xref: /linux/fs/ocfs2/inode.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
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 #include <cluster/masklog.h>
35 
36 #include "ocfs2.h"
37 
38 #include "alloc.h"
39 #include "dir.h"
40 #include "blockcheck.h"
41 #include "dlmglue.h"
42 #include "extent_map.h"
43 #include "file.h"
44 #include "heartbeat.h"
45 #include "inode.h"
46 #include "journal.h"
47 #include "namei.h"
48 #include "suballoc.h"
49 #include "super.h"
50 #include "symlink.h"
51 #include "sysfile.h"
52 #include "uptodate.h"
53 #include "xattr.h"
54 #include "refcounttree.h"
55 #include "ocfs2_trace.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 	journal_t *journal = OCFS2_SB(sb)->journal->j_journal;
134 
135 	trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
136 			       sysfile_type);
137 
138 	/* Ok. By now we've either got the offsets passed to us by the
139 	 * caller, or we just pulled them off the bh. Lets do some
140 	 * sanity checks to make sure they're OK. */
141 	if (blkno == 0) {
142 		inode = ERR_PTR(-EINVAL);
143 		mlog_errno(PTR_ERR(inode));
144 		goto bail;
145 	}
146 
147 	args.fi_blkno = blkno;
148 	args.fi_flags = flags;
149 	args.fi_ino = ino_from_blkno(sb, blkno);
150 	args.fi_sysfile_type = sysfile_type;
151 
152 	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
153 			     ocfs2_init_locked_inode, &args);
154 	/* inode was *not* in the inode cache. 2.6.x requires
155 	 * us to do our own read_inode call and unlock it
156 	 * afterwards. */
157 	if (inode == NULL) {
158 		inode = ERR_PTR(-ENOMEM);
159 		mlog_errno(PTR_ERR(inode));
160 		goto bail;
161 	}
162 	trace_ocfs2_iget5_locked(inode->i_state);
163 	if (inode->i_state & I_NEW) {
164 		ocfs2_read_locked_inode(inode, &args);
165 		unlock_new_inode(inode);
166 	}
167 	if (is_bad_inode(inode)) {
168 		iput(inode);
169 		inode = ERR_PTR(-ESTALE);
170 		goto bail;
171 	}
172 
173 	/*
174 	 * Set transaction id's of transactions that have to be committed
175 	 * to finish f[data]sync. We set them to currently running transaction
176 	 * as we cannot be sure that the inode or some of its metadata isn't
177 	 * part of the transaction - the inode could have been reclaimed and
178 	 * now it is reread from disk.
179 	 */
180 	if (journal) {
181 		transaction_t *transaction;
182 		tid_t tid;
183 		struct ocfs2_inode_info *oi = OCFS2_I(inode);
184 
185 		read_lock(&journal->j_state_lock);
186 		if (journal->j_running_transaction)
187 			transaction = journal->j_running_transaction;
188 		else
189 			transaction = journal->j_committing_transaction;
190 		if (transaction)
191 			tid = transaction->t_tid;
192 		else
193 			tid = journal->j_commit_sequence;
194 		read_unlock(&journal->j_state_lock);
195 		oi->i_sync_tid = tid;
196 		oi->i_datasync_tid = tid;
197 	}
198 
199 bail:
200 	if (!IS_ERR(inode)) {
201 		trace_ocfs2_iget_end(inode,
202 			(unsigned long long)OCFS2_I(inode)->ip_blkno);
203 	}
204 
205 	return inode;
206 }
207 
208 
209 /*
210  * here's how inodes get read from disk:
211  * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
212  * found? : return the in-memory inode
213  * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
214  */
215 
216 static int ocfs2_find_actor(struct inode *inode, void *opaque)
217 {
218 	struct ocfs2_find_inode_args *args = NULL;
219 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
220 	int ret = 0;
221 
222 	args = opaque;
223 
224 	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
225 
226 	trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
227 
228 	if (oi->ip_blkno != args->fi_blkno)
229 		goto bail;
230 
231 	ret = 1;
232 bail:
233 	return ret;
234 }
235 
236 /*
237  * initialize the new inode, but don't do anything that would cause
238  * us to sleep.
239  * return 0 on success, 1 on failure
240  */
241 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
242 {
243 	struct ocfs2_find_inode_args *args = opaque;
244 	static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
245 				     ocfs2_file_ip_alloc_sem_key;
246 
247 	inode->i_ino = args->fi_ino;
248 	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
249 	if (args->fi_sysfile_type != 0)
250 		lockdep_set_class(&inode->i_mutex,
251 			&ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
252 	if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
253 	    args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
254 	    args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
255 	    args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
256 		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
257 				  &ocfs2_quota_ip_alloc_sem_key);
258 	else
259 		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
260 				  &ocfs2_file_ip_alloc_sem_key);
261 
262 	return 0;
263 }
264 
265 void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
266 			  int create_ino)
267 {
268 	struct super_block *sb;
269 	struct ocfs2_super *osb;
270 	int use_plocks = 1;
271 
272 	sb = inode->i_sb;
273 	osb = OCFS2_SB(sb);
274 
275 	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
276 	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
277 		use_plocks = 0;
278 
279 	/*
280 	 * These have all been checked by ocfs2_read_inode_block() or set
281 	 * by ocfs2_mknod_locked(), so a failure is a code bug.
282 	 */
283 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
284 						cannot create a superblock
285 						inode today.  change if
286 						that is needed. */
287 	BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
288 	BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
289 
290 
291 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
292 	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
293 	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
294 
295 	inode->i_version = 1;
296 	inode->i_generation = le32_to_cpu(fe->i_generation);
297 	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
298 	inode->i_mode = le16_to_cpu(fe->i_mode);
299 	i_uid_write(inode, le32_to_cpu(fe->i_uid));
300 	i_gid_write(inode, le32_to_cpu(fe->i_gid));
301 
302 	/* Fast symlinks will have i_size but no allocated clusters. */
303 	if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
304 		inode->i_blocks = 0;
305 		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
306 	} else {
307 		inode->i_blocks = ocfs2_inode_sector_count(inode);
308 		inode->i_mapping->a_ops = &ocfs2_aops;
309 	}
310 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
311 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
312 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
313 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
314 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
315 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
316 
317 	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
318 		mlog(ML_ERROR,
319 		     "ip_blkno %llu != i_blkno %llu!\n",
320 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
321 		     (unsigned long long)le64_to_cpu(fe->i_blkno));
322 
323 	set_nlink(inode, ocfs2_read_links_count(fe));
324 
325 	trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
326 				   le32_to_cpu(fe->i_flags));
327 	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
328 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
329 		inode->i_flags |= S_NOQUOTA;
330 	}
331 
332 	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
333 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
334 	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
335 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
336 	} else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
337 		inode->i_flags |= S_NOQUOTA;
338 	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
339 		/* we can't actually hit this as read_inode can't
340 		 * handle superblocks today ;-) */
341 		BUG();
342 	}
343 
344 	switch (inode->i_mode & S_IFMT) {
345 	    case S_IFREG:
346 		    if (use_plocks)
347 			    inode->i_fop = &ocfs2_fops;
348 		    else
349 			    inode->i_fop = &ocfs2_fops_no_plocks;
350 		    inode->i_op = &ocfs2_file_iops;
351 		    i_size_write(inode, le64_to_cpu(fe->i_size));
352 		    break;
353 	    case S_IFDIR:
354 		    inode->i_op = &ocfs2_dir_iops;
355 		    if (use_plocks)
356 			    inode->i_fop = &ocfs2_dops;
357 		    else
358 			    inode->i_fop = &ocfs2_dops_no_plocks;
359 		    i_size_write(inode, le64_to_cpu(fe->i_size));
360 		    OCFS2_I(inode)->ip_dir_lock_gen = 1;
361 		    break;
362 	    case S_IFLNK:
363 		    inode->i_op = &ocfs2_symlink_inode_operations;
364 		    i_size_write(inode, le64_to_cpu(fe->i_size));
365 		    break;
366 	    default:
367 		    inode->i_op = &ocfs2_special_file_iops;
368 		    init_special_inode(inode, inode->i_mode,
369 				       inode->i_rdev);
370 		    break;
371 	}
372 
373 	if (create_ino) {
374 		inode->i_ino = ino_from_blkno(inode->i_sb,
375 			       le64_to_cpu(fe->i_blkno));
376 
377 		/*
378 		 * If we ever want to create system files from kernel,
379 		 * the generation argument to
380 		 * ocfs2_inode_lock_res_init() will have to change.
381 		 */
382 		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
383 
384 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
385 					  OCFS2_LOCK_TYPE_META, 0, inode);
386 
387 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
388 					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
389 	}
390 
391 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
392 				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
393 				  inode);
394 
395 	ocfs2_set_inode_flags(inode);
396 
397 	OCFS2_I(inode)->ip_last_used_slot = 0;
398 	OCFS2_I(inode)->ip_last_used_group = 0;
399 
400 	if (S_ISDIR(inode->i_mode))
401 		ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
402 				    OCFS2_RESV_FLAG_DIR);
403 }
404 
405 static int ocfs2_read_locked_inode(struct inode *inode,
406 				   struct ocfs2_find_inode_args *args)
407 {
408 	struct super_block *sb;
409 	struct ocfs2_super *osb;
410 	struct ocfs2_dinode *fe;
411 	struct buffer_head *bh = NULL;
412 	int status, can_lock;
413 	u32 generation = 0;
414 
415 	status = -EINVAL;
416 	sb = inode->i_sb;
417 	osb = OCFS2_SB(sb);
418 
419 	/*
420 	 * To improve performance of cold-cache inode stats, we take
421 	 * the cluster lock here if possible.
422 	 *
423 	 * Generally, OCFS2 never trusts the contents of an inode
424 	 * unless it's holding a cluster lock, so taking it here isn't
425 	 * a correctness issue as much as it is a performance
426 	 * improvement.
427 	 *
428 	 * There are three times when taking the lock is not a good idea:
429 	 *
430 	 * 1) During startup, before we have initialized the DLM.
431 	 *
432 	 * 2) If we are reading certain system files which never get
433 	 *    cluster locks (local alloc, truncate log).
434 	 *
435 	 * 3) If the process doing the iget() is responsible for
436 	 *    orphan dir recovery. We're holding the orphan dir lock and
437 	 *    can get into a deadlock with another process on another
438 	 *    node in ->delete_inode().
439 	 *
440 	 * #1 and #2 can be simply solved by never taking the lock
441 	 * here for system files (which are the only type we read
442 	 * during mount). It's a heavier approach, but our main
443 	 * concern is user-accessible files anyway.
444 	 *
445 	 * #3 works itself out because we'll eventually take the
446 	 * cluster lock before trusting anything anyway.
447 	 */
448 	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
449 		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
450 		&& !ocfs2_mount_local(osb);
451 
452 	trace_ocfs2_read_locked_inode(
453 		(unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
454 
455 	/*
456 	 * To maintain backwards compatibility with older versions of
457 	 * ocfs2-tools, we still store the generation value for system
458 	 * files. The only ones that actually matter to userspace are
459 	 * the journals, but it's easier and inexpensive to just flag
460 	 * all system files similarly.
461 	 */
462 	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
463 		generation = osb->fs_generation;
464 
465 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
466 				  OCFS2_LOCK_TYPE_META,
467 				  generation, inode);
468 
469 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
470 				  OCFS2_LOCK_TYPE_OPEN,
471 				  0, inode);
472 
473 	if (can_lock) {
474 		status = ocfs2_open_lock(inode);
475 		if (status) {
476 			make_bad_inode(inode);
477 			mlog_errno(status);
478 			return status;
479 		}
480 		status = ocfs2_inode_lock(inode, NULL, 0);
481 		if (status) {
482 			make_bad_inode(inode);
483 			mlog_errno(status);
484 			return status;
485 		}
486 	}
487 
488 	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
489 		status = ocfs2_try_open_lock(inode, 0);
490 		if (status) {
491 			make_bad_inode(inode);
492 			return status;
493 		}
494 	}
495 
496 	if (can_lock) {
497 		status = ocfs2_read_inode_block_full(inode, &bh,
498 						     OCFS2_BH_IGNORE_CACHE);
499 	} else {
500 		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
501 		/*
502 		 * If buffer is in jbd, then its checksum may not have been
503 		 * computed as yet.
504 		 */
505 		if (!status && !buffer_jbd(bh))
506 			status = ocfs2_validate_inode_block(osb->sb, bh);
507 	}
508 	if (status < 0) {
509 		mlog_errno(status);
510 		goto bail;
511 	}
512 
513 	status = -EINVAL;
514 	fe = (struct ocfs2_dinode *) bh->b_data;
515 
516 	/*
517 	 * This is a code bug. Right now the caller needs to
518 	 * understand whether it is asking for a system file inode or
519 	 * not so the proper lock names can be built.
520 	 */
521 	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
522 			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
523 			"Inode %llu: system file state is ambigous\n",
524 			(unsigned long long)args->fi_blkno);
525 
526 	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
527 	    S_ISBLK(le16_to_cpu(fe->i_mode)))
528 		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
529 
530 	ocfs2_populate_inode(inode, fe, 0);
531 
532 	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
533 
534 	status = 0;
535 
536 bail:
537 	if (can_lock)
538 		ocfs2_inode_unlock(inode, 0);
539 
540 	if (status < 0)
541 		make_bad_inode(inode);
542 
543 	brelse(bh);
544 
545 	return status;
546 }
547 
548 void ocfs2_sync_blockdev(struct super_block *sb)
549 {
550 	sync_blockdev(sb->s_bdev);
551 }
552 
553 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
554 				     struct inode *inode,
555 				     struct buffer_head *fe_bh)
556 {
557 	int status = 0;
558 	struct ocfs2_dinode *fe;
559 	handle_t *handle = NULL;
560 
561 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
562 
563 	/*
564 	 * This check will also skip truncate of inodes with inline
565 	 * data and fast symlinks.
566 	 */
567 	if (fe->i_clusters) {
568 		if (ocfs2_should_order_data(inode))
569 			ocfs2_begin_ordered_truncate(inode, 0);
570 
571 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
572 		if (IS_ERR(handle)) {
573 			status = PTR_ERR(handle);
574 			handle = NULL;
575 			mlog_errno(status);
576 			goto out;
577 		}
578 
579 		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
580 						 fe_bh,
581 						 OCFS2_JOURNAL_ACCESS_WRITE);
582 		if (status < 0) {
583 			mlog_errno(status);
584 			goto out;
585 		}
586 
587 		i_size_write(inode, 0);
588 
589 		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
590 		if (status < 0) {
591 			mlog_errno(status);
592 			goto out;
593 		}
594 
595 		ocfs2_commit_trans(osb, handle);
596 		handle = NULL;
597 
598 		status = ocfs2_commit_truncate(osb, inode, fe_bh);
599 		if (status < 0) {
600 			mlog_errno(status);
601 			goto out;
602 		}
603 	}
604 
605 out:
606 	if (handle)
607 		ocfs2_commit_trans(osb, handle);
608 	return status;
609 }
610 
611 static int ocfs2_remove_inode(struct inode *inode,
612 			      struct buffer_head *di_bh,
613 			      struct inode *orphan_dir_inode,
614 			      struct buffer_head *orphan_dir_bh)
615 {
616 	int status;
617 	struct inode *inode_alloc_inode = NULL;
618 	struct buffer_head *inode_alloc_bh = NULL;
619 	handle_t *handle;
620 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
621 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
622 
623 	inode_alloc_inode =
624 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
625 					    le16_to_cpu(di->i_suballoc_slot));
626 	if (!inode_alloc_inode) {
627 		status = -ENOENT;
628 		mlog_errno(status);
629 		goto bail;
630 	}
631 
632 	mutex_lock(&inode_alloc_inode->i_mutex);
633 	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
634 	if (status < 0) {
635 		mutex_unlock(&inode_alloc_inode->i_mutex);
636 
637 		mlog_errno(status);
638 		goto bail;
639 	}
640 
641 	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
642 				   ocfs2_quota_trans_credits(inode->i_sb));
643 	if (IS_ERR(handle)) {
644 		status = PTR_ERR(handle);
645 		mlog_errno(status);
646 		goto bail_unlock;
647 	}
648 
649 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
650 		status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
651 					  orphan_dir_bh, false);
652 		if (status < 0) {
653 			mlog_errno(status);
654 			goto bail_commit;
655 		}
656 	}
657 
658 	/* set the inodes dtime */
659 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
660 					 OCFS2_JOURNAL_ACCESS_WRITE);
661 	if (status < 0) {
662 		mlog_errno(status);
663 		goto bail_commit;
664 	}
665 
666 	di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
667 	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
668 	ocfs2_journal_dirty(handle, di_bh);
669 
670 	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
671 	dquot_free_inode(inode);
672 
673 	status = ocfs2_free_dinode(handle, inode_alloc_inode,
674 				   inode_alloc_bh, di);
675 	if (status < 0)
676 		mlog_errno(status);
677 
678 bail_commit:
679 	ocfs2_commit_trans(osb, handle);
680 bail_unlock:
681 	ocfs2_inode_unlock(inode_alloc_inode, 1);
682 	mutex_unlock(&inode_alloc_inode->i_mutex);
683 	brelse(inode_alloc_bh);
684 bail:
685 	iput(inode_alloc_inode);
686 
687 	return status;
688 }
689 
690 /*
691  * Serialize with orphan dir recovery. If the process doing
692  * recovery on this orphan dir does an iget() with the dir
693  * i_mutex held, we'll deadlock here. Instead we detect this
694  * and exit early - recovery will wipe this inode for us.
695  */
696 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
697 					     int slot)
698 {
699 	int ret = 0;
700 
701 	spin_lock(&osb->osb_lock);
702 	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
703 		ret = -EDEADLK;
704 		goto out;
705 	}
706 	/* This signals to the orphan recovery process that it should
707 	 * wait for us to handle the wipe. */
708 	osb->osb_orphan_wipes[slot]++;
709 out:
710 	spin_unlock(&osb->osb_lock);
711 	trace_ocfs2_check_orphan_recovery_state(slot, ret);
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 = -ENOENT;
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 	trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
823 					     (unsigned long long)oi->ip_blkno,
824 					     oi->ip_flags);
825 
826 	/* We shouldn't be getting here for the root directory
827 	 * inode.. */
828 	if (inode == osb->root_inode) {
829 		mlog(ML_ERROR, "Skipping delete of root inode.\n");
830 		goto bail;
831 	}
832 
833 	/*
834 	 * If we're coming from downconvert_thread we can't go into our own
835 	 * voting [hello, deadlock city!] so we cannot delete the inode. But
836 	 * since we dropped last inode ref when downconverting dentry lock,
837 	 * we cannot have the file open and thus the node doing unlink will
838 	 * take care of deleting the inode.
839 	 */
840 	if (current == osb->dc_task)
841 		goto bail;
842 
843 	spin_lock(&oi->ip_lock);
844 	/* OCFS2 *never* deletes system files. This should technically
845 	 * never get here as system file inodes should always have a
846 	 * positive link count. */
847 	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
848 		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
849 		     (unsigned long long)oi->ip_blkno);
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, reason = 0;
869 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
870 	struct ocfs2_dinode *di;
871 
872 	*wipe = 0;
873 
874 	trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
875 					   inode->i_nlink);
876 
877 	/* While we were waiting for the cluster lock in
878 	 * ocfs2_delete_inode, another node might have asked to delete
879 	 * the inode. Recheck our flags to catch this. */
880 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
881 		reason = 1;
882 		goto bail;
883 	}
884 
885 	/* Now that we have an up to date inode, we can double check
886 	 * the link count. */
887 	if (inode->i_nlink)
888 		goto bail;
889 
890 	/* Do some basic inode verification... */
891 	di = (struct ocfs2_dinode *) di_bh->b_data;
892 	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
893 	    !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
894 		/*
895 		 * Inodes in the orphan dir must have ORPHANED_FL.  The only
896 		 * inodes that come back out of the orphan dir are reflink
897 		 * targets. A reflink target may be moved out of the orphan
898 		 * dir between the time we scan the directory and the time we
899 		 * process it. This would lead to HAS_REFCOUNT_FL being set but
900 		 * ORPHANED_FL not.
901 		 */
902 		if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
903 			reason = 2;
904 			goto bail;
905 		}
906 
907 		/* for lack of a better error? */
908 		status = -EEXIST;
909 		mlog(ML_ERROR,
910 		     "Inode %llu (on-disk %llu) not orphaned! "
911 		     "Disk flags  0x%x, inode flags 0x%x\n",
912 		     (unsigned long long)oi->ip_blkno,
913 		     (unsigned long long)le64_to_cpu(di->i_blkno),
914 		     le32_to_cpu(di->i_flags), oi->ip_flags);
915 		goto bail;
916 	}
917 
918 	/* has someone already deleted us?! baaad... */
919 	if (di->i_dtime) {
920 		status = -EEXIST;
921 		mlog_errno(status);
922 		goto bail;
923 	}
924 
925 	/*
926 	 * This is how ocfs2 determines whether an inode is still live
927 	 * within the cluster. Every node takes a shared read lock on
928 	 * the inode open lock in ocfs2_read_locked_inode(). When we
929 	 * get to ->delete_inode(), each node tries to convert it's
930 	 * lock to an exclusive. Trylocks are serialized by the inode
931 	 * meta data lock. If the upconvert succeeds, we know the inode
932 	 * is no longer live and can be deleted.
933 	 *
934 	 * Though we call this with the meta data lock held, the
935 	 * trylock keeps us from ABBA deadlock.
936 	 */
937 	status = ocfs2_try_open_lock(inode, 1);
938 	if (status == -EAGAIN) {
939 		status = 0;
940 		reason = 3;
941 		goto bail;
942 	}
943 	if (status < 0) {
944 		mlog_errno(status);
945 		goto bail;
946 	}
947 
948 	*wipe = 1;
949 	trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
950 
951 bail:
952 	trace_ocfs2_query_inode_wipe_end(status, reason);
953 	return status;
954 }
955 
956 /* Support function for ocfs2_delete_inode. Will help us keep the
957  * inode data in a consistent state for clear_inode. Always truncates
958  * pages, optionally sync's them first. */
959 static void ocfs2_cleanup_delete_inode(struct inode *inode,
960 				       int sync_data)
961 {
962 	trace_ocfs2_cleanup_delete_inode(
963 		(unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
964 	if (sync_data)
965 		filemap_write_and_wait(inode->i_mapping);
966 	truncate_inode_pages_final(&inode->i_data);
967 }
968 
969 static void ocfs2_delete_inode(struct inode *inode)
970 {
971 	int wipe, status;
972 	sigset_t oldset;
973 	struct buffer_head *di_bh = NULL;
974 
975 	trace_ocfs2_delete_inode(inode->i_ino,
976 				 (unsigned long long)OCFS2_I(inode)->ip_blkno,
977 				 is_bad_inode(inode));
978 
979 	/* When we fail in read_inode() we mark inode as bad. The second test
980 	 * catches the case when inode allocation fails before allocating
981 	 * a block for inode. */
982 	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
983 		goto bail;
984 
985 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
986 		/* It's probably not necessary to truncate_inode_pages
987 		 * here but we do it for safety anyway (it will most
988 		 * likely be a no-op anyway) */
989 		ocfs2_cleanup_delete_inode(inode, 0);
990 		goto bail;
991 	}
992 
993 	dquot_initialize(inode);
994 
995 	/* We want to block signals in delete_inode as the lock and
996 	 * messaging paths may return us -ERESTARTSYS. Which would
997 	 * cause us to exit early, resulting in inodes being orphaned
998 	 * forever. */
999 	ocfs2_block_signals(&oldset);
1000 
1001 	/*
1002 	 * Synchronize us against ocfs2_get_dentry. We take this in
1003 	 * shared mode so that all nodes can still concurrently
1004 	 * process deletes.
1005 	 */
1006 	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1007 	if (status < 0) {
1008 		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1009 		ocfs2_cleanup_delete_inode(inode, 0);
1010 		goto bail_unblock;
1011 	}
1012 	/* Lock down the inode. This gives us an up to date view of
1013 	 * it's metadata (for verification), and allows us to
1014 	 * serialize delete_inode on multiple nodes.
1015 	 *
1016 	 * Even though we might be doing a truncate, we don't take the
1017 	 * allocation lock here as it won't be needed - nobody will
1018 	 * have the file open.
1019 	 */
1020 	status = ocfs2_inode_lock(inode, &di_bh, 1);
1021 	if (status < 0) {
1022 		if (status != -ENOENT)
1023 			mlog_errno(status);
1024 		ocfs2_cleanup_delete_inode(inode, 0);
1025 		goto bail_unlock_nfs_sync;
1026 	}
1027 
1028 	/* Query the cluster. This will be the final decision made
1029 	 * before we go ahead and wipe the inode. */
1030 	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1031 	if (!wipe || status < 0) {
1032 		/* Error and remote inode busy both mean we won't be
1033 		 * removing the inode, so they take almost the same
1034 		 * path. */
1035 		if (status < 0)
1036 			mlog_errno(status);
1037 
1038 		/* Someone in the cluster has disallowed a wipe of
1039 		 * this inode, or it was never completely
1040 		 * orphaned. Write out the pages and exit now. */
1041 		ocfs2_cleanup_delete_inode(inode, 1);
1042 		goto bail_unlock_inode;
1043 	}
1044 
1045 	ocfs2_cleanup_delete_inode(inode, 0);
1046 
1047 	status = ocfs2_wipe_inode(inode, di_bh);
1048 	if (status < 0) {
1049 		if (status != -EDEADLK)
1050 			mlog_errno(status);
1051 		goto bail_unlock_inode;
1052 	}
1053 
1054 	/*
1055 	 * Mark the inode as successfully deleted.
1056 	 *
1057 	 * This is important for ocfs2_clear_inode() as it will check
1058 	 * this flag and skip any checkpointing work
1059 	 *
1060 	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1061 	 * the LVB for other nodes.
1062 	 */
1063 	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1064 
1065 bail_unlock_inode:
1066 	ocfs2_inode_unlock(inode, 1);
1067 	brelse(di_bh);
1068 
1069 bail_unlock_nfs_sync:
1070 	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1071 
1072 bail_unblock:
1073 	ocfs2_unblock_signals(&oldset);
1074 bail:
1075 	return;
1076 }
1077 
1078 static void ocfs2_clear_inode(struct inode *inode)
1079 {
1080 	int status;
1081 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1082 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1083 
1084 	clear_inode(inode);
1085 	trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1086 				inode->i_nlink);
1087 
1088 	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1089 			"Inode=%lu\n", inode->i_ino);
1090 
1091 	dquot_drop(inode);
1092 
1093 	/* To preven remote deletes we hold open lock before, now it
1094 	 * is time to unlock PR and EX open locks. */
1095 	ocfs2_open_unlock(inode);
1096 
1097 	/* Do these before all the other work so that we don't bounce
1098 	 * the downconvert thread while waiting to destroy the locks. */
1099 	ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres);
1100 	ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres);
1101 	ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres);
1102 
1103 	ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1104 			   &oi->ip_la_data_resv);
1105 	ocfs2_resv_init_once(&oi->ip_la_data_resv);
1106 
1107 	/* We very well may get a clear_inode before all an inodes
1108 	 * metadata has hit disk. Of course, we can't drop any cluster
1109 	 * locks until the journal has finished with it. The only
1110 	 * exception here are successfully wiped inodes - their
1111 	 * metadata can now be considered to be part of the system
1112 	 * inodes from which it came. */
1113 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1114 		ocfs2_checkpoint_inode(inode);
1115 
1116 	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1117 			"Clear inode of %llu, inode has io markers\n",
1118 			(unsigned long long)oi->ip_blkno);
1119 
1120 	ocfs2_extent_map_trunc(inode, 0);
1121 
1122 	status = ocfs2_drop_inode_locks(inode);
1123 	if (status < 0)
1124 		mlog_errno(status);
1125 
1126 	ocfs2_lock_res_free(&oi->ip_rw_lockres);
1127 	ocfs2_lock_res_free(&oi->ip_inode_lockres);
1128 	ocfs2_lock_res_free(&oi->ip_open_lockres);
1129 
1130 	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1131 
1132 	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1133 			"Clear inode of %llu, inode has %u cache items\n",
1134 			(unsigned long long)oi->ip_blkno,
1135 			INODE_CACHE(inode)->ci_num_cached);
1136 
1137 	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1138 			"Clear inode of %llu, inode has a bad flag\n",
1139 			(unsigned long long)oi->ip_blkno);
1140 
1141 	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1142 			"Clear inode of %llu, inode is locked\n",
1143 			(unsigned long long)oi->ip_blkno);
1144 
1145 	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1146 			"Clear inode of %llu, io_mutex is locked\n",
1147 			(unsigned long long)oi->ip_blkno);
1148 	mutex_unlock(&oi->ip_io_mutex);
1149 
1150 	/*
1151 	 * down_trylock() returns 0, down_write_trylock() returns 1
1152 	 * kernel 1, world 0
1153 	 */
1154 	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1155 			"Clear inode of %llu, alloc_sem is locked\n",
1156 			(unsigned long long)oi->ip_blkno);
1157 	up_write(&oi->ip_alloc_sem);
1158 
1159 	mlog_bug_on_msg(oi->ip_open_count,
1160 			"Clear inode of %llu has open count %d\n",
1161 			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
1162 
1163 	/* Clear all other flags. */
1164 	oi->ip_flags = 0;
1165 	oi->ip_dir_start_lookup = 0;
1166 	oi->ip_blkno = 0ULL;
1167 
1168 	/*
1169 	 * ip_jinode is used to track txns against this inode. We ensure that
1170 	 * the journal is flushed before journal shutdown. Thus it is safe to
1171 	 * have inodes get cleaned up after journal shutdown.
1172 	 */
1173 	jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1174 				       &oi->ip_jinode);
1175 }
1176 
1177 void ocfs2_evict_inode(struct inode *inode)
1178 {
1179 	if (!inode->i_nlink ||
1180 	    (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1181 		ocfs2_delete_inode(inode);
1182 	} else {
1183 		truncate_inode_pages_final(&inode->i_data);
1184 	}
1185 	ocfs2_clear_inode(inode);
1186 }
1187 
1188 /* Called under inode_lock, with no more references on the
1189  * struct inode, so it's safe here to check the flags field
1190  * and to manipulate i_nlink without any other locks. */
1191 int ocfs2_drop_inode(struct inode *inode)
1192 {
1193 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1194 	int res;
1195 
1196 	trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1197 				inode->i_nlink, oi->ip_flags);
1198 
1199 	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1200 		res = 1;
1201 	else
1202 		res = generic_drop_inode(inode);
1203 
1204 	return res;
1205 }
1206 
1207 /*
1208  * This is called from our getattr.
1209  */
1210 int ocfs2_inode_revalidate(struct dentry *dentry)
1211 {
1212 	struct inode *inode = d_inode(dentry);
1213 	int status = 0;
1214 
1215 	trace_ocfs2_inode_revalidate(inode,
1216 		inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1217 		inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1218 
1219 	if (!inode) {
1220 		status = -ENOENT;
1221 		goto bail;
1222 	}
1223 
1224 	spin_lock(&OCFS2_I(inode)->ip_lock);
1225 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1226 		spin_unlock(&OCFS2_I(inode)->ip_lock);
1227 		status = -ENOENT;
1228 		goto bail;
1229 	}
1230 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1231 
1232 	/* Let ocfs2_inode_lock do the work of updating our struct
1233 	 * inode for us. */
1234 	status = ocfs2_inode_lock(inode, NULL, 0);
1235 	if (status < 0) {
1236 		if (status != -ENOENT)
1237 			mlog_errno(status);
1238 		goto bail;
1239 	}
1240 	ocfs2_inode_unlock(inode, 0);
1241 bail:
1242 	return status;
1243 }
1244 
1245 /*
1246  * Updates a disk inode from a
1247  * struct inode.
1248  * Only takes ip_lock.
1249  */
1250 int ocfs2_mark_inode_dirty(handle_t *handle,
1251 			   struct inode *inode,
1252 			   struct buffer_head *bh)
1253 {
1254 	int status;
1255 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1256 
1257 	trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1258 
1259 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1260 					 OCFS2_JOURNAL_ACCESS_WRITE);
1261 	if (status < 0) {
1262 		mlog_errno(status);
1263 		goto leave;
1264 	}
1265 
1266 	spin_lock(&OCFS2_I(inode)->ip_lock);
1267 	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1268 	ocfs2_get_inode_flags(OCFS2_I(inode));
1269 	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1270 	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1271 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1272 
1273 	fe->i_size = cpu_to_le64(i_size_read(inode));
1274 	ocfs2_set_links_count(fe, inode->i_nlink);
1275 	fe->i_uid = cpu_to_le32(i_uid_read(inode));
1276 	fe->i_gid = cpu_to_le32(i_gid_read(inode));
1277 	fe->i_mode = cpu_to_le16(inode->i_mode);
1278 	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1279 	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1280 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1281 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1282 	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1283 	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1284 
1285 	ocfs2_journal_dirty(handle, bh);
1286 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1287 leave:
1288 	return status;
1289 }
1290 
1291 /*
1292  *
1293  * Updates a struct inode from a disk inode.
1294  * does no i/o, only takes ip_lock.
1295  */
1296 void ocfs2_refresh_inode(struct inode *inode,
1297 			 struct ocfs2_dinode *fe)
1298 {
1299 	spin_lock(&OCFS2_I(inode)->ip_lock);
1300 
1301 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1302 	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1303 	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1304 	ocfs2_set_inode_flags(inode);
1305 	i_size_write(inode, le64_to_cpu(fe->i_size));
1306 	set_nlink(inode, ocfs2_read_links_count(fe));
1307 	i_uid_write(inode, le32_to_cpu(fe->i_uid));
1308 	i_gid_write(inode, le32_to_cpu(fe->i_gid));
1309 	inode->i_mode = le16_to_cpu(fe->i_mode);
1310 	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1311 		inode->i_blocks = 0;
1312 	else
1313 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1314 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1315 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1316 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1317 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1318 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1319 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1320 
1321 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1322 }
1323 
1324 int ocfs2_validate_inode_block(struct super_block *sb,
1325 			       struct buffer_head *bh)
1326 {
1327 	int rc;
1328 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1329 
1330 	trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1331 
1332 	BUG_ON(!buffer_uptodate(bh));
1333 
1334 	/*
1335 	 * If the ecc fails, we return the error but otherwise
1336 	 * leave the filesystem running.  We know any error is
1337 	 * local to this block.
1338 	 */
1339 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1340 	if (rc) {
1341 		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1342 		     (unsigned long long)bh->b_blocknr);
1343 		goto bail;
1344 	}
1345 
1346 	/*
1347 	 * Errors after here are fatal.
1348 	 */
1349 
1350 	rc = -EINVAL;
1351 
1352 	if (!OCFS2_IS_VALID_DINODE(di)) {
1353 		ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1354 			    (unsigned long long)bh->b_blocknr, 7,
1355 			    di->i_signature);
1356 		goto bail;
1357 	}
1358 
1359 	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1360 		ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1361 			    (unsigned long long)bh->b_blocknr,
1362 			    (unsigned long long)le64_to_cpu(di->i_blkno));
1363 		goto bail;
1364 	}
1365 
1366 	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1367 		ocfs2_error(sb,
1368 			    "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1369 			    (unsigned long long)bh->b_blocknr);
1370 		goto bail;
1371 	}
1372 
1373 	if (le32_to_cpu(di->i_fs_generation) !=
1374 	    OCFS2_SB(sb)->fs_generation) {
1375 		ocfs2_error(sb,
1376 			    "Invalid dinode #%llu: fs_generation is %u\n",
1377 			    (unsigned long long)bh->b_blocknr,
1378 			    le32_to_cpu(di->i_fs_generation));
1379 		goto bail;
1380 	}
1381 
1382 	rc = 0;
1383 
1384 bail:
1385 	return rc;
1386 }
1387 
1388 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1389 				int flags)
1390 {
1391 	int rc;
1392 	struct buffer_head *tmp = *bh;
1393 
1394 	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1395 			       1, &tmp, flags, ocfs2_validate_inode_block);
1396 
1397 	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1398 	if (!rc && !*bh)
1399 		*bh = tmp;
1400 
1401 	return rc;
1402 }
1403 
1404 int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1405 {
1406 	return ocfs2_read_inode_block_full(inode, bh, 0);
1407 }
1408 
1409 
1410 static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1411 {
1412 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1413 
1414 	return oi->ip_blkno;
1415 }
1416 
1417 static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1418 {
1419 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1420 
1421 	return oi->vfs_inode.i_sb;
1422 }
1423 
1424 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1425 {
1426 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1427 
1428 	spin_lock(&oi->ip_lock);
1429 }
1430 
1431 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1432 {
1433 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1434 
1435 	spin_unlock(&oi->ip_lock);
1436 }
1437 
1438 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1439 {
1440 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1441 
1442 	mutex_lock(&oi->ip_io_mutex);
1443 }
1444 
1445 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1446 {
1447 	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1448 
1449 	mutex_unlock(&oi->ip_io_mutex);
1450 }
1451 
1452 const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1453 	.co_owner		= ocfs2_inode_cache_owner,
1454 	.co_get_super		= ocfs2_inode_cache_get_super,
1455 	.co_cache_lock		= ocfs2_inode_cache_lock,
1456 	.co_cache_unlock	= ocfs2_inode_cache_unlock,
1457 	.co_io_lock		= ocfs2_inode_cache_io_lock,
1458 	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
1459 };
1460 
1461