xref: /linux/fs/ocfs2/inode.c (revision c145211d1f9e2ef19e7b4c2b943f68366daa97af)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * inode.c
5  *
6  * vfs' aops, fops, dops and iops
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
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.
14  *
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.
19  *
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.
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/types.h>
28 #include <linux/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/quotaops.h>
31 
32 #include <asm/byteorder.h>
33 
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
36 
37 #include "ocfs2.h"
38 
39 #include "alloc.h"
40 #include "dir.h"
41 #include "blockcheck.h"
42 #include "dlmglue.h"
43 #include "extent_map.h"
44 #include "file.h"
45 #include "heartbeat.h"
46 #include "inode.h"
47 #include "journal.h"
48 #include "namei.h"
49 #include "suballoc.h"
50 #include "super.h"
51 #include "symlink.h"
52 #include "sysfile.h"
53 #include "uptodate.h"
54 #include "xattr.h"
55 #include "refcounttree.h"
56 
57 #include "buffer_head_io.h"
58 
59 struct ocfs2_find_inode_args
60 {
61 	u64		fi_blkno;
62 	unsigned long	fi_ino;
63 	unsigned int	fi_flags;
64 	unsigned int	fi_sysfile_type;
65 };
66 
67 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
68 
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,
74 				    struct inode *inode,
75 				    struct buffer_head *fe_bh);
76 
77 void ocfs2_set_inode_flags(struct inode *inode)
78 {
79 	unsigned int flags = OCFS2_I(inode)->ip_attr;
80 
81 	inode->i_flags &= ~(S_IMMUTABLE |
82 		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
83 
84 	if (flags & OCFS2_IMMUTABLE_FL)
85 		inode->i_flags |= S_IMMUTABLE;
86 
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;
95 }
96 
97 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
98 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
99 {
100 	unsigned int flags = oi->vfs_inode.i_flags;
101 
102 	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
103 			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
104 	if (flags & S_SYNC)
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;
114 }
115 
116 struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
117 {
118 	struct ocfs2_find_inode_args args;
119 
120 	args.fi_blkno = blkno;
121 	args.fi_flags = 0;
122 	args.fi_ino = ino_from_blkno(sb, blkno);
123 	args.fi_sysfile_type = 0;
124 
125 	return ilookup5(sb, blkno, ocfs2_find_actor, &args);
126 }
127 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
128 			 int sysfile_type)
129 {
130 	struct inode *inode = NULL;
131 	struct super_block *sb = osb->sb;
132 	struct ocfs2_find_inode_args args;
133 
134 	mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
135 
136 	/* Ok. By now we've either got the offsets passed to us by the
137 	 * caller, or we just pulled them off the bh. Lets do some
138 	 * sanity checks to make sure they're OK. */
139 	if (blkno == 0) {
140 		inode = ERR_PTR(-EINVAL);
141 		mlog_errno(PTR_ERR(inode));
142 		goto bail;
143 	}
144 
145 	args.fi_blkno = blkno;
146 	args.fi_flags = flags;
147 	args.fi_ino = ino_from_blkno(sb, blkno);
148 	args.fi_sysfile_type = sysfile_type;
149 
150 	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
151 			     ocfs2_init_locked_inode, &args);
152 	/* inode was *not* in the inode cache. 2.6.x requires
153 	 * us to do our own read_inode call and unlock it
154 	 * afterwards. */
155 	if (inode && inode->i_state & I_NEW) {
156 		mlog(0, "Inode was not in inode cache, reading it.\n");
157 		ocfs2_read_locked_inode(inode, &args);
158 		unlock_new_inode(inode);
159 	}
160 	if (inode == NULL) {
161 		inode = ERR_PTR(-ENOMEM);
162 		mlog_errno(PTR_ERR(inode));
163 		goto bail;
164 	}
165 	if (is_bad_inode(inode)) {
166 		iput(inode);
167 		inode = ERR_PTR(-ESTALE);
168 		goto bail;
169 	}
170 
171 bail:
172 	if (!IS_ERR(inode)) {
173 		mlog(0, "returning inode with number %llu\n",
174 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
175 		mlog_exit_ptr(inode);
176 	}
177 
178 	return inode;
179 }
180 
181 
182 /*
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
187  */
188 
189 static int ocfs2_find_actor(struct inode *inode, void *opaque)
190 {
191 	struct ocfs2_find_inode_args *args = NULL;
192 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
193 	int ret = 0;
194 
195 	mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
196 
197 	args = opaque;
198 
199 	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
200 
201 	if (oi->ip_blkno != args->fi_blkno)
202 		goto bail;
203 
204 	ret = 1;
205 bail:
206 	mlog_exit(ret);
207 	return ret;
208 }
209 
210 /*
211  * initialize the new inode, but don't do anything that would cause
212  * us to sleep.
213  * return 0 on success, 1 on failure
214  */
215 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
216 {
217 	struct ocfs2_find_inode_args *args = opaque;
218 	static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
219 				     ocfs2_file_ip_alloc_sem_key;
220 
221 	mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
222 
223 	inode->i_ino = args->fi_ino;
224 	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
225 	if (args->fi_sysfile_type != 0)
226 		lockdep_set_class(&inode->i_mutex,
227 			&ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
228 	if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
229 	    args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
230 	    args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
231 	    args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
232 		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233 				  &ocfs2_quota_ip_alloc_sem_key);
234 	else
235 		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
236 				  &ocfs2_file_ip_alloc_sem_key);
237 
238 	mlog_exit(0);
239 	return 0;
240 }
241 
242 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
243 			  int create_ino)
244 {
245 	struct super_block *sb;
246 	struct ocfs2_super *osb;
247 	int use_plocks = 1;
248 
249 	mlog_entry("(0x%p, size:%llu)\n", inode,
250 		   (unsigned long long)le64_to_cpu(fe->i_size));
251 
252 	sb = inode->i_sb;
253 	osb = OCFS2_SB(sb);
254 
255 	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
256 	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
257 		use_plocks = 0;
258 
259 	/*
260 	 * These have all been checked by ocfs2_read_inode_block() or set
261 	 * by ocfs2_mknod_locked(), so a failure is a code bug.
262 	 */
263 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
264 						cannot create a superblock
265 						inode today.  change if
266 						that is needed. */
267 	BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
268 	BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
269 
270 
271 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
272 	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
273 	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
274 
275 	inode->i_version = 1;
276 	inode->i_generation = le32_to_cpu(fe->i_generation);
277 	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
278 	inode->i_mode = le16_to_cpu(fe->i_mode);
279 	inode->i_uid = le32_to_cpu(fe->i_uid);
280 	inode->i_gid = le32_to_cpu(fe->i_gid);
281 
282 	/* Fast symlinks will have i_size but no allocated clusters. */
283 	if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
284 		inode->i_blocks = 0;
285 	else
286 		inode->i_blocks = ocfs2_inode_sector_count(inode);
287 	inode->i_mapping->a_ops = &ocfs2_aops;
288 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
289 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
290 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
291 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
292 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
293 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
294 
295 	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
296 		mlog(ML_ERROR,
297 		     "ip_blkno %llu != i_blkno %llu!\n",
298 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
299 		     (unsigned long long)le64_to_cpu(fe->i_blkno));
300 
301 	inode->i_nlink = ocfs2_read_links_count(fe);
302 
303 	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
304 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
305 		inode->i_flags |= S_NOQUOTA;
306 	}
307 
308 	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
309 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
310 		mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
311 	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
312 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
313 	} else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
314 		inode->i_flags |= S_NOQUOTA;
315 	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
316 		mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
317 		/* we can't actually hit this as read_inode can't
318 		 * handle superblocks today ;-) */
319 		BUG();
320 	}
321 
322 	switch (inode->i_mode & S_IFMT) {
323 	    case S_IFREG:
324 		    if (use_plocks)
325 			    inode->i_fop = &ocfs2_fops;
326 		    else
327 			    inode->i_fop = &ocfs2_fops_no_plocks;
328 		    inode->i_op = &ocfs2_file_iops;
329 		    i_size_write(inode, le64_to_cpu(fe->i_size));
330 		    break;
331 	    case S_IFDIR:
332 		    inode->i_op = &ocfs2_dir_iops;
333 		    if (use_plocks)
334 			    inode->i_fop = &ocfs2_dops;
335 		    else
336 			    inode->i_fop = &ocfs2_dops_no_plocks;
337 		    i_size_write(inode, le64_to_cpu(fe->i_size));
338 		    break;
339 	    case S_IFLNK:
340 		    if (ocfs2_inode_is_fast_symlink(inode))
341 			inode->i_op = &ocfs2_fast_symlink_inode_operations;
342 		    else
343 			inode->i_op = &ocfs2_symlink_inode_operations;
344 		    i_size_write(inode, le64_to_cpu(fe->i_size));
345 		    break;
346 	    default:
347 		    inode->i_op = &ocfs2_special_file_iops;
348 		    init_special_inode(inode, inode->i_mode,
349 				       inode->i_rdev);
350 		    break;
351 	}
352 
353 	if (create_ino) {
354 		inode->i_ino = ino_from_blkno(inode->i_sb,
355 			       le64_to_cpu(fe->i_blkno));
356 
357 		/*
358 		 * If we ever want to create system files from kernel,
359 		 * the generation argument to
360 		 * ocfs2_inode_lock_res_init() will have to change.
361 		 */
362 		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
363 
364 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
365 					  OCFS2_LOCK_TYPE_META, 0, inode);
366 
367 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
368 					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
369 	}
370 
371 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
372 				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
373 				  inode);
374 
375 	ocfs2_set_inode_flags(inode);
376 
377 	OCFS2_I(inode)->ip_last_used_slot = 0;
378 	OCFS2_I(inode)->ip_last_used_group = 0;
379 	mlog_exit_void();
380 }
381 
382 static int ocfs2_read_locked_inode(struct inode *inode,
383 				   struct ocfs2_find_inode_args *args)
384 {
385 	struct super_block *sb;
386 	struct ocfs2_super *osb;
387 	struct ocfs2_dinode *fe;
388 	struct buffer_head *bh = NULL;
389 	int status, can_lock;
390 	u32 generation = 0;
391 
392 	mlog_entry("(0x%p, 0x%p)\n", inode, args);
393 
394 	status = -EINVAL;
395 	if (inode == NULL || inode->i_sb == NULL) {
396 		mlog(ML_ERROR, "bad inode\n");
397 		return status;
398 	}
399 	sb = inode->i_sb;
400 	osb = OCFS2_SB(sb);
401 
402 	if (!args) {
403 		mlog(ML_ERROR, "bad inode args\n");
404 		make_bad_inode(inode);
405 		return status;
406 	}
407 
408 	/*
409 	 * To improve performance of cold-cache inode stats, we take
410 	 * the cluster lock here if possible.
411 	 *
412 	 * Generally, OCFS2 never trusts the contents of an inode
413 	 * unless it's holding a cluster lock, so taking it here isn't
414 	 * a correctness issue as much as it is a performance
415 	 * improvement.
416 	 *
417 	 * There are three times when taking the lock is not a good idea:
418 	 *
419 	 * 1) During startup, before we have initialized the DLM.
420 	 *
421 	 * 2) If we are reading certain system files which never get
422 	 *    cluster locks (local alloc, truncate log).
423 	 *
424 	 * 3) If the process doing the iget() is responsible for
425 	 *    orphan dir recovery. We're holding the orphan dir lock and
426 	 *    can get into a deadlock with another process on another
427 	 *    node in ->delete_inode().
428 	 *
429 	 * #1 and #2 can be simply solved by never taking the lock
430 	 * here for system files (which are the only type we read
431 	 * during mount). It's a heavier approach, but our main
432 	 * concern is user-accesible files anyway.
433 	 *
434 	 * #3 works itself out because we'll eventually take the
435 	 * cluster lock before trusting anything anyway.
436 	 */
437 	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
438 		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
439 		&& !ocfs2_mount_local(osb);
440 
441 	/*
442 	 * To maintain backwards compatibility with older versions of
443 	 * ocfs2-tools, we still store the generation value for system
444 	 * files. The only ones that actually matter to userspace are
445 	 * the journals, but it's easier and inexpensive to just flag
446 	 * all system files similarly.
447 	 */
448 	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
449 		generation = osb->fs_generation;
450 
451 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
452 				  OCFS2_LOCK_TYPE_META,
453 				  generation, inode);
454 
455 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
456 				  OCFS2_LOCK_TYPE_OPEN,
457 				  0, inode);
458 
459 	if (can_lock) {
460 		status = ocfs2_open_lock(inode);
461 		if (status) {
462 			make_bad_inode(inode);
463 			mlog_errno(status);
464 			return status;
465 		}
466 		status = ocfs2_inode_lock(inode, NULL, 0);
467 		if (status) {
468 			make_bad_inode(inode);
469 			mlog_errno(status);
470 			return status;
471 		}
472 	}
473 
474 	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
475 		status = ocfs2_try_open_lock(inode, 0);
476 		if (status) {
477 			make_bad_inode(inode);
478 			return status;
479 		}
480 	}
481 
482 	if (can_lock) {
483 		status = ocfs2_read_inode_block_full(inode, &bh,
484 						     OCFS2_BH_IGNORE_CACHE);
485 	} else {
486 		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
487 		if (!status)
488 			status = ocfs2_validate_inode_block(osb->sb, bh);
489 	}
490 	if (status < 0) {
491 		mlog_errno(status);
492 		goto bail;
493 	}
494 
495 	status = -EINVAL;
496 	fe = (struct ocfs2_dinode *) bh->b_data;
497 
498 	/*
499 	 * This is a code bug. Right now the caller needs to
500 	 * understand whether it is asking for a system file inode or
501 	 * not so the proper lock names can be built.
502 	 */
503 	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
504 			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
505 			"Inode %llu: system file state is ambigous\n",
506 			(unsigned long long)args->fi_blkno);
507 
508 	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
509 	    S_ISBLK(le16_to_cpu(fe->i_mode)))
510 		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
511 
512 	ocfs2_populate_inode(inode, fe, 0);
513 
514 	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
515 
516 	status = 0;
517 
518 bail:
519 	if (can_lock)
520 		ocfs2_inode_unlock(inode, 0);
521 
522 	if (status < 0)
523 		make_bad_inode(inode);
524 
525 	if (args && bh)
526 		brelse(bh);
527 
528 	mlog_exit(status);
529 	return status;
530 }
531 
532 void ocfs2_sync_blockdev(struct super_block *sb)
533 {
534 	sync_blockdev(sb->s_bdev);
535 }
536 
537 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
538 				     struct inode *inode,
539 				     struct buffer_head *fe_bh)
540 {
541 	int status = 0;
542 	struct ocfs2_truncate_context *tc = NULL;
543 	struct ocfs2_dinode *fe;
544 	handle_t *handle = NULL;
545 
546 	mlog_entry_void();
547 
548 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
549 
550 	/*
551 	 * This check will also skip truncate of inodes with inline
552 	 * data and fast symlinks.
553 	 */
554 	if (fe->i_clusters) {
555 		if (ocfs2_should_order_data(inode))
556 			ocfs2_begin_ordered_truncate(inode, 0);
557 
558 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
559 		if (IS_ERR(handle)) {
560 			status = PTR_ERR(handle);
561 			handle = NULL;
562 			mlog_errno(status);
563 			goto out;
564 		}
565 
566 		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
567 						 fe_bh,
568 						 OCFS2_JOURNAL_ACCESS_WRITE);
569 		if (status < 0) {
570 			mlog_errno(status);
571 			goto out;
572 		}
573 
574 		i_size_write(inode, 0);
575 
576 		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
577 		if (status < 0) {
578 			mlog_errno(status);
579 			goto out;
580 		}
581 
582 		ocfs2_commit_trans(osb, handle);
583 		handle = NULL;
584 
585 		status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
586 		if (status < 0) {
587 			mlog_errno(status);
588 			goto out;
589 		}
590 
591 		status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
592 		if (status < 0) {
593 			mlog_errno(status);
594 			goto out;
595 		}
596 	}
597 
598 out:
599 	if (handle)
600 		ocfs2_commit_trans(osb, handle);
601 	mlog_exit(status);
602 	return status;
603 }
604 
605 static int ocfs2_remove_inode(struct inode *inode,
606 			      struct buffer_head *di_bh,
607 			      struct inode *orphan_dir_inode,
608 			      struct buffer_head *orphan_dir_bh)
609 {
610 	int status;
611 	struct inode *inode_alloc_inode = NULL;
612 	struct buffer_head *inode_alloc_bh = NULL;
613 	handle_t *handle;
614 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
615 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
616 
617 	inode_alloc_inode =
618 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
619 					    le16_to_cpu(di->i_suballoc_slot));
620 	if (!inode_alloc_inode) {
621 		status = -EEXIST;
622 		mlog_errno(status);
623 		goto bail;
624 	}
625 
626 	mutex_lock(&inode_alloc_inode->i_mutex);
627 	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
628 	if (status < 0) {
629 		mutex_unlock(&inode_alloc_inode->i_mutex);
630 
631 		mlog_errno(status);
632 		goto bail;
633 	}
634 
635 	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
636 				   ocfs2_quota_trans_credits(inode->i_sb));
637 	if (IS_ERR(handle)) {
638 		status = PTR_ERR(handle);
639 		mlog_errno(status);
640 		goto bail_unlock;
641 	}
642 
643 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
644 		status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
645 					  orphan_dir_bh);
646 		if (status < 0) {
647 			mlog_errno(status);
648 			goto bail_commit;
649 		}
650 	}
651 
652 	/* set the inodes dtime */
653 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
654 					 OCFS2_JOURNAL_ACCESS_WRITE);
655 	if (status < 0) {
656 		mlog_errno(status);
657 		goto bail_commit;
658 	}
659 
660 	di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
661 	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
662 
663 	status = ocfs2_journal_dirty(handle, di_bh);
664 	if (status < 0) {
665 		mlog_errno(status);
666 		goto bail_commit;
667 	}
668 
669 	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
670 	dquot_free_inode(inode);
671 
672 	status = ocfs2_free_dinode(handle, inode_alloc_inode,
673 				   inode_alloc_bh, di);
674 	if (status < 0)
675 		mlog_errno(status);
676 
677 bail_commit:
678 	ocfs2_commit_trans(osb, handle);
679 bail_unlock:
680 	ocfs2_inode_unlock(inode_alloc_inode, 1);
681 	mutex_unlock(&inode_alloc_inode->i_mutex);
682 	brelse(inode_alloc_bh);
683 bail:
684 	iput(inode_alloc_inode);
685 
686 	return status;
687 }
688 
689 /*
690  * Serialize with orphan dir recovery. If the process doing
691  * recovery on this orphan dir does an iget() with the dir
692  * i_mutex held, we'll deadlock here. Instead we detect this
693  * and exit early - recovery will wipe this inode for us.
694  */
695 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
696 					     int slot)
697 {
698 	int ret = 0;
699 
700 	spin_lock(&osb->osb_lock);
701 	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
702 		mlog(0, "Recovery is happening on orphan dir %d, will skip "
703 		     "this inode\n", slot);
704 		ret = -EDEADLK;
705 		goto out;
706 	}
707 	/* This signals to the orphan recovery process that it should
708 	 * wait for us to handle the wipe. */
709 	osb->osb_orphan_wipes[slot]++;
710 out:
711 	spin_unlock(&osb->osb_lock);
712 	return ret;
713 }
714 
715 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
716 					 int slot)
717 {
718 	spin_lock(&osb->osb_lock);
719 	osb->osb_orphan_wipes[slot]--;
720 	spin_unlock(&osb->osb_lock);
721 
722 	wake_up(&osb->osb_wipe_event);
723 }
724 
725 static int ocfs2_wipe_inode(struct inode *inode,
726 			    struct buffer_head *di_bh)
727 {
728 	int status, orphaned_slot = -1;
729 	struct inode *orphan_dir_inode = NULL;
730 	struct buffer_head *orphan_dir_bh = NULL;
731 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
732 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
733 
734 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
735 		orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
736 
737 		status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
738 		if (status)
739 			return status;
740 
741 		orphan_dir_inode = ocfs2_get_system_file_inode(osb,
742 							       ORPHAN_DIR_SYSTEM_INODE,
743 							       orphaned_slot);
744 		if (!orphan_dir_inode) {
745 			status = -EEXIST;
746 			mlog_errno(status);
747 			goto bail;
748 		}
749 
750 		/* Lock the orphan dir. The lock will be held for the entire
751 		 * delete_inode operation. We do this now to avoid races with
752 		 * recovery completion on other nodes. */
753 		mutex_lock(&orphan_dir_inode->i_mutex);
754 		status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
755 		if (status < 0) {
756 			mutex_unlock(&orphan_dir_inode->i_mutex);
757 
758 			mlog_errno(status);
759 			goto bail;
760 		}
761 	}
762 
763 	/* we do this while holding the orphan dir lock because we
764 	 * don't want recovery being run from another node to try an
765 	 * inode delete underneath us -- this will result in two nodes
766 	 * truncating the same file! */
767 	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
768 	if (status < 0) {
769 		mlog_errno(status);
770 		goto bail_unlock_dir;
771 	}
772 
773 	/* Remove any dir index tree */
774 	if (S_ISDIR(inode->i_mode)) {
775 		status = ocfs2_dx_dir_truncate(inode, di_bh);
776 		if (status) {
777 			mlog_errno(status);
778 			goto bail_unlock_dir;
779 		}
780 	}
781 
782 	/*Free extended attribute resources associated with this inode.*/
783 	status = ocfs2_xattr_remove(inode, di_bh);
784 	if (status < 0) {
785 		mlog_errno(status);
786 		goto bail_unlock_dir;
787 	}
788 
789 	status = ocfs2_remove_refcount_tree(inode, di_bh);
790 	if (status < 0) {
791 		mlog_errno(status);
792 		goto bail_unlock_dir;
793 	}
794 
795 	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
796 				    orphan_dir_bh);
797 	if (status < 0)
798 		mlog_errno(status);
799 
800 bail_unlock_dir:
801 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
802 		return status;
803 
804 	ocfs2_inode_unlock(orphan_dir_inode, 1);
805 	mutex_unlock(&orphan_dir_inode->i_mutex);
806 	brelse(orphan_dir_bh);
807 bail:
808 	iput(orphan_dir_inode);
809 	ocfs2_signal_wipe_completion(osb, orphaned_slot);
810 
811 	return status;
812 }
813 
814 /* There is a series of simple checks that should be done before a
815  * trylock is even considered. Encapsulate those in this function. */
816 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
817 {
818 	int ret = 0;
819 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
820 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
821 
822 	/* We shouldn't be getting here for the root directory
823 	 * inode.. */
824 	if (inode == osb->root_inode) {
825 		mlog(ML_ERROR, "Skipping delete of root inode.\n");
826 		goto bail;
827 	}
828 
829 	/* If we're coming from downconvert_thread we can't go into our own
830 	 * voting [hello, deadlock city!], so unforuntately we just
831 	 * have to skip deleting this guy. That's OK though because
832 	 * the node who's doing the actual deleting should handle it
833 	 * anyway. */
834 	if (current == osb->dc_task) {
835 		mlog(0, "Skipping delete of %lu because we're currently "
836 		     "in downconvert\n", inode->i_ino);
837 		goto bail;
838 	}
839 
840 	spin_lock(&oi->ip_lock);
841 	/* OCFS2 *never* deletes system files. This should technically
842 	 * never get here as system file inodes should always have a
843 	 * positive link count. */
844 	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
845 		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
846 		     (unsigned long long)oi->ip_blkno);
847 		goto bail_unlock;
848 	}
849 
850 	/* If we have allowd wipe of this inode for another node, it
851 	 * will be marked here so we can safely skip it. Recovery will
852 	 * cleanup any inodes we might inadvertantly skip here. */
853 	if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
854 		mlog(0, "Skipping delete of %lu because another node "
855 		     "has done this for us.\n", inode->i_ino);
856 		goto bail_unlock;
857 	}
858 
859 	ret = 1;
860 bail_unlock:
861 	spin_unlock(&oi->ip_lock);
862 bail:
863 	return ret;
864 }
865 
866 /* Query the cluster to determine whether we should wipe an inode from
867  * disk or not.
868  *
869  * Requires the inode to have the cluster lock. */
870 static int ocfs2_query_inode_wipe(struct inode *inode,
871 				  struct buffer_head *di_bh,
872 				  int *wipe)
873 {
874 	int status = 0;
875 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
876 	struct ocfs2_dinode *di;
877 
878 	*wipe = 0;
879 
880 	/* While we were waiting for the cluster lock in
881 	 * ocfs2_delete_inode, another node might have asked to delete
882 	 * the inode. Recheck our flags to catch this. */
883 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
884 		mlog(0, "Skipping delete of %llu because flags changed\n",
885 		     (unsigned long long)oi->ip_blkno);
886 		goto bail;
887 	}
888 
889 	/* Now that we have an up to date inode, we can double check
890 	 * the link count. */
891 	if (inode->i_nlink) {
892 		mlog(0, "Skipping delete of %llu because nlink = %u\n",
893 		     (unsigned long long)oi->ip_blkno, inode->i_nlink);
894 		goto bail;
895 	}
896 
897 	/* Do some basic inode verification... */
898 	di = (struct ocfs2_dinode *) di_bh->b_data;
899 	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
900 	    !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
901 		/*
902 		 * Inodes in the orphan dir must have ORPHANED_FL.  The only
903 		 * inodes that come back out of the orphan dir are reflink
904 		 * targets. A reflink target may be moved out of the orphan
905 		 * dir between the time we scan the directory and the time we
906 		 * process it. This would lead to HAS_REFCOUNT_FL being set but
907 		 * ORPHANED_FL not.
908 		 */
909 		if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
910 			mlog(0, "Reflinked inode %llu is no longer orphaned.  "
911 			     "it shouldn't be deleted\n",
912 			     (unsigned long long)oi->ip_blkno);
913 			goto bail;
914 		}
915 
916 		/* for lack of a better error? */
917 		status = -EEXIST;
918 		mlog(ML_ERROR,
919 		     "Inode %llu (on-disk %llu) not orphaned! "
920 		     "Disk flags  0x%x, inode flags 0x%x\n",
921 		     (unsigned long long)oi->ip_blkno,
922 		     (unsigned long long)le64_to_cpu(di->i_blkno),
923 		     le32_to_cpu(di->i_flags), oi->ip_flags);
924 		goto bail;
925 	}
926 
927 	/* has someone already deleted us?! baaad... */
928 	if (di->i_dtime) {
929 		status = -EEXIST;
930 		mlog_errno(status);
931 		goto bail;
932 	}
933 
934 	/*
935 	 * This is how ocfs2 determines whether an inode is still live
936 	 * within the cluster. Every node takes a shared read lock on
937 	 * the inode open lock in ocfs2_read_locked_inode(). When we
938 	 * get to ->delete_inode(), each node tries to convert it's
939 	 * lock to an exclusive. Trylocks are serialized by the inode
940 	 * meta data lock. If the upconvert suceeds, we know the inode
941 	 * is no longer live and can be deleted.
942 	 *
943 	 * Though we call this with the meta data lock held, the
944 	 * trylock keeps us from ABBA deadlock.
945 	 */
946 	status = ocfs2_try_open_lock(inode, 1);
947 	if (status == -EAGAIN) {
948 		status = 0;
949 		mlog(0, "Skipping delete of %llu because it is in use on "
950 		     "other nodes\n", (unsigned long long)oi->ip_blkno);
951 		goto bail;
952 	}
953 	if (status < 0) {
954 		mlog_errno(status);
955 		goto bail;
956 	}
957 
958 	*wipe = 1;
959 	mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
960 	     (unsigned long long)oi->ip_blkno,
961 	     le16_to_cpu(di->i_orphaned_slot));
962 
963 bail:
964 	return status;
965 }
966 
967 /* Support function for ocfs2_delete_inode. Will help us keep the
968  * inode data in a consistent state for clear_inode. Always truncates
969  * pages, optionally sync's them first. */
970 static void ocfs2_cleanup_delete_inode(struct inode *inode,
971 				       int sync_data)
972 {
973 	mlog(0, "Cleanup inode %llu, sync = %d\n",
974 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
975 	if (sync_data)
976 		write_inode_now(inode, 1);
977 	truncate_inode_pages(&inode->i_data, 0);
978 }
979 
980 void ocfs2_delete_inode(struct inode *inode)
981 {
982 	int wipe, status;
983 	sigset_t blocked, oldset;
984 	struct buffer_head *di_bh = NULL;
985 
986 	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
987 
988 	/* When we fail in read_inode() we mark inode as bad. The second test
989 	 * catches the case when inode allocation fails before allocating
990 	 * a block for inode. */
991 	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
992 		mlog(0, "Skipping delete of bad inode\n");
993 		goto bail;
994 	}
995 
996 	dquot_initialize(inode);
997 
998 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
999 		/* It's probably not necessary to truncate_inode_pages
1000 		 * here but we do it for safety anyway (it will most
1001 		 * likely be a no-op anyway) */
1002 		ocfs2_cleanup_delete_inode(inode, 0);
1003 		goto bail;
1004 	}
1005 
1006 	/* We want to block signals in delete_inode as the lock and
1007 	 * messaging paths may return us -ERESTARTSYS. Which would
1008 	 * cause us to exit early, resulting in inodes being orphaned
1009 	 * forever. */
1010 	sigfillset(&blocked);
1011 	status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
1012 	if (status < 0) {
1013 		mlog_errno(status);
1014 		ocfs2_cleanup_delete_inode(inode, 1);
1015 		goto bail;
1016 	}
1017 
1018 	/*
1019 	 * Synchronize us against ocfs2_get_dentry. We take this in
1020 	 * shared mode so that all nodes can still concurrently
1021 	 * process deletes.
1022 	 */
1023 	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1024 	if (status < 0) {
1025 		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1026 		ocfs2_cleanup_delete_inode(inode, 0);
1027 		goto bail_unblock;
1028 	}
1029 	/* Lock down the inode. This gives us an up to date view of
1030 	 * it's metadata (for verification), and allows us to
1031 	 * serialize delete_inode on multiple nodes.
1032 	 *
1033 	 * Even though we might be doing a truncate, we don't take the
1034 	 * allocation lock here as it won't be needed - nobody will
1035 	 * have the file open.
1036 	 */
1037 	status = ocfs2_inode_lock(inode, &di_bh, 1);
1038 	if (status < 0) {
1039 		if (status != -ENOENT)
1040 			mlog_errno(status);
1041 		ocfs2_cleanup_delete_inode(inode, 0);
1042 		goto bail_unlock_nfs_sync;
1043 	}
1044 
1045 	/* Query the cluster. This will be the final decision made
1046 	 * before we go ahead and wipe the inode. */
1047 	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1048 	if (!wipe || status < 0) {
1049 		/* Error and remote inode busy both mean we won't be
1050 		 * removing the inode, so they take almost the same
1051 		 * path. */
1052 		if (status < 0)
1053 			mlog_errno(status);
1054 
1055 		/* Someone in the cluster has disallowed a wipe of
1056 		 * this inode, or it was never completely
1057 		 * orphaned. Write out the pages and exit now. */
1058 		ocfs2_cleanup_delete_inode(inode, 1);
1059 		goto bail_unlock_inode;
1060 	}
1061 
1062 	ocfs2_cleanup_delete_inode(inode, 0);
1063 
1064 	status = ocfs2_wipe_inode(inode, di_bh);
1065 	if (status < 0) {
1066 		if (status != -EDEADLK)
1067 			mlog_errno(status);
1068 		goto bail_unlock_inode;
1069 	}
1070 
1071 	/*
1072 	 * Mark the inode as successfully deleted.
1073 	 *
1074 	 * This is important for ocfs2_clear_inode() as it will check
1075 	 * this flag and skip any checkpointing work
1076 	 *
1077 	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1078 	 * the LVB for other nodes.
1079 	 */
1080 	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1081 
1082 bail_unlock_inode:
1083 	ocfs2_inode_unlock(inode, 1);
1084 	brelse(di_bh);
1085 
1086 bail_unlock_nfs_sync:
1087 	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1088 
1089 bail_unblock:
1090 	status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1091 	if (status < 0)
1092 		mlog_errno(status);
1093 bail:
1094 	clear_inode(inode);
1095 	mlog_exit_void();
1096 }
1097 
1098 void ocfs2_clear_inode(struct inode *inode)
1099 {
1100 	int status;
1101 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1102 
1103 	mlog_entry_void();
1104 
1105 	if (!inode)
1106 		goto bail;
1107 
1108 	mlog(0, "Clearing inode: %llu, nlink = %u\n",
1109 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1110 
1111 	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1112 			"Inode=%lu\n", inode->i_ino);
1113 
1114 	dquot_drop(inode);
1115 
1116 	/* To preven remote deletes we hold open lock before, now it
1117 	 * is time to unlock PR and EX open locks. */
1118 	ocfs2_open_unlock(inode);
1119 
1120 	/* Do these before all the other work so that we don't bounce
1121 	 * the downconvert thread while waiting to destroy the locks. */
1122 	ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1123 	ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1124 	ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1125 
1126 	/* We very well may get a clear_inode before all an inodes
1127 	 * metadata has hit disk. Of course, we can't drop any cluster
1128 	 * locks until the journal has finished with it. The only
1129 	 * exception here are successfully wiped inodes - their
1130 	 * metadata can now be considered to be part of the system
1131 	 * inodes from which it came. */
1132 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1133 		ocfs2_checkpoint_inode(inode);
1134 
1135 	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1136 			"Clear inode of %llu, inode has io markers\n",
1137 			(unsigned long long)oi->ip_blkno);
1138 
1139 	ocfs2_extent_map_trunc(inode, 0);
1140 
1141 	status = ocfs2_drop_inode_locks(inode);
1142 	if (status < 0)
1143 		mlog_errno(status);
1144 
1145 	ocfs2_lock_res_free(&oi->ip_rw_lockres);
1146 	ocfs2_lock_res_free(&oi->ip_inode_lockres);
1147 	ocfs2_lock_res_free(&oi->ip_open_lockres);
1148 
1149 	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1150 
1151 	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1152 			"Clear inode of %llu, inode has %u cache items\n",
1153 			(unsigned long long)oi->ip_blkno,
1154 			INODE_CACHE(inode)->ci_num_cached);
1155 
1156 	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1157 			"Clear inode of %llu, inode has a bad flag\n",
1158 			(unsigned long long)oi->ip_blkno);
1159 
1160 	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1161 			"Clear inode of %llu, inode is locked\n",
1162 			(unsigned long long)oi->ip_blkno);
1163 
1164 	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1165 			"Clear inode of %llu, io_mutex is locked\n",
1166 			(unsigned long long)oi->ip_blkno);
1167 	mutex_unlock(&oi->ip_io_mutex);
1168 
1169 	/*
1170 	 * down_trylock() returns 0, down_write_trylock() returns 1
1171 	 * kernel 1, world 0
1172 	 */
1173 	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1174 			"Clear inode of %llu, alloc_sem is locked\n",
1175 			(unsigned long long)oi->ip_blkno);
1176 	up_write(&oi->ip_alloc_sem);
1177 
1178 	mlog_bug_on_msg(oi->ip_open_count,
1179 			"Clear inode of %llu has open count %d\n",
1180 			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
1181 
1182 	/* Clear all other flags. */
1183 	oi->ip_flags = 0;
1184 	oi->ip_dir_start_lookup = 0;
1185 	oi->ip_blkno = 0ULL;
1186 
1187 	/*
1188 	 * ip_jinode is used to track txns against this inode. We ensure that
1189 	 * the journal is flushed before journal shutdown. Thus it is safe to
1190 	 * have inodes get cleaned up after journal shutdown.
1191 	 */
1192 	jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1193 				       &oi->ip_jinode);
1194 
1195 bail:
1196 	mlog_exit_void();
1197 }
1198 
1199 /* Called under inode_lock, with no more references on the
1200  * struct inode, so it's safe here to check the flags field
1201  * and to manipulate i_nlink without any other locks. */
1202 void ocfs2_drop_inode(struct inode *inode)
1203 {
1204 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1205 
1206 	mlog_entry_void();
1207 
1208 	mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1209 	     (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1210 
1211 	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1212 		generic_delete_inode(inode);
1213 	else
1214 		generic_drop_inode(inode);
1215 
1216 	mlog_exit_void();
1217 }
1218 
1219 /*
1220  * This is called from our getattr.
1221  */
1222 int ocfs2_inode_revalidate(struct dentry *dentry)
1223 {
1224 	struct inode *inode = dentry->d_inode;
1225 	int status = 0;
1226 
1227 	mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1228 		   inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1229 
1230 	if (!inode) {
1231 		mlog(0, "eep, no inode!\n");
1232 		status = -ENOENT;
1233 		goto bail;
1234 	}
1235 
1236 	spin_lock(&OCFS2_I(inode)->ip_lock);
1237 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1238 		spin_unlock(&OCFS2_I(inode)->ip_lock);
1239 		mlog(0, "inode deleted!\n");
1240 		status = -ENOENT;
1241 		goto bail;
1242 	}
1243 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1244 
1245 	/* Let ocfs2_inode_lock do the work of updating our struct
1246 	 * inode for us. */
1247 	status = ocfs2_inode_lock(inode, NULL, 0);
1248 	if (status < 0) {
1249 		if (status != -ENOENT)
1250 			mlog_errno(status);
1251 		goto bail;
1252 	}
1253 	ocfs2_inode_unlock(inode, 0);
1254 bail:
1255 	mlog_exit(status);
1256 
1257 	return status;
1258 }
1259 
1260 /*
1261  * Updates a disk inode from a
1262  * struct inode.
1263  * Only takes ip_lock.
1264  */
1265 int ocfs2_mark_inode_dirty(handle_t *handle,
1266 			   struct inode *inode,
1267 			   struct buffer_head *bh)
1268 {
1269 	int status;
1270 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1271 
1272 	mlog_entry("(inode %llu)\n",
1273 		   (unsigned long long)OCFS2_I(inode)->ip_blkno);
1274 
1275 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1276 					 OCFS2_JOURNAL_ACCESS_WRITE);
1277 	if (status < 0) {
1278 		mlog_errno(status);
1279 		goto leave;
1280 	}
1281 
1282 	spin_lock(&OCFS2_I(inode)->ip_lock);
1283 	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1284 	ocfs2_get_inode_flags(OCFS2_I(inode));
1285 	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1286 	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1287 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1288 
1289 	fe->i_size = cpu_to_le64(i_size_read(inode));
1290 	ocfs2_set_links_count(fe, inode->i_nlink);
1291 	fe->i_uid = cpu_to_le32(inode->i_uid);
1292 	fe->i_gid = cpu_to_le32(inode->i_gid);
1293 	fe->i_mode = cpu_to_le16(inode->i_mode);
1294 	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1295 	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1296 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1297 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1298 	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1299 	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1300 
1301 	status = ocfs2_journal_dirty(handle, bh);
1302 	if (status < 0)
1303 		mlog_errno(status);
1304 
1305 	status = 0;
1306 leave:
1307 
1308 	mlog_exit(status);
1309 	return status;
1310 }
1311 
1312 /*
1313  *
1314  * Updates a struct inode from a disk inode.
1315  * does no i/o, only takes ip_lock.
1316  */
1317 void ocfs2_refresh_inode(struct inode *inode,
1318 			 struct ocfs2_dinode *fe)
1319 {
1320 	spin_lock(&OCFS2_I(inode)->ip_lock);
1321 
1322 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1323 	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1324 	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1325 	ocfs2_set_inode_flags(inode);
1326 	i_size_write(inode, le64_to_cpu(fe->i_size));
1327 	inode->i_nlink = ocfs2_read_links_count(fe);
1328 	inode->i_uid = le32_to_cpu(fe->i_uid);
1329 	inode->i_gid = le32_to_cpu(fe->i_gid);
1330 	inode->i_mode = le16_to_cpu(fe->i_mode);
1331 	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1332 		inode->i_blocks = 0;
1333 	else
1334 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1335 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1336 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1337 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1338 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1339 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1340 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1341 
1342 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1343 }
1344 
1345 int ocfs2_validate_inode_block(struct super_block *sb,
1346 			       struct buffer_head *bh)
1347 {
1348 	int rc;
1349 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1350 
1351 	mlog(0, "Validating dinode %llu\n",
1352 	     (unsigned long long)bh->b_blocknr);
1353 
1354 	BUG_ON(!buffer_uptodate(bh));
1355 
1356 	/*
1357 	 * If the ecc fails, we return the error but otherwise
1358 	 * leave the filesystem running.  We know any error is
1359 	 * local to this block.
1360 	 */
1361 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1362 	if (rc) {
1363 		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1364 		     (unsigned long long)bh->b_blocknr);
1365 		goto bail;
1366 	}
1367 
1368 	/*
1369 	 * Errors after here are fatal.
1370 	 */
1371 
1372 	rc = -EINVAL;
1373 
1374 	if (!OCFS2_IS_VALID_DINODE(di)) {
1375 		ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1376 			    (unsigned long long)bh->b_blocknr, 7,
1377 			    di->i_signature);
1378 		goto bail;
1379 	}
1380 
1381 	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1382 		ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1383 			    (unsigned long long)bh->b_blocknr,
1384 			    (unsigned long long)le64_to_cpu(di->i_blkno));
1385 		goto bail;
1386 	}
1387 
1388 	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1389 		ocfs2_error(sb,
1390 			    "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1391 			    (unsigned long long)bh->b_blocknr);
1392 		goto bail;
1393 	}
1394 
1395 	if (le32_to_cpu(di->i_fs_generation) !=
1396 	    OCFS2_SB(sb)->fs_generation) {
1397 		ocfs2_error(sb,
1398 			    "Invalid dinode #%llu: fs_generation is %u\n",
1399 			    (unsigned long long)bh->b_blocknr,
1400 			    le32_to_cpu(di->i_fs_generation));
1401 		goto bail;
1402 	}
1403 
1404 	rc = 0;
1405 
1406 bail:
1407 	return rc;
1408 }
1409 
1410 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1411 				int flags)
1412 {
1413 	int rc;
1414 	struct buffer_head *tmp = *bh;
1415 
1416 	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1417 			       1, &tmp, flags, ocfs2_validate_inode_block);
1418 
1419 	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1420 	if (!rc && !*bh)
1421 		*bh = tmp;
1422 
1423 	return rc;
1424 }
1425 
1426 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1427 {
1428 	return ocfs2_read_inode_block_full(inode, bh, 0);
1429 }
1430 
1431 
1432 static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1433 {
1434 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1435 
1436 	return oi->ip_blkno;
1437 }
1438 
1439 static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1440 {
1441 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1442 
1443 	return oi->vfs_inode.i_sb;
1444 }
1445 
1446 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1447 {
1448 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1449 
1450 	spin_lock(&oi->ip_lock);
1451 }
1452 
1453 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1454 {
1455 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1456 
1457 	spin_unlock(&oi->ip_lock);
1458 }
1459 
1460 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1461 {
1462 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1463 
1464 	mutex_lock(&oi->ip_io_mutex);
1465 }
1466 
1467 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1468 {
1469 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1470 
1471 	mutex_unlock(&oi->ip_io_mutex);
1472 }
1473 
1474 const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1475 	.co_owner		= ocfs2_inode_cache_owner,
1476 	.co_get_super		= ocfs2_inode_cache_get_super,
1477 	.co_cache_lock		= ocfs2_inode_cache_lock,
1478 	.co_cache_unlock	= ocfs2_inode_cache_unlock,
1479 	.co_io_lock		= ocfs2_inode_cache_io_lock,
1480 	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
1481 };
1482 
1483