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>
42 #include "blockcheck.h"
44 #include "extent_map.h"
46 #include "heartbeat.h"
56 #include "refcounttree.h"
58 #include "buffer_head_io.h"
60 struct ocfs2_find_inode_args
64 unsigned int fi_flags;
65 unsigned int fi_sysfile_type;
68 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
70 static int ocfs2_read_locked_inode(struct inode *inode,
71 struct ocfs2_find_inode_args *args);
72 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
73 static int ocfs2_find_actor(struct inode *inode, void *opaque);
74 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
76 struct buffer_head *fe_bh);
78 void ocfs2_set_inode_flags(struct inode *inode)
80 unsigned int flags = OCFS2_I(inode)->ip_attr;
82 inode->i_flags &= ~(S_IMMUTABLE |
83 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
85 if (flags & OCFS2_IMMUTABLE_FL)
86 inode->i_flags |= S_IMMUTABLE;
88 if (flags & OCFS2_SYNC_FL)
89 inode->i_flags |= S_SYNC;
90 if (flags & OCFS2_APPEND_FL)
91 inode->i_flags |= S_APPEND;
92 if (flags & OCFS2_NOATIME_FL)
93 inode->i_flags |= S_NOATIME;
94 if (flags & OCFS2_DIRSYNC_FL)
95 inode->i_flags |= S_DIRSYNC;
98 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
99 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
101 unsigned int flags = oi->vfs_inode.i_flags;
103 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
104 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
106 oi->ip_attr |= OCFS2_SYNC_FL;
107 if (flags & S_APPEND)
108 oi->ip_attr |= OCFS2_APPEND_FL;
109 if (flags & S_IMMUTABLE)
110 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
111 if (flags & S_NOATIME)
112 oi->ip_attr |= OCFS2_NOATIME_FL;
113 if (flags & S_DIRSYNC)
114 oi->ip_attr |= OCFS2_DIRSYNC_FL;
117 struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
119 struct ocfs2_find_inode_args args;
121 args.fi_blkno = blkno;
123 args.fi_ino = ino_from_blkno(sb, blkno);
124 args.fi_sysfile_type = 0;
126 return ilookup5(sb, blkno, ocfs2_find_actor, &args);
128 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
131 struct inode *inode = NULL;
132 struct super_block *sb = osb->sb;
133 struct ocfs2_find_inode_args args;
135 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
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
156 if (inode && inode->i_state & I_NEW) {
157 mlog(0, "Inode was not in inode cache, reading it.\n");
158 ocfs2_read_locked_inode(inode, &args);
159 unlock_new_inode(inode);
162 inode = ERR_PTR(-ENOMEM);
163 mlog_errno(PTR_ERR(inode));
166 if (is_bad_inode(inode)) {
168 inode = ERR_PTR(-ESTALE);
173 if (!IS_ERR(inode)) {
174 mlog(0, "returning inode with number %llu\n",
175 (unsigned long long)OCFS2_I(inode)->ip_blkno);
176 mlog_exit_ptr(inode);
184 * here's how inodes get read from disk:
185 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
186 * found? : return the in-memory inode
187 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
190 static int ocfs2_find_actor(struct inode *inode, void *opaque)
192 struct ocfs2_find_inode_args *args = NULL;
193 struct ocfs2_inode_info *oi = OCFS2_I(inode);
196 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
200 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
202 if (oi->ip_blkno != args->fi_blkno)
212 * initialize the new inode, but don't do anything that would cause
214 * return 0 on success, 1 on failure
216 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
218 struct ocfs2_find_inode_args *args = opaque;
219 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
220 ocfs2_file_ip_alloc_sem_key;
222 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
224 inode->i_ino = args->fi_ino;
225 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
226 if (args->fi_sysfile_type != 0)
227 lockdep_set_class(&inode->i_mutex,
228 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
229 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
230 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
231 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
232 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
233 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
234 &ocfs2_quota_ip_alloc_sem_key);
236 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
237 &ocfs2_file_ip_alloc_sem_key);
243 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
246 struct super_block *sb;
247 struct ocfs2_super *osb;
250 mlog_entry("(0x%p, size:%llu)\n", inode,
251 (unsigned long long)le64_to_cpu(fe->i_size));
256 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
257 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
261 * These have all been checked by ocfs2_read_inode_block() or set
262 * by ocfs2_mknod_locked(), so a failure is a code bug.
264 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
265 cannot create a superblock
266 inode today. change if
268 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
269 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
272 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
273 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
274 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
276 inode->i_version = 1;
277 inode->i_generation = le32_to_cpu(fe->i_generation);
278 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
279 inode->i_mode = le16_to_cpu(fe->i_mode);
280 inode->i_uid = le32_to_cpu(fe->i_uid);
281 inode->i_gid = le32_to_cpu(fe->i_gid);
283 /* Fast symlinks will have i_size but no allocated clusters. */
284 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
287 inode->i_blocks = ocfs2_inode_sector_count(inode);
288 inode->i_mapping->a_ops = &ocfs2_aops;
289 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
290 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
291 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
292 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
293 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
294 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
296 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
298 "ip_blkno %llu != i_blkno %llu!\n",
299 (unsigned long long)OCFS2_I(inode)->ip_blkno,
300 (unsigned long long)le64_to_cpu(fe->i_blkno));
302 inode->i_nlink = ocfs2_read_links_count(fe);
304 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
305 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
306 inode->i_flags |= S_NOQUOTA;
309 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
310 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
311 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
312 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
313 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
314 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
315 inode->i_flags |= S_NOQUOTA;
316 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
317 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
318 /* we can't actually hit this as read_inode can't
319 * handle superblocks today ;-) */
323 switch (inode->i_mode & S_IFMT) {
326 inode->i_fop = &ocfs2_fops;
328 inode->i_fop = &ocfs2_fops_no_plocks;
329 inode->i_op = &ocfs2_file_iops;
330 i_size_write(inode, le64_to_cpu(fe->i_size));
333 inode->i_op = &ocfs2_dir_iops;
335 inode->i_fop = &ocfs2_dops;
337 inode->i_fop = &ocfs2_dops_no_plocks;
338 i_size_write(inode, le64_to_cpu(fe->i_size));
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_entry("(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-accesible 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 status = ocfs2_validate_inode_block(osb->sb, bh);
501 fe = (struct ocfs2_dinode *) bh->b_data;
504 * This is a code bug. Right now the caller needs to
505 * understand whether it is asking for a system file inode or
506 * not so the proper lock names can be built.
508 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
509 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
510 "Inode %llu: system file state is ambigous\n",
511 (unsigned long long)args->fi_blkno);
513 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
514 S_ISBLK(le16_to_cpu(fe->i_mode)))
515 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
517 ocfs2_populate_inode(inode, fe, 0);
519 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
525 ocfs2_inode_unlock(inode, 0);
528 make_bad_inode(inode);
537 void ocfs2_sync_blockdev(struct super_block *sb)
539 sync_blockdev(sb->s_bdev);
542 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
544 struct buffer_head *fe_bh)
547 struct ocfs2_dinode *fe;
548 handle_t *handle = NULL;
552 fe = (struct ocfs2_dinode *) fe_bh->b_data;
555 * This check will also skip truncate of inodes with inline
556 * data and fast symlinks.
558 if (fe->i_clusters) {
559 if (ocfs2_should_order_data(inode))
560 ocfs2_begin_ordered_truncate(inode, 0);
562 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
563 if (IS_ERR(handle)) {
564 status = PTR_ERR(handle);
569 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
571 OCFS2_JOURNAL_ACCESS_WRITE);
577 i_size_write(inode, 0);
579 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
585 ocfs2_commit_trans(osb, handle);
588 status = ocfs2_commit_truncate(osb, inode, fe_bh);
597 ocfs2_commit_trans(osb, handle);
602 static int ocfs2_remove_inode(struct inode *inode,
603 struct buffer_head *di_bh,
604 struct inode *orphan_dir_inode,
605 struct buffer_head *orphan_dir_bh)
608 struct inode *inode_alloc_inode = NULL;
609 struct buffer_head *inode_alloc_bh = NULL;
611 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
612 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
615 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
616 le16_to_cpu(di->i_suballoc_slot));
617 if (!inode_alloc_inode) {
623 mutex_lock(&inode_alloc_inode->i_mutex);
624 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
626 mutex_unlock(&inode_alloc_inode->i_mutex);
632 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
633 ocfs2_quota_trans_credits(inode->i_sb));
634 if (IS_ERR(handle)) {
635 status = PTR_ERR(handle);
640 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
647 /* set the inodes dtime */
648 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
649 OCFS2_JOURNAL_ACCESS_WRITE);
655 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
656 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
657 ocfs2_journal_dirty(handle, di_bh);
659 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
660 dquot_free_inode(inode);
662 status = ocfs2_free_dinode(handle, inode_alloc_inode,
668 ocfs2_commit_trans(osb, handle);
670 ocfs2_inode_unlock(inode_alloc_inode, 1);
671 mutex_unlock(&inode_alloc_inode->i_mutex);
672 brelse(inode_alloc_bh);
674 iput(inode_alloc_inode);
680 * Serialize with orphan dir recovery. If the process doing
681 * recovery on this orphan dir does an iget() with the dir
682 * i_mutex held, we'll deadlock here. Instead we detect this
683 * and exit early - recovery will wipe this inode for us.
685 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
690 spin_lock(&osb->osb_lock);
691 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
692 mlog(0, "Recovery is happening on orphan dir %d, will skip "
693 "this inode\n", slot);
697 /* This signals to the orphan recovery process that it should
698 * wait for us to handle the wipe. */
699 osb->osb_orphan_wipes[slot]++;
701 spin_unlock(&osb->osb_lock);
705 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
708 spin_lock(&osb->osb_lock);
709 osb->osb_orphan_wipes[slot]--;
710 spin_unlock(&osb->osb_lock);
712 wake_up(&osb->osb_wipe_event);
715 static int ocfs2_wipe_inode(struct inode *inode,
716 struct buffer_head *di_bh)
718 int status, orphaned_slot;
719 struct inode *orphan_dir_inode = NULL;
720 struct buffer_head *orphan_dir_bh = NULL;
721 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
722 struct ocfs2_dinode *di;
724 di = (struct ocfs2_dinode *) di_bh->b_data;
725 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
727 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
731 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
732 ORPHAN_DIR_SYSTEM_INODE,
734 if (!orphan_dir_inode) {
740 /* Lock the orphan dir. The lock will be held for the entire
741 * delete_inode operation. We do this now to avoid races with
742 * recovery completion on other nodes. */
743 mutex_lock(&orphan_dir_inode->i_mutex);
744 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
746 mutex_unlock(&orphan_dir_inode->i_mutex);
752 /* we do this while holding the orphan dir lock because we
753 * don't want recovery being run from another node to try an
754 * inode delete underneath us -- this will result in two nodes
755 * truncating the same file! */
756 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
759 goto bail_unlock_dir;
762 /* Remove any dir index tree */
763 if (S_ISDIR(inode->i_mode)) {
764 status = ocfs2_dx_dir_truncate(inode, di_bh);
767 goto bail_unlock_dir;
771 /*Free extended attribute resources associated with this inode.*/
772 status = ocfs2_xattr_remove(inode, di_bh);
775 goto bail_unlock_dir;
778 status = ocfs2_remove_refcount_tree(inode, di_bh);
781 goto bail_unlock_dir;
784 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
790 ocfs2_inode_unlock(orphan_dir_inode, 1);
791 mutex_unlock(&orphan_dir_inode->i_mutex);
792 brelse(orphan_dir_bh);
794 iput(orphan_dir_inode);
795 ocfs2_signal_wipe_completion(osb, orphaned_slot);
800 /* There is a series of simple checks that should be done before a
801 * trylock is even considered. Encapsulate those in this function. */
802 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
805 struct ocfs2_inode_info *oi = OCFS2_I(inode);
806 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
808 /* We shouldn't be getting here for the root directory
810 if (inode == osb->root_inode) {
811 mlog(ML_ERROR, "Skipping delete of root inode.\n");
815 /* If we're coming from downconvert_thread we can't go into our own
816 * voting [hello, deadlock city!], so unforuntately we just
817 * have to skip deleting this guy. That's OK though because
818 * the node who's doing the actual deleting should handle it
820 if (current == osb->dc_task) {
821 mlog(0, "Skipping delete of %lu because we're currently "
822 "in downconvert\n", inode->i_ino);
826 spin_lock(&oi->ip_lock);
827 /* OCFS2 *never* deletes system files. This should technically
828 * never get here as system file inodes should always have a
829 * positive link count. */
830 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
831 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
832 (unsigned long long)oi->ip_blkno);
836 /* If we have allowd wipe of this inode for another node, it
837 * will be marked here so we can safely skip it. Recovery will
838 * cleanup any inodes we might inadvertantly skip here. */
839 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
840 mlog(0, "Skipping delete of %lu because another node "
841 "has done this for us.\n", inode->i_ino);
847 spin_unlock(&oi->ip_lock);
852 /* Query the cluster to determine whether we should wipe an inode from
855 * Requires the inode to have the cluster lock. */
856 static int ocfs2_query_inode_wipe(struct inode *inode,
857 struct buffer_head *di_bh,
861 struct ocfs2_inode_info *oi = OCFS2_I(inode);
862 struct ocfs2_dinode *di;
866 /* While we were waiting for the cluster lock in
867 * ocfs2_delete_inode, another node might have asked to delete
868 * the inode. Recheck our flags to catch this. */
869 if (!ocfs2_inode_is_valid_to_delete(inode)) {
870 mlog(0, "Skipping delete of %llu because flags changed\n",
871 (unsigned long long)oi->ip_blkno);
875 /* Now that we have an up to date inode, we can double check
877 if (inode->i_nlink) {
878 mlog(0, "Skipping delete of %llu because nlink = %u\n",
879 (unsigned long long)oi->ip_blkno, inode->i_nlink);
883 /* Do some basic inode verification... */
884 di = (struct ocfs2_dinode *) di_bh->b_data;
885 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
886 /* for lack of a better error? */
889 "Inode %llu (on-disk %llu) not orphaned! "
890 "Disk flags 0x%x, inode flags 0x%x\n",
891 (unsigned long long)oi->ip_blkno,
892 (unsigned long long)le64_to_cpu(di->i_blkno),
893 le32_to_cpu(di->i_flags), oi->ip_flags);
897 /* has someone already deleted us?! baaad... */
905 * This is how ocfs2 determines whether an inode is still live
906 * within the cluster. Every node takes a shared read lock on
907 * the inode open lock in ocfs2_read_locked_inode(). When we
908 * get to ->delete_inode(), each node tries to convert it's
909 * lock to an exclusive. Trylocks are serialized by the inode
910 * meta data lock. If the upconvert suceeds, we know the inode
911 * is no longer live and can be deleted.
913 * Though we call this with the meta data lock held, the
914 * trylock keeps us from ABBA deadlock.
916 status = ocfs2_try_open_lock(inode, 1);
917 if (status == -EAGAIN) {
919 mlog(0, "Skipping delete of %llu because it is in use on "
920 "other nodes\n", (unsigned long long)oi->ip_blkno);
929 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
930 (unsigned long long)oi->ip_blkno,
931 le16_to_cpu(di->i_orphaned_slot));
937 /* Support function for ocfs2_delete_inode. Will help us keep the
938 * inode data in a consistent state for clear_inode. Always truncates
939 * pages, optionally sync's them first. */
940 static void ocfs2_cleanup_delete_inode(struct inode *inode,
943 mlog(0, "Cleanup inode %llu, sync = %d\n",
944 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
946 write_inode_now(inode, 1);
947 truncate_inode_pages(&inode->i_data, 0);
950 void ocfs2_delete_inode(struct inode *inode)
954 struct buffer_head *di_bh = NULL;
956 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
958 /* When we fail in read_inode() we mark inode as bad. The second test
959 * catches the case when inode allocation fails before allocating
960 * a block for inode. */
961 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
962 mlog(0, "Skipping delete of bad inode\n");
966 dquot_initialize(inode);
968 if (!ocfs2_inode_is_valid_to_delete(inode)) {
969 /* It's probably not necessary to truncate_inode_pages
970 * here but we do it for safety anyway (it will most
971 * likely be a no-op anyway) */
972 ocfs2_cleanup_delete_inode(inode, 0);
976 /* We want to block signals in delete_inode as the lock and
977 * messaging paths may return us -ERESTARTSYS. Which would
978 * cause us to exit early, resulting in inodes being orphaned
980 ocfs2_block_signals(&oldset);
983 * Synchronize us against ocfs2_get_dentry. We take this in
984 * shared mode so that all nodes can still concurrently
987 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
989 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
990 ocfs2_cleanup_delete_inode(inode, 0);
993 /* Lock down the inode. This gives us an up to date view of
994 * it's metadata (for verification), and allows us to
995 * serialize delete_inode on multiple nodes.
997 * Even though we might be doing a truncate, we don't take the
998 * allocation lock here as it won't be needed - nobody will
999 * have the file open.
1001 status = ocfs2_inode_lock(inode, &di_bh, 1);
1003 if (status != -ENOENT)
1005 ocfs2_cleanup_delete_inode(inode, 0);
1006 goto bail_unlock_nfs_sync;
1009 /* Query the cluster. This will be the final decision made
1010 * before we go ahead and wipe the inode. */
1011 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1012 if (!wipe || status < 0) {
1013 /* Error and remote inode busy both mean we won't be
1014 * removing the inode, so they take almost the same
1019 /* Someone in the cluster has disallowed a wipe of
1020 * this inode, or it was never completely
1021 * orphaned. Write out the pages and exit now. */
1022 ocfs2_cleanup_delete_inode(inode, 1);
1023 goto bail_unlock_inode;
1026 ocfs2_cleanup_delete_inode(inode, 0);
1028 status = ocfs2_wipe_inode(inode, di_bh);
1030 if (status != -EDEADLK)
1032 goto bail_unlock_inode;
1036 * Mark the inode as successfully deleted.
1038 * This is important for ocfs2_clear_inode() as it will check
1039 * this flag and skip any checkpointing work
1041 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1042 * the LVB for other nodes.
1044 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1047 ocfs2_inode_unlock(inode, 1);
1050 bail_unlock_nfs_sync:
1051 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1054 ocfs2_unblock_signals(&oldset);
1060 void ocfs2_clear_inode(struct inode *inode)
1063 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1070 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1071 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1073 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1074 "Inode=%lu\n", inode->i_ino);
1078 /* To preven remote deletes we hold open lock before, now it
1079 * is time to unlock PR and EX open locks. */
1080 ocfs2_open_unlock(inode);
1082 /* Do these before all the other work so that we don't bounce
1083 * the downconvert thread while waiting to destroy the locks. */
1084 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1085 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1086 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1088 ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1089 &oi->ip_la_data_resv);
1090 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1092 /* We very well may get a clear_inode before all an inodes
1093 * metadata has hit disk. Of course, we can't drop any cluster
1094 * locks until the journal has finished with it. The only
1095 * exception here are successfully wiped inodes - their
1096 * metadata can now be considered to be part of the system
1097 * inodes from which it came. */
1098 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1099 ocfs2_checkpoint_inode(inode);
1101 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1102 "Clear inode of %llu, inode has io markers\n",
1103 (unsigned long long)oi->ip_blkno);
1105 ocfs2_extent_map_trunc(inode, 0);
1107 status = ocfs2_drop_inode_locks(inode);
1111 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1112 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1113 ocfs2_lock_res_free(&oi->ip_open_lockres);
1115 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1117 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1118 "Clear inode of %llu, inode has %u cache items\n",
1119 (unsigned long long)oi->ip_blkno,
1120 INODE_CACHE(inode)->ci_num_cached);
1122 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1123 "Clear inode of %llu, inode has a bad flag\n",
1124 (unsigned long long)oi->ip_blkno);
1126 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1127 "Clear inode of %llu, inode is locked\n",
1128 (unsigned long long)oi->ip_blkno);
1130 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1131 "Clear inode of %llu, io_mutex is locked\n",
1132 (unsigned long long)oi->ip_blkno);
1133 mutex_unlock(&oi->ip_io_mutex);
1136 * down_trylock() returns 0, down_write_trylock() returns 1
1139 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1140 "Clear inode of %llu, alloc_sem is locked\n",
1141 (unsigned long long)oi->ip_blkno);
1142 up_write(&oi->ip_alloc_sem);
1144 mlog_bug_on_msg(oi->ip_open_count,
1145 "Clear inode of %llu has open count %d\n",
1146 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1148 /* Clear all other flags. */
1150 oi->ip_dir_start_lookup = 0;
1151 oi->ip_blkno = 0ULL;
1154 * ip_jinode is used to track txns against this inode. We ensure that
1155 * the journal is flushed before journal shutdown. Thus it is safe to
1156 * have inodes get cleaned up after journal shutdown.
1158 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1165 /* Called under inode_lock, with no more references on the
1166 * struct inode, so it's safe here to check the flags field
1167 * and to manipulate i_nlink without any other locks. */
1168 void ocfs2_drop_inode(struct inode *inode)
1170 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1174 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1175 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1177 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1178 generic_delete_inode(inode);
1180 generic_drop_inode(inode);
1186 * This is called from our getattr.
1188 int ocfs2_inode_revalidate(struct dentry *dentry)
1190 struct inode *inode = dentry->d_inode;
1193 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1194 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1197 mlog(0, "eep, no inode!\n");
1202 spin_lock(&OCFS2_I(inode)->ip_lock);
1203 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1204 spin_unlock(&OCFS2_I(inode)->ip_lock);
1205 mlog(0, "inode deleted!\n");
1209 spin_unlock(&OCFS2_I(inode)->ip_lock);
1211 /* Let ocfs2_inode_lock do the work of updating our struct
1213 status = ocfs2_inode_lock(inode, NULL, 0);
1215 if (status != -ENOENT)
1219 ocfs2_inode_unlock(inode, 0);
1227 * Updates a disk inode from a
1229 * Only takes ip_lock.
1231 int ocfs2_mark_inode_dirty(handle_t *handle,
1232 struct inode *inode,
1233 struct buffer_head *bh)
1236 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1238 mlog_entry("(inode %llu)\n",
1239 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1241 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1242 OCFS2_JOURNAL_ACCESS_WRITE);
1248 spin_lock(&OCFS2_I(inode)->ip_lock);
1249 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1250 ocfs2_get_inode_flags(OCFS2_I(inode));
1251 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1252 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1253 spin_unlock(&OCFS2_I(inode)->ip_lock);
1255 fe->i_size = cpu_to_le64(i_size_read(inode));
1256 ocfs2_set_links_count(fe, inode->i_nlink);
1257 fe->i_uid = cpu_to_le32(inode->i_uid);
1258 fe->i_gid = cpu_to_le32(inode->i_gid);
1259 fe->i_mode = cpu_to_le16(inode->i_mode);
1260 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1261 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1262 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1263 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1264 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1265 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1267 ocfs2_journal_dirty(handle, bh);
1275 * Updates a struct inode from a disk inode.
1276 * does no i/o, only takes ip_lock.
1278 void ocfs2_refresh_inode(struct inode *inode,
1279 struct ocfs2_dinode *fe)
1281 spin_lock(&OCFS2_I(inode)->ip_lock);
1283 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1284 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1285 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1286 ocfs2_set_inode_flags(inode);
1287 i_size_write(inode, le64_to_cpu(fe->i_size));
1288 inode->i_nlink = ocfs2_read_links_count(fe);
1289 inode->i_uid = le32_to_cpu(fe->i_uid);
1290 inode->i_gid = le32_to_cpu(fe->i_gid);
1291 inode->i_mode = le16_to_cpu(fe->i_mode);
1292 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1293 inode->i_blocks = 0;
1295 inode->i_blocks = ocfs2_inode_sector_count(inode);
1296 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1297 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1298 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1299 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1300 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1301 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1303 spin_unlock(&OCFS2_I(inode)->ip_lock);
1306 int ocfs2_validate_inode_block(struct super_block *sb,
1307 struct buffer_head *bh)
1310 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1312 mlog(0, "Validating dinode %llu\n",
1313 (unsigned long long)bh->b_blocknr);
1315 BUG_ON(!buffer_uptodate(bh));
1318 * If the ecc fails, we return the error but otherwise
1319 * leave the filesystem running. We know any error is
1320 * local to this block.
1322 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1324 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1325 (unsigned long long)bh->b_blocknr);
1330 * Errors after here are fatal.
1335 if (!OCFS2_IS_VALID_DINODE(di)) {
1336 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1337 (unsigned long long)bh->b_blocknr, 7,
1342 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1343 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1344 (unsigned long long)bh->b_blocknr,
1345 (unsigned long long)le64_to_cpu(di->i_blkno));
1349 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1351 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1352 (unsigned long long)bh->b_blocknr);
1356 if (le32_to_cpu(di->i_fs_generation) !=
1357 OCFS2_SB(sb)->fs_generation) {
1359 "Invalid dinode #%llu: fs_generation is %u\n",
1360 (unsigned long long)bh->b_blocknr,
1361 le32_to_cpu(di->i_fs_generation));
1371 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1375 struct buffer_head *tmp = *bh;
1377 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1378 1, &tmp, flags, ocfs2_validate_inode_block);
1380 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1387 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1389 return ocfs2_read_inode_block_full(inode, bh, 0);
1393 static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1395 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1397 return oi->ip_blkno;
1400 static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1402 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1404 return oi->vfs_inode.i_sb;
1407 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1409 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1411 spin_lock(&oi->ip_lock);
1414 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1416 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1418 spin_unlock(&oi->ip_lock);
1421 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1423 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1425 mutex_lock(&oi->ip_io_mutex);
1428 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1430 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1432 mutex_unlock(&oi->ip_io_mutex);
1435 const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1436 .co_owner = ocfs2_inode_cache_owner,
1437 .co_get_super = ocfs2_inode_cache_get_super,
1438 .co_cache_lock = ocfs2_inode_cache_lock,
1439 .co_cache_unlock = ocfs2_inode_cache_unlock,
1440 .co_io_lock = ocfs2_inode_cache_io_lock,
1441 .co_io_unlock = ocfs2_inode_cache_io_unlock,