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 #include <cluster/masklog.h>
40 #include "blockcheck.h"
42 #include "extent_map.h"
44 #include "heartbeat.h"
54 #include "refcounttree.h"
55 #include "ocfs2_trace.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 trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
137 /* Ok. By now we've either got the offsets passed to us by the
138 * caller, or we just pulled them off the bh. Lets do some
139 * sanity checks to make sure they're OK. */
141 inode = ERR_PTR(-EINVAL);
142 mlog_errno(PTR_ERR(inode));
146 args.fi_blkno = blkno;
147 args.fi_flags = flags;
148 args.fi_ino = ino_from_blkno(sb, blkno);
149 args.fi_sysfile_type = sysfile_type;
151 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
152 ocfs2_init_locked_inode, &args);
153 /* inode was *not* in the inode cache. 2.6.x requires
154 * us to do our own read_inode call and unlock it
157 inode = ERR_PTR(-ENOMEM);
158 mlog_errno(PTR_ERR(inode));
161 trace_ocfs2_iget5_locked(inode->i_state);
162 if (inode->i_state & I_NEW) {
163 ocfs2_read_locked_inode(inode, &args);
164 unlock_new_inode(inode);
166 if (is_bad_inode(inode)) {
168 inode = ERR_PTR(-ESTALE);
173 if (!IS_ERR(inode)) {
174 trace_ocfs2_iget_end(inode,
175 (unsigned long long)OCFS2_I(inode)->ip_blkno);
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);
197 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
199 trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
201 if (oi->ip_blkno != args->fi_blkno)
210 * initialize the new inode, but don't do anything that would cause
212 * return 0 on success, 1 on failure
214 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
216 struct ocfs2_find_inode_args *args = opaque;
217 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
218 ocfs2_file_ip_alloc_sem_key;
220 inode->i_ino = args->fi_ino;
221 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
222 if (args->fi_sysfile_type != 0)
223 lockdep_set_class(&inode->i_mutex,
224 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
225 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
226 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
227 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
228 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
229 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
230 &ocfs2_quota_ip_alloc_sem_key);
232 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233 &ocfs2_file_ip_alloc_sem_key);
238 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
241 struct super_block *sb;
242 struct ocfs2_super *osb;
248 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
249 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
253 * These have all been checked by ocfs2_read_inode_block() or set
254 * by ocfs2_mknod_locked(), so a failure is a code bug.
256 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
257 cannot create a superblock
258 inode today. change if
260 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
261 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
264 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
265 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
266 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
268 inode->i_version = 1;
269 inode->i_generation = le32_to_cpu(fe->i_generation);
270 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
271 inode->i_mode = le16_to_cpu(fe->i_mode);
272 i_uid_write(inode, le32_to_cpu(fe->i_uid));
273 i_gid_write(inode, le32_to_cpu(fe->i_gid));
275 /* Fast symlinks will have i_size but no allocated clusters. */
276 if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
278 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
280 inode->i_blocks = ocfs2_inode_sector_count(inode);
281 inode->i_mapping->a_ops = &ocfs2_aops;
283 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
284 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
285 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
286 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
287 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
288 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
290 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
292 "ip_blkno %llu != i_blkno %llu!\n",
293 (unsigned long long)OCFS2_I(inode)->ip_blkno,
294 (unsigned long long)le64_to_cpu(fe->i_blkno));
296 set_nlink(inode, ocfs2_read_links_count(fe));
298 trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
299 le32_to_cpu(fe->i_flags));
300 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
301 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
302 inode->i_flags |= S_NOQUOTA;
305 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
306 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
307 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
308 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
309 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
310 inode->i_flags |= S_NOQUOTA;
311 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
312 /* we can't actually hit this as read_inode can't
313 * handle superblocks today ;-) */
317 switch (inode->i_mode & S_IFMT) {
320 inode->i_fop = &ocfs2_fops;
322 inode->i_fop = &ocfs2_fops_no_plocks;
323 inode->i_op = &ocfs2_file_iops;
324 i_size_write(inode, le64_to_cpu(fe->i_size));
327 inode->i_op = &ocfs2_dir_iops;
329 inode->i_fop = &ocfs2_dops;
331 inode->i_fop = &ocfs2_dops_no_plocks;
332 i_size_write(inode, le64_to_cpu(fe->i_size));
333 OCFS2_I(inode)->ip_dir_lock_gen = 1;
336 inode->i_op = &ocfs2_symlink_inode_operations;
337 i_size_write(inode, le64_to_cpu(fe->i_size));
340 inode->i_op = &ocfs2_special_file_iops;
341 init_special_inode(inode, inode->i_mode,
347 inode->i_ino = ino_from_blkno(inode->i_sb,
348 le64_to_cpu(fe->i_blkno));
351 * If we ever want to create system files from kernel,
352 * the generation argument to
353 * ocfs2_inode_lock_res_init() will have to change.
355 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
357 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
358 OCFS2_LOCK_TYPE_META, 0, inode);
360 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
361 OCFS2_LOCK_TYPE_OPEN, 0, inode);
364 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
365 OCFS2_LOCK_TYPE_RW, inode->i_generation,
368 ocfs2_set_inode_flags(inode);
370 OCFS2_I(inode)->ip_last_used_slot = 0;
371 OCFS2_I(inode)->ip_last_used_group = 0;
373 if (S_ISDIR(inode->i_mode))
374 ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
375 OCFS2_RESV_FLAG_DIR);
378 static int ocfs2_read_locked_inode(struct inode *inode,
379 struct ocfs2_find_inode_args *args)
381 struct super_block *sb;
382 struct ocfs2_super *osb;
383 struct ocfs2_dinode *fe;
384 struct buffer_head *bh = NULL;
385 int status, can_lock;
389 if (inode == NULL || inode->i_sb == NULL) {
390 mlog(ML_ERROR, "bad inode\n");
397 mlog(ML_ERROR, "bad inode args\n");
398 make_bad_inode(inode);
403 * To improve performance of cold-cache inode stats, we take
404 * the cluster lock here if possible.
406 * Generally, OCFS2 never trusts the contents of an inode
407 * unless it's holding a cluster lock, so taking it here isn't
408 * a correctness issue as much as it is a performance
411 * There are three times when taking the lock is not a good idea:
413 * 1) During startup, before we have initialized the DLM.
415 * 2) If we are reading certain system files which never get
416 * cluster locks (local alloc, truncate log).
418 * 3) If the process doing the iget() is responsible for
419 * orphan dir recovery. We're holding the orphan dir lock and
420 * can get into a deadlock with another process on another
421 * node in ->delete_inode().
423 * #1 and #2 can be simply solved by never taking the lock
424 * here for system files (which are the only type we read
425 * during mount). It's a heavier approach, but our main
426 * concern is user-accessible files anyway.
428 * #3 works itself out because we'll eventually take the
429 * cluster lock before trusting anything anyway.
431 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
432 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
433 && !ocfs2_mount_local(osb);
435 trace_ocfs2_read_locked_inode(
436 (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
439 * To maintain backwards compatibility with older versions of
440 * ocfs2-tools, we still store the generation value for system
441 * files. The only ones that actually matter to userspace are
442 * the journals, but it's easier and inexpensive to just flag
443 * all system files similarly.
445 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
446 generation = osb->fs_generation;
448 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
449 OCFS2_LOCK_TYPE_META,
452 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
453 OCFS2_LOCK_TYPE_OPEN,
457 status = ocfs2_open_lock(inode);
459 make_bad_inode(inode);
463 status = ocfs2_inode_lock(inode, NULL, 0);
465 make_bad_inode(inode);
471 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
472 status = ocfs2_try_open_lock(inode, 0);
474 make_bad_inode(inode);
480 status = ocfs2_read_inode_block_full(inode, &bh,
481 OCFS2_BH_IGNORE_CACHE);
483 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
485 * If buffer is in jbd, then its checksum may not have been
488 if (!status && !buffer_jbd(bh))
489 status = ocfs2_validate_inode_block(osb->sb, bh);
497 fe = (struct ocfs2_dinode *) bh->b_data;
500 * This is a code bug. Right now the caller needs to
501 * understand whether it is asking for a system file inode or
502 * not so the proper lock names can be built.
504 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
505 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
506 "Inode %llu: system file state is ambigous\n",
507 (unsigned long long)args->fi_blkno);
509 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
510 S_ISBLK(le16_to_cpu(fe->i_mode)))
511 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
513 ocfs2_populate_inode(inode, fe, 0);
515 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
521 ocfs2_inode_unlock(inode, 0);
524 make_bad_inode(inode);
532 void ocfs2_sync_blockdev(struct super_block *sb)
534 sync_blockdev(sb->s_bdev);
537 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
539 struct buffer_head *fe_bh)
542 struct ocfs2_dinode *fe;
543 handle_t *handle = NULL;
545 fe = (struct ocfs2_dinode *) fe_bh->b_data;
548 * This check will also skip truncate of inodes with inline
549 * data and fast symlinks.
551 if (fe->i_clusters) {
552 if (ocfs2_should_order_data(inode))
553 ocfs2_begin_ordered_truncate(inode, 0);
555 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
556 if (IS_ERR(handle)) {
557 status = PTR_ERR(handle);
563 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
565 OCFS2_JOURNAL_ACCESS_WRITE);
571 i_size_write(inode, 0);
573 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
579 ocfs2_commit_trans(osb, handle);
582 status = ocfs2_commit_truncate(osb, inode, fe_bh);
591 ocfs2_commit_trans(osb, handle);
595 static int ocfs2_remove_inode(struct inode *inode,
596 struct buffer_head *di_bh,
597 struct inode *orphan_dir_inode,
598 struct buffer_head *orphan_dir_bh)
601 struct inode *inode_alloc_inode = NULL;
602 struct buffer_head *inode_alloc_bh = NULL;
604 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
605 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
608 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
609 le16_to_cpu(di->i_suballoc_slot));
610 if (!inode_alloc_inode) {
616 mutex_lock(&inode_alloc_inode->i_mutex);
617 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
619 mutex_unlock(&inode_alloc_inode->i_mutex);
625 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
626 ocfs2_quota_trans_credits(inode->i_sb));
627 if (IS_ERR(handle)) {
628 status = PTR_ERR(handle);
633 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
634 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
642 /* set the inodes dtime */
643 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
644 OCFS2_JOURNAL_ACCESS_WRITE);
650 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
651 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
652 ocfs2_journal_dirty(handle, di_bh);
654 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
655 dquot_free_inode(inode);
657 status = ocfs2_free_dinode(handle, inode_alloc_inode,
663 ocfs2_commit_trans(osb, handle);
665 ocfs2_inode_unlock(inode_alloc_inode, 1);
666 mutex_unlock(&inode_alloc_inode->i_mutex);
667 brelse(inode_alloc_bh);
669 iput(inode_alloc_inode);
675 * Serialize with orphan dir recovery. If the process doing
676 * recovery on this orphan dir does an iget() with the dir
677 * i_mutex held, we'll deadlock here. Instead we detect this
678 * and exit early - recovery will wipe this inode for us.
680 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
685 spin_lock(&osb->osb_lock);
686 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
690 /* This signals to the orphan recovery process that it should
691 * wait for us to handle the wipe. */
692 osb->osb_orphan_wipes[slot]++;
694 spin_unlock(&osb->osb_lock);
695 trace_ocfs2_check_orphan_recovery_state(slot, ret);
699 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
702 spin_lock(&osb->osb_lock);
703 osb->osb_orphan_wipes[slot]--;
704 spin_unlock(&osb->osb_lock);
706 wake_up(&osb->osb_wipe_event);
709 static int ocfs2_wipe_inode(struct inode *inode,
710 struct buffer_head *di_bh)
712 int status, orphaned_slot = -1;
713 struct inode *orphan_dir_inode = NULL;
714 struct buffer_head *orphan_dir_bh = NULL;
715 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
716 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
718 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
719 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
721 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
725 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
726 ORPHAN_DIR_SYSTEM_INODE,
728 if (!orphan_dir_inode) {
734 /* Lock the orphan dir. The lock will be held for the entire
735 * delete_inode operation. We do this now to avoid races with
736 * recovery completion on other nodes. */
737 mutex_lock(&orphan_dir_inode->i_mutex);
738 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
740 mutex_unlock(&orphan_dir_inode->i_mutex);
747 /* we do this while holding the orphan dir lock because we
748 * don't want recovery being run from another node to try an
749 * inode delete underneath us -- this will result in two nodes
750 * truncating the same file! */
751 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
754 goto bail_unlock_dir;
757 /* Remove any dir index tree */
758 if (S_ISDIR(inode->i_mode)) {
759 status = ocfs2_dx_dir_truncate(inode, di_bh);
762 goto bail_unlock_dir;
766 /*Free extended attribute resources associated with this inode.*/
767 status = ocfs2_xattr_remove(inode, di_bh);
770 goto bail_unlock_dir;
773 status = ocfs2_remove_refcount_tree(inode, di_bh);
776 goto bail_unlock_dir;
779 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
785 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
788 ocfs2_inode_unlock(orphan_dir_inode, 1);
789 mutex_unlock(&orphan_dir_inode->i_mutex);
790 brelse(orphan_dir_bh);
792 iput(orphan_dir_inode);
793 ocfs2_signal_wipe_completion(osb, orphaned_slot);
798 /* There is a series of simple checks that should be done before a
799 * trylock is even considered. Encapsulate those in this function. */
800 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
803 struct ocfs2_inode_info *oi = OCFS2_I(inode);
804 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
806 trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
807 (unsigned long long)oi->ip_blkno,
810 /* We shouldn't be getting here for the root directory
812 if (inode == osb->root_inode) {
813 mlog(ML_ERROR, "Skipping delete of root inode.\n");
817 /* If we're coming from downconvert_thread we can't go into our own
818 * voting [hello, deadlock city!], so unforuntately we just
819 * have to skip deleting this guy. That's OK though because
820 * the node who's doing the actual deleting should handle it
822 if (current == osb->dc_task)
825 spin_lock(&oi->ip_lock);
826 /* OCFS2 *never* deletes system files. This should technically
827 * never get here as system file inodes should always have a
828 * positive link count. */
829 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
830 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
831 (unsigned long long)oi->ip_blkno);
835 /* If we have allowd wipe of this inode for another node, it
836 * will be marked here so we can safely skip it. Recovery will
837 * cleanup any inodes we might inadvertently skip here. */
838 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE)
843 spin_unlock(&oi->ip_lock);
848 /* Query the cluster to determine whether we should wipe an inode from
851 * Requires the inode to have the cluster lock. */
852 static int ocfs2_query_inode_wipe(struct inode *inode,
853 struct buffer_head *di_bh,
856 int status = 0, reason = 0;
857 struct ocfs2_inode_info *oi = OCFS2_I(inode);
858 struct ocfs2_dinode *di;
862 trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
865 /* While we were waiting for the cluster lock in
866 * ocfs2_delete_inode, another node might have asked to delete
867 * the inode. Recheck our flags to catch this. */
868 if (!ocfs2_inode_is_valid_to_delete(inode)) {
873 /* Now that we have an up to date inode, we can double check
878 /* Do some basic inode verification... */
879 di = (struct ocfs2_dinode *) di_bh->b_data;
880 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
881 !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
883 * Inodes in the orphan dir must have ORPHANED_FL. The only
884 * inodes that come back out of the orphan dir are reflink
885 * targets. A reflink target may be moved out of the orphan
886 * dir between the time we scan the directory and the time we
887 * process it. This would lead to HAS_REFCOUNT_FL being set but
890 if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
895 /* for lack of a better error? */
898 "Inode %llu (on-disk %llu) not orphaned! "
899 "Disk flags 0x%x, inode flags 0x%x\n",
900 (unsigned long long)oi->ip_blkno,
901 (unsigned long long)le64_to_cpu(di->i_blkno),
902 le32_to_cpu(di->i_flags), oi->ip_flags);
906 /* has someone already deleted us?! baaad... */
914 * This is how ocfs2 determines whether an inode is still live
915 * within the cluster. Every node takes a shared read lock on
916 * the inode open lock in ocfs2_read_locked_inode(). When we
917 * get to ->delete_inode(), each node tries to convert it's
918 * lock to an exclusive. Trylocks are serialized by the inode
919 * meta data lock. If the upconvert succeeds, we know the inode
920 * is no longer live and can be deleted.
922 * Though we call this with the meta data lock held, the
923 * trylock keeps us from ABBA deadlock.
925 status = ocfs2_try_open_lock(inode, 1);
926 if (status == -EAGAIN) {
937 trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
940 trace_ocfs2_query_inode_wipe_end(status, reason);
944 /* Support function for ocfs2_delete_inode. Will help us keep the
945 * inode data in a consistent state for clear_inode. Always truncates
946 * pages, optionally sync's them first. */
947 static void ocfs2_cleanup_delete_inode(struct inode *inode,
950 trace_ocfs2_cleanup_delete_inode(
951 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
953 filemap_write_and_wait(inode->i_mapping);
954 truncate_inode_pages(&inode->i_data, 0);
957 static void ocfs2_delete_inode(struct inode *inode)
961 struct buffer_head *di_bh = NULL;
963 trace_ocfs2_delete_inode(inode->i_ino,
964 (unsigned long long)OCFS2_I(inode)->ip_blkno,
965 is_bad_inode(inode));
967 /* When we fail in read_inode() we mark inode as bad. The second test
968 * catches the case when inode allocation fails before allocating
969 * a block for inode. */
970 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
973 dquot_initialize(inode);
975 if (!ocfs2_inode_is_valid_to_delete(inode)) {
976 /* It's probably not necessary to truncate_inode_pages
977 * here but we do it for safety anyway (it will most
978 * likely be a no-op anyway) */
979 ocfs2_cleanup_delete_inode(inode, 0);
983 /* We want to block signals in delete_inode as the lock and
984 * messaging paths may return us -ERESTARTSYS. Which would
985 * cause us to exit early, resulting in inodes being orphaned
987 ocfs2_block_signals(&oldset);
990 * Synchronize us against ocfs2_get_dentry. We take this in
991 * shared mode so that all nodes can still concurrently
994 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
996 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
997 ocfs2_cleanup_delete_inode(inode, 0);
1000 /* Lock down the inode. This gives us an up to date view of
1001 * it's metadata (for verification), and allows us to
1002 * serialize delete_inode on multiple nodes.
1004 * Even though we might be doing a truncate, we don't take the
1005 * allocation lock here as it won't be needed - nobody will
1006 * have the file open.
1008 status = ocfs2_inode_lock(inode, &di_bh, 1);
1010 if (status != -ENOENT)
1012 ocfs2_cleanup_delete_inode(inode, 0);
1013 goto bail_unlock_nfs_sync;
1016 /* Query the cluster. This will be the final decision made
1017 * before we go ahead and wipe the inode. */
1018 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1019 if (!wipe || status < 0) {
1020 /* Error and remote inode busy both mean we won't be
1021 * removing the inode, so they take almost the same
1026 /* Someone in the cluster has disallowed a wipe of
1027 * this inode, or it was never completely
1028 * orphaned. Write out the pages and exit now. */
1029 ocfs2_cleanup_delete_inode(inode, 1);
1030 goto bail_unlock_inode;
1033 ocfs2_cleanup_delete_inode(inode, 0);
1035 status = ocfs2_wipe_inode(inode, di_bh);
1037 if (status != -EDEADLK)
1039 goto bail_unlock_inode;
1043 * Mark the inode as successfully deleted.
1045 * This is important for ocfs2_clear_inode() as it will check
1046 * this flag and skip any checkpointing work
1048 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1049 * the LVB for other nodes.
1051 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1054 ocfs2_inode_unlock(inode, 1);
1057 bail_unlock_nfs_sync:
1058 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1061 ocfs2_unblock_signals(&oldset);
1066 static void ocfs2_clear_inode(struct inode *inode)
1069 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1072 trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1075 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1076 "Inode=%lu\n", inode->i_ino);
1080 /* To preven remote deletes we hold open lock before, now it
1081 * is time to unlock PR and EX open locks. */
1082 ocfs2_open_unlock(inode);
1084 /* Do these before all the other work so that we don't bounce
1085 * the downconvert thread while waiting to destroy the locks. */
1086 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1087 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1088 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1090 ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1091 &oi->ip_la_data_resv);
1092 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1094 /* We very well may get a clear_inode before all an inodes
1095 * metadata has hit disk. Of course, we can't drop any cluster
1096 * locks until the journal has finished with it. The only
1097 * exception here are successfully wiped inodes - their
1098 * metadata can now be considered to be part of the system
1099 * inodes from which it came. */
1100 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1101 ocfs2_checkpoint_inode(inode);
1103 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1104 "Clear inode of %llu, inode has io markers\n",
1105 (unsigned long long)oi->ip_blkno);
1107 ocfs2_extent_map_trunc(inode, 0);
1109 status = ocfs2_drop_inode_locks(inode);
1113 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1114 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1115 ocfs2_lock_res_free(&oi->ip_open_lockres);
1117 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1119 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1120 "Clear inode of %llu, inode has %u cache items\n",
1121 (unsigned long long)oi->ip_blkno,
1122 INODE_CACHE(inode)->ci_num_cached);
1124 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1125 "Clear inode of %llu, inode has a bad flag\n",
1126 (unsigned long long)oi->ip_blkno);
1128 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1129 "Clear inode of %llu, inode is locked\n",
1130 (unsigned long long)oi->ip_blkno);
1132 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1133 "Clear inode of %llu, io_mutex is locked\n",
1134 (unsigned long long)oi->ip_blkno);
1135 mutex_unlock(&oi->ip_io_mutex);
1138 * down_trylock() returns 0, down_write_trylock() returns 1
1141 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1142 "Clear inode of %llu, alloc_sem is locked\n",
1143 (unsigned long long)oi->ip_blkno);
1144 up_write(&oi->ip_alloc_sem);
1146 mlog_bug_on_msg(oi->ip_open_count,
1147 "Clear inode of %llu has open count %d\n",
1148 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1150 /* Clear all other flags. */
1152 oi->ip_dir_start_lookup = 0;
1153 oi->ip_blkno = 0ULL;
1156 * ip_jinode is used to track txns against this inode. We ensure that
1157 * the journal is flushed before journal shutdown. Thus it is safe to
1158 * have inodes get cleaned up after journal shutdown.
1160 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1164 void ocfs2_evict_inode(struct inode *inode)
1166 if (!inode->i_nlink ||
1167 (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1168 ocfs2_delete_inode(inode);
1170 truncate_inode_pages(&inode->i_data, 0);
1172 ocfs2_clear_inode(inode);
1175 /* Called under inode_lock, with no more references on the
1176 * struct inode, so it's safe here to check the flags field
1177 * and to manipulate i_nlink without any other locks. */
1178 int ocfs2_drop_inode(struct inode *inode)
1180 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1183 trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1184 inode->i_nlink, oi->ip_flags);
1186 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1189 res = generic_drop_inode(inode);
1195 * This is called from our getattr.
1197 int ocfs2_inode_revalidate(struct dentry *dentry)
1199 struct inode *inode = dentry->d_inode;
1202 trace_ocfs2_inode_revalidate(inode,
1203 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1204 inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1211 spin_lock(&OCFS2_I(inode)->ip_lock);
1212 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1213 spin_unlock(&OCFS2_I(inode)->ip_lock);
1217 spin_unlock(&OCFS2_I(inode)->ip_lock);
1219 /* Let ocfs2_inode_lock do the work of updating our struct
1221 status = ocfs2_inode_lock(inode, NULL, 0);
1223 if (status != -ENOENT)
1227 ocfs2_inode_unlock(inode, 0);
1233 * Updates a disk inode from a
1235 * Only takes ip_lock.
1237 int ocfs2_mark_inode_dirty(handle_t *handle,
1238 struct inode *inode,
1239 struct buffer_head *bh)
1242 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1244 trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1246 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1247 OCFS2_JOURNAL_ACCESS_WRITE);
1253 spin_lock(&OCFS2_I(inode)->ip_lock);
1254 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1255 ocfs2_get_inode_flags(OCFS2_I(inode));
1256 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1257 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1258 spin_unlock(&OCFS2_I(inode)->ip_lock);
1260 fe->i_size = cpu_to_le64(i_size_read(inode));
1261 ocfs2_set_links_count(fe, inode->i_nlink);
1262 fe->i_uid = cpu_to_le32(i_uid_read(inode));
1263 fe->i_gid = cpu_to_le32(i_gid_read(inode));
1264 fe->i_mode = cpu_to_le16(inode->i_mode);
1265 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1266 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1267 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1268 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1269 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1270 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1272 ocfs2_journal_dirty(handle, bh);
1279 * Updates a struct inode from a disk inode.
1280 * does no i/o, only takes ip_lock.
1282 void ocfs2_refresh_inode(struct inode *inode,
1283 struct ocfs2_dinode *fe)
1285 spin_lock(&OCFS2_I(inode)->ip_lock);
1287 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1288 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1289 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1290 ocfs2_set_inode_flags(inode);
1291 i_size_write(inode, le64_to_cpu(fe->i_size));
1292 set_nlink(inode, ocfs2_read_links_count(fe));
1293 i_uid_write(inode, le32_to_cpu(fe->i_uid));
1294 i_gid_write(inode, le32_to_cpu(fe->i_gid));
1295 inode->i_mode = le16_to_cpu(fe->i_mode);
1296 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1297 inode->i_blocks = 0;
1299 inode->i_blocks = ocfs2_inode_sector_count(inode);
1300 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1301 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1302 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1303 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1304 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1305 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1307 spin_unlock(&OCFS2_I(inode)->ip_lock);
1310 int ocfs2_validate_inode_block(struct super_block *sb,
1311 struct buffer_head *bh)
1314 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1316 trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1318 BUG_ON(!buffer_uptodate(bh));
1321 * If the ecc fails, we return the error but otherwise
1322 * leave the filesystem running. We know any error is
1323 * local to this block.
1325 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1327 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1328 (unsigned long long)bh->b_blocknr);
1333 * Errors after here are fatal.
1338 if (!OCFS2_IS_VALID_DINODE(di)) {
1339 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1340 (unsigned long long)bh->b_blocknr, 7,
1345 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1346 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1347 (unsigned long long)bh->b_blocknr,
1348 (unsigned long long)le64_to_cpu(di->i_blkno));
1352 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1354 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1355 (unsigned long long)bh->b_blocknr);
1359 if (le32_to_cpu(di->i_fs_generation) !=
1360 OCFS2_SB(sb)->fs_generation) {
1362 "Invalid dinode #%llu: fs_generation is %u\n",
1363 (unsigned long long)bh->b_blocknr,
1364 le32_to_cpu(di->i_fs_generation));
1374 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1378 struct buffer_head *tmp = *bh;
1380 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1381 1, &tmp, flags, ocfs2_validate_inode_block);
1383 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1390 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1392 return ocfs2_read_inode_block_full(inode, bh, 0);
1396 static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1398 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1400 return oi->ip_blkno;
1403 static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1405 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1407 return oi->vfs_inode.i_sb;
1410 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1412 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1414 spin_lock(&oi->ip_lock);
1417 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1419 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1421 spin_unlock(&oi->ip_lock);
1424 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1426 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1428 mutex_lock(&oi->ip_io_mutex);
1431 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1433 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1435 mutex_unlock(&oi->ip_io_mutex);
1438 const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1439 .co_owner = ocfs2_inode_cache_owner,
1440 .co_get_super = ocfs2_inode_cache_get_super,
1441 .co_cache_lock = ocfs2_inode_cache_lock,
1442 .co_cache_unlock = ocfs2_inode_cache_unlock,
1443 .co_io_lock = ocfs2_inode_cache_io_lock,
1444 .co_io_unlock = ocfs2_inode_cache_io_unlock,