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>
32 #include <linux/vs_tag.h>
34 #include <asm/byteorder.h>
36 #define MLOG_MASK_PREFIX ML_INODE
37 #include <cluster/masklog.h>
43 #include "extent_map.h"
45 #include "heartbeat.h"
57 #include "buffer_head_io.h"
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 void ocfs2_set_inode_flags(struct inode *inode)
76 unsigned int flags = OCFS2_I(inode)->ip_attr;
78 inode->i_flags &= ~(S_IMMUTABLE |
79 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
81 if (flags & OCFS2_IMMUTABLE_FL)
82 inode->i_flags |= S_IMMUTABLE;
83 if (flags & OCFS2_IUNLINK_FL)
84 inode->i_flags |= S_IUNLINK;
85 if (flags & OCFS2_BARRIER_FL)
86 inode->i_flags |= S_BARRIER;
88 if (flags & OCFS2_SYNC_FL)
89 inode->i_flags |= S_SYNC;
90 if (flags & OCFS2_APPEND_FL)
91 inode->i_flags |= S_APPEND;
92 if (flags & OCFS2_NOATIME_FL)
93 inode->i_flags |= S_NOATIME;
94 if (flags & OCFS2_DIRSYNC_FL)
95 inode->i_flags |= S_DIRSYNC;
98 int ocfs2_sync_flags(struct inode *inode)
100 unsigned int oldflags, newflags;
102 oldflags = OCFS2_I(inode)->ip_flags;
103 newflags = oldflags & ~(OCFS2_IMMUTABLE_FL |
104 OCFS2_IUNLINK_FL | OCFS2_BARRIER_FL);
106 if (IS_IMMUTABLE(inode))
107 newflags |= OCFS2_IMMUTABLE_FL;
108 if (IS_IUNLINK(inode))
109 newflags |= OCFS2_IUNLINK_FL;
110 if (IS_BARRIER(inode))
111 newflags |= OCFS2_BARRIER_FL;
113 if (oldflags ^ newflags)
114 return ocfs2_set_inode_attr(inode,
115 newflags, OCFS2_FL_MASK);
119 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
123 struct ocfs2_find_inode_args args;
125 /* ocfs2_ilookup_for_vote should *only* be called from the
127 BUG_ON(current != osb->vote_task);
129 args.fi_blkno = blkno;
130 args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
132 args.fi_flags |= OCFS2_FI_FLAG_DELETE;
133 args.fi_ino = ino_from_blkno(osb->sb, blkno);
134 return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
137 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
139 struct inode *inode = NULL;
140 struct super_block *sb = osb->sb;
141 struct ocfs2_find_inode_args args;
143 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
145 /* Ok. By now we've either got the offsets passed to us by the
146 * caller, or we just pulled them off the bh. Lets do some
147 * sanity checks to make sure they're OK. */
149 inode = ERR_PTR(-EINVAL);
150 mlog_errno(PTR_ERR(inode));
154 args.fi_blkno = blkno;
155 args.fi_flags = flags;
156 args.fi_ino = ino_from_blkno(sb, blkno);
158 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
159 ocfs2_init_locked_inode, &args);
160 /* inode was *not* in the inode cache. 2.6.x requires
161 * us to do our own read_inode call and unlock it
163 if (inode && inode->i_state & I_NEW) {
164 mlog(0, "Inode was not in inode cache, reading it.\n");
165 ocfs2_read_locked_inode(inode, &args);
166 unlock_new_inode(inode);
169 inode = ERR_PTR(-ENOMEM);
170 mlog_errno(PTR_ERR(inode));
173 if (is_bad_inode(inode)) {
175 inode = ERR_PTR(-ESTALE);
180 if (!IS_ERR(inode)) {
181 mlog(0, "returning inode with number %llu\n",
182 (unsigned long long)OCFS2_I(inode)->ip_blkno);
183 mlog_exit_ptr(inode);
191 * here's how inodes get read from disk:
192 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
193 * found? : return the in-memory inode
194 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
197 static int ocfs2_find_actor(struct inode *inode, void *opaque)
199 struct ocfs2_find_inode_args *args = NULL;
200 struct ocfs2_inode_info *oi = OCFS2_I(inode);
203 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
207 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
209 if (oi->ip_blkno != args->fi_blkno)
212 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
213 * ocfs2_ilookup_for_vote which won't create an inode for one
214 * that isn't found. The vote thread which doesn't want to get
215 * an inode which is in the process of going away - otherwise
216 * the call to __wait_on_freeing_inode in find_inode_fast will
217 * cause it to deadlock on an inode which may be waiting on a
218 * vote (or lock release) in delete_inode */
219 if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
220 (inode->i_state & (I_FREEING|I_CLEAR))) {
221 /* As stated above, we're not going to return an
222 * inode. In the case of a delete vote, the voting
223 * code is going to signal the other node to go
224 * ahead. Mark that state here, so this freeing inode
225 * has the state when it gets to delete_inode. */
226 if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
227 spin_lock(&oi->ip_lock);
228 ocfs2_mark_inode_remotely_deleted(inode);
229 spin_unlock(&oi->ip_lock);
241 * initialize the new inode, but don't do anything that would cause
243 * return 0 on success, 1 on failure
245 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
247 struct ocfs2_find_inode_args *args = opaque;
249 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
251 inode->i_ino = args->fi_ino;
252 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
258 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
261 struct super_block *sb;
262 struct ocfs2_super *osb;
263 int status = -EINVAL;
267 mlog_entry("(0x%p, size:%llu)\n", inode,
268 (unsigned long long)fe->i_size);
273 /* this means that read_inode cannot create a superblock inode
274 * today. change if needed. */
275 if (!OCFS2_IS_VALID_DINODE(fe) ||
276 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
277 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
278 "signature = %.*s, flags = 0x%x\n",
280 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
281 fe->i_signature, le32_to_cpu(fe->i_flags));
285 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
286 mlog(ML_ERROR, "file entry generation does not match "
287 "superblock! osb->fs_generation=%x, "
288 "fe->i_fs_generation=%x\n",
289 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
293 inode->i_version = 1;
294 inode->i_generation = le32_to_cpu(fe->i_generation);
295 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
296 inode->i_mode = le16_to_cpu(fe->i_mode);
297 uid = le32_to_cpu(fe->i_uid);
298 gid = le32_to_cpu(fe->i_gid);
299 inode->i_uid = INOTAG_UID(DX_TAG(inode), uid, gid);
300 inode->i_gid = INOTAG_GID(DX_TAG(inode), uid, gid);
301 inode->i_tag = INOTAG_TAG(DX_TAG(inode), uid, gid,
302 /* le16_to_cpu(raw_inode->i_raw_tag)i */ 0);
304 /* Fast symlinks will have i_size but no allocated clusters. */
305 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
309 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
310 inode->i_mapping->a_ops = &ocfs2_aops;
311 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
312 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
313 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
314 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
315 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
316 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
318 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
320 "ip_blkno %llu != i_blkno %llu!\n",
321 (unsigned long long)OCFS2_I(inode)->ip_blkno,
322 (unsigned long long)fe->i_blkno);
324 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
325 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
326 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
328 inode->i_nlink = le16_to_cpu(fe->i_links_count);
330 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
331 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
333 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
334 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
335 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
336 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
337 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
338 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
339 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
340 /* we can't actually hit this as read_inode can't
341 * handle superblocks today ;-) */
345 switch (inode->i_mode & S_IFMT) {
347 inode->i_fop = &ocfs2_fops;
348 inode->i_op = &ocfs2_file_iops;
349 i_size_write(inode, le64_to_cpu(fe->i_size));
352 inode->i_op = &ocfs2_dir_iops;
353 inode->i_fop = &ocfs2_dops;
354 i_size_write(inode, le64_to_cpu(fe->i_size));
357 if (ocfs2_inode_is_fast_symlink(inode))
358 inode->i_op = &ocfs2_fast_symlink_inode_operations;
360 inode->i_op = &ocfs2_symlink_inode_operations;
361 i_size_write(inode, le64_to_cpu(fe->i_size));
364 inode->i_op = &ocfs2_special_file_iops;
365 init_special_inode(inode, inode->i_mode,
371 inode->i_ino = ino_from_blkno(inode->i_sb,
372 le64_to_cpu(fe->i_blkno));
375 * If we ever want to create system files from kernel,
376 * the generation argument to
377 * ocfs2_inode_lock_res_init() will have to change.
379 BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
381 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
382 OCFS2_LOCK_TYPE_META, 0, inode);
385 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
386 OCFS2_LOCK_TYPE_RW, inode->i_generation,
389 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
390 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
393 ocfs2_set_inode_flags(inode);
401 static int ocfs2_read_locked_inode(struct inode *inode,
402 struct ocfs2_find_inode_args *args)
404 struct super_block *sb;
405 struct ocfs2_super *osb;
406 struct ocfs2_dinode *fe;
407 struct buffer_head *bh = NULL;
408 int status, can_lock;
411 mlog_entry("(0x%p, 0x%p)\n", inode, args);
414 if (inode == NULL || inode->i_sb == NULL) {
415 mlog(ML_ERROR, "bad inode\n");
422 mlog(ML_ERROR, "bad inode args\n");
423 make_bad_inode(inode);
428 * To improve performance of cold-cache inode stats, we take
429 * the cluster lock here if possible.
431 * Generally, OCFS2 never trusts the contents of an inode
432 * unless it's holding a cluster lock, so taking it here isn't
433 * a correctness issue as much as it is a performance
436 * There are three times when taking the lock is not a good idea:
438 * 1) During startup, before we have initialized the DLM.
440 * 2) If we are reading certain system files which never get
441 * cluster locks (local alloc, truncate log).
443 * 3) If the process doing the iget() is responsible for
444 * orphan dir recovery. We're holding the orphan dir lock and
445 * can get into a deadlock with another process on another
446 * node in ->delete_inode().
448 * #1 and #2 can be simply solved by never taking the lock
449 * here for system files (which are the only type we read
450 * during mount). It's a heavier approach, but our main
451 * concern is user-accesible files anyway.
453 * #3 works itself out because we'll eventually take the
454 * cluster lock before trusting anything anyway.
456 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
457 && !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK)
458 && !ocfs2_mount_local(osb);
461 * To maintain backwards compatibility with older versions of
462 * ocfs2-tools, we still store the generation value for system
463 * files. The only ones that actually matter to userspace are
464 * the journals, but it's easier and inexpensive to just flag
465 * all system files similarly.
467 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
468 generation = osb->fs_generation;
470 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
471 OCFS2_LOCK_TYPE_META,
475 status = ocfs2_meta_lock(inode, NULL, 0);
477 make_bad_inode(inode);
483 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
484 can_lock ? inode : NULL);
491 fe = (struct ocfs2_dinode *) bh->b_data;
492 if (!OCFS2_IS_VALID_DINODE(fe)) {
493 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
494 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
499 * This is a code bug. Right now the caller needs to
500 * understand whether it is asking for a system file inode or
501 * not so the proper lock names can be built.
503 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
504 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
505 "Inode %llu: system file state is ambigous\n",
506 (unsigned long long)args->fi_blkno);
508 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
509 S_ISBLK(le16_to_cpu(fe->i_mode)))
510 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
512 if (ocfs2_populate_inode(inode, fe, 0) < 0)
515 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
521 ocfs2_meta_unlock(inode, 0);
524 make_bad_inode(inode);
533 void ocfs2_sync_blockdev(struct super_block *sb)
535 sync_blockdev(sb->s_bdev);
538 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
540 struct buffer_head *fe_bh)
543 handle_t *handle = NULL;
544 struct ocfs2_truncate_context *tc = NULL;
545 struct ocfs2_dinode *fe;
549 fe = (struct ocfs2_dinode *) fe_bh->b_data;
551 /* zero allocation, zero truncate :) */
555 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
556 if (IS_ERR(handle)) {
557 status = PTR_ERR(handle);
563 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
569 ocfs2_commit_trans(osb, handle);
572 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
578 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
585 ocfs2_commit_trans(osb, handle);
591 static int ocfs2_remove_inode(struct inode *inode,
592 struct buffer_head *di_bh,
593 struct inode *orphan_dir_inode,
594 struct buffer_head *orphan_dir_bh)
597 struct inode *inode_alloc_inode = NULL;
598 struct buffer_head *inode_alloc_bh = NULL;
600 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
601 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
604 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
605 le16_to_cpu(di->i_suballoc_slot));
606 if (!inode_alloc_inode) {
612 mutex_lock(&inode_alloc_inode->i_mutex);
613 status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
615 mutex_unlock(&inode_alloc_inode->i_mutex);
621 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
622 if (IS_ERR(handle)) {
623 status = PTR_ERR(handle);
628 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
635 /* set the inodes dtime */
636 status = ocfs2_journal_access(handle, inode, di_bh,
637 OCFS2_JOURNAL_ACCESS_WRITE);
643 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
644 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
646 status = ocfs2_journal_dirty(handle, di_bh);
652 ocfs2_remove_from_cache(inode, di_bh);
654 status = ocfs2_free_dinode(handle, inode_alloc_inode,
660 ocfs2_commit_trans(osb, handle);
662 ocfs2_meta_unlock(inode_alloc_inode, 1);
663 mutex_unlock(&inode_alloc_inode->i_mutex);
664 brelse(inode_alloc_bh);
666 iput(inode_alloc_inode);
672 * Serialize with orphan dir recovery. If the process doing
673 * recovery on this orphan dir does an iget() with the dir
674 * i_mutex held, we'll deadlock here. Instead we detect this
675 * and exit early - recovery will wipe this inode for us.
677 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
682 spin_lock(&osb->osb_lock);
683 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
684 mlog(0, "Recovery is happening on orphan dir %d, will skip "
685 "this inode\n", slot);
689 /* This signals to the orphan recovery process that it should
690 * wait for us to handle the wipe. */
691 osb->osb_orphan_wipes[slot]++;
693 spin_unlock(&osb->osb_lock);
697 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
700 spin_lock(&osb->osb_lock);
701 osb->osb_orphan_wipes[slot]--;
702 spin_unlock(&osb->osb_lock);
704 wake_up(&osb->osb_wipe_event);
707 static int ocfs2_wipe_inode(struct inode *inode,
708 struct buffer_head *di_bh)
710 int status, orphaned_slot;
711 struct inode *orphan_dir_inode = NULL;
712 struct buffer_head *orphan_dir_bh = NULL;
713 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
715 /* We've already voted on this so it should be readonly - no
716 * spinlock needed. */
717 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
719 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
723 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
724 ORPHAN_DIR_SYSTEM_INODE,
726 if (!orphan_dir_inode) {
732 /* Lock the orphan dir. The lock will be held for the entire
733 * delete_inode operation. We do this now to avoid races with
734 * recovery completion on other nodes. */
735 mutex_lock(&orphan_dir_inode->i_mutex);
736 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
738 mutex_unlock(&orphan_dir_inode->i_mutex);
744 /* we do this while holding the orphan dir lock because we
745 * don't want recovery being run from another node to vote for
746 * an inode delete on us -- this will result in two nodes
747 * truncating the same file! */
748 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
751 goto bail_unlock_dir;
754 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
760 ocfs2_meta_unlock(orphan_dir_inode, 1);
761 mutex_unlock(&orphan_dir_inode->i_mutex);
762 brelse(orphan_dir_bh);
764 iput(orphan_dir_inode);
765 ocfs2_signal_wipe_completion(osb, orphaned_slot);
770 /* There is a series of simple checks that should be done before a
771 * vote is even considered. Encapsulate those in this function. */
772 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
775 struct ocfs2_inode_info *oi = OCFS2_I(inode);
776 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
778 /* We shouldn't be getting here for the root directory
780 if (inode == osb->root_inode) {
781 mlog(ML_ERROR, "Skipping delete of root inode.\n");
785 /* If we're coming from process_vote we can't go into our own
786 * voting [hello, deadlock city!], so unforuntately we just
787 * have to skip deleting this guy. That's OK though because
788 * the node who's doing the actual deleting should handle it
790 if (current == osb->vote_task) {
791 mlog(0, "Skipping delete of %lu because we're currently "
792 "in process_vote\n", inode->i_ino);
796 spin_lock(&oi->ip_lock);
797 /* OCFS2 *never* deletes system files. This should technically
798 * never get here as system file inodes should always have a
799 * positive link count. */
800 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
801 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
802 (unsigned long long)oi->ip_blkno);
806 /* If we have voted "yes" on the wipe of this inode for
807 * another node, it will be marked here so we can safely skip
808 * it. Recovery will cleanup any inodes we might inadvertantly
810 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
811 mlog(0, "Skipping delete of %lu because another node "
812 "has done this for us.\n", inode->i_ino);
818 spin_unlock(&oi->ip_lock);
823 /* Query the cluster to determine whether we should wipe an inode from
826 * Requires the inode to have the cluster lock. */
827 static int ocfs2_query_inode_wipe(struct inode *inode,
828 struct buffer_head *di_bh,
832 struct ocfs2_inode_info *oi = OCFS2_I(inode);
833 struct ocfs2_dinode *di;
837 /* While we were waiting for the cluster lock in
838 * ocfs2_delete_inode, another node might have asked to delete
839 * the inode. Recheck our flags to catch this. */
840 if (!ocfs2_inode_is_valid_to_delete(inode)) {
841 mlog(0, "Skipping delete of %llu because flags changed\n",
842 (unsigned long long)oi->ip_blkno);
846 /* Now that we have an up to date inode, we can double check
848 if (inode->i_nlink) {
849 mlog(0, "Skipping delete of %llu because nlink = %u\n",
850 (unsigned long long)oi->ip_blkno, inode->i_nlink);
854 /* Do some basic inode verification... */
855 di = (struct ocfs2_dinode *) di_bh->b_data;
856 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
857 /* for lack of a better error? */
860 "Inode %llu (on-disk %llu) not orphaned! "
861 "Disk flags 0x%x, inode flags 0x%x\n",
862 (unsigned long long)oi->ip_blkno,
863 (unsigned long long)di->i_blkno, di->i_flags,
868 /* has someone already deleted us?! baaad... */
875 status = ocfs2_request_delete_vote(inode);
876 /* -EBUSY means that other nodes are still using the
877 * inode. We're done here though, so avoid doing anything on
878 * disk and let them worry about deleting it. */
879 if (status == -EBUSY) {
881 mlog(0, "Skipping delete of %llu because it is in use on"
882 "other nodes\n", (unsigned long long)oi->ip_blkno);
890 spin_lock(&oi->ip_lock);
891 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
892 /* Nobody knew which slot this inode was orphaned
893 * into. This may happen during node death and
894 * recovery knows how to clean it up so we can safely
895 * ignore this inode for now on. */
896 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
897 (unsigned long long)oi->ip_blkno);
901 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
902 (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
904 spin_unlock(&oi->ip_lock);
910 /* Support function for ocfs2_delete_inode. Will help us keep the
911 * inode data in a consistent state for clear_inode. Always truncates
912 * pages, optionally sync's them first. */
913 static void ocfs2_cleanup_delete_inode(struct inode *inode,
916 mlog(0, "Cleanup inode %llu, sync = %d\n",
917 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
919 write_inode_now(inode, 1);
920 truncate_inode_pages(&inode->i_data, 0);
923 void ocfs2_delete_inode(struct inode *inode)
926 sigset_t blocked, oldset;
927 struct buffer_head *di_bh = NULL;
929 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
931 if (is_bad_inode(inode)) {
932 mlog(0, "Skipping delete of bad inode\n");
936 if (!ocfs2_inode_is_valid_to_delete(inode)) {
937 /* It's probably not necessary to truncate_inode_pages
938 * here but we do it for safety anyway (it will most
939 * likely be a no-op anyway) */
940 ocfs2_cleanup_delete_inode(inode, 0);
944 /* We want to block signals in delete_inode as the lock and
945 * messaging paths may return us -ERESTARTSYS. Which would
946 * cause us to exit early, resulting in inodes being orphaned
948 sigfillset(&blocked);
949 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
952 ocfs2_cleanup_delete_inode(inode, 1);
956 /* Lock down the inode. This gives us an up to date view of
957 * it's metadata (for verification), and allows us to
958 * serialize delete_inode votes.
960 * Even though we might be doing a truncate, we don't take the
961 * allocation lock here as it won't be needed - nobody will
962 * have the file open.
964 status = ocfs2_meta_lock(inode, &di_bh, 1);
966 if (status != -ENOENT)
968 ocfs2_cleanup_delete_inode(inode, 0);
972 /* Query the cluster. This will be the final decision made
973 * before we go ahead and wipe the inode. */
974 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
975 if (!wipe || status < 0) {
976 /* Error and inode busy vote both mean we won't be
977 * removing the inode, so they take almost the same
982 /* Someone in the cluster has voted to not wipe this
983 * inode, or it was never completely orphaned. Write
984 * out the pages and exit now. */
985 ocfs2_cleanup_delete_inode(inode, 1);
986 goto bail_unlock_inode;
989 ocfs2_cleanup_delete_inode(inode, 0);
991 status = ocfs2_wipe_inode(inode, di_bh);
993 if (status != -EDEADLK)
995 goto bail_unlock_inode;
999 * Mark the inode as successfully deleted.
1001 * This is important for ocfs2_clear_inode() as it will check
1002 * this flag and skip any checkpointing work
1004 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1005 * the LVB for other nodes.
1007 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1010 ocfs2_meta_unlock(inode, 1);
1013 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1021 void ocfs2_clear_inode(struct inode *inode)
1024 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1031 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1032 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1034 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1035 "Inode=%lu\n", inode->i_ino);
1037 /* Do these before all the other work so that we don't bounce
1038 * the vote thread while waiting to destroy the locks. */
1039 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1040 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
1041 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
1043 /* We very well may get a clear_inode before all an inodes
1044 * metadata has hit disk. Of course, we can't drop any cluster
1045 * locks until the journal has finished with it. The only
1046 * exception here are successfully wiped inodes - their
1047 * metadata can now be considered to be part of the system
1048 * inodes from which it came. */
1049 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1050 ocfs2_checkpoint_inode(inode);
1052 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1053 "Clear inode of %llu, inode has io markers\n",
1054 (unsigned long long)oi->ip_blkno);
1056 ocfs2_extent_map_drop(inode, 0);
1057 ocfs2_extent_map_init(inode);
1059 status = ocfs2_drop_inode_locks(inode);
1063 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1064 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1065 ocfs2_lock_res_free(&oi->ip_data_lockres);
1067 ocfs2_metadata_cache_purge(inode);
1069 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1070 "Clear inode of %llu, inode has %u cache items\n",
1071 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1073 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1074 "Clear inode of %llu, inode has a bad flag\n",
1075 (unsigned long long)oi->ip_blkno);
1077 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1078 "Clear inode of %llu, inode is locked\n",
1079 (unsigned long long)oi->ip_blkno);
1081 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1082 "Clear inode of %llu, io_mutex is locked\n",
1083 (unsigned long long)oi->ip_blkno);
1084 mutex_unlock(&oi->ip_io_mutex);
1087 * down_trylock() returns 0, down_write_trylock() returns 1
1090 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1091 "Clear inode of %llu, alloc_sem is locked\n",
1092 (unsigned long long)oi->ip_blkno);
1093 up_write(&oi->ip_alloc_sem);
1095 mlog_bug_on_msg(oi->ip_open_count,
1096 "Clear inode of %llu has open count %d\n",
1097 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1099 /* Clear all other flags. */
1100 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1101 oi->ip_created_trans = 0;
1102 oi->ip_last_trans = 0;
1103 oi->ip_dir_start_lookup = 0;
1104 oi->ip_blkno = 0ULL;
1110 /* Called under inode_lock, with no more references on the
1111 * struct inode, so it's safe here to check the flags field
1112 * and to manipulate i_nlink without any other locks. */
1113 void ocfs2_drop_inode(struct inode *inode)
1115 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1119 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1120 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1122 /* Testing ip_orphaned_slot here wouldn't work because we may
1123 * not have gotten a delete_inode vote from any other nodes
1125 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1126 generic_delete_inode(inode);
1128 generic_drop_inode(inode);
1134 * TODO: this should probably be merged into ocfs2_get_block
1136 * However, you now need to pay attention to the cont_prepare_write()
1137 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1138 * expects never to extend).
1140 struct buffer_head *ocfs2_bread(struct inode *inode,
1141 int block, int *err, int reada)
1143 struct buffer_head *bh = NULL;
1146 int readflags = OCFS2_BH_CACHED;
1149 readflags |= OCFS2_BH_READAHEAD;
1151 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1152 i_size_read(inode)) {
1157 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1164 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1184 * This is called from our getattr.
1186 int ocfs2_inode_revalidate(struct dentry *dentry)
1188 struct inode *inode = dentry->d_inode;
1191 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1192 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1195 mlog(0, "eep, no inode!\n");
1200 spin_lock(&OCFS2_I(inode)->ip_lock);
1201 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1202 spin_unlock(&OCFS2_I(inode)->ip_lock);
1203 mlog(0, "inode deleted!\n");
1207 spin_unlock(&OCFS2_I(inode)->ip_lock);
1209 /* Let ocfs2_meta_lock do the work of updating our struct
1211 status = ocfs2_meta_lock(inode, NULL, 0);
1213 if (status != -ENOENT)
1217 ocfs2_meta_unlock(inode, 0);
1225 * Updates a disk inode from a
1227 * Only takes ip_lock.
1229 int ocfs2_mark_inode_dirty(handle_t *handle,
1230 struct inode *inode,
1231 struct buffer_head *bh)
1234 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1236 mlog_entry("(inode %llu)\n",
1237 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1239 status = ocfs2_journal_access(handle, inode, bh,
1240 OCFS2_JOURNAL_ACCESS_WRITE);
1246 spin_lock(&OCFS2_I(inode)->ip_lock);
1247 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1248 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1249 spin_unlock(&OCFS2_I(inode)->ip_lock);
1251 fe->i_size = cpu_to_le64(i_size_read(inode));
1252 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1253 fe->i_uid = cpu_to_le32(TAGINO_UID(DX_TAG(inode),
1254 inode->i_uid, inode->i_tag));
1255 fe->i_gid = cpu_to_le32(TAGINO_GID(DX_TAG(inode),
1256 inode->i_gid, inode->i_tag));
1257 /* i_tag = = cpu_to_le16(inode->i_tag); */
1258 fe->i_mode = cpu_to_le16(inode->i_mode);
1259 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1260 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1261 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1262 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1263 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1264 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1266 status = ocfs2_journal_dirty(handle, bh);
1279 * Updates a struct inode from a disk inode.
1280 * does no i/o, only takes ip_lock.
1282 void ocfs2_refresh_inode(struct inode *inode,
1283 struct ocfs2_dinode *fe)
1288 spin_lock(&OCFS2_I(inode)->ip_lock);
1290 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1291 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1292 /* OCFS2_I(inode)->ip_flags &= ~OCFS2_FL_MASK;
1293 OCFS2_I(inode)->ip_flags |= le32_to_cpu(fe->i_flags) & OCFS2_FL_MASK; */
1294 ocfs2_set_inode_flags(inode);
1295 i_size_write(inode, le64_to_cpu(fe->i_size));
1296 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1297 uid = le32_to_cpu(fe->i_uid);
1298 gid = le32_to_cpu(fe->i_gid);
1299 inode->i_uid = INOTAG_UID(DX_TAG(inode), uid, gid);
1300 inode->i_gid = INOTAG_GID(DX_TAG(inode), uid, gid);
1301 inode->i_tag = INOTAG_TAG(DX_TAG(inode), uid, gid,
1302 /* le16_to_cpu(raw_inode->i_raw_tag)i */ 0);
1303 inode->i_mode = le16_to_cpu(fe->i_mode);
1304 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1305 inode->i_blocks = 0;
1307 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1308 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1309 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1310 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1311 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1312 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1313 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1315 spin_unlock(&OCFS2_I(inode)->ip_lock);