1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * vfs' aops, fops, dops and iops
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/smp_lock.h>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
42 #include "extent_map.h"
44 #include "heartbeat.h"
55 #include "buffer_head_io.h"
57 struct ocfs2_find_inode_args
61 unsigned int fi_flags;
64 static int ocfs2_read_locked_inode(struct inode *inode,
65 struct ocfs2_find_inode_args *args);
66 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
67 static int ocfs2_find_actor(struct inode *inode, void *opaque);
68 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
70 struct buffer_head *fe_bh);
72 void ocfs2_set_inode_flags(struct inode *inode)
74 unsigned int flags = OCFS2_I(inode)->ip_attr;
76 inode->i_flags &= ~(S_IMMUTABLE |
77 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
79 if (flags & OCFS2_IMMUTABLE_FL)
80 inode->i_flags |= S_IMMUTABLE;
82 if (flags & OCFS2_SYNC_FL)
83 inode->i_flags |= S_SYNC;
84 if (flags & OCFS2_APPEND_FL)
85 inode->i_flags |= S_APPEND;
86 if (flags & OCFS2_NOATIME_FL)
87 inode->i_flags |= S_NOATIME;
88 if (flags & OCFS2_DIRSYNC_FL)
89 inode->i_flags |= S_DIRSYNC;
92 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
96 struct ocfs2_find_inode_args args;
98 /* ocfs2_ilookup_for_vote should *only* be called from the
100 BUG_ON(current != osb->vote_task);
102 args.fi_blkno = blkno;
103 args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
105 args.fi_flags |= OCFS2_FI_FLAG_DELETE;
106 args.fi_ino = ino_from_blkno(osb->sb, blkno);
107 return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
110 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
112 struct inode *inode = NULL;
113 struct super_block *sb = osb->sb;
114 struct ocfs2_find_inode_args args;
116 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
118 /* Ok. By now we've either got the offsets passed to us by the
119 * caller, or we just pulled them off the bh. Lets do some
120 * sanity checks to make sure they're OK. */
122 inode = ERR_PTR(-EINVAL);
123 mlog_errno(PTR_ERR(inode));
127 args.fi_blkno = blkno;
128 args.fi_flags = flags;
129 args.fi_ino = ino_from_blkno(sb, blkno);
131 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
132 ocfs2_init_locked_inode, &args);
133 /* inode was *not* in the inode cache. 2.6.x requires
134 * us to do our own read_inode call and unlock it
136 if (inode && inode->i_state & I_NEW) {
137 mlog(0, "Inode was not in inode cache, reading it.\n");
138 ocfs2_read_locked_inode(inode, &args);
139 unlock_new_inode(inode);
142 inode = ERR_PTR(-ENOMEM);
143 mlog_errno(PTR_ERR(inode));
146 if (is_bad_inode(inode)) {
148 inode = ERR_PTR(-ESTALE);
149 mlog_errno(PTR_ERR(inode));
154 if (!IS_ERR(inode)) {
155 mlog(0, "returning inode with number %llu\n",
156 (unsigned long long)OCFS2_I(inode)->ip_blkno);
157 mlog_exit_ptr(inode);
159 mlog_errno(PTR_ERR(inode));
166 * here's how inodes get read from disk:
167 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
168 * found? : return the in-memory inode
169 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
172 static int ocfs2_find_actor(struct inode *inode, void *opaque)
174 struct ocfs2_find_inode_args *args = NULL;
175 struct ocfs2_inode_info *oi = OCFS2_I(inode);
178 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
182 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
184 if (oi->ip_blkno != args->fi_blkno)
187 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
188 * ocfs2_ilookup_for_vote which won't create an inode for one
189 * that isn't found. The vote thread which doesn't want to get
190 * an inode which is in the process of going away - otherwise
191 * the call to __wait_on_freeing_inode in find_inode_fast will
192 * cause it to deadlock on an inode which may be waiting on a
193 * vote (or lock release) in delete_inode */
194 if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
195 (inode->i_state & (I_FREEING|I_CLEAR))) {
196 /* As stated above, we're not going to return an
197 * inode. In the case of a delete vote, the voting
198 * code is going to signal the other node to go
199 * ahead. Mark that state here, so this freeing inode
200 * has the state when it gets to delete_inode. */
201 if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
202 spin_lock(&oi->ip_lock);
203 ocfs2_mark_inode_remotely_deleted(inode);
204 spin_unlock(&oi->ip_lock);
216 * initialize the new inode, but don't do anything that would cause
218 * return 0 on success, 1 on failure
220 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
222 struct ocfs2_find_inode_args *args = opaque;
224 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
226 inode->i_ino = args->fi_ino;
227 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
233 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
236 struct super_block *sb;
237 struct ocfs2_super *osb;
238 int status = -EINVAL;
240 mlog_entry("(0x%p, size:%llu)\n", inode,
241 (unsigned long long)fe->i_size);
246 /* this means that read_inode cannot create a superblock inode
247 * today. change if needed. */
248 if (!OCFS2_IS_VALID_DINODE(fe) ||
249 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
250 mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
251 "signature = %.*s, flags = 0x%x\n",
253 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
254 fe->i_signature, le32_to_cpu(fe->i_flags));
258 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
259 mlog(ML_ERROR, "file entry generation does not match "
260 "superblock! osb->fs_generation=%x, "
261 "fe->i_fs_generation=%x\n",
262 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
266 inode->i_version = 1;
267 inode->i_generation = le32_to_cpu(fe->i_generation);
268 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
269 inode->i_mode = le16_to_cpu(fe->i_mode);
270 inode->i_uid = le32_to_cpu(fe->i_uid);
271 inode->i_gid = le32_to_cpu(fe->i_gid);
273 /* Fast symlinks will have i_size but no allocated clusters. */
274 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
278 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
279 inode->i_mapping->a_ops = &ocfs2_aops;
280 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
281 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
282 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
283 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
284 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
285 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
287 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
289 "ip_blkno %llu != i_blkno %llu!\n",
290 (unsigned long long)OCFS2_I(inode)->ip_blkno,
291 (unsigned long long)fe->i_blkno);
293 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
294 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
295 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
297 inode->i_nlink = le16_to_cpu(fe->i_links_count);
299 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
300 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
302 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
303 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
304 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
305 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
306 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
307 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
308 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
309 /* we can't actually hit this as read_inode can't
310 * handle superblocks today ;-) */
314 switch (inode->i_mode & S_IFMT) {
316 inode->i_fop = &ocfs2_fops;
317 inode->i_op = &ocfs2_file_iops;
318 i_size_write(inode, le64_to_cpu(fe->i_size));
321 inode->i_op = &ocfs2_dir_iops;
322 inode->i_fop = &ocfs2_dops;
323 i_size_write(inode, le64_to_cpu(fe->i_size));
326 if (ocfs2_inode_is_fast_symlink(inode))
327 inode->i_op = &ocfs2_fast_symlink_inode_operations;
329 inode->i_op = &ocfs2_symlink_inode_operations;
330 i_size_write(inode, le64_to_cpu(fe->i_size));
333 inode->i_op = &ocfs2_special_file_iops;
334 init_special_inode(inode, inode->i_mode,
340 inode->i_ino = ino_from_blkno(inode->i_sb,
341 le64_to_cpu(fe->i_blkno));
344 * If we ever want to create system files from kernel,
345 * the generation argument to
346 * ocfs2_inode_lock_res_init() will have to change.
348 BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
350 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
351 OCFS2_LOCK_TYPE_META, 0, inode);
354 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
355 OCFS2_LOCK_TYPE_RW, inode->i_generation,
358 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
359 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
362 ocfs2_set_inode_flags(inode);
363 inode->i_flags |= S_NOATIME;
371 static int ocfs2_read_locked_inode(struct inode *inode,
372 struct ocfs2_find_inode_args *args)
374 struct super_block *sb;
375 struct ocfs2_super *osb;
376 struct ocfs2_dinode *fe;
377 struct buffer_head *bh = NULL;
378 int status, can_lock;
381 mlog_entry("(0x%p, 0x%p)\n", inode, args);
384 if (inode == NULL || inode->i_sb == NULL) {
385 mlog(ML_ERROR, "bad inode\n");
392 mlog(ML_ERROR, "bad inode args\n");
393 make_bad_inode(inode);
398 * To improve performance of cold-cache inode stats, we take
399 * the cluster lock here if possible.
401 * Generally, OCFS2 never trusts the contents of an inode
402 * unless it's holding a cluster lock, so taking it here isn't
403 * a correctness issue as much as it is a performance
406 * There are three times when taking the lock is not a good idea:
408 * 1) During startup, before we have initialized the DLM.
410 * 2) If we are reading certain system files which never get
411 * cluster locks (local alloc, truncate log).
413 * 3) If the process doing the iget() is responsible for
414 * orphan dir recovery. We're holding the orphan dir lock and
415 * can get into a deadlock with another process on another
416 * node in ->delete_inode().
418 * #1 and #2 can be simply solved by never taking the lock
419 * here for system files (which are the only type we read
420 * during mount). It's a heavier approach, but our main
421 * concern is user-accesible files anyway.
423 * #3 works itself out because we'll eventually take the
424 * cluster lock before trusting anything anyway.
426 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
427 && !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK);
430 * To maintain backwards compatibility with older versions of
431 * ocfs2-tools, we still store the generation value for system
432 * files. The only ones that actually matter to userspace are
433 * the journals, but it's easier and inexpensive to just flag
434 * all system files similarly.
436 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
437 generation = osb->fs_generation;
439 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
440 OCFS2_LOCK_TYPE_META,
444 status = ocfs2_meta_lock(inode, NULL, NULL, 0);
446 make_bad_inode(inode);
452 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
453 can_lock ? inode : NULL);
460 fe = (struct ocfs2_dinode *) bh->b_data;
461 if (!OCFS2_IS_VALID_DINODE(fe)) {
462 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
463 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
468 * This is a code bug. Right now the caller needs to
469 * understand whether it is asking for a system file inode or
470 * not so the proper lock names can be built.
472 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
473 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
474 "Inode %llu: system file state is ambigous\n",
475 (unsigned long long)args->fi_blkno);
477 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
478 S_ISBLK(le16_to_cpu(fe->i_mode)))
479 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
481 if (ocfs2_populate_inode(inode, fe, 0) < 0) {
482 mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
483 (unsigned long long)fe->i_blkno, inode->i_ino);
487 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
493 ocfs2_meta_unlock(inode, 0);
496 make_bad_inode(inode);
505 void ocfs2_sync_blockdev(struct super_block *sb)
507 sync_blockdev(sb->s_bdev);
510 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
512 struct buffer_head *fe_bh)
515 struct ocfs2_journal_handle *handle = NULL;
516 struct ocfs2_truncate_context *tc = NULL;
517 struct ocfs2_dinode *fe;
521 fe = (struct ocfs2_dinode *) fe_bh->b_data;
523 /* zero allocation, zero truncate :) */
527 handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS);
528 if (IS_ERR(handle)) {
529 status = PTR_ERR(handle);
535 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
541 ocfs2_commit_trans(handle);
544 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
550 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
557 ocfs2_commit_trans(handle);
563 static int ocfs2_remove_inode(struct inode *inode,
564 struct buffer_head *di_bh,
565 struct inode *orphan_dir_inode,
566 struct buffer_head *orphan_dir_bh)
569 struct inode *inode_alloc_inode = NULL;
570 struct buffer_head *inode_alloc_bh = NULL;
571 struct ocfs2_journal_handle *handle;
572 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
573 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
576 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
577 le16_to_cpu(di->i_suballoc_slot));
578 if (!inode_alloc_inode) {
584 mutex_lock(&inode_alloc_inode->i_mutex);
585 status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1);
587 mutex_unlock(&inode_alloc_inode->i_mutex);
593 handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS);
594 if (IS_ERR(handle)) {
595 status = PTR_ERR(handle);
600 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
607 /* set the inodes dtime */
608 status = ocfs2_journal_access(handle, inode, di_bh,
609 OCFS2_JOURNAL_ACCESS_WRITE);
615 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
616 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
618 status = ocfs2_journal_dirty(handle, di_bh);
624 ocfs2_remove_from_cache(inode, di_bh);
626 status = ocfs2_free_dinode(handle, inode_alloc_inode,
632 ocfs2_commit_trans(handle);
634 ocfs2_meta_unlock(inode_alloc_inode, 1);
635 mutex_unlock(&inode_alloc_inode->i_mutex);
636 brelse(inode_alloc_bh);
638 iput(inode_alloc_inode);
644 * Serialize with orphan dir recovery. If the process doing
645 * recovery on this orphan dir does an iget() with the dir
646 * i_mutex held, we'll deadlock here. Instead we detect this
647 * and exit early - recovery will wipe this inode for us.
649 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
654 spin_lock(&osb->osb_lock);
655 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
656 mlog(0, "Recovery is happening on orphan dir %d, will skip "
657 "this inode\n", slot);
661 /* This signals to the orphan recovery process that it should
662 * wait for us to handle the wipe. */
663 osb->osb_orphan_wipes[slot]++;
665 spin_unlock(&osb->osb_lock);
669 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
672 spin_lock(&osb->osb_lock);
673 osb->osb_orphan_wipes[slot]--;
674 spin_unlock(&osb->osb_lock);
676 wake_up(&osb->osb_wipe_event);
679 static int ocfs2_wipe_inode(struct inode *inode,
680 struct buffer_head *di_bh)
682 int status, orphaned_slot;
683 struct inode *orphan_dir_inode = NULL;
684 struct buffer_head *orphan_dir_bh = NULL;
685 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
687 /* We've already voted on this so it should be readonly - no
688 * spinlock needed. */
689 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
691 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
695 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
696 ORPHAN_DIR_SYSTEM_INODE,
698 if (!orphan_dir_inode) {
704 /* Lock the orphan dir. The lock will be held for the entire
705 * delete_inode operation. We do this now to avoid races with
706 * recovery completion on other nodes. */
707 mutex_lock(&orphan_dir_inode->i_mutex);
708 status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
710 mutex_unlock(&orphan_dir_inode->i_mutex);
716 /* we do this while holding the orphan dir lock because we
717 * don't want recovery being run from another node to vote for
718 * an inode delete on us -- this will result in two nodes
719 * truncating the same file! */
720 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
723 goto bail_unlock_dir;
726 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
732 ocfs2_meta_unlock(orphan_dir_inode, 1);
733 mutex_unlock(&orphan_dir_inode->i_mutex);
734 brelse(orphan_dir_bh);
736 iput(orphan_dir_inode);
737 ocfs2_signal_wipe_completion(osb, orphaned_slot);
742 /* There is a series of simple checks that should be done before a
743 * vote is even considered. Encapsulate those in this function. */
744 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
747 struct ocfs2_inode_info *oi = OCFS2_I(inode);
748 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
750 /* We shouldn't be getting here for the root directory
752 if (inode == osb->root_inode) {
753 mlog(ML_ERROR, "Skipping delete of root inode.\n");
757 /* If we're coming from process_vote we can't go into our own
758 * voting [hello, deadlock city!], so unforuntately we just
759 * have to skip deleting this guy. That's OK though because
760 * the node who's doing the actual deleting should handle it
762 if (current == osb->vote_task) {
763 mlog(0, "Skipping delete of %lu because we're currently "
764 "in process_vote\n", inode->i_ino);
768 spin_lock(&oi->ip_lock);
769 /* OCFS2 *never* deletes system files. This should technically
770 * never get here as system file inodes should always have a
771 * positive link count. */
772 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
773 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
774 (unsigned long long)oi->ip_blkno);
778 /* If we have voted "yes" on the wipe of this inode for
779 * another node, it will be marked here so we can safely skip
780 * it. Recovery will cleanup any inodes we might inadvertantly
782 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
783 mlog(0, "Skipping delete of %lu because another node "
784 "has done this for us.\n", inode->i_ino);
790 spin_unlock(&oi->ip_lock);
795 /* Query the cluster to determine whether we should wipe an inode from
798 * Requires the inode to have the cluster lock. */
799 static int ocfs2_query_inode_wipe(struct inode *inode,
800 struct buffer_head *di_bh,
804 struct ocfs2_inode_info *oi = OCFS2_I(inode);
805 struct ocfs2_dinode *di;
809 /* While we were waiting for the cluster lock in
810 * ocfs2_delete_inode, another node might have asked to delete
811 * the inode. Recheck our flags to catch this. */
812 if (!ocfs2_inode_is_valid_to_delete(inode)) {
813 mlog(0, "Skipping delete of %llu because flags changed\n",
814 (unsigned long long)oi->ip_blkno);
818 /* Now that we have an up to date inode, we can double check
820 if (inode->i_nlink) {
821 mlog(0, "Skipping delete of %llu because nlink = %u\n",
822 (unsigned long long)oi->ip_blkno, inode->i_nlink);
826 /* Do some basic inode verification... */
827 di = (struct ocfs2_dinode *) di_bh->b_data;
828 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
829 /* for lack of a better error? */
832 "Inode %llu (on-disk %llu) not orphaned! "
833 "Disk flags 0x%x, inode flags 0x%x\n",
834 (unsigned long long)oi->ip_blkno,
835 (unsigned long long)di->i_blkno, di->i_flags,
840 /* has someone already deleted us?! baaad... */
847 status = ocfs2_request_delete_vote(inode);
848 /* -EBUSY means that other nodes are still using the
849 * inode. We're done here though, so avoid doing anything on
850 * disk and let them worry about deleting it. */
851 if (status == -EBUSY) {
853 mlog(0, "Skipping delete of %llu because it is in use on"
854 "other nodes\n", (unsigned long long)oi->ip_blkno);
862 spin_lock(&oi->ip_lock);
863 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
864 /* Nobody knew which slot this inode was orphaned
865 * into. This may happen during node death and
866 * recovery knows how to clean it up so we can safely
867 * ignore this inode for now on. */
868 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
869 (unsigned long long)oi->ip_blkno);
873 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
874 (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
876 spin_unlock(&oi->ip_lock);
882 /* Support function for ocfs2_delete_inode. Will help us keep the
883 * inode data in a consistent state for clear_inode. Always truncates
884 * pages, optionally sync's them first. */
885 static void ocfs2_cleanup_delete_inode(struct inode *inode,
888 mlog(0, "Cleanup inode %llu, sync = %d\n",
889 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
891 write_inode_now(inode, 1);
892 truncate_inode_pages(&inode->i_data, 0);
895 void ocfs2_delete_inode(struct inode *inode)
898 sigset_t blocked, oldset;
899 struct buffer_head *di_bh = NULL;
901 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
903 if (is_bad_inode(inode)) {
904 mlog(0, "Skipping delete of bad inode\n");
908 if (!ocfs2_inode_is_valid_to_delete(inode)) {
909 /* It's probably not necessary to truncate_inode_pages
910 * here but we do it for safety anyway (it will most
911 * likely be a no-op anyway) */
912 ocfs2_cleanup_delete_inode(inode, 0);
916 /* We want to block signals in delete_inode as the lock and
917 * messaging paths may return us -ERESTARTSYS. Which would
918 * cause us to exit early, resulting in inodes being orphaned
920 sigfillset(&blocked);
921 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
924 ocfs2_cleanup_delete_inode(inode, 1);
928 /* Lock down the inode. This gives us an up to date view of
929 * it's metadata (for verification), and allows us to
930 * serialize delete_inode votes.
932 * Even though we might be doing a truncate, we don't take the
933 * allocation lock here as it won't be needed - nobody will
934 * have the file open.
936 status = ocfs2_meta_lock(inode, NULL, &di_bh, 1);
938 if (status != -ENOENT)
940 ocfs2_cleanup_delete_inode(inode, 0);
944 /* Query the cluster. This will be the final decision made
945 * before we go ahead and wipe the inode. */
946 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
947 if (!wipe || status < 0) {
948 /* Error and inode busy vote both mean we won't be
949 * removing the inode, so they take almost the same
954 /* Someone in the cluster has voted to not wipe this
955 * inode, or it was never completely orphaned. Write
956 * out the pages and exit now. */
957 ocfs2_cleanup_delete_inode(inode, 1);
958 goto bail_unlock_inode;
961 ocfs2_cleanup_delete_inode(inode, 0);
963 status = ocfs2_wipe_inode(inode, di_bh);
965 if (status != -EDEADLK)
967 goto bail_unlock_inode;
971 * Mark the inode as successfully deleted.
973 * This is important for ocfs2_clear_inode() as it will check
974 * this flag and skip any checkpointing work
976 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
977 * the LVB for other nodes.
979 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
982 ocfs2_meta_unlock(inode, 1);
985 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
993 void ocfs2_clear_inode(struct inode *inode)
996 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1003 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1004 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1006 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1007 "Inode=%lu\n", inode->i_ino);
1009 /* Do these before all the other work so that we don't bounce
1010 * the vote thread while waiting to destroy the locks. */
1011 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1012 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
1013 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
1015 /* We very well may get a clear_inode before all an inodes
1016 * metadata has hit disk. Of course, we can't drop any cluster
1017 * locks until the journal has finished with it. The only
1018 * exception here are successfully wiped inodes - their
1019 * metadata can now be considered to be part of the system
1020 * inodes from which it came. */
1021 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1022 ocfs2_checkpoint_inode(inode);
1024 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1025 "Clear inode of %llu, inode has io markers\n",
1026 (unsigned long long)oi->ip_blkno);
1028 ocfs2_extent_map_drop(inode, 0);
1029 ocfs2_extent_map_init(inode);
1031 status = ocfs2_drop_inode_locks(inode);
1035 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1036 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1037 ocfs2_lock_res_free(&oi->ip_data_lockres);
1039 ocfs2_metadata_cache_purge(inode);
1041 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1042 "Clear inode of %llu, inode has %u cache items\n",
1043 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1045 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1046 "Clear inode of %llu, inode has a bad flag\n",
1047 (unsigned long long)oi->ip_blkno);
1049 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1050 "Clear inode of %llu, inode is locked\n",
1051 (unsigned long long)oi->ip_blkno);
1053 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1054 "Clear inode of %llu, io_mutex is locked\n",
1055 (unsigned long long)oi->ip_blkno);
1056 mutex_unlock(&oi->ip_io_mutex);
1059 * down_trylock() returns 0, down_write_trylock() returns 1
1062 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1063 "Clear inode of %llu, alloc_sem is locked\n",
1064 (unsigned long long)oi->ip_blkno);
1065 up_write(&oi->ip_alloc_sem);
1067 mlog_bug_on_msg(oi->ip_open_count,
1068 "Clear inode of %llu has open count %d\n",
1069 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1070 mlog_bug_on_msg(!list_empty(&oi->ip_handle_list),
1071 "Clear inode of %llu has non empty handle list\n",
1072 (unsigned long long)oi->ip_blkno);
1073 mlog_bug_on_msg(oi->ip_handle,
1074 "Clear inode of %llu has non empty handle pointer\n",
1075 (unsigned long long)oi->ip_blkno);
1077 /* Clear all other flags. */
1078 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1079 oi->ip_created_trans = 0;
1080 oi->ip_last_trans = 0;
1081 oi->ip_dir_start_lookup = 0;
1082 oi->ip_blkno = 0ULL;
1088 /* Called under inode_lock, with no more references on the
1089 * struct inode, so it's safe here to check the flags field
1090 * and to manipulate i_nlink without any other locks. */
1091 void ocfs2_drop_inode(struct inode *inode)
1093 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1097 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1098 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1100 /* Testing ip_orphaned_slot here wouldn't work because we may
1101 * not have gotten a delete_inode vote from any other nodes
1103 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1104 generic_delete_inode(inode);
1106 generic_drop_inode(inode);
1112 * TODO: this should probably be merged into ocfs2_get_block
1114 * However, you now need to pay attention to the cont_prepare_write()
1115 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1116 * expects never to extend).
1118 struct buffer_head *ocfs2_bread(struct inode *inode,
1119 int block, int *err, int reada)
1121 struct buffer_head *bh = NULL;
1124 int readflags = OCFS2_BH_CACHED;
1127 readflags |= OCFS2_BH_READAHEAD;
1129 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1130 i_size_read(inode)) {
1135 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1142 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1162 * This is called from our getattr.
1164 int ocfs2_inode_revalidate(struct dentry *dentry)
1166 struct inode *inode = dentry->d_inode;
1169 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1170 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1173 mlog(0, "eep, no inode!\n");
1178 spin_lock(&OCFS2_I(inode)->ip_lock);
1179 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1180 spin_unlock(&OCFS2_I(inode)->ip_lock);
1181 mlog(0, "inode deleted!\n");
1185 spin_unlock(&OCFS2_I(inode)->ip_lock);
1187 /* Let ocfs2_meta_lock do the work of updating our struct
1189 status = ocfs2_meta_lock(inode, NULL, NULL, 0);
1191 if (status != -ENOENT)
1195 ocfs2_meta_unlock(inode, 0);
1203 * Updates a disk inode from a
1205 * Only takes ip_lock.
1207 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle,
1208 struct inode *inode,
1209 struct buffer_head *bh)
1212 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1214 mlog_entry("(inode %llu)\n",
1215 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1217 status = ocfs2_journal_access(handle, inode, bh,
1218 OCFS2_JOURNAL_ACCESS_WRITE);
1224 spin_lock(&OCFS2_I(inode)->ip_lock);
1225 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1226 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1227 spin_unlock(&OCFS2_I(inode)->ip_lock);
1229 fe->i_size = cpu_to_le64(i_size_read(inode));
1230 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1231 fe->i_uid = cpu_to_le32(inode->i_uid);
1232 fe->i_gid = cpu_to_le32(inode->i_gid);
1233 fe->i_mode = cpu_to_le16(inode->i_mode);
1234 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1235 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1236 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1237 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1238 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1239 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1241 status = ocfs2_journal_dirty(handle, bh);
1254 * Updates a struct inode from a disk inode.
1255 * does no i/o, only takes ip_lock.
1257 void ocfs2_refresh_inode(struct inode *inode,
1258 struct ocfs2_dinode *fe)
1260 spin_lock(&OCFS2_I(inode)->ip_lock);
1262 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1263 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1264 ocfs2_set_inode_flags(inode);
1265 i_size_write(inode, le64_to_cpu(fe->i_size));
1266 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1267 inode->i_uid = le32_to_cpu(fe->i_uid);
1268 inode->i_gid = le32_to_cpu(fe->i_gid);
1269 inode->i_mode = le16_to_cpu(fe->i_mode);
1270 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1271 inode->i_blocks = 0;
1273 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1274 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1275 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1276 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1277 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1278 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1279 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1281 spin_unlock(&OCFS2_I(inode)->ip_lock);