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/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/quotaops.h>
32 #include <asm/byteorder.h>
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
41 #include "blockcheck.h"
43 #include "extent_map.h"
45 #include "heartbeat.h"
55 #include "refcounttree.h"
57 #include "buffer_head_io.h"
59 struct ocfs2_find_inode_args
63 unsigned int fi_flags;
64 unsigned int fi_sysfile_type;
67 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
69 static int ocfs2_read_locked_inode(struct inode *inode,
70 struct ocfs2_find_inode_args *args);
71 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
72 static int ocfs2_find_actor(struct inode *inode, void *opaque);
73 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
75 struct buffer_head *fe_bh);
77 void ocfs2_set_inode_flags(struct inode *inode)
79 unsigned int flags = OCFS2_I(inode)->ip_attr;
81 inode->i_flags &= ~(S_IMMUTABLE |
82 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
84 if (flags & OCFS2_IMMUTABLE_FL)
85 inode->i_flags |= S_IMMUTABLE;
87 if (flags & OCFS2_SYNC_FL)
88 inode->i_flags |= S_SYNC;
89 if (flags & OCFS2_APPEND_FL)
90 inode->i_flags |= S_APPEND;
91 if (flags & OCFS2_NOATIME_FL)
92 inode->i_flags |= S_NOATIME;
93 if (flags & OCFS2_DIRSYNC_FL)
94 inode->i_flags |= S_DIRSYNC;
97 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
98 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
100 unsigned int flags = oi->vfs_inode.i_flags;
102 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
103 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
105 oi->ip_attr |= OCFS2_SYNC_FL;
106 if (flags & S_APPEND)
107 oi->ip_attr |= OCFS2_APPEND_FL;
108 if (flags & S_IMMUTABLE)
109 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
110 if (flags & S_NOATIME)
111 oi->ip_attr |= OCFS2_NOATIME_FL;
112 if (flags & S_DIRSYNC)
113 oi->ip_attr |= OCFS2_DIRSYNC_FL;
116 struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
118 struct ocfs2_find_inode_args args;
120 args.fi_blkno = blkno;
122 args.fi_ino = ino_from_blkno(sb, blkno);
123 args.fi_sysfile_type = 0;
125 return ilookup5(sb, blkno, ocfs2_find_actor, &args);
127 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
130 struct inode *inode = NULL;
131 struct super_block *sb = osb->sb;
132 struct ocfs2_find_inode_args args;
134 mlog(0, "(blkno = %llu)\n", (unsigned long long)blkno);
136 /* Ok. By now we've either got the offsets passed to us by the
137 * caller, or we just pulled them off the bh. Lets do some
138 * sanity checks to make sure they're OK. */
140 inode = ERR_PTR(-EINVAL);
141 mlog_errno(PTR_ERR(inode));
145 args.fi_blkno = blkno;
146 args.fi_flags = flags;
147 args.fi_ino = ino_from_blkno(sb, blkno);
148 args.fi_sysfile_type = sysfile_type;
150 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
151 ocfs2_init_locked_inode, &args);
152 /* inode was *not* in the inode cache. 2.6.x requires
153 * us to do our own read_inode call and unlock it
155 if (inode && inode->i_state & I_NEW) {
156 mlog(0, "Inode was not in inode cache, reading it.\n");
157 ocfs2_read_locked_inode(inode, &args);
158 unlock_new_inode(inode);
161 inode = ERR_PTR(-ENOMEM);
162 mlog_errno(PTR_ERR(inode));
165 if (is_bad_inode(inode)) {
167 inode = ERR_PTR(-ESTALE);
172 if (!IS_ERR(inode)) {
173 mlog(0, "returning inode with number %llu\n",
174 (unsigned long long)OCFS2_I(inode)->ip_blkno);
175 mlog_exit_ptr(inode);
183 * here's how inodes get read from disk:
184 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
185 * found? : return the in-memory inode
186 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
189 static int ocfs2_find_actor(struct inode *inode, void *opaque)
191 struct ocfs2_find_inode_args *args = NULL;
192 struct ocfs2_inode_info *oi = OCFS2_I(inode);
195 mlog(0, "(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
199 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
201 if (oi->ip_blkno != args->fi_blkno)
211 * initialize the new inode, but don't do anything that would cause
213 * return 0 on success, 1 on failure
215 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
217 struct ocfs2_find_inode_args *args = opaque;
218 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
219 ocfs2_file_ip_alloc_sem_key;
221 mlog(0, "inode = %p, opaque = %p\n", inode, opaque);
223 inode->i_ino = args->fi_ino;
224 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
225 if (args->fi_sysfile_type != 0)
226 lockdep_set_class(&inode->i_mutex,
227 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
228 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
229 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
230 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
231 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
232 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233 &ocfs2_quota_ip_alloc_sem_key);
235 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
236 &ocfs2_file_ip_alloc_sem_key);
242 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
245 struct super_block *sb;
246 struct ocfs2_super *osb;
249 mlog(0, "(0x%p, size:%llu)\n", inode,
250 (unsigned long long)le64_to_cpu(fe->i_size));
255 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
256 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
260 * These have all been checked by ocfs2_read_inode_block() or set
261 * by ocfs2_mknod_locked(), so a failure is a code bug.
263 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
264 cannot create a superblock
265 inode today. change if
267 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
268 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
271 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
272 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
273 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
275 inode->i_version = 1;
276 inode->i_generation = le32_to_cpu(fe->i_generation);
277 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
278 inode->i_mode = le16_to_cpu(fe->i_mode);
279 inode->i_uid = le32_to_cpu(fe->i_uid);
280 inode->i_gid = le32_to_cpu(fe->i_gid);
282 /* Fast symlinks will have i_size but no allocated clusters. */
283 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
286 inode->i_blocks = ocfs2_inode_sector_count(inode);
287 inode->i_mapping->a_ops = &ocfs2_aops;
288 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
289 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
290 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
291 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
292 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
293 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
295 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
297 "ip_blkno %llu != i_blkno %llu!\n",
298 (unsigned long long)OCFS2_I(inode)->ip_blkno,
299 (unsigned long long)le64_to_cpu(fe->i_blkno));
301 inode->i_nlink = ocfs2_read_links_count(fe);
303 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
304 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
305 inode->i_flags |= S_NOQUOTA;
308 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
309 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
310 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
311 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
312 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
313 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
314 inode->i_flags |= S_NOQUOTA;
315 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
316 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
317 /* we can't actually hit this as read_inode can't
318 * handle superblocks today ;-) */
322 switch (inode->i_mode & S_IFMT) {
325 inode->i_fop = &ocfs2_fops;
327 inode->i_fop = &ocfs2_fops_no_plocks;
328 inode->i_op = &ocfs2_file_iops;
329 i_size_write(inode, le64_to_cpu(fe->i_size));
332 inode->i_op = &ocfs2_dir_iops;
334 inode->i_fop = &ocfs2_dops;
336 inode->i_fop = &ocfs2_dops_no_plocks;
337 i_size_write(inode, le64_to_cpu(fe->i_size));
338 OCFS2_I(inode)->ip_dir_lock_gen = 1;
341 if (ocfs2_inode_is_fast_symlink(inode))
342 inode->i_op = &ocfs2_fast_symlink_inode_operations;
344 inode->i_op = &ocfs2_symlink_inode_operations;
345 i_size_write(inode, le64_to_cpu(fe->i_size));
348 inode->i_op = &ocfs2_special_file_iops;
349 init_special_inode(inode, inode->i_mode,
355 inode->i_ino = ino_from_blkno(inode->i_sb,
356 le64_to_cpu(fe->i_blkno));
359 * If we ever want to create system files from kernel,
360 * the generation argument to
361 * ocfs2_inode_lock_res_init() will have to change.
363 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
365 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
366 OCFS2_LOCK_TYPE_META, 0, inode);
368 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
369 OCFS2_LOCK_TYPE_OPEN, 0, inode);
372 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
373 OCFS2_LOCK_TYPE_RW, inode->i_generation,
376 ocfs2_set_inode_flags(inode);
378 OCFS2_I(inode)->ip_last_used_slot = 0;
379 OCFS2_I(inode)->ip_last_used_group = 0;
381 if (S_ISDIR(inode->i_mode))
382 ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
383 OCFS2_RESV_FLAG_DIR);
387 static int ocfs2_read_locked_inode(struct inode *inode,
388 struct ocfs2_find_inode_args *args)
390 struct super_block *sb;
391 struct ocfs2_super *osb;
392 struct ocfs2_dinode *fe;
393 struct buffer_head *bh = NULL;
394 int status, can_lock;
397 mlog(0, "(0x%p, 0x%p)\n", inode, args);
400 if (inode == NULL || inode->i_sb == NULL) {
401 mlog(ML_ERROR, "bad inode\n");
408 mlog(ML_ERROR, "bad inode args\n");
409 make_bad_inode(inode);
414 * To improve performance of cold-cache inode stats, we take
415 * the cluster lock here if possible.
417 * Generally, OCFS2 never trusts the contents of an inode
418 * unless it's holding a cluster lock, so taking it here isn't
419 * a correctness issue as much as it is a performance
422 * There are three times when taking the lock is not a good idea:
424 * 1) During startup, before we have initialized the DLM.
426 * 2) If we are reading certain system files which never get
427 * cluster locks (local alloc, truncate log).
429 * 3) If the process doing the iget() is responsible for
430 * orphan dir recovery. We're holding the orphan dir lock and
431 * can get into a deadlock with another process on another
432 * node in ->delete_inode().
434 * #1 and #2 can be simply solved by never taking the lock
435 * here for system files (which are the only type we read
436 * during mount). It's a heavier approach, but our main
437 * concern is user-accessible files anyway.
439 * #3 works itself out because we'll eventually take the
440 * cluster lock before trusting anything anyway.
442 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
443 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
444 && !ocfs2_mount_local(osb);
447 * To maintain backwards compatibility with older versions of
448 * ocfs2-tools, we still store the generation value for system
449 * files. The only ones that actually matter to userspace are
450 * the journals, but it's easier and inexpensive to just flag
451 * all system files similarly.
453 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
454 generation = osb->fs_generation;
456 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
457 OCFS2_LOCK_TYPE_META,
460 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
461 OCFS2_LOCK_TYPE_OPEN,
465 status = ocfs2_open_lock(inode);
467 make_bad_inode(inode);
471 status = ocfs2_inode_lock(inode, NULL, 0);
473 make_bad_inode(inode);
479 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
480 status = ocfs2_try_open_lock(inode, 0);
482 make_bad_inode(inode);
488 status = ocfs2_read_inode_block_full(inode, &bh,
489 OCFS2_BH_IGNORE_CACHE);
491 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
493 * If buffer is in jbd, then its checksum may not have been
496 if (!status && !buffer_jbd(bh))
497 status = ocfs2_validate_inode_block(osb->sb, bh);
505 fe = (struct ocfs2_dinode *) bh->b_data;
508 * This is a code bug. Right now the caller needs to
509 * understand whether it is asking for a system file inode or
510 * not so the proper lock names can be built.
512 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
513 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
514 "Inode %llu: system file state is ambigous\n",
515 (unsigned long long)args->fi_blkno);
517 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
518 S_ISBLK(le16_to_cpu(fe->i_mode)))
519 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
521 ocfs2_populate_inode(inode, fe, 0);
523 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
529 ocfs2_inode_unlock(inode, 0);
532 make_bad_inode(inode);
541 void ocfs2_sync_blockdev(struct super_block *sb)
543 sync_blockdev(sb->s_bdev);
546 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
548 struct buffer_head *fe_bh)
551 struct ocfs2_dinode *fe;
552 handle_t *handle = NULL;
554 fe = (struct ocfs2_dinode *) fe_bh->b_data;
557 * This check will also skip truncate of inodes with inline
558 * data and fast symlinks.
560 if (fe->i_clusters) {
561 if (ocfs2_should_order_data(inode))
562 ocfs2_begin_ordered_truncate(inode, 0);
564 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
565 if (IS_ERR(handle)) {
566 status = PTR_ERR(handle);
572 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
574 OCFS2_JOURNAL_ACCESS_WRITE);
580 i_size_write(inode, 0);
582 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
588 ocfs2_commit_trans(osb, handle);
591 status = ocfs2_commit_truncate(osb, inode, fe_bh);
600 ocfs2_commit_trans(osb, handle);
605 static int ocfs2_remove_inode(struct inode *inode,
606 struct buffer_head *di_bh,
607 struct inode *orphan_dir_inode,
608 struct buffer_head *orphan_dir_bh)
611 struct inode *inode_alloc_inode = NULL;
612 struct buffer_head *inode_alloc_bh = NULL;
614 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
615 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
618 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
619 le16_to_cpu(di->i_suballoc_slot));
620 if (!inode_alloc_inode) {
626 mutex_lock(&inode_alloc_inode->i_mutex);
627 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
629 mutex_unlock(&inode_alloc_inode->i_mutex);
635 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
636 ocfs2_quota_trans_credits(inode->i_sb));
637 if (IS_ERR(handle)) {
638 status = PTR_ERR(handle);
643 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
644 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
652 /* set the inodes dtime */
653 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
654 OCFS2_JOURNAL_ACCESS_WRITE);
660 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
661 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
662 ocfs2_journal_dirty(handle, di_bh);
664 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
665 dquot_free_inode(inode);
667 status = ocfs2_free_dinode(handle, inode_alloc_inode,
673 ocfs2_commit_trans(osb, handle);
675 ocfs2_inode_unlock(inode_alloc_inode, 1);
676 mutex_unlock(&inode_alloc_inode->i_mutex);
677 brelse(inode_alloc_bh);
679 iput(inode_alloc_inode);
685 * Serialize with orphan dir recovery. If the process doing
686 * recovery on this orphan dir does an iget() with the dir
687 * i_mutex held, we'll deadlock here. Instead we detect this
688 * and exit early - recovery will wipe this inode for us.
690 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
695 spin_lock(&osb->osb_lock);
696 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
697 mlog(0, "Recovery is happening on orphan dir %d, will skip "
698 "this inode\n", slot);
702 /* This signals to the orphan recovery process that it should
703 * wait for us to handle the wipe. */
704 osb->osb_orphan_wipes[slot]++;
706 spin_unlock(&osb->osb_lock);
710 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
713 spin_lock(&osb->osb_lock);
714 osb->osb_orphan_wipes[slot]--;
715 spin_unlock(&osb->osb_lock);
717 wake_up(&osb->osb_wipe_event);
720 static int ocfs2_wipe_inode(struct inode *inode,
721 struct buffer_head *di_bh)
723 int status, orphaned_slot = -1;
724 struct inode *orphan_dir_inode = NULL;
725 struct buffer_head *orphan_dir_bh = NULL;
726 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
727 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
729 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
730 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
732 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
736 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
737 ORPHAN_DIR_SYSTEM_INODE,
739 if (!orphan_dir_inode) {
745 /* Lock the orphan dir. The lock will be held for the entire
746 * delete_inode operation. We do this now to avoid races with
747 * recovery completion on other nodes. */
748 mutex_lock(&orphan_dir_inode->i_mutex);
749 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
751 mutex_unlock(&orphan_dir_inode->i_mutex);
758 /* we do this while holding the orphan dir lock because we
759 * don't want recovery being run from another node to try an
760 * inode delete underneath us -- this will result in two nodes
761 * truncating the same file! */
762 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
765 goto bail_unlock_dir;
768 /* Remove any dir index tree */
769 if (S_ISDIR(inode->i_mode)) {
770 status = ocfs2_dx_dir_truncate(inode, di_bh);
773 goto bail_unlock_dir;
777 /*Free extended attribute resources associated with this inode.*/
778 status = ocfs2_xattr_remove(inode, di_bh);
781 goto bail_unlock_dir;
784 status = ocfs2_remove_refcount_tree(inode, di_bh);
787 goto bail_unlock_dir;
790 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
796 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
799 ocfs2_inode_unlock(orphan_dir_inode, 1);
800 mutex_unlock(&orphan_dir_inode->i_mutex);
801 brelse(orphan_dir_bh);
803 iput(orphan_dir_inode);
804 ocfs2_signal_wipe_completion(osb, orphaned_slot);
809 /* There is a series of simple checks that should be done before a
810 * trylock is even considered. Encapsulate those in this function. */
811 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
814 struct ocfs2_inode_info *oi = OCFS2_I(inode);
815 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
817 /* We shouldn't be getting here for the root directory
819 if (inode == osb->root_inode) {
820 mlog(ML_ERROR, "Skipping delete of root inode.\n");
824 /* If we're coming from downconvert_thread we can't go into our own
825 * voting [hello, deadlock city!], so unforuntately we just
826 * have to skip deleting this guy. That's OK though because
827 * the node who's doing the actual deleting should handle it
829 if (current == osb->dc_task) {
830 mlog(0, "Skipping delete of %lu because we're currently "
831 "in downconvert\n", inode->i_ino);
835 spin_lock(&oi->ip_lock);
836 /* OCFS2 *never* deletes system files. This should technically
837 * never get here as system file inodes should always have a
838 * positive link count. */
839 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
840 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
841 (unsigned long long)oi->ip_blkno);
845 /* If we have allowd wipe of this inode for another node, it
846 * will be marked here so we can safely skip it. Recovery will
847 * cleanup any inodes we might inadvertantly skip here. */
848 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
849 mlog(0, "Skipping delete of %lu because another node "
850 "has done this for us.\n", inode->i_ino);
856 spin_unlock(&oi->ip_lock);
861 /* Query the cluster to determine whether we should wipe an inode from
864 * Requires the inode to have the cluster lock. */
865 static int ocfs2_query_inode_wipe(struct inode *inode,
866 struct buffer_head *di_bh,
870 struct ocfs2_inode_info *oi = OCFS2_I(inode);
871 struct ocfs2_dinode *di;
875 /* While we were waiting for the cluster lock in
876 * ocfs2_delete_inode, another node might have asked to delete
877 * the inode. Recheck our flags to catch this. */
878 if (!ocfs2_inode_is_valid_to_delete(inode)) {
879 mlog(0, "Skipping delete of %llu because flags changed\n",
880 (unsigned long long)oi->ip_blkno);
884 /* Now that we have an up to date inode, we can double check
886 if (inode->i_nlink) {
887 mlog(0, "Skipping delete of %llu because nlink = %u\n",
888 (unsigned long long)oi->ip_blkno, inode->i_nlink);
892 /* Do some basic inode verification... */
893 di = (struct ocfs2_dinode *) di_bh->b_data;
894 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
895 !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
897 * Inodes in the orphan dir must have ORPHANED_FL. The only
898 * inodes that come back out of the orphan dir are reflink
899 * targets. A reflink target may be moved out of the orphan
900 * dir between the time we scan the directory and the time we
901 * process it. This would lead to HAS_REFCOUNT_FL being set but
904 if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
905 mlog(0, "Reflinked inode %llu is no longer orphaned. "
906 "it shouldn't be deleted\n",
907 (unsigned long long)oi->ip_blkno);
911 /* for lack of a better error? */
914 "Inode %llu (on-disk %llu) not orphaned! "
915 "Disk flags 0x%x, inode flags 0x%x\n",
916 (unsigned long long)oi->ip_blkno,
917 (unsigned long long)le64_to_cpu(di->i_blkno),
918 le32_to_cpu(di->i_flags), oi->ip_flags);
922 /* has someone already deleted us?! baaad... */
930 * This is how ocfs2 determines whether an inode is still live
931 * within the cluster. Every node takes a shared read lock on
932 * the inode open lock in ocfs2_read_locked_inode(). When we
933 * get to ->delete_inode(), each node tries to convert it's
934 * lock to an exclusive. Trylocks are serialized by the inode
935 * meta data lock. If the upconvert suceeds, we know the inode
936 * is no longer live and can be deleted.
938 * Though we call this with the meta data lock held, the
939 * trylock keeps us from ABBA deadlock.
941 status = ocfs2_try_open_lock(inode, 1);
942 if (status == -EAGAIN) {
944 mlog(0, "Skipping delete of %llu because it is in use on "
945 "other nodes\n", (unsigned long long)oi->ip_blkno);
954 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
955 (unsigned long long)oi->ip_blkno,
956 le16_to_cpu(di->i_orphaned_slot));
962 /* Support function for ocfs2_delete_inode. Will help us keep the
963 * inode data in a consistent state for clear_inode. Always truncates
964 * pages, optionally sync's them first. */
965 static void ocfs2_cleanup_delete_inode(struct inode *inode,
968 mlog(0, "Cleanup inode %llu, sync = %d\n",
969 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
971 write_inode_now(inode, 1);
972 truncate_inode_pages(&inode->i_data, 0);
975 static void ocfs2_delete_inode(struct inode *inode)
979 struct buffer_head *di_bh = NULL;
981 mlog(0, "(inode->i_ino = %lu)\n", inode->i_ino);
983 /* When we fail in read_inode() we mark inode as bad. The second test
984 * catches the case when inode allocation fails before allocating
985 * a block for inode. */
986 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
987 mlog(0, "Skipping delete of bad inode\n");
991 dquot_initialize(inode);
993 if (!ocfs2_inode_is_valid_to_delete(inode)) {
994 /* It's probably not necessary to truncate_inode_pages
995 * here but we do it for safety anyway (it will most
996 * likely be a no-op anyway) */
997 ocfs2_cleanup_delete_inode(inode, 0);
1001 /* We want to block signals in delete_inode as the lock and
1002 * messaging paths may return us -ERESTARTSYS. Which would
1003 * cause us to exit early, resulting in inodes being orphaned
1005 ocfs2_block_signals(&oldset);
1008 * Synchronize us against ocfs2_get_dentry. We take this in
1009 * shared mode so that all nodes can still concurrently
1012 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1014 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1015 ocfs2_cleanup_delete_inode(inode, 0);
1018 /* Lock down the inode. This gives us an up to date view of
1019 * it's metadata (for verification), and allows us to
1020 * serialize delete_inode on multiple nodes.
1022 * Even though we might be doing a truncate, we don't take the
1023 * allocation lock here as it won't be needed - nobody will
1024 * have the file open.
1026 status = ocfs2_inode_lock(inode, &di_bh, 1);
1028 if (status != -ENOENT)
1030 ocfs2_cleanup_delete_inode(inode, 0);
1031 goto bail_unlock_nfs_sync;
1034 /* Query the cluster. This will be the final decision made
1035 * before we go ahead and wipe the inode. */
1036 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1037 if (!wipe || status < 0) {
1038 /* Error and remote inode busy both mean we won't be
1039 * removing the inode, so they take almost the same
1044 /* Someone in the cluster has disallowed a wipe of
1045 * this inode, or it was never completely
1046 * orphaned. Write out the pages and exit now. */
1047 ocfs2_cleanup_delete_inode(inode, 1);
1048 goto bail_unlock_inode;
1051 ocfs2_cleanup_delete_inode(inode, 0);
1053 status = ocfs2_wipe_inode(inode, di_bh);
1055 if (status != -EDEADLK)
1057 goto bail_unlock_inode;
1061 * Mark the inode as successfully deleted.
1063 * This is important for ocfs2_clear_inode() as it will check
1064 * this flag and skip any checkpointing work
1066 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1067 * the LVB for other nodes.
1069 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1072 ocfs2_inode_unlock(inode, 1);
1075 bail_unlock_nfs_sync:
1076 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1079 ocfs2_unblock_signals(&oldset);
1084 static void ocfs2_clear_inode(struct inode *inode)
1087 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1089 end_writeback(inode);
1090 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1091 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1093 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1094 "Inode=%lu\n", inode->i_ino);
1098 /* To preven remote deletes we hold open lock before, now it
1099 * is time to unlock PR and EX open locks. */
1100 ocfs2_open_unlock(inode);
1102 /* Do these before all the other work so that we don't bounce
1103 * the downconvert thread while waiting to destroy the locks. */
1104 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1105 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1106 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1108 ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1109 &oi->ip_la_data_resv);
1110 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1112 /* We very well may get a clear_inode before all an inodes
1113 * metadata has hit disk. Of course, we can't drop any cluster
1114 * locks until the journal has finished with it. The only
1115 * exception here are successfully wiped inodes - their
1116 * metadata can now be considered to be part of the system
1117 * inodes from which it came. */
1118 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1119 ocfs2_checkpoint_inode(inode);
1121 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1122 "Clear inode of %llu, inode has io markers\n",
1123 (unsigned long long)oi->ip_blkno);
1125 ocfs2_extent_map_trunc(inode, 0);
1127 status = ocfs2_drop_inode_locks(inode);
1131 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1132 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1133 ocfs2_lock_res_free(&oi->ip_open_lockres);
1135 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1137 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1138 "Clear inode of %llu, inode has %u cache items\n",
1139 (unsigned long long)oi->ip_blkno,
1140 INODE_CACHE(inode)->ci_num_cached);
1142 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1143 "Clear inode of %llu, inode has a bad flag\n",
1144 (unsigned long long)oi->ip_blkno);
1146 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1147 "Clear inode of %llu, inode is locked\n",
1148 (unsigned long long)oi->ip_blkno);
1150 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1151 "Clear inode of %llu, io_mutex is locked\n",
1152 (unsigned long long)oi->ip_blkno);
1153 mutex_unlock(&oi->ip_io_mutex);
1156 * down_trylock() returns 0, down_write_trylock() returns 1
1159 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1160 "Clear inode of %llu, alloc_sem is locked\n",
1161 (unsigned long long)oi->ip_blkno);
1162 up_write(&oi->ip_alloc_sem);
1164 mlog_bug_on_msg(oi->ip_open_count,
1165 "Clear inode of %llu has open count %d\n",
1166 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1168 /* Clear all other flags. */
1170 oi->ip_dir_start_lookup = 0;
1171 oi->ip_blkno = 0ULL;
1174 * ip_jinode is used to track txns against this inode. We ensure that
1175 * the journal is flushed before journal shutdown. Thus it is safe to
1176 * have inodes get cleaned up after journal shutdown.
1178 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1184 void ocfs2_evict_inode(struct inode *inode)
1186 if (!inode->i_nlink ||
1187 (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1188 ocfs2_delete_inode(inode);
1190 truncate_inode_pages(&inode->i_data, 0);
1192 ocfs2_clear_inode(inode);
1195 /* Called under inode_lock, with no more references on the
1196 * struct inode, so it's safe here to check the flags field
1197 * and to manipulate i_nlink without any other locks. */
1198 int ocfs2_drop_inode(struct inode *inode)
1200 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1203 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1204 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1206 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1209 res = generic_drop_inode(inode);
1216 * This is called from our getattr.
1218 int ocfs2_inode_revalidate(struct dentry *dentry)
1220 struct inode *inode = dentry->d_inode;
1223 mlog(0, "(inode = 0x%p, ino = %llu)\n", inode,
1224 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1227 mlog(0, "eep, no inode!\n");
1232 spin_lock(&OCFS2_I(inode)->ip_lock);
1233 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1234 spin_unlock(&OCFS2_I(inode)->ip_lock);
1235 mlog(0, "inode deleted!\n");
1239 spin_unlock(&OCFS2_I(inode)->ip_lock);
1241 /* Let ocfs2_inode_lock do the work of updating our struct
1243 status = ocfs2_inode_lock(inode, NULL, 0);
1245 if (status != -ENOENT)
1249 ocfs2_inode_unlock(inode, 0);
1257 * Updates a disk inode from a
1259 * Only takes ip_lock.
1261 int ocfs2_mark_inode_dirty(handle_t *handle,
1262 struct inode *inode,
1263 struct buffer_head *bh)
1266 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1268 mlog(0, "(inode %llu)\n",
1269 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1271 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1272 OCFS2_JOURNAL_ACCESS_WRITE);
1278 spin_lock(&OCFS2_I(inode)->ip_lock);
1279 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1280 ocfs2_get_inode_flags(OCFS2_I(inode));
1281 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1282 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1283 spin_unlock(&OCFS2_I(inode)->ip_lock);
1285 fe->i_size = cpu_to_le64(i_size_read(inode));
1286 ocfs2_set_links_count(fe, inode->i_nlink);
1287 fe->i_uid = cpu_to_le32(inode->i_uid);
1288 fe->i_gid = cpu_to_le32(inode->i_gid);
1289 fe->i_mode = cpu_to_le16(inode->i_mode);
1290 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1291 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1292 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1293 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1294 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1295 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1297 ocfs2_journal_dirty(handle, bh);
1305 * Updates a struct inode from a disk inode.
1306 * does no i/o, only takes ip_lock.
1308 void ocfs2_refresh_inode(struct inode *inode,
1309 struct ocfs2_dinode *fe)
1311 spin_lock(&OCFS2_I(inode)->ip_lock);
1313 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1314 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1315 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1316 ocfs2_set_inode_flags(inode);
1317 i_size_write(inode, le64_to_cpu(fe->i_size));
1318 inode->i_nlink = ocfs2_read_links_count(fe);
1319 inode->i_uid = le32_to_cpu(fe->i_uid);
1320 inode->i_gid = le32_to_cpu(fe->i_gid);
1321 inode->i_mode = le16_to_cpu(fe->i_mode);
1322 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1323 inode->i_blocks = 0;
1325 inode->i_blocks = ocfs2_inode_sector_count(inode);
1326 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1327 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1328 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1329 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1330 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1331 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1333 spin_unlock(&OCFS2_I(inode)->ip_lock);
1336 int ocfs2_validate_inode_block(struct super_block *sb,
1337 struct buffer_head *bh)
1340 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1342 mlog(0, "Validating dinode %llu\n",
1343 (unsigned long long)bh->b_blocknr);
1345 BUG_ON(!buffer_uptodate(bh));
1348 * If the ecc fails, we return the error but otherwise
1349 * leave the filesystem running. We know any error is
1350 * local to this block.
1352 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1354 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1355 (unsigned long long)bh->b_blocknr);
1360 * Errors after here are fatal.
1365 if (!OCFS2_IS_VALID_DINODE(di)) {
1366 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1367 (unsigned long long)bh->b_blocknr, 7,
1372 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1373 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1374 (unsigned long long)bh->b_blocknr,
1375 (unsigned long long)le64_to_cpu(di->i_blkno));
1379 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1381 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1382 (unsigned long long)bh->b_blocknr);
1386 if (le32_to_cpu(di->i_fs_generation) !=
1387 OCFS2_SB(sb)->fs_generation) {
1389 "Invalid dinode #%llu: fs_generation is %u\n",
1390 (unsigned long long)bh->b_blocknr,
1391 le32_to_cpu(di->i_fs_generation));
1401 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1405 struct buffer_head *tmp = *bh;
1407 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1408 1, &tmp, flags, ocfs2_validate_inode_block);
1410 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1417 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1419 return ocfs2_read_inode_block_full(inode, bh, 0);
1423 static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1425 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1427 return oi->ip_blkno;
1430 static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1432 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1434 return oi->vfs_inode.i_sb;
1437 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1439 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1441 spin_lock(&oi->ip_lock);
1444 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1446 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1448 spin_unlock(&oi->ip_lock);
1451 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1453 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1455 mutex_lock(&oi->ip_io_mutex);
1458 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1460 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1462 mutex_unlock(&oi->ip_io_mutex);
1465 const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1466 .co_owner = ocfs2_inode_cache_owner,
1467 .co_get_super = ocfs2_inode_cache_get_super,
1468 .co_cache_lock = ocfs2_inode_cache_lock,
1469 .co_cache_unlock = ocfs2_inode_cache_unlock,
1470 .co_io_lock = ocfs2_inode_cache_io_lock,
1471 .co_io_unlock = ocfs2_inode_cache_io_unlock,