->ip_blkno);
980 /* Clear all other flags. */
981 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
982 oi->ip_created_trans = 0;
983 oi->ip_last_trans = 0;
984 oi->ip_dir_start_lookup = 0;
991 /* Called under inode_lock, with no more references on the
992 * struct inode, so it's safe here to check the flags field
993 * and to manipulate i_nlink without any other locks. */
994 void ocfs2_drop_inode(struct inode *inode)
996 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1000 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1001 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1003 /* Testing ip_orphaned_slot here wouldn't work because we may
1004 * not have gotten a delete_inode vote from any other nodes
1006 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
1007 mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
1011 generic_drop_inode(inode);
1017 * TODO: this should probably be merged into ocfs2_get_block
1019 * However, you now need to pay attention to the cont_prepare_write()
1020 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1021 * expects never to extend).
1023 struct buffer_head *ocfs2_bread(struct inode *inode,
1024 int block, int *err, int reada)
1026 struct buffer_head *bh = NULL;
1029 int readflags = OCFS2_BH_CACHED;
1032 /* only turn this on if we know we can deal with read_block
1033 * returning nothing */
1035 readflags |= OCFS2_BH_READAHEAD;
1038 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1039 i_size_read(inode)) {
1044 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1051 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1071 * This is called from our getattr.
1073 int ocfs2_inode_revalidate(struct dentry *dentry)
1075 struct inode *inode = dentry->d_inode;
1078 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1079 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1082 mlog(0, "eep, no inode!\n");
1087 spin_lock(&OCFS2_I(inode)->ip_lock);
1088 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1089 spin_unlock(&OCFS2_I(inode)->ip_lock);
1090 mlog(0, "inode deleted!\n");
1094 spin_unlock(&OCFS2_I(inode)->ip_lock);
1096 /* Let ocfs2_meta_lock do the work of updating our struct
1098 status = ocfs2_meta_lock(inode, NULL, NULL, 0);
1100 if (status != -ENOENT)
1104 ocfs2_meta_unlock(inode, 0);
1112 * Updates a disk inode from a
1114 * Only takes ip_lock.
1116 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle,
1117 struct inode *inode,
1118 struct buffer_head *bh)
1121 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1123 mlog_entry("(inode %llu)\n",
1124 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1126 status = ocfs2_journal_access(handle, inode, bh,
1127 OCFS2_JOURNAL_ACCESS_WRITE);
1133 spin_lock(&OCFS2_I(inode)->ip_lock);
1134 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1135 spin_unlock(&OCFS2_I(inode)->ip_lock);
1137 fe->i_size = cpu_to_le64(i_size_read(inode));
1138 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1139 fe->i_uid = cpu_to_le32(inode->i_uid);
1140 fe->i_gid = cpu_to_le32(inode->i_gid);
1141 fe->i_mode = cpu_to_le16(inode->i_mode);
1142 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1143 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1144 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1145 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1146 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1147 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1149 status = ocfs2_journal_dirty(handle, bh);
1162 * Updates a struct inode from a disk inode.
1163 * does no i/o, only takes ip_lock.
1165 void ocfs2_refresh_inode(struct inode *inode,
1166 struct ocfs2_dinode *fe)
1168 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1170 spin_lock(&OCFS2_I(inode)->ip_lock);
1172 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1173 i_size_write(inode, le64_to_cpu(fe->i_size));
1174 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1175 inode->i_uid = le32_to_cpu(fe->i_uid);
1176 inode->i_gid = le32_to_cpu(fe->i_gid);
1177 inode->i_mode = le16_to_cpu(fe->i_mode);
1178 inode->i_blksize = (u32) osb->s_clustersize;
1179 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1180 inode->i_blocks = 0;
1182 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1183 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1184 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1185 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1186 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1187 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1188 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1190 spin_unlock(&OCFS2_I(inode)->ip_lock);