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/quotaops.h>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
41 #include "blockcheck.h"
43 #include "extent_map.h"
45 #include "heartbeat.h"
56 #include "buffer_head_io.h"
58 struct ocfs2_find_inode_args
62 unsigned int fi_flags;
63 unsigned int fi_sysfile_type;
66 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
68 static int ocfs2_read_locked_inode(struct inode *inode,
69 struct ocfs2_find_inode_args *args);
70 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
71 static int ocfs2_find_actor(struct inode *inode, void *opaque);
72 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
74 struct buffer_head *fe_bh);
76 void ocfs2_set_inode_flags(struct inode *inode)
78 unsigned int flags = OCFS2_I(inode)->ip_attr;
80 inode->i_flags &= ~(S_IMMUTABLE |
81 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
83 if (flags & OCFS2_IMMUTABLE_FL)
84 inode->i_flags |= S_IMMUTABLE;
86 if (flags & OCFS2_SYNC_FL)
87 inode->i_flags |= S_SYNC;
88 if (flags & OCFS2_APPEND_FL)
89 inode->i_flags |= S_APPEND;
90 if (flags & OCFS2_NOATIME_FL)
91 inode->i_flags |= S_NOATIME;
92 if (flags & OCFS2_DIRSYNC_FL)
93 inode->i_flags |= S_DIRSYNC;
96 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
97 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
99 unsigned int flags = oi->vfs_inode.i_flags;
101 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
102 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
104 oi->ip_attr |= OCFS2_SYNC_FL;
105 if (flags & S_APPEND)
106 oi->ip_attr |= OCFS2_APPEND_FL;
107 if (flags & S_IMMUTABLE)
108 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
109 if (flags & S_NOATIME)
110 oi->ip_attr |= OCFS2_NOATIME_FL;
111 if (flags & S_DIRSYNC)
112 oi->ip_attr |= OCFS2_DIRSYNC_FL;
115 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
118 struct inode *inode = NULL;
119 struct super_block *sb = osb->sb;
120 struct ocfs2_find_inode_args args;
122 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
124 /* Ok. By now we've either got the offsets passed to us by the
125 * caller, or we just pulled them off the bh. Lets do some
126 * sanity checks to make sure they're OK. */
128 inode = ERR_PTR(-EINVAL);
129 mlog_errno(PTR_ERR(inode));
133 args.fi_blkno = blkno;
134 args.fi_flags = flags;
135 args.fi_ino = ino_from_blkno(sb, blkno);
136 args.fi_sysfile_type = sysfile_type;
138 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
139 ocfs2_init_locked_inode, &args);
140 /* inode was *not* in the inode cache. 2.6.x requires
141 * us to do our own read_inode call and unlock it
143 if (inode && inode->i_state & I_NEW) {
144 mlog(0, "Inode was not in inode cache, reading it.\n");
145 ocfs2_read_locked_inode(inode, &args);
146 unlock_new_inode(inode);
149 inode = ERR_PTR(-ENOMEM);
150 mlog_errno(PTR_ERR(inode));
153 if (is_bad_inode(inode)) {
155 inode = ERR_PTR(-ESTALE);
160 if (!IS_ERR(inode)) {
161 mlog(0, "returning inode with number %llu\n",
162 (unsigned long long)OCFS2_I(inode)->ip_blkno);
163 mlog_exit_ptr(inode);
171 * here's how inodes get read from disk:
172 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
173 * found? : return the in-memory inode
174 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
177 static int ocfs2_find_actor(struct inode *inode, void *opaque)
179 struct ocfs2_find_inode_args *args = NULL;
180 struct ocfs2_inode_info *oi = OCFS2_I(inode);
183 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
187 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
189 if (oi->ip_blkno != args->fi_blkno)
199 * initialize the new inode, but don't do anything that would cause
201 * return 0 on success, 1 on failure
203 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
205 struct ocfs2_find_inode_args *args = opaque;
207 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
209 inode->i_ino = args->fi_ino;
210 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
211 if (args->fi_sysfile_type != 0)
212 lockdep_set_class(&inode->i_mutex,
213 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
219 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
222 struct super_block *sb;
223 struct ocfs2_super *osb;
226 mlog_entry("(0x%p, size:%llu)\n", inode,
227 (unsigned long long)le64_to_cpu(fe->i_size));
232 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
233 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
237 * These have all been checked by ocfs2_read_inode_block() or set
238 * by ocfs2_mknod_locked(), so a failure is a code bug.
240 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
241 cannot create a superblock
242 inode today. change if
244 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
245 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
248 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
249 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
250 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
252 inode->i_version = 1;
253 inode->i_generation = le32_to_cpu(fe->i_generation);
254 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
255 inode->i_mode = le16_to_cpu(fe->i_mode);
256 inode->i_uid = le32_to_cpu(fe->i_uid);
257 inode->i_gid = le32_to_cpu(fe->i_gid);
259 /* Fast symlinks will have i_size but no allocated clusters. */
260 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
263 inode->i_blocks = ocfs2_inode_sector_count(inode);
264 inode->i_mapping->a_ops = &ocfs2_aops;
265 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
266 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
267 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
268 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
269 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
270 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
272 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
274 "ip_blkno %llu != i_blkno %llu!\n",
275 (unsigned long long)OCFS2_I(inode)->ip_blkno,
276 (unsigned long long)le64_to_cpu(fe->i_blkno));
278 inode->i_nlink = le16_to_cpu(fe->i_links_count);
280 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
281 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
282 inode->i_flags |= S_NOQUOTA;
285 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
286 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
287 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
288 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
289 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
290 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
291 inode->i_flags |= S_NOQUOTA;
292 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
293 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
294 /* we can't actually hit this as read_inode can't
295 * handle superblocks today ;-) */
299 switch (inode->i_mode & S_IFMT) {
302 inode->i_fop = &ocfs2_fops;
304 inode->i_fop = &ocfs2_fops_no_plocks;
305 inode->i_op = &ocfs2_file_iops;
306 i_size_write(inode, le64_to_cpu(fe->i_size));
309 inode->i_op = &ocfs2_dir_iops;
311 inode->i_fop = &ocfs2_dops;
313 inode->i_fop = &ocfs2_dops_no_plocks;
314 i_size_write(inode, le64_to_cpu(fe->i_size));
317 if (ocfs2_inode_is_fast_symlink(inode))
318 inode->i_op = &ocfs2_fast_symlink_inode_operations;
320 inode->i_op = &ocfs2_symlink_inode_operations;
321 i_size_write(inode, le64_to_cpu(fe->i_size));
324 inode->i_op = &ocfs2_special_file_iops;
325 init_special_inode(inode, inode->i_mode,
331 inode->i_ino = ino_from_blkno(inode->i_sb,
332 le64_to_cpu(fe->i_blkno));
335 * If we ever want to create system files from kernel,
336 * the generation argument to
337 * ocfs2_inode_lock_res_init() will have to change.
339 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
341 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
342 OCFS2_LOCK_TYPE_META, 0, inode);
344 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
345 OCFS2_LOCK_TYPE_OPEN, 0, inode);
348 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
349 OCFS2_LOCK_TYPE_RW, inode->i_generation,
352 ocfs2_set_inode_flags(inode);
357 static int ocfs2_read_locked_inode(struct inode *inode,
358 struct ocfs2_find_inode_args *args)
360 struct super_block *sb;
361 struct ocfs2_super *osb;
362 struct ocfs2_dinode *fe;
363 struct buffer_head *bh = NULL;
364 int status, can_lock;
367 mlog_entry("(0x%p, 0x%p)\n", inode, args);
370 if (inode == NULL || inode->i_sb == NULL) {
371 mlog(ML_ERROR, "bad inode\n");
378 mlog(ML_ERROR, "bad inode args\n");
379 make_bad_inode(inode);
384 * To improve performance of cold-cache inode stats, we take
385 * the cluster lock here if possible.
387 * Generally, OCFS2 never trusts the contents of an inode
388 * unless it's holding a cluster lock, so taking it here isn't
389 * a correctness issue as much as it is a performance
392 * There are three times when taking the lock is not a good idea:
394 * 1) During startup, before we have initialized the DLM.
396 * 2) If we are reading certain system files which never get
397 * cluster locks (local alloc, truncate log).
399 * 3) If the process doing the iget() is responsible for
400 * orphan dir recovery. We're holding the orphan dir lock and
401 * can get into a deadlock with another process on another
402 * node in ->delete_inode().
404 * #1 and #2 can be simply solved by never taking the lock
405 * here for system files (which are the only type we read
406 * during mount). It's a heavier approach, but our main
407 * concern is user-accesible files anyway.
409 * #3 works itself out because we'll eventually take the
410 * cluster lock before trusting anything anyway.
412 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
413 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
414 && !ocfs2_mount_local(osb);
417 * To maintain backwards compatibility with older versions of
418 * ocfs2-tools, we still store the generation value for system
419 * files. The only ones that actually matter to userspace are
420 * the journals, but it's easier and inexpensive to just flag
421 * all system files similarly.
423 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
424 generation = osb->fs_generation;
426 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
427 OCFS2_LOCK_TYPE_META,
430 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
431 OCFS2_LOCK_TYPE_OPEN,
435 status = ocfs2_open_lock(inode);
437 make_bad_inode(inode);
441 status = ocfs2_inode_lock(inode, NULL, 0);
443 make_bad_inode(inode);
449 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
450 status = ocfs2_try_open_lock(inode, 0);
452 make_bad_inode(inode);
458 status = ocfs2_read_inode_block_full(inode, &bh,
459 OCFS2_BH_IGNORE_CACHE);
461 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
463 status = ocfs2_validate_inode_block(osb->sb, bh);
471 fe = (struct ocfs2_dinode *) bh->b_data;
474 * This is a code bug. Right now the caller needs to
475 * understand whether it is asking for a system file inode or
476 * not so the proper lock names can be built.
478 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
479 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
480 "Inode %llu: system file state is ambigous\n",
481 (unsigned long long)args->fi_blkno);
483 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
484 S_ISBLK(le16_to_cpu(fe->i_mode)))
485 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
487 ocfs2_populate_inode(inode, fe, 0);
489 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
495 ocfs2_inode_unlock(inode, 0);
498 make_bad_inode(inode);
507 void ocfs2_sync_blockdev(struct super_block *sb)
509 sync_blockdev(sb->s_bdev);
512 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
514 struct buffer_head *fe_bh)
517 struct ocfs2_truncate_context *tc = NULL;
518 struct ocfs2_dinode *fe;
519 handle_t *handle = NULL;
523 fe = (struct ocfs2_dinode *) fe_bh->b_data;
526 * This check will also skip truncate of inodes with inline
527 * data and fast symlinks.
529 if (fe->i_clusters) {
530 if (ocfs2_should_order_data(inode))
531 ocfs2_begin_ordered_truncate(inode, 0);
533 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
534 if (IS_ERR(handle)) {
535 status = PTR_ERR(handle);
540 status = ocfs2_journal_access_di(handle, inode, fe_bh,
541 OCFS2_JOURNAL_ACCESS_WRITE);
547 i_size_write(inode, 0);
549 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
555 ocfs2_commit_trans(osb, handle);
558 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
564 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
573 ocfs2_commit_trans(osb, handle);
578 static int ocfs2_remove_inode(struct inode *inode,
579 struct buffer_head *di_bh,
580 struct inode *orphan_dir_inode,
581 struct buffer_head *orphan_dir_bh)
584 struct inode *inode_alloc_inode = NULL;
585 struct buffer_head *inode_alloc_bh = NULL;
587 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
588 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
591 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
592 le16_to_cpu(di->i_suballoc_slot));
593 if (!inode_alloc_inode) {
599 mutex_lock(&inode_alloc_inode->i_mutex);
600 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
602 mutex_unlock(&inode_alloc_inode->i_mutex);
608 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
609 ocfs2_quota_trans_credits(inode->i_sb));
610 if (IS_ERR(handle)) {
611 status = PTR_ERR(handle);
616 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
623 /* set the inodes dtime */
624 status = ocfs2_journal_access_di(handle, inode, di_bh,
625 OCFS2_JOURNAL_ACCESS_WRITE);
631 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
632 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
634 status = ocfs2_journal_dirty(handle, di_bh);
640 ocfs2_remove_from_cache(inode, di_bh);
641 vfs_dq_free_inode(inode);
643 status = ocfs2_free_dinode(handle, inode_alloc_inode,
649 ocfs2_commit_trans(osb, handle);
651 ocfs2_inode_unlock(inode_alloc_inode, 1);
652 mutex_unlock(&inode_alloc_inode->i_mutex);
653 brelse(inode_alloc_bh);
655 iput(inode_alloc_inode);
661 * Serialize with orphan dir recovery. If the process doing
662 * recovery on this orphan dir does an iget() with the dir
663 * i_mutex held, we'll deadlock here. Instead we detect this
664 * and exit early - recovery will wipe this inode for us.
666 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
671 spin_lock(&osb->osb_lock);
672 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
673 mlog(0, "Recovery is happening on orphan dir %d, will skip "
674 "this inode\n", slot);
678 /* This signals to the orphan recovery process that it should
679 * wait for us to handle the wipe. */
680 osb->osb_orphan_wipes[slot]++;
682 spin_unlock(&osb->osb_lock);
686 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
689 spin_lock(&osb->osb_lock);
690 osb->osb_orphan_wipes[slot]--;
691 spin_unlock(&osb->osb_lock);
693 wake_up(&osb->osb_wipe_event);
696 static int ocfs2_wipe_inode(struct inode *inode,
697 struct buffer_head *di_bh)
699 int status, orphaned_slot;
700 struct inode *orphan_dir_inode = NULL;
701 struct buffer_head *orphan_dir_bh = NULL;
702 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
703 struct ocfs2_dinode *di;
705 di = (struct ocfs2_dinode *) di_bh->b_data;
706 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
708 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
712 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
713 ORPHAN_DIR_SYSTEM_INODE,
715 if (!orphan_dir_inode) {
721 /* Lock the orphan dir. The lock will be held for the entire
722 * delete_inode operation. We do this now to avoid races with
723 * recovery completion on other nodes. */
724 mutex_lock(&orphan_dir_inode->i_mutex);
725 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
727 mutex_unlock(&orphan_dir_inode->i_mutex);
733 /* we do this while holding the orphan dir lock because we
734 * don't want recovery being run from another node to try an
735 * inode delete underneath us -- this will result in two nodes
736 * truncating the same file! */
737 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
740 goto bail_unlock_dir;
743 /*Free extended attribute resources associated with this inode.*/
744 status = ocfs2_xattr_remove(inode, di_bh);
747 goto bail_unlock_dir;
750 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
756 ocfs2_inode_unlock(orphan_dir_inode, 1);
757 mutex_unlock(&orphan_dir_inode->i_mutex);
758 brelse(orphan_dir_bh);
760 iput(orphan_dir_inode);
761 ocfs2_signal_wipe_completion(osb, orphaned_slot);
766 /* There is a series of simple checks that should be done before a
767 * trylock is even considered. Encapsulate those in this function. */
768 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
771 struct ocfs2_inode_info *oi = OCFS2_I(inode);
772 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
774 /* We shouldn't be getting here for the root directory
776 if (inode == osb->root_inode) {
777 mlog(ML_ERROR, "Skipping delete of root inode.\n");
781 /* If we're coming from downconvert_thread we can't go into our own
782 * voting [hello, deadlock city!], so unforuntately we just
783 * have to skip deleting this guy. That's OK though because
784 * the node who's doing the actual deleting should handle it
786 if (current == osb->dc_task) {
787 mlog(0, "Skipping delete of %lu because we're currently "
788 "in downconvert\n", inode->i_ino);
792 spin_lock(&oi->ip_lock);
793 /* OCFS2 *never* deletes system files. This should technically
794 * never get here as system file inodes should always have a
795 * positive link count. */
796 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
797 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
798 (unsigned long long)oi->ip_blkno);
802 /* If we have allowd wipe of this inode for another node, it
803 * will be marked here so we can safely skip it. Recovery will
804 * cleanup any inodes we might inadvertantly skip here. */
805 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
806 mlog(0, "Skipping delete of %lu because another node "
807 "has done this for us.\n", inode->i_ino);
813 spin_unlock(&oi->ip_lock);
818 /* Query the cluster to determine whether we should wipe an inode from
821 * Requires the inode to have the cluster lock. */
822 static int ocfs2_query_inode_wipe(struct inode *inode,
823 struct buffer_head *di_bh,
827 struct ocfs2_inode_info *oi = OCFS2_I(inode);
828 struct ocfs2_dinode *di;
832 /* While we were waiting for the cluster lock in
833 * ocfs2_delete_inode, another node might have asked to delete
834 * the inode. Recheck our flags to catch this. */
835 if (!ocfs2_inode_is_valid_to_delete(inode)) {
836 mlog(0, "Skipping delete of %llu because flags changed\n",
837 (unsigned long long)oi->ip_blkno);
841 /* Now that we have an up to date inode, we can double check
843 if (inode->i_nlink) {
844 mlog(0, "Skipping delete of %llu because nlink = %u\n",
845 (unsigned long long)oi->ip_blkno, inode->i_nlink);
849 /* Do some basic inode verification... */
850 di = (struct ocfs2_dinode *) di_bh->b_data;
851 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
852 /* for lack of a better error? */
855 "Inode %llu (on-disk %llu) not orphaned! "
856 "Disk flags 0x%x, inode flags 0x%x\n",
857 (unsigned long long)oi->ip_blkno,
858 (unsigned long long)le64_to_cpu(di->i_blkno),
859 le32_to_cpu(di->i_flags), oi->ip_flags);
863 /* has someone already deleted us?! baaad... */
871 * This is how ocfs2 determines whether an inode is still live
872 * within the cluster. Every node takes a shared read lock on
873 * the inode open lock in ocfs2_read_locked_inode(). When we
874 * get to ->delete_inode(), each node tries to convert it's
875 * lock to an exclusive. Trylocks are serialized by the inode
876 * meta data lock. If the upconvert suceeds, we know the inode
877 * is no longer live and can be deleted.
879 * Though we call this with the meta data lock held, the
880 * trylock keeps us from ABBA deadlock.
882 status = ocfs2_try_open_lock(inode, 1);
883 if (status == -EAGAIN) {
885 mlog(0, "Skipping delete of %llu because it is in use on "
886 "other nodes\n", (unsigned long long)oi->ip_blkno);
895 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
896 (unsigned long long)oi->ip_blkno,
897 le16_to_cpu(di->i_orphaned_slot));
903 /* Support function for ocfs2_delete_inode. Will help us keep the
904 * inode data in a consistent state for clear_inode. Always truncates
905 * pages, optionally sync's them first. */
906 static void ocfs2_cleanup_delete_inode(struct inode *inode,
909 mlog(0, "Cleanup inode %llu, sync = %d\n",
910 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
912 write_inode_now(inode, 1);
913 truncate_inode_pages(&inode->i_data, 0);
916 void ocfs2_delete_inode(struct inode *inode)
919 sigset_t blocked, oldset;
920 struct buffer_head *di_bh = NULL;
922 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
924 /* When we fail in read_inode() we mark inode as bad. The second test
925 * catches the case when inode allocation fails before allocating
926 * a block for inode. */
927 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
928 mlog(0, "Skipping delete of bad inode\n");
932 if (!ocfs2_inode_is_valid_to_delete(inode)) {
933 /* It's probably not necessary to truncate_inode_pages
934 * here but we do it for safety anyway (it will most
935 * likely be a no-op anyway) */
936 ocfs2_cleanup_delete_inode(inode, 0);
940 /* We want to block signals in delete_inode as the lock and
941 * messaging paths may return us -ERESTARTSYS. Which would
942 * cause us to exit early, resulting in inodes being orphaned
944 sigfillset(&blocked);
945 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
948 ocfs2_cleanup_delete_inode(inode, 1);
952 /* Lock down the inode. This gives us an up to date view of
953 * it's metadata (for verification), and allows us to
954 * serialize delete_inode on multiple nodes.
956 * Even though we might be doing a truncate, we don't take the
957 * allocation lock here as it won't be needed - nobody will
958 * have the file open.
960 status = ocfs2_inode_lock(inode, &di_bh, 1);
962 if (status != -ENOENT)
964 ocfs2_cleanup_delete_inode(inode, 0);
968 /* Query the cluster. This will be the final decision made
969 * before we go ahead and wipe the inode. */
970 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
971 if (!wipe || status < 0) {
972 /* Error and remote inode busy both mean we won't be
973 * removing the inode, so they take almost the same
978 /* Someone in the cluster has disallowed a wipe of
979 * this inode, or it was never completely
980 * orphaned. Write out the pages and exit now. */
981 ocfs2_cleanup_delete_inode(inode, 1);
982 goto bail_unlock_inode;
985 ocfs2_cleanup_delete_inode(inode, 0);
987 status = ocfs2_wipe_inode(inode, di_bh);
989 if (status != -EDEADLK)
991 goto bail_unlock_inode;
995 * Mark the inode as successfully deleted.
997 * This is important for ocfs2_clear_inode() as it will check
998 * this flag and skip any checkpointing work
1000 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1001 * the LVB for other nodes.
1003 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1006 ocfs2_inode_unlock(inode, 1);
1009 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1017 void ocfs2_clear_inode(struct inode *inode)
1020 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1027 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1028 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1030 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1031 "Inode=%lu\n", inode->i_ino);
1033 /* To preven remote deletes we hold open lock before, now it
1034 * is time to unlock PR and EX open locks. */
1035 ocfs2_open_unlock(inode);
1037 /* Do these before all the other work so that we don't bounce
1038 * the downconvert thread while waiting to destroy the locks. */
1039 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1040 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1041 ocfs2_mark_lockres_freeing(&oi->ip_open_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_trunc(inode, 0);
1058 status = ocfs2_drop_inode_locks(inode);
1062 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1063 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1064 ocfs2_lock_res_free(&oi->ip_open_lockres);
1066 ocfs2_metadata_cache_purge(inode);
1068 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1069 "Clear inode of %llu, inode has %u cache items\n",
1070 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1072 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1073 "Clear inode of %llu, inode has a bad flag\n",
1074 (unsigned long long)oi->ip_blkno);
1076 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1077 "Clear inode of %llu, inode is locked\n",
1078 (unsigned long long)oi->ip_blkno);
1080 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1081 "Clear inode of %llu, io_mutex is locked\n",
1082 (unsigned long long)oi->ip_blkno);
1083 mutex_unlock(&oi->ip_io_mutex);
1086 * down_trylock() returns 0, down_write_trylock() returns 1
1089 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1090 "Clear inode of %llu, alloc_sem is locked\n",
1091 (unsigned long long)oi->ip_blkno);
1092 up_write(&oi->ip_alloc_sem);
1094 mlog_bug_on_msg(oi->ip_open_count,
1095 "Clear inode of %llu has open count %d\n",
1096 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1098 /* Clear all other flags. */
1099 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1100 oi->ip_created_trans = 0;
1101 oi->ip_last_trans = 0;
1102 oi->ip_dir_start_lookup = 0;
1103 oi->ip_blkno = 0ULL;
1106 * ip_jinode is used to track txns against this inode. We ensure that
1107 * the journal is flushed before journal shutdown. Thus it is safe to
1108 * have inodes get cleaned up after journal shutdown.
1110 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1117 /* Called under inode_lock, with no more references on the
1118 * struct inode, so it's safe here to check the flags field
1119 * and to manipulate i_nlink without any other locks. */
1120 void ocfs2_drop_inode(struct inode *inode)
1122 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1126 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1127 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1129 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1130 generic_delete_inode(inode);
1132 generic_drop_inode(inode);
1138 * This is called from our getattr.
1140 int ocfs2_inode_revalidate(struct dentry *dentry)
1142 struct inode *inode = dentry->d_inode;
1145 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1146 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1149 mlog(0, "eep, no inode!\n");
1154 spin_lock(&OCFS2_I(inode)->ip_lock);
1155 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1156 spin_unlock(&OCFS2_I(inode)->ip_lock);
1157 mlog(0, "inode deleted!\n");
1161 spin_unlock(&OCFS2_I(inode)->ip_lock);
1163 /* Let ocfs2_inode_lock do the work of updating our struct
1165 status = ocfs2_inode_lock(inode, NULL, 0);
1167 if (status != -ENOENT)
1171 ocfs2_inode_unlock(inode, 0);
1179 * Updates a disk inode from a
1181 * Only takes ip_lock.
1183 int ocfs2_mark_inode_dirty(handle_t *handle,
1184 struct inode *inode,
1185 struct buffer_head *bh)
1188 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1190 mlog_entry("(inode %llu)\n",
1191 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1193 status = ocfs2_journal_access_di(handle, inode, bh,
1194 OCFS2_JOURNAL_ACCESS_WRITE);
1200 spin_lock(&OCFS2_I(inode)->ip_lock);
1201 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1202 ocfs2_get_inode_flags(OCFS2_I(inode));
1203 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1204 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1205 spin_unlock(&OCFS2_I(inode)->ip_lock);
1207 fe->i_size = cpu_to_le64(i_size_read(inode));
1208 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1209 fe->i_uid = cpu_to_le32(inode->i_uid);
1210 fe->i_gid = cpu_to_le32(inode->i_gid);
1211 fe->i_mode = cpu_to_le16(inode->i_mode);
1212 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1213 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1214 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1215 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1216 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1217 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1219 status = ocfs2_journal_dirty(handle, bh);
1232 * Updates a struct inode from a disk inode.
1233 * does no i/o, only takes ip_lock.
1235 void ocfs2_refresh_inode(struct inode *inode,
1236 struct ocfs2_dinode *fe)
1238 spin_lock(&OCFS2_I(inode)->ip_lock);
1240 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1241 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1242 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1243 ocfs2_set_inode_flags(inode);
1244 i_size_write(inode, le64_to_cpu(fe->i_size));
1245 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1246 inode->i_uid = le32_to_cpu(fe->i_uid);
1247 inode->i_gid = le32_to_cpu(fe->i_gid);
1248 inode->i_mode = le16_to_cpu(fe->i_mode);
1249 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1250 inode->i_blocks = 0;
1252 inode->i_blocks = ocfs2_inode_sector_count(inode);
1253 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1254 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1255 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1256 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1257 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1258 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1260 spin_unlock(&OCFS2_I(inode)->ip_lock);
1263 int ocfs2_validate_inode_block(struct super_block *sb,
1264 struct buffer_head *bh)
1267 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1269 mlog(0, "Validating dinode %llu\n",
1270 (unsigned long long)bh->b_blocknr);
1272 BUG_ON(!buffer_uptodate(bh));
1275 * If the ecc fails, we return the error but otherwise
1276 * leave the filesystem running. We know any error is
1277 * local to this block.
1279 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1281 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1282 (unsigned long long)bh->b_blocknr);
1287 * Errors after here are fatal.
1292 if (!OCFS2_IS_VALID_DINODE(di)) {
1293 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1294 (unsigned long long)bh->b_blocknr, 7,
1299 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1300 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1301 (unsigned long long)bh->b_blocknr,
1302 (unsigned long long)le64_to_cpu(di->i_blkno));
1306 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1308 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1309 (unsigned long long)bh->b_blocknr);
1313 if (le32_to_cpu(di->i_fs_generation) !=
1314 OCFS2_SB(sb)->fs_generation) {
1316 "Invalid dinode #%llu: fs_generation is %u\n",
1317 (unsigned long long)bh->b_blocknr,
1318 le32_to_cpu(di->i_fs_generation));
1328 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1332 struct buffer_head *tmp = *bh;
1334 rc = ocfs2_read_blocks(inode, OCFS2_I(inode)->ip_blkno, 1, &tmp,
1335 flags, ocfs2_validate_inode_block);
1337 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1344 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1346 return ocfs2_read_inode_block_full(inode, bh, 0);