xref: /openbmc/linux/fs/ocfs2/inode.c (revision 8b2c0dba)
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/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 
33 #include <asm/byteorder.h>
34 
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
37 
38 #include "ocfs2.h"
39 
40 #include "alloc.h"
41 #include "dir.h"
42 #include "blockcheck.h"
43 #include "dlmglue.h"
44 #include "extent_map.h"
45 #include "file.h"
46 #include "heartbeat.h"
47 #include "inode.h"
48 #include "journal.h"
49 #include "namei.h"
50 #include "suballoc.h"
51 #include "super.h"
52 #include "symlink.h"
53 #include "sysfile.h"
54 #include "uptodate.h"
55 #include "xattr.h"
56 #include "refcounttree.h"
57 
58 #include "buffer_head_io.h"
59 
60 struct ocfs2_find_inode_args
61 {
62 	u64		fi_blkno;
63 	unsigned long	fi_ino;
64 	unsigned int	fi_flags;
65 	unsigned int	fi_sysfile_type;
66 };
67 
68 static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
69 
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,
75 				    struct inode *inode,
76 				    struct buffer_head *fe_bh);
77 
78 void ocfs2_set_inode_flags(struct inode *inode)
79 {
80 	unsigned int flags = OCFS2_I(inode)->ip_attr;
81 
82 	inode->i_flags &= ~(S_IMMUTABLE |
83 		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
84 
85 	if (flags & OCFS2_IMMUTABLE_FL)
86 		inode->i_flags |= S_IMMUTABLE;
87 
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;
96 }
97 
98 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
99 void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
100 {
101 	unsigned int flags = oi->vfs_inode.i_flags;
102 
103 	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
104 			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
105 	if (flags & S_SYNC)
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;
115 }
116 
117 struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
118 {
119 	struct ocfs2_find_inode_args args;
120 
121 	args.fi_blkno = blkno;
122 	args.fi_flags = 0;
123 	args.fi_ino = ino_from_blkno(sb, blkno);
124 	args.fi_sysfile_type = 0;
125 
126 	return ilookup5(sb, blkno, ocfs2_find_actor, &args);
127 }
128 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
129 			 int sysfile_type)
130 {
131 	struct inode *inode = NULL;
132 	struct super_block *sb = osb->sb;
133 	struct ocfs2_find_inode_args args;
134 
135 	mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
136 
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. */
140 	if (blkno == 0) {
141 		inode = ERR_PTR(-EINVAL);
142 		mlog_errno(PTR_ERR(inode));
143 		goto bail;
144 	}
145 
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;
150 
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
155 	 * afterwards. */
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);
160 	}
161 	if (inode == NULL) {
162 		inode = ERR_PTR(-ENOMEM);
163 		mlog_errno(PTR_ERR(inode));
164 		goto bail;
165 	}
166 	if (is_bad_inode(inode)) {
167 		iput(inode);
168 		inode = ERR_PTR(-ESTALE);
169 		goto bail;
170 	}
171 
172 bail:
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);
177 	}
178 
179 	return inode;
180 }
181 
182 
183 /*
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
188  */
189 
190 static int ocfs2_find_actor(struct inode *inode, void *opaque)
191 {
192 	struct ocfs2_find_inode_args *args = NULL;
193 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
194 	int ret = 0;
195 
196 	mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
197 
198 	args = opaque;
199 
200 	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
201 
202 	if (oi->ip_blkno != args->fi_blkno)
203 		goto bail;
204 
205 	ret = 1;
206 bail:
207 	mlog_exit(ret);
208 	return ret;
209 }
210 
211 /*
212  * initialize the new inode, but don't do anything that would cause
213  * us to sleep.
214  * return 0 on success, 1 on failure
215  */
216 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
217 {
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;
221 
222 	mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
223 
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);
235 	else
236 		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
237 				  &ocfs2_file_ip_alloc_sem_key);
238 
239 	mlog_exit(0);
240 	return 0;
241 }
242 
243 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
244 			  int create_ino)
245 {
246 	struct super_block *sb;
247 	struct ocfs2_super *osb;
248 	int use_plocks = 1;
249 
250 	mlog_entry("(0x%p, size:%llu)\n", inode,
251 		   (unsigned long long)le64_to_cpu(fe->i_size));
252 
253 	sb = inode->i_sb;
254 	osb = OCFS2_SB(sb);
255 
256 	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
257 	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
258 		use_plocks = 0;
259 
260 	/*
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.
263 	 */
264 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
265 						cannot create a superblock
266 						inode today.  change if
267 						that is needed. */
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);
270 
271 
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);
275 
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);
282 
283 	/* Fast symlinks will have i_size but no allocated clusters. */
284 	if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
285 		inode->i_blocks = 0;
286 	else
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);
295 
296 	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
297 		mlog(ML_ERROR,
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));
301 
302 	inode->i_nlink = ocfs2_read_links_count(fe);
303 
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;
307 	}
308 
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 ;-) */
320 		BUG();
321 	}
322 
323 	switch (inode->i_mode & S_IFMT) {
324 	    case S_IFREG:
325 		    if (use_plocks)
326 			    inode->i_fop = &ocfs2_fops;
327 		    else
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));
331 		    break;
332 	    case S_IFDIR:
333 		    inode->i_op = &ocfs2_dir_iops;
334 		    if (use_plocks)
335 			    inode->i_fop = &ocfs2_dops;
336 		    else
337 			    inode->i_fop = &ocfs2_dops_no_plocks;
338 		    i_size_write(inode, le64_to_cpu(fe->i_size));
339 		    break;
340 	    case S_IFLNK:
341 		    if (ocfs2_inode_is_fast_symlink(inode))
342 			inode->i_op = &ocfs2_fast_symlink_inode_operations;
343 		    else
344 			inode->i_op = &ocfs2_symlink_inode_operations;
345 		    i_size_write(inode, le64_to_cpu(fe->i_size));
346 		    break;
347 	    default:
348 		    inode->i_op = &ocfs2_special_file_iops;
349 		    init_special_inode(inode, inode->i_mode,
350 				       inode->i_rdev);
351 		    break;
352 	}
353 
354 	if (create_ino) {
355 		inode->i_ino = ino_from_blkno(inode->i_sb,
356 			       le64_to_cpu(fe->i_blkno));
357 
358 		/*
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.
362 		 */
363 		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
364 
365 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
366 					  OCFS2_LOCK_TYPE_META, 0, inode);
367 
368 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
369 					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
370 	}
371 
372 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
373 				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
374 				  inode);
375 
376 	ocfs2_set_inode_flags(inode);
377 
378 	OCFS2_I(inode)->ip_last_used_slot = 0;
379 	OCFS2_I(inode)->ip_last_used_group = 0;
380 	mlog_exit_void();
381 }
382 
383 static int ocfs2_read_locked_inode(struct inode *inode,
384 				   struct ocfs2_find_inode_args *args)
385 {
386 	struct super_block *sb;
387 	struct ocfs2_super *osb;
388 	struct ocfs2_dinode *fe;
389 	struct buffer_head *bh = NULL;
390 	int status, can_lock;
391 	u32 generation = 0;
392 
393 	mlog_entry("(0x%p, 0x%p)\n", inode, args);
394 
395 	status = -EINVAL;
396 	if (inode == NULL || inode->i_sb == NULL) {
397 		mlog(ML_ERROR, "bad inode\n");
398 		return status;
399 	}
400 	sb = inode->i_sb;
401 	osb = OCFS2_SB(sb);
402 
403 	if (!args) {
404 		mlog(ML_ERROR, "bad inode args\n");
405 		make_bad_inode(inode);
406 		return status;
407 	}
408 
409 	/*
410 	 * To improve performance of cold-cache inode stats, we take
411 	 * the cluster lock here if possible.
412 	 *
413 	 * Generally, OCFS2 never trusts the contents of an inode
414 	 * unless it's holding a cluster lock, so taking it here isn't
415 	 * a correctness issue as much as it is a performance
416 	 * improvement.
417 	 *
418 	 * There are three times when taking the lock is not a good idea:
419 	 *
420 	 * 1) During startup, before we have initialized the DLM.
421 	 *
422 	 * 2) If we are reading certain system files which never get
423 	 *    cluster locks (local alloc, truncate log).
424 	 *
425 	 * 3) If the process doing the iget() is responsible for
426 	 *    orphan dir recovery. We're holding the orphan dir lock and
427 	 *    can get into a deadlock with another process on another
428 	 *    node in ->delete_inode().
429 	 *
430 	 * #1 and #2 can be simply solved by never taking the lock
431 	 * here for system files (which are the only type we read
432 	 * during mount). It's a heavier approach, but our main
433 	 * concern is user-accesible files anyway.
434 	 *
435 	 * #3 works itself out because we'll eventually take the
436 	 * cluster lock before trusting anything anyway.
437 	 */
438 	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
439 		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
440 		&& !ocfs2_mount_local(osb);
441 
442 	/*
443 	 * To maintain backwards compatibility with older versions of
444 	 * ocfs2-tools, we still store the generation value for system
445 	 * files. The only ones that actually matter to userspace are
446 	 * the journals, but it's easier and inexpensive to just flag
447 	 * all system files similarly.
448 	 */
449 	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
450 		generation = osb->fs_generation;
451 
452 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
453 				  OCFS2_LOCK_TYPE_META,
454 				  generation, inode);
455 
456 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
457 				  OCFS2_LOCK_TYPE_OPEN,
458 				  0, inode);
459 
460 	if (can_lock) {
461 		status = ocfs2_open_lock(inode);
462 		if (status) {
463 			make_bad_inode(inode);
464 			mlog_errno(status);
465 			return status;
466 		}
467 		status = ocfs2_inode_lock(inode, NULL, 0);
468 		if (status) {
469 			make_bad_inode(inode);
470 			mlog_errno(status);
471 			return status;
472 		}
473 	}
474 
475 	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
476 		status = ocfs2_try_open_lock(inode, 0);
477 		if (status) {
478 			make_bad_inode(inode);
479 			return status;
480 		}
481 	}
482 
483 	if (can_lock) {
484 		status = ocfs2_read_inode_block_full(inode, &bh,
485 						     OCFS2_BH_IGNORE_CACHE);
486 	} else {
487 		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
488 		if (!status)
489 			status = ocfs2_validate_inode_block(osb->sb, bh);
490 	}
491 	if (status < 0) {
492 		mlog_errno(status);
493 		goto bail;
494 	}
495 
496 	status = -EINVAL;
497 	fe = (struct ocfs2_dinode *) bh->b_data;
498 
499 	/*
500 	 * This is a code bug. Right now the caller needs to
501 	 * understand whether it is asking for a system file inode or
502 	 * not so the proper lock names can be built.
503 	 */
504 	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
505 			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
506 			"Inode %llu: system file state is ambigous\n",
507 			(unsigned long long)args->fi_blkno);
508 
509 	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
510 	    S_ISBLK(le16_to_cpu(fe->i_mode)))
511 		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
512 
513 	ocfs2_populate_inode(inode, fe, 0);
514 
515 	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
516 
517 	status = 0;
518 
519 bail:
520 	if (can_lock)
521 		ocfs2_inode_unlock(inode, 0);
522 
523 	if (status < 0)
524 		make_bad_inode(inode);
525 
526 	if (args && bh)
527 		brelse(bh);
528 
529 	mlog_exit(status);
530 	return status;
531 }
532 
533 void ocfs2_sync_blockdev(struct super_block *sb)
534 {
535 	sync_blockdev(sb->s_bdev);
536 }
537 
538 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
539 				     struct inode *inode,
540 				     struct buffer_head *fe_bh)
541 {
542 	int status = 0;
543 	struct ocfs2_truncate_context *tc = NULL;
544 	struct ocfs2_dinode *fe;
545 	handle_t *handle = NULL;
546 
547 	mlog_entry_void();
548 
549 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
550 
551 	/*
552 	 * This check will also skip truncate of inodes with inline
553 	 * data and fast symlinks.
554 	 */
555 	if (fe->i_clusters) {
556 		if (ocfs2_should_order_data(inode))
557 			ocfs2_begin_ordered_truncate(inode, 0);
558 
559 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
560 		if (IS_ERR(handle)) {
561 			status = PTR_ERR(handle);
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 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
644 				  orphan_dir_bh);
645 	if (status < 0) {
646 		mlog_errno(status);
647 		goto bail_commit;
648 	}
649 
650 	/* set the inodes dtime */
651 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
652 					 OCFS2_JOURNAL_ACCESS_WRITE);
653 	if (status < 0) {
654 		mlog_errno(status);
655 		goto bail_commit;
656 	}
657 
658 	di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
659 	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
660 
661 	status = ocfs2_journal_dirty(handle, di_bh);
662 	if (status < 0) {
663 		mlog_errno(status);
664 		goto bail_commit;
665 	}
666 
667 	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
668 	vfs_dq_free_inode(inode);
669 
670 	status = ocfs2_free_dinode(handle, inode_alloc_inode,
671 				   inode_alloc_bh, di);
672 	if (status < 0)
673 		mlog_errno(status);
674 
675 bail_commit:
676 	ocfs2_commit_trans(osb, handle);
677 bail_unlock:
678 	ocfs2_inode_unlock(inode_alloc_inode, 1);
679 	mutex_unlock(&inode_alloc_inode->i_mutex);
680 	brelse(inode_alloc_bh);
681 bail:
682 	iput(inode_alloc_inode);
683 
684 	return status;
685 }
686 
687 /*
688  * Serialize with orphan dir recovery. If the process doing
689  * recovery on this orphan dir does an iget() with the dir
690  * i_mutex held, we'll deadlock here. Instead we detect this
691  * and exit early - recovery will wipe this inode for us.
692  */
693 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
694 					     int slot)
695 {
696 	int ret = 0;
697 
698 	spin_lock(&osb->osb_lock);
699 	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
700 		mlog(0, "Recovery is happening on orphan dir %d, will skip "
701 		     "this inode\n", slot);
702 		ret = -EDEADLK;
703 		goto out;
704 	}
705 	/* This signals to the orphan recovery process that it should
706 	 * wait for us to handle the wipe. */
707 	osb->osb_orphan_wipes[slot]++;
708 out:
709 	spin_unlock(&osb->osb_lock);
710 	return ret;
711 }
712 
713 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
714 					 int slot)
715 {
716 	spin_lock(&osb->osb_lock);
717 	osb->osb_orphan_wipes[slot]--;
718 	spin_unlock(&osb->osb_lock);
719 
720 	wake_up(&osb->osb_wipe_event);
721 }
722 
723 static int ocfs2_wipe_inode(struct inode *inode,
724 			    struct buffer_head *di_bh)
725 {
726 	int status, orphaned_slot;
727 	struct inode *orphan_dir_inode = NULL;
728 	struct buffer_head *orphan_dir_bh = NULL;
729 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
730 	struct ocfs2_dinode *di;
731 
732 	di = (struct ocfs2_dinode *) di_bh->b_data;
733 	orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
734 
735 	status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
736 	if (status)
737 		return status;
738 
739 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
740 						       ORPHAN_DIR_SYSTEM_INODE,
741 						       orphaned_slot);
742 	if (!orphan_dir_inode) {
743 		status = -EEXIST;
744 		mlog_errno(status);
745 		goto bail;
746 	}
747 
748 	/* Lock the orphan dir. The lock will be held for the entire
749 	 * delete_inode operation. We do this now to avoid races with
750 	 * recovery completion on other nodes. */
751 	mutex_lock(&orphan_dir_inode->i_mutex);
752 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
753 	if (status < 0) {
754 		mutex_unlock(&orphan_dir_inode->i_mutex);
755 
756 		mlog_errno(status);
757 		goto bail;
758 	}
759 
760 	/* we do this while holding the orphan dir lock because we
761 	 * don't want recovery being run from another node to try an
762 	 * inode delete underneath us -- this will result in two nodes
763 	 * truncating the same file! */
764 	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
765 	if (status < 0) {
766 		mlog_errno(status);
767 		goto bail_unlock_dir;
768 	}
769 
770 	/* Remove any dir index tree */
771 	if (S_ISDIR(inode->i_mode)) {
772 		status = ocfs2_dx_dir_truncate(inode, di_bh);
773 		if (status) {
774 			mlog_errno(status);
775 			goto bail_unlock_dir;
776 		}
777 	}
778 
779 	/*Free extended attribute resources associated with this inode.*/
780 	status = ocfs2_xattr_remove(inode, di_bh);
781 	if (status < 0) {
782 		mlog_errno(status);
783 		goto bail_unlock_dir;
784 	}
785 
786 	status = ocfs2_remove_refcount_tree(inode, di_bh);
787 	if (status < 0) {
788 		mlog_errno(status);
789 		goto bail_unlock_dir;
790 	}
791 
792 	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
793 				    orphan_dir_bh);
794 	if (status < 0)
795 		mlog_errno(status);
796 
797 bail_unlock_dir:
798 	ocfs2_inode_unlock(orphan_dir_inode, 1);
799 	mutex_unlock(&orphan_dir_inode->i_mutex);
800 	brelse(orphan_dir_bh);
801 bail:
802 	iput(orphan_dir_inode);
803 	ocfs2_signal_wipe_completion(osb, orphaned_slot);
804 
805 	return status;
806 }
807 
808 /* There is a series of simple checks that should be done before a
809  * trylock is even considered. Encapsulate those in this function. */
810 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
811 {
812 	int ret = 0;
813 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
814 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
815 
816 	/* We shouldn't be getting here for the root directory
817 	 * inode.. */
818 	if (inode == osb->root_inode) {
819 		mlog(ML_ERROR, "Skipping delete of root inode.\n");
820 		goto bail;
821 	}
822 
823 	/* If we're coming from downconvert_thread we can't go into our own
824 	 * voting [hello, deadlock city!], so unforuntately we just
825 	 * have to skip deleting this guy. That's OK though because
826 	 * the node who's doing the actual deleting should handle it
827 	 * anyway. */
828 	if (current == osb->dc_task) {
829 		mlog(0, "Skipping delete of %lu because we're currently "
830 		     "in downconvert\n", inode->i_ino);
831 		goto bail;
832 	}
833 
834 	spin_lock(&oi->ip_lock);
835 	/* OCFS2 *never* deletes system files. This should technically
836 	 * never get here as system file inodes should always have a
837 	 * positive link count. */
838 	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
839 		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
840 		     (unsigned long long)oi->ip_blkno);
841 		goto bail_unlock;
842 	}
843 
844 	/* If we have allowd wipe of this inode for another node, it
845 	 * will be marked here so we can safely skip it. Recovery will
846 	 * cleanup any inodes we might inadvertantly skip here. */
847 	if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
848 		mlog(0, "Skipping delete of %lu because another node "
849 		     "has done this for us.\n", inode->i_ino);
850 		goto bail_unlock;
851 	}
852 
853 	ret = 1;
854 bail_unlock:
855 	spin_unlock(&oi->ip_lock);
856 bail:
857 	return ret;
858 }
859 
860 /* Query the cluster to determine whether we should wipe an inode from
861  * disk or not.
862  *
863  * Requires the inode to have the cluster lock. */
864 static int ocfs2_query_inode_wipe(struct inode *inode,
865 				  struct buffer_head *di_bh,
866 				  int *wipe)
867 {
868 	int status = 0;
869 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
870 	struct ocfs2_dinode *di;
871 
872 	*wipe = 0;
873 
874 	/* While we were waiting for the cluster lock in
875 	 * ocfs2_delete_inode, another node might have asked to delete
876 	 * the inode. Recheck our flags to catch this. */
877 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
878 		mlog(0, "Skipping delete of %llu because flags changed\n",
879 		     (unsigned long long)oi->ip_blkno);
880 		goto bail;
881 	}
882 
883 	/* Now that we have an up to date inode, we can double check
884 	 * the link count. */
885 	if (inode->i_nlink) {
886 		mlog(0, "Skipping delete of %llu because nlink = %u\n",
887 		     (unsigned long long)oi->ip_blkno, inode->i_nlink);
888 		goto bail;
889 	}
890 
891 	/* Do some basic inode verification... */
892 	di = (struct ocfs2_dinode *) di_bh->b_data;
893 	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
894 		/* for lack of a better error? */
895 		status = -EEXIST;
896 		mlog(ML_ERROR,
897 		     "Inode %llu (on-disk %llu) not orphaned! "
898 		     "Disk flags  0x%x, inode flags 0x%x\n",
899 		     (unsigned long long)oi->ip_blkno,
900 		     (unsigned long long)le64_to_cpu(di->i_blkno),
901 		     le32_to_cpu(di->i_flags), oi->ip_flags);
902 		goto bail;
903 	}
904 
905 	/* has someone already deleted us?! baaad... */
906 	if (di->i_dtime) {
907 		status = -EEXIST;
908 		mlog_errno(status);
909 		goto bail;
910 	}
911 
912 	/*
913 	 * This is how ocfs2 determines whether an inode is still live
914 	 * within the cluster. Every node takes a shared read lock on
915 	 * the inode open lock in ocfs2_read_locked_inode(). When we
916 	 * get to ->delete_inode(), each node tries to convert it's
917 	 * lock to an exclusive. Trylocks are serialized by the inode
918 	 * meta data lock. If the upconvert suceeds, we know the inode
919 	 * is no longer live and can be deleted.
920 	 *
921 	 * Though we call this with the meta data lock held, the
922 	 * trylock keeps us from ABBA deadlock.
923 	 */
924 	status = ocfs2_try_open_lock(inode, 1);
925 	if (status == -EAGAIN) {
926 		status = 0;
927 		mlog(0, "Skipping delete of %llu because it is in use on "
928 		     "other nodes\n", (unsigned long long)oi->ip_blkno);
929 		goto bail;
930 	}
931 	if (status < 0) {
932 		mlog_errno(status);
933 		goto bail;
934 	}
935 
936 	*wipe = 1;
937 	mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
938 	     (unsigned long long)oi->ip_blkno,
939 	     le16_to_cpu(di->i_orphaned_slot));
940 
941 bail:
942 	return status;
943 }
944 
945 /* Support function for ocfs2_delete_inode. Will help us keep the
946  * inode data in a consistent state for clear_inode. Always truncates
947  * pages, optionally sync's them first. */
948 static void ocfs2_cleanup_delete_inode(struct inode *inode,
949 				       int sync_data)
950 {
951 	mlog(0, "Cleanup inode %llu, sync = %d\n",
952 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
953 	if (sync_data)
954 		write_inode_now(inode, 1);
955 	truncate_inode_pages(&inode->i_data, 0);
956 }
957 
958 void ocfs2_delete_inode(struct inode *inode)
959 {
960 	int wipe, status;
961 	sigset_t blocked, oldset;
962 	struct buffer_head *di_bh = NULL;
963 
964 	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
965 
966 	/* When we fail in read_inode() we mark inode as bad. The second test
967 	 * catches the case when inode allocation fails before allocating
968 	 * a block for inode. */
969 	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
970 		mlog(0, "Skipping delete of bad inode\n");
971 		goto bail;
972 	}
973 
974 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
975 		/* It's probably not necessary to truncate_inode_pages
976 		 * here but we do it for safety anyway (it will most
977 		 * likely be a no-op anyway) */
978 		ocfs2_cleanup_delete_inode(inode, 0);
979 		goto bail;
980 	}
981 
982 	/* We want to block signals in delete_inode as the lock and
983 	 * messaging paths may return us -ERESTARTSYS. Which would
984 	 * cause us to exit early, resulting in inodes being orphaned
985 	 * forever. */
986 	sigfillset(&blocked);
987 	status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
988 	if (status < 0) {
989 		mlog_errno(status);
990 		ocfs2_cleanup_delete_inode(inode, 1);
991 		goto bail;
992 	}
993 
994 	/*
995 	 * Synchronize us against ocfs2_get_dentry. We take this in
996 	 * shared mode so that all nodes can still concurrently
997 	 * process deletes.
998 	 */
999 	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1000 	if (status < 0) {
1001 		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1002 		ocfs2_cleanup_delete_inode(inode, 0);
1003 		goto bail_unblock;
1004 	}
1005 	/* Lock down the inode. This gives us an up to date view of
1006 	 * it's metadata (for verification), and allows us to
1007 	 * serialize delete_inode on multiple nodes.
1008 	 *
1009 	 * Even though we might be doing a truncate, we don't take the
1010 	 * allocation lock here as it won't be needed - nobody will
1011 	 * have the file open.
1012 	 */
1013 	status = ocfs2_inode_lock(inode, &di_bh, 1);
1014 	if (status < 0) {
1015 		if (status != -ENOENT)
1016 			mlog_errno(status);
1017 		ocfs2_cleanup_delete_inode(inode, 0);
1018 		goto bail_unlock_nfs_sync;
1019 	}
1020 
1021 	/* Query the cluster. This will be the final decision made
1022 	 * before we go ahead and wipe the inode. */
1023 	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1024 	if (!wipe || status < 0) {
1025 		/* Error and remote inode busy both mean we won't be
1026 		 * removing the inode, so they take almost the same
1027 		 * path. */
1028 		if (status < 0)
1029 			mlog_errno(status);
1030 
1031 		/* Someone in the cluster has disallowed a wipe of
1032 		 * this inode, or it was never completely
1033 		 * orphaned. Write out the pages and exit now. */
1034 		ocfs2_cleanup_delete_inode(inode, 1);
1035 		goto bail_unlock_inode;
1036 	}
1037 
1038 	ocfs2_cleanup_delete_inode(inode, 0);
1039 
1040 	status = ocfs2_wipe_inode(inode, di_bh);
1041 	if (status < 0) {
1042 		if (status != -EDEADLK)
1043 			mlog_errno(status);
1044 		goto bail_unlock_inode;
1045 	}
1046 
1047 	/*
1048 	 * Mark the inode as successfully deleted.
1049 	 *
1050 	 * This is important for ocfs2_clear_inode() as it will check
1051 	 * this flag and skip any checkpointing work
1052 	 *
1053 	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1054 	 * the LVB for other nodes.
1055 	 */
1056 	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1057 
1058 bail_unlock_inode:
1059 	ocfs2_inode_unlock(inode, 1);
1060 	brelse(di_bh);
1061 
1062 bail_unlock_nfs_sync:
1063 	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1064 
1065 bail_unblock:
1066 	status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1067 	if (status < 0)
1068 		mlog_errno(status);
1069 bail:
1070 	clear_inode(inode);
1071 	mlog_exit_void();
1072 }
1073 
1074 void ocfs2_clear_inode(struct inode *inode)
1075 {
1076 	int status;
1077 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1078 
1079 	mlog_entry_void();
1080 
1081 	if (!inode)
1082 		goto bail;
1083 
1084 	mlog(0, "Clearing inode: %llu, nlink = %u\n",
1085 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1086 
1087 	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1088 			"Inode=%lu\n", inode->i_ino);
1089 
1090 	/* To preven remote deletes we hold open lock before, now it
1091 	 * is time to unlock PR and EX open locks. */
1092 	ocfs2_open_unlock(inode);
1093 
1094 	/* Do these before all the other work so that we don't bounce
1095 	 * the downconvert thread while waiting to destroy the locks. */
1096 	ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1097 	ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1098 	ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1099 
1100 	/* We very well may get a clear_inode before all an inodes
1101 	 * metadata has hit disk. Of course, we can't drop any cluster
1102 	 * locks until the journal has finished with it. The only
1103 	 * exception here are successfully wiped inodes - their
1104 	 * metadata can now be considered to be part of the system
1105 	 * inodes from which it came. */
1106 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1107 		ocfs2_checkpoint_inode(inode);
1108 
1109 	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1110 			"Clear inode of %llu, inode has io markers\n",
1111 			(unsigned long long)oi->ip_blkno);
1112 
1113 	ocfs2_extent_map_trunc(inode, 0);
1114 
1115 	status = ocfs2_drop_inode_locks(inode);
1116 	if (status < 0)
1117 		mlog_errno(status);
1118 
1119 	ocfs2_lock_res_free(&oi->ip_rw_lockres);
1120 	ocfs2_lock_res_free(&oi->ip_inode_lockres);
1121 	ocfs2_lock_res_free(&oi->ip_open_lockres);
1122 
1123 	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1124 
1125 	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1126 			"Clear inode of %llu, inode has %u cache items\n",
1127 			(unsigned long long)oi->ip_blkno,
1128 			INODE_CACHE(inode)->ci_num_cached);
1129 
1130 	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1131 			"Clear inode of %llu, inode has a bad flag\n",
1132 			(unsigned long long)oi->ip_blkno);
1133 
1134 	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1135 			"Clear inode of %llu, inode is locked\n",
1136 			(unsigned long long)oi->ip_blkno);
1137 
1138 	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1139 			"Clear inode of %llu, io_mutex is locked\n",
1140 			(unsigned long long)oi->ip_blkno);
1141 	mutex_unlock(&oi->ip_io_mutex);
1142 
1143 	/*
1144 	 * down_trylock() returns 0, down_write_trylock() returns 1
1145 	 * kernel 1, world 0
1146 	 */
1147 	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1148 			"Clear inode of %llu, alloc_sem is locked\n",
1149 			(unsigned long long)oi->ip_blkno);
1150 	up_write(&oi->ip_alloc_sem);
1151 
1152 	mlog_bug_on_msg(oi->ip_open_count,
1153 			"Clear inode of %llu has open count %d\n",
1154 			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
1155 
1156 	/* Clear all other flags. */
1157 	oi->ip_flags = 0;
1158 	oi->ip_dir_start_lookup = 0;
1159 	oi->ip_blkno = 0ULL;
1160 
1161 	/*
1162 	 * ip_jinode is used to track txns against this inode. We ensure that
1163 	 * the journal is flushed before journal shutdown. Thus it is safe to
1164 	 * have inodes get cleaned up after journal shutdown.
1165 	 */
1166 	jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1167 				       &oi->ip_jinode);
1168 
1169 bail:
1170 	mlog_exit_void();
1171 }
1172 
1173 /* Called under inode_lock, with no more references on the
1174  * struct inode, so it's safe here to check the flags field
1175  * and to manipulate i_nlink without any other locks. */
1176 void ocfs2_drop_inode(struct inode *inode)
1177 {
1178 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1179 
1180 	mlog_entry_void();
1181 
1182 	mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1183 	     (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1184 
1185 	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1186 		generic_delete_inode(inode);
1187 	else
1188 		generic_drop_inode(inode);
1189 
1190 	mlog_exit_void();
1191 }
1192 
1193 /*
1194  * This is called from our getattr.
1195  */
1196 int ocfs2_inode_revalidate(struct dentry *dentry)
1197 {
1198 	struct inode *inode = dentry->d_inode;
1199 	int status = 0;
1200 
1201 	mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1202 		   inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1203 
1204 	if (!inode) {
1205 		mlog(0, "eep, no inode!\n");
1206 		status = -ENOENT;
1207 		goto bail;
1208 	}
1209 
1210 	spin_lock(&OCFS2_I(inode)->ip_lock);
1211 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1212 		spin_unlock(&OCFS2_I(inode)->ip_lock);
1213 		mlog(0, "inode deleted!\n");
1214 		status = -ENOENT;
1215 		goto bail;
1216 	}
1217 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1218 
1219 	/* Let ocfs2_inode_lock do the work of updating our struct
1220 	 * inode for us. */
1221 	status = ocfs2_inode_lock(inode, NULL, 0);
1222 	if (status < 0) {
1223 		if (status != -ENOENT)
1224 			mlog_errno(status);
1225 		goto bail;
1226 	}
1227 	ocfs2_inode_unlock(inode, 0);
1228 bail:
1229 	mlog_exit(status);
1230 
1231 	return status;
1232 }
1233 
1234 /*
1235  * Updates a disk inode from a
1236  * struct inode.
1237  * Only takes ip_lock.
1238  */
1239 int ocfs2_mark_inode_dirty(handle_t *handle,
1240 			   struct inode *inode,
1241 			   struct buffer_head *bh)
1242 {
1243 	int status;
1244 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1245 
1246 	mlog_entry("(inode %llu)\n",
1247 		   (unsigned long long)OCFS2_I(inode)->ip_blkno);
1248 
1249 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1250 					 OCFS2_JOURNAL_ACCESS_WRITE);
1251 	if (status < 0) {
1252 		mlog_errno(status);
1253 		goto leave;
1254 	}
1255 
1256 	spin_lock(&OCFS2_I(inode)->ip_lock);
1257 	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1258 	ocfs2_get_inode_flags(OCFS2_I(inode));
1259 	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1260 	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1261 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1262 
1263 	fe->i_size = cpu_to_le64(i_size_read(inode));
1264 	ocfs2_set_links_count(fe, inode->i_nlink);
1265 	fe->i_uid = cpu_to_le32(inode->i_uid);
1266 	fe->i_gid = cpu_to_le32(inode->i_gid);
1267 	fe->i_mode = cpu_to_le16(inode->i_mode);
1268 	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1269 	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1270 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1271 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1272 	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1273 	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1274 
1275 	status = ocfs2_journal_dirty(handle, bh);
1276 	if (status < 0)
1277 		mlog_errno(status);
1278 
1279 	status = 0;
1280 leave:
1281 
1282 	mlog_exit(status);
1283 	return status;
1284 }
1285 
1286 /*
1287  *
1288  * Updates a struct inode from a disk inode.
1289  * does no i/o, only takes ip_lock.
1290  */
1291 void ocfs2_refresh_inode(struct inode *inode,
1292 			 struct ocfs2_dinode *fe)
1293 {
1294 	spin_lock(&OCFS2_I(inode)->ip_lock);
1295 
1296 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1297 	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1298 	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1299 	ocfs2_set_inode_flags(inode);
1300 	i_size_write(inode, le64_to_cpu(fe->i_size));
1301 	inode->i_nlink = ocfs2_read_links_count(fe);
1302 	inode->i_uid = le32_to_cpu(fe->i_uid);
1303 	inode->i_gid = le32_to_cpu(fe->i_gid);
1304 	inode->i_mode = le16_to_cpu(fe->i_mode);
1305 	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1306 		inode->i_blocks = 0;
1307 	else
1308 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1309 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1310 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1311 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1312 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1313 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1314 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1315 
1316 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1317 }
1318 
1319 int ocfs2_validate_inode_block(struct super_block *sb,
1320 			       struct buffer_head *bh)
1321 {
1322 	int rc;
1323 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1324 
1325 	mlog(0, "Validating dinode %llu\n",
1326 	     (unsigned long long)bh->b_blocknr);
1327 
1328 	BUG_ON(!buffer_uptodate(bh));
1329 
1330 	/*
1331 	 * If the ecc fails, we return the error but otherwise
1332 	 * leave the filesystem running.  We know any error is
1333 	 * local to this block.
1334 	 */
1335 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1336 	if (rc) {
1337 		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1338 		     (unsigned long long)bh->b_blocknr);
1339 		goto bail;
1340 	}
1341 
1342 	/*
1343 	 * Errors after here are fatal.
1344 	 */
1345 
1346 	rc = -EINVAL;
1347 
1348 	if (!OCFS2_IS_VALID_DINODE(di)) {
1349 		ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1350 			    (unsigned long long)bh->b_blocknr, 7,
1351 			    di->i_signature);
1352 		goto bail;
1353 	}
1354 
1355 	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1356 		ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1357 			    (unsigned long long)bh->b_blocknr,
1358 			    (unsigned long long)le64_to_cpu(di->i_blkno));
1359 		goto bail;
1360 	}
1361 
1362 	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1363 		ocfs2_error(sb,
1364 			    "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1365 			    (unsigned long long)bh->b_blocknr);
1366 		goto bail;
1367 	}
1368 
1369 	if (le32_to_cpu(di->i_fs_generation) !=
1370 	    OCFS2_SB(sb)->fs_generation) {
1371 		ocfs2_error(sb,
1372 			    "Invalid dinode #%llu: fs_generation is %u\n",
1373 			    (unsigned long long)bh->b_blocknr,
1374 			    le32_to_cpu(di->i_fs_generation));
1375 		goto bail;
1376 	}
1377 
1378 	rc = 0;
1379 
1380 bail:
1381 	return rc;
1382 }
1383 
1384 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1385 				int flags)
1386 {
1387 	int rc;
1388 	struct buffer_head *tmp = *bh;
1389 
1390 	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1391 			       1, &tmp, flags, ocfs2_validate_inode_block);
1392 
1393 	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1394 	if (!rc && !*bh)
1395 		*bh = tmp;
1396 
1397 	return rc;
1398 }
1399 
1400 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1401 {
1402 	return ocfs2_read_inode_block_full(inode, bh, 0);
1403 }
1404 
1405 
1406 static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1407 {
1408 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1409 
1410 	return oi->ip_blkno;
1411 }
1412 
1413 static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1414 {
1415 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1416 
1417 	return oi->vfs_inode.i_sb;
1418 }
1419 
1420 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1421 {
1422 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1423 
1424 	spin_lock(&oi->ip_lock);
1425 }
1426 
1427 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1428 {
1429 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1430 
1431 	spin_unlock(&oi->ip_lock);
1432 }
1433 
1434 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1435 {
1436 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1437 
1438 	mutex_lock(&oi->ip_io_mutex);
1439 }
1440 
1441 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1442 {
1443 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1444 
1445 	mutex_unlock(&oi->ip_io_mutex);
1446 }
1447 
1448 const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1449 	.co_owner		= ocfs2_inode_cache_owner,
1450 	.co_get_super		= ocfs2_inode_cache_get_super,
1451 	.co_cache_lock		= ocfs2_inode_cache_lock,
1452 	.co_cache_unlock	= ocfs2_inode_cache_unlock,
1453 	.co_io_lock		= ocfs2_inode_cache_io_lock,
1454 	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
1455 };
1456 
1457