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>
32 #include <asm/byteorder.h>
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
41 #include "extent_map.h"
43 #include "heartbeat.h"
53 #include "buffer_head_io.h"
55 struct ocfs2_find_inode_args
59 unsigned int fi_flags;
60 unsigned int fi_sysfile_type;
63 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
65 static int ocfs2_read_locked_inode(struct inode *inode,
66 struct ocfs2_find_inode_args *args);
67 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
68 static int ocfs2_find_actor(struct inode *inode, void *opaque);
69 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
71 struct buffer_head *fe_bh);
73 void ocfs2_set_inode_flags(struct inode *inode)
75 unsigned int flags = OCFS2_I(inode)->ip_attr;
77 inode->i_flags &= ~(S_IMMUTABLE |
78 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
80 if (flags & OCFS2_IMMUTABLE_FL)
81 inode->i_flags |= S_IMMUTABLE;
83 if (flags & OCFS2_SYNC_FL)
84 inode->i_flags |= S_SYNC;
85 if (flags & OCFS2_APPEND_FL)
86 inode->i_flags |= S_APPEND;
87 if (flags & OCFS2_NOATIME_FL)
88 inode->i_flags |= S_NOATIME;
89 if (flags & OCFS2_DIRSYNC_FL)
90 inode->i_flags |= S_DIRSYNC;
93 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
94 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
96 unsigned int flags = oi->vfs_inode.i_flags;
98 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
99 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
101 oi->ip_attr |= OCFS2_SYNC_FL;
102 if (flags & S_APPEND)
103 oi->ip_attr |= OCFS2_APPEND_FL;
104 if (flags & S_IMMUTABLE)
105 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
106 if (flags & S_NOATIME)
107 oi->ip_attr |= OCFS2_NOATIME_FL;
108 if (flags & S_DIRSYNC)
109 oi->ip_attr |= OCFS2_DIRSYNC_FL;
112 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
115 struct inode *inode = NULL;
116 struct super_block *sb = osb->sb;
117 struct ocfs2_find_inode_args args;
119 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
121 /* Ok. By now we've either got the offsets passed to us by the
122 * caller, or we just pulled them off the bh. Lets do some
123 * sanity checks to make sure they're OK. */
125 inode = ERR_PTR(-EINVAL);
126 mlog_errno(PTR_ERR(inode));
130 args.fi_blkno = blkno;
131 args.fi_flags = flags;
132 args.fi_ino = ino_from_blkno(sb, blkno);
133 args.fi_sysfile_type = sysfile_type;
135 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
136 ocfs2_init_locked_inode, &args);
137 /* inode was *not* in the inode cache. 2.6.x requires
138 * us to do our own read_inode call and unlock it
140 if (inode && inode->i_state & I_NEW) {
141 mlog(0, "Inode was not in inode cache, reading it.\n");
142 ocfs2_read_locked_inode(inode, &args);
143 unlock_new_inode(inode);
146 inode = ERR_PTR(-ENOMEM);
147 mlog_errno(PTR_ERR(inode));
150 if (is_bad_inode(inode)) {
152 inode = ERR_PTR(-ESTALE);
157 if (!IS_ERR(inode)) {
158 mlog(0, "returning inode with number %llu\n",
159 (unsigned long long)OCFS2_I(inode)->ip_blkno);
160 mlog_exit_ptr(inode);
168 * here's how inodes get read from disk:
169 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
170 * found? : return the in-memory inode
171 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
174 static int ocfs2_find_actor(struct inode *inode, void *opaque)
176 struct ocfs2_find_inode_args *args = NULL;
177 struct ocfs2_inode_info *oi = OCFS2_I(inode);
180 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
184 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
186 if (oi->ip_blkno != args->fi_blkno)
196 * initialize the new inode, but don't do anything that would cause
198 * return 0 on success, 1 on failure
200 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
202 struct ocfs2_find_inode_args *args = opaque;
204 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
206 inode->i_ino = args->fi_ino;
207 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
208 if (args->fi_sysfile_type != 0)
209 lockdep_set_class(&inode->i_mutex,
210 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
216 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
219 struct super_block *sb;
220 struct ocfs2_super *osb;
221 int status = -EINVAL;
224 mlog_entry("(0x%p, size:%llu)\n", inode,
225 (unsigned long long)le64_to_cpu(fe->i_size));
230 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
231 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
234 /* this means that read_inode cannot create a superblock inode
235 * today. change if needed. */
236 if (!OCFS2_IS_VALID_DINODE(fe) ||
237 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
238 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
239 "signature = %.*s, flags = 0x%x\n",
241 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
242 fe->i_signature, le32_to_cpu(fe->i_flags));
246 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
247 mlog(ML_ERROR, "file entry generation does not match "
248 "superblock! osb->fs_generation=%x, "
249 "fe->i_fs_generation=%x\n",
250 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
254 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
255 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
256 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
258 inode->i_version = 1;
259 inode->i_generation = le32_to_cpu(fe->i_generation);
260 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
261 inode->i_mode = le16_to_cpu(fe->i_mode);
262 inode->i_uid = le32_to_cpu(fe->i_uid);
263 inode->i_gid = le32_to_cpu(fe->i_gid);
265 /* Fast symlinks will have i_size but no allocated clusters. */
266 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
269 inode->i_blocks = ocfs2_inode_sector_count(inode);
270 inode->i_mapping->a_ops = &ocfs2_aops;
271 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
272 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
273 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
274 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
275 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
276 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
278 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
280 "ip_blkno %llu != i_blkno %llu!\n",
281 (unsigned long long)OCFS2_I(inode)->ip_blkno,
282 (unsigned long long)le64_to_cpu(fe->i_blkno));
284 inode->i_nlink = le16_to_cpu(fe->i_links_count);
286 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
287 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
289 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
290 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
291 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
292 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
293 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
294 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
295 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
296 /* we can't actually hit this as read_inode can't
297 * handle superblocks today ;-) */
301 switch (inode->i_mode & S_IFMT) {
304 inode->i_fop = &ocfs2_fops;
306 inode->i_fop = &ocfs2_fops_no_plocks;
307 inode->i_op = &ocfs2_file_iops;
308 i_size_write(inode, le64_to_cpu(fe->i_size));
311 inode->i_op = &ocfs2_dir_iops;
313 inode->i_fop = &ocfs2_dops;
315 inode->i_fop = &ocfs2_dops_no_plocks;
316 i_size_write(inode, le64_to_cpu(fe->i_size));
319 if (ocfs2_inode_is_fast_symlink(inode))
320 inode->i_op = &ocfs2_fast_symlink_inode_operations;
322 inode->i_op = &ocfs2_symlink_inode_operations;
323 i_size_write(inode, le64_to_cpu(fe->i_size));
326 inode->i_op = &ocfs2_special_file_iops;
327 init_special_inode(inode, inode->i_mode,
333 inode->i_ino = ino_from_blkno(inode->i_sb,
334 le64_to_cpu(fe->i_blkno));
337 * If we ever want to create system files from kernel,
338 * the generation argument to
339 * ocfs2_inode_lock_res_init() will have to change.
341 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
343 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
344 OCFS2_LOCK_TYPE_META, 0, inode);
346 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
347 OCFS2_LOCK_TYPE_OPEN, 0, inode);
350 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
351 OCFS2_LOCK_TYPE_RW, inode->i_generation,
354 ocfs2_set_inode_flags(inode);
362 static int ocfs2_read_locked_inode(struct inode *inode,
363 struct ocfs2_find_inode_args *args)
365 struct super_block *sb;
366 struct ocfs2_super *osb;
367 struct ocfs2_dinode *fe;
368 struct buffer_head *bh = NULL;
369 int status, can_lock;
372 mlog_entry("(0x%p, 0x%p)\n", inode, args);
375 if (inode == NULL || inode->i_sb == NULL) {
376 mlog(ML_ERROR, "bad inode\n");
383 mlog(ML_ERROR, "bad inode args\n");
384 make_bad_inode(inode);
389 * To improve performance of cold-cache inode stats, we take
390 * the cluster lock here if possible.
392 * Generally, OCFS2 never trusts the contents of an inode
393 * unless it's holding a cluster lock, so taking it here isn't
394 * a correctness issue as much as it is a performance
397 * There are three times when taking the lock is not a good idea:
399 * 1) During startup, before we have initialized the DLM.
401 * 2) If we are reading certain system files which never get
402 * cluster locks (local alloc, truncate log).
404 * 3) If the process doing the iget() is responsible for
405 * orphan dir recovery. We're holding the orphan dir lock and
406 * can get into a deadlock with another process on another
407 * node in ->delete_inode().
409 * #1 and #2 can be simply solved by never taking the lock
410 * here for system files (which are the only type we read
411 * during mount). It's a heavier approach, but our main
412 * concern is user-accesible files anyway.
414 * #3 works itself out because we'll eventually take the
415 * cluster lock before trusting anything anyway.
417 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
418 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
419 && !ocfs2_mount_local(osb);
422 * To maintain backwards compatibility with older versions of
423 * ocfs2-tools, we still store the generation value for system
424 * files. The only ones that actually matter to userspace are
425 * the journals, but it's easier and inexpensive to just flag
426 * all system files similarly.
428 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
429 generation = osb->fs_generation;
431 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
432 OCFS2_LOCK_TYPE_META,
435 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
436 OCFS2_LOCK_TYPE_OPEN,
440 status = ocfs2_open_lock(inode);
442 make_bad_inode(inode);
446 status = ocfs2_inode_lock(inode, NULL, 0);
448 make_bad_inode(inode);
454 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
455 status = ocfs2_try_open_lock(inode, 0);
457 make_bad_inode(inode);
462 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
463 can_lock ? inode : NULL);
470 fe = (struct ocfs2_dinode *) bh->b_data;
471 if (!OCFS2_IS_VALID_DINODE(fe)) {
472 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
473 (unsigned long long)args->fi_blkno, 7,
479 * This is a code bug. Right now the caller needs to
480 * understand whether it is asking for a system file inode or
481 * not so the proper lock names can be built.
483 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
484 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
485 "Inode %llu: system file state is ambigous\n",
486 (unsigned long long)args->fi_blkno);
488 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
489 S_ISBLK(le16_to_cpu(fe->i_mode)))
490 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
492 if (ocfs2_populate_inode(inode, fe, 0) < 0)
495 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
501 ocfs2_inode_unlock(inode, 0);
504 make_bad_inode(inode);
513 void ocfs2_sync_blockdev(struct super_block *sb)
515 sync_blockdev(sb->s_bdev);
518 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
520 struct buffer_head *fe_bh)
523 struct ocfs2_truncate_context *tc = NULL;
524 struct ocfs2_dinode *fe;
525 handle_t *handle = NULL;
529 fe = (struct ocfs2_dinode *) fe_bh->b_data;
532 * This check will also skip truncate of inodes with inline
533 * data and fast symlinks.
535 if (fe->i_clusters) {
536 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
537 if (IS_ERR(handle)) {
538 status = PTR_ERR(handle);
543 status = ocfs2_journal_access(handle, inode, fe_bh,
544 OCFS2_JOURNAL_ACCESS_WRITE);
550 i_size_write(inode, 0);
552 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
558 ocfs2_commit_trans(osb, handle);
561 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
567 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
576 ocfs2_commit_trans(osb, handle);
581 static int ocfs2_remove_inode(struct inode *inode,
582 struct buffer_head *di_bh,
583 struct inode *orphan_dir_inode,
584 struct buffer_head *orphan_dir_bh)
587 struct inode *inode_alloc_inode = NULL;
588 struct buffer_head *inode_alloc_bh = NULL;
590 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
591 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
594 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
595 le16_to_cpu(di->i_suballoc_slot));
596 if (!inode_alloc_inode) {
602 mutex_lock(&inode_alloc_inode->i_mutex);
603 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
605 mutex_unlock(&inode_alloc_inode->i_mutex);
611 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
612 if (IS_ERR(handle)) {
613 status = PTR_ERR(handle);
618 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
625 /* set the inodes dtime */
626 status = ocfs2_journal_access(handle, inode, di_bh,
627 OCFS2_JOURNAL_ACCESS_WRITE);
633 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
634 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
636 status = ocfs2_journal_dirty(handle, di_bh);
642 ocfs2_remove_from_cache(inode, di_bh);
644 status = ocfs2_free_dinode(handle, inode_alloc_inode,
650 ocfs2_commit_trans(osb, handle);
652 ocfs2_inode_unlock(inode_alloc_inode, 1);
653 mutex_unlock(&inode_alloc_inode->i_mutex);
654 brelse(inode_alloc_bh);
656 iput(inode_alloc_inode);
662 * Serialize with orphan dir recovery. If the process doing
663 * recovery on this orphan dir does an iget() with the dir
664 * i_mutex held, we'll deadlock here. Instead we detect this
665 * and exit early - recovery will wipe this inode for us.
667 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
672 spin_lock(&osb->osb_lock);
673 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
674 mlog(0, "Recovery is happening on orphan dir %d, will skip "
675 "this inode\n", slot);
679 /* This signals to the orphan recovery process that it should
680 * wait for us to handle the wipe. */
681 osb->osb_orphan_wipes[slot]++;
683 spin_unlock(&osb->osb_lock);
687 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
690 spin_lock(&osb->osb_lock);
691 osb->osb_orphan_wipes[slot]--;
692 spin_unlock(&osb->osb_lock);
694 wake_up(&osb->osb_wipe_event);
697 static int ocfs2_wipe_inode(struct inode *inode,
698 struct buffer_head *di_bh)
700 int status, orphaned_slot;
701 struct inode *orphan_dir_inode = NULL;
702 struct buffer_head *orphan_dir_bh = NULL;
703 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
704 struct ocfs2_dinode *di;
706 di = (struct ocfs2_dinode *) di_bh->b_data;
707 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
709 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
713 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
714 ORPHAN_DIR_SYSTEM_INODE,
716 if (!orphan_dir_inode) {
722 /* Lock the orphan dir. The lock will be held for the entire
723 * delete_inode operation. We do this now to avoid races with
724 * recovery completion on other nodes. */
725 mutex_lock(&orphan_dir_inode->i_mutex);
726 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
728 mutex_unlock(&orphan_dir_inode->i_mutex);
734 /* we do this while holding the orphan dir lock because we
735 * don't want recovery being run from another node to try an
736 * inode delete underneath us -- this will result in two nodes
737 * truncating the same file! */
738 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
741 goto bail_unlock_dir;
744 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
750 ocfs2_inode_unlock(orphan_dir_inode, 1);
751 mutex_unlock(&orphan_dir_inode->i_mutex);
752 brelse(orphan_dir_bh);
754 iput(orphan_dir_inode);
755 ocfs2_signal_wipe_completion(osb, orphaned_slot);
760 /* There is a series of simple checks that should be done before a
761 * trylock is even considered. Encapsulate those in this function. */
762 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
765 struct ocfs2_inode_info *oi = OCFS2_I(inode);
766 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
768 /* We shouldn't be getting here for the root directory
770 if (inode == osb->root_inode) {
771 mlog(ML_ERROR, "Skipping delete of root inode.\n");
775 /* If we're coming from downconvert_thread we can't go into our own
776 * voting [hello, deadlock city!], so unforuntately we just
777 * have to skip deleting this guy. That's OK though because
778 * the node who's doing the actual deleting should handle it
780 if (current == osb->dc_task) {
781 mlog(0, "Skipping delete of %lu because we're currently "
782 "in downconvert\n", inode->i_ino);
786 spin_lock(&oi->ip_lock);
787 /* OCFS2 *never* deletes system files. This should technically
788 * never get here as system file inodes should always have a
789 * positive link count. */
790 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
791 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
792 (unsigned long long)oi->ip_blkno);
796 /* If we have allowd wipe of this inode for another node, it
797 * will be marked here so we can safely skip it. Recovery will
798 * cleanup any inodes we might inadvertantly skip here. */
799 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
800 mlog(0, "Skipping delete of %lu because another node "
801 "has done this for us.\n", inode->i_ino);
807 spin_unlock(&oi->ip_lock);
812 /* Query the cluster to determine whether we should wipe an inode from
815 * Requires the inode to have the cluster lock. */
816 static int ocfs2_query_inode_wipe(struct inode *inode,
817 struct buffer_head *di_bh,
821 struct ocfs2_inode_info *oi = OCFS2_I(inode);
822 struct ocfs2_dinode *di;
826 /* While we were waiting for the cluster lock in
827 * ocfs2_delete_inode, another node might have asked to delete
828 * the inode. Recheck our flags to catch this. */
829 if (!ocfs2_inode_is_valid_to_delete(inode)) {
830 mlog(0, "Skipping delete of %llu because flags changed\n",
831 (unsigned long long)oi->ip_blkno);
835 /* Now that we have an up to date inode, we can double check
837 if (inode->i_nlink) {
838 mlog(0, "Skipping delete of %llu because nlink = %u\n",
839 (unsigned long long)oi->ip_blkno, inode->i_nlink);
843 /* Do some basic inode verification... */
844 di = (struct ocfs2_dinode *) di_bh->b_data;
845 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
846 /* for lack of a better error? */
849 "Inode %llu (on-disk %llu) not orphaned! "
850 "Disk flags 0x%x, inode flags 0x%x\n",
851 (unsigned long long)oi->ip_blkno,
852 (unsigned long long)le64_to_cpu(di->i_blkno),
853 le32_to_cpu(di->i_flags), oi->ip_flags);
857 /* has someone already deleted us?! baaad... */
865 * This is how ocfs2 determines whether an inode is still live
866 * within the cluster. Every node takes a shared read lock on
867 * the inode open lock in ocfs2_read_locked_inode(). When we
868 * get to ->delete_inode(), each node tries to convert it's
869 * lock to an exclusive. Trylocks are serialized by the inode
870 * meta data lock. If the upconvert suceeds, we know the inode
871 * is no longer live and can be deleted.
873 * Though we call this with the meta data lock held, the
874 * trylock keeps us from ABBA deadlock.
876 status = ocfs2_try_open_lock(inode, 1);
877 if (status == -EAGAIN) {
879 mlog(0, "Skipping delete of %llu because it is in use on "
880 "other nodes\n", (unsigned long long)oi->ip_blkno);
889 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
890 (unsigned long long)oi->ip_blkno,
891 le16_to_cpu(di->i_orphaned_slot));
897 /* Support function for ocfs2_delete_inode. Will help us keep the
898 * inode data in a consistent state for clear_inode. Always truncates
899 * pages, optionally sync's them first. */
900 static void ocfs2_cleanup_delete_inode(struct inode *inode,
903 mlog(0, "Cleanup inode %llu, sync = %d\n",
904 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
906 write_inode_now(inode, 1);
907 truncate_inode_pages(&inode->i_data, 0);
910 void ocfs2_delete_inode(struct inode *inode)
913 sigset_t blocked, oldset;
914 struct buffer_head *di_bh = NULL;
916 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
918 if (is_bad_inode(inode)) {
919 mlog(0, "Skipping delete of bad inode\n");
923 if (!ocfs2_inode_is_valid_to_delete(inode)) {
924 /* It's probably not necessary to truncate_inode_pages
925 * here but we do it for safety anyway (it will most
926 * likely be a no-op anyway) */
927 ocfs2_cleanup_delete_inode(inode, 0);
931 /* We want to block signals in delete_inode as the lock and
932 * messaging paths may return us -ERESTARTSYS. Which would
933 * cause us to exit early, resulting in inodes being orphaned
935 sigfillset(&blocked);
936 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
939 ocfs2_cleanup_delete_inode(inode, 1);
943 /* Lock down the inode. This gives us an up to date view of
944 * it's metadata (for verification), and allows us to
945 * serialize delete_inode on multiple nodes.
947 * Even though we might be doing a truncate, we don't take the
948 * allocation lock here as it won't be needed - nobody will
949 * have the file open.
951 status = ocfs2_inode_lock(inode, &di_bh, 1);
953 if (status != -ENOENT)
955 ocfs2_cleanup_delete_inode(inode, 0);
959 /* Query the cluster. This will be the final decision made
960 * before we go ahead and wipe the inode. */
961 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
962 if (!wipe || status < 0) {
963 /* Error and remote inode busy both mean we won't be
964 * removing the inode, so they take almost the same
969 /* Someone in the cluster has disallowed a wipe of
970 * this inode, or it was never completely
971 * orphaned. Write out the pages and exit now. */
972 ocfs2_cleanup_delete_inode(inode, 1);
973 goto bail_unlock_inode;
976 ocfs2_cleanup_delete_inode(inode, 0);
978 status = ocfs2_wipe_inode(inode, di_bh);
980 if (status != -EDEADLK)
982 goto bail_unlock_inode;
986 * Mark the inode as successfully deleted.
988 * This is important for ocfs2_clear_inode() as it will check
989 * this flag and skip any checkpointing work
991 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
992 * the LVB for other nodes.
994 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
997 ocfs2_inode_unlock(inode, 1);
1000 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1008 void ocfs2_clear_inode(struct inode *inode)
1011 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1018 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1019 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1021 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1022 "Inode=%lu\n", inode->i_ino);
1024 /* To preven remote deletes we hold open lock before, now it
1025 * is time to unlock PR and EX open locks. */
1026 ocfs2_open_unlock(inode);
1028 /* Do these before all the other work so that we don't bounce
1029 * the downconvert thread while waiting to destroy the locks. */
1030 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1031 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1032 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1034 /* We very well may get a clear_inode before all an inodes
1035 * metadata has hit disk. Of course, we can't drop any cluster
1036 * locks until the journal has finished with it. The only
1037 * exception here are successfully wiped inodes - their
1038 * metadata can now be considered to be part of the system
1039 * inodes from which it came. */
1040 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1041 ocfs2_checkpoint_inode(inode);
1043 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1044 "Clear inode of %llu, inode has io markers\n",
1045 (unsigned long long)oi->ip_blkno);
1047 ocfs2_extent_map_trunc(inode, 0);
1049 status = ocfs2_drop_inode_locks(inode);
1053 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1054 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1055 ocfs2_lock_res_free(&oi->ip_open_lockres);
1057 ocfs2_metadata_cache_purge(inode);
1059 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
1060 "Clear inode of %llu, inode has %u cache items\n",
1061 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
1063 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
1064 "Clear inode of %llu, inode has a bad flag\n",
1065 (unsigned long long)oi->ip_blkno);
1067 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1068 "Clear inode of %llu, inode is locked\n",
1069 (unsigned long long)oi->ip_blkno);
1071 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1072 "Clear inode of %llu, io_mutex is locked\n",
1073 (unsigned long long)oi->ip_blkno);
1074 mutex_unlock(&oi->ip_io_mutex);
1077 * down_trylock() returns 0, down_write_trylock() returns 1
1080 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1081 "Clear inode of %llu, alloc_sem is locked\n",
1082 (unsigned long long)oi->ip_blkno);
1083 up_write(&oi->ip_alloc_sem);
1085 mlog_bug_on_msg(oi->ip_open_count,
1086 "Clear inode of %llu has open count %d\n",
1087 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1089 /* Clear all other flags. */
1090 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1091 oi->ip_created_trans = 0;
1092 oi->ip_last_trans = 0;
1093 oi->ip_dir_start_lookup = 0;
1094 oi->ip_blkno = 0ULL;
1100 /* Called under inode_lock, with no more references on the
1101 * struct inode, so it's safe here to check the flags field
1102 * and to manipulate i_nlink without any other locks. */
1103 void ocfs2_drop_inode(struct inode *inode)
1105 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1109 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1110 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1112 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1113 generic_delete_inode(inode);
1115 generic_drop_inode(inode);
1121 * TODO: this should probably be merged into ocfs2_get_block
1123 * However, you now need to pay attention to the cont_prepare_write()
1124 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1125 * expects never to extend).
1127 struct buffer_head *ocfs2_bread(struct inode *inode,
1128 int block, int *err, int reada)
1130 struct buffer_head *bh = NULL;
1133 int readflags = OCFS2_BH_CACHED;
1136 readflags |= OCFS2_BH_READAHEAD;
1138 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1139 i_size_read(inode)) {
1144 down_read(&OCFS2_I(inode)->ip_alloc_sem);
1145 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1147 up_read(&OCFS2_I(inode)->ip_alloc_sem);
1153 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1173 * This is called from our getattr.
1175 int ocfs2_inode_revalidate(struct dentry *dentry)
1177 struct inode *inode = dentry->d_inode;
1180 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1181 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1184 mlog(0, "eep, no inode!\n");
1189 spin_lock(&OCFS2_I(inode)->ip_lock);
1190 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1191 spin_unlock(&OCFS2_I(inode)->ip_lock);
1192 mlog(0, "inode deleted!\n");
1196 spin_unlock(&OCFS2_I(inode)->ip_lock);
1198 /* Let ocfs2_inode_lock do the work of updating our struct
1200 status = ocfs2_inode_lock(inode, NULL, 0);
1202 if (status != -ENOENT)
1206 ocfs2_inode_unlock(inode, 0);
1214 * Updates a disk inode from a
1216 * Only takes ip_lock.
1218 int ocfs2_mark_inode_dirty(handle_t *handle,
1219 struct inode *inode,
1220 struct buffer_head *bh)
1223 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1225 mlog_entry("(inode %llu)\n",
1226 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1228 status = ocfs2_journal_access(handle, inode, bh,
1229 OCFS2_JOURNAL_ACCESS_WRITE);
1235 spin_lock(&OCFS2_I(inode)->ip_lock);
1236 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1237 ocfs2_get_inode_flags(OCFS2_I(inode));
1238 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1239 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1240 spin_unlock(&OCFS2_I(inode)->ip_lock);
1242 fe->i_size = cpu_to_le64(i_size_read(inode));
1243 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1244 fe->i_uid = cpu_to_le32(inode->i_uid);
1245 fe->i_gid = cpu_to_le32(inode->i_gid);
1246 fe->i_mode = cpu_to_le16(inode->i_mode);
1247 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1248 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1249 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1250 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1251 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1252 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1254 status = ocfs2_journal_dirty(handle, bh);
1267 * Updates a struct inode from a disk inode.
1268 * does no i/o, only takes ip_lock.
1270 void ocfs2_refresh_inode(struct inode *inode,
1271 struct ocfs2_dinode *fe)
1273 spin_lock(&OCFS2_I(inode)->ip_lock);
1275 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1276 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1277 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1278 ocfs2_set_inode_flags(inode);
1279 i_size_write(inode, le64_to_cpu(fe->i_size));
1280 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1281 inode->i_uid = le32_to_cpu(fe->i_uid);
1282 inode->i_gid = le32_to_cpu(fe->i_gid);
1283 inode->i_mode = le16_to_cpu(fe->i_mode);
1284 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1285 inode->i_blocks = 0;
1287 inode->i_blocks = ocfs2_inode_sector_count(inode);
1288 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1289 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1290 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1291 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1292 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1293 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1295 spin_unlock(&OCFS2_I(inode)->ip_lock);