Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * inode.c
6 *
7 * vfs' aops, fops, dops and iops
8 *
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 */
11
12#include <linux/fs.h>
13#include <linux/types.h>
14#include <linux/highmem.h>
15#include <linux/pagemap.h>
16#include <linux/quotaops.h>
17#include <linux/iversion.h>
18
19#include <asm/byteorder.h>
20
21#include <cluster/masklog.h>
22
23#include "ocfs2.h"
24
25#include "alloc.h"
26#include "dir.h"
27#include "blockcheck.h"
28#include "dlmglue.h"
29#include "extent_map.h"
30#include "file.h"
31#include "heartbeat.h"
32#include "inode.h"
33#include "journal.h"
34#include "namei.h"
35#include "suballoc.h"
36#include "super.h"
37#include "symlink.h"
38#include "sysfile.h"
39#include "uptodate.h"
40#include "xattr.h"
41#include "refcounttree.h"
42#include "ocfs2_trace.h"
43#include "filecheck.h"
44
45#include "buffer_head_io.h"
46
47struct ocfs2_find_inode_args
48{
49 u64 fi_blkno;
50 unsigned long fi_ino;
51 unsigned int fi_flags;
52 unsigned int fi_sysfile_type;
53};
54
55static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
56
57static int ocfs2_read_locked_inode(struct inode *inode,
58 struct ocfs2_find_inode_args *args);
59static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
60static int ocfs2_find_actor(struct inode *inode, void *opaque);
61static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
62 struct inode *inode,
63 struct buffer_head *fe_bh);
64
65static int ocfs2_filecheck_read_inode_block_full(struct inode *inode,
66 struct buffer_head **bh,
67 int flags, int type);
68static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
69 struct buffer_head *bh);
70static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
71 struct buffer_head *bh);
72
73void ocfs2_set_inode_flags(struct inode *inode)
74{
75 unsigned int flags = OCFS2_I(inode)->ip_attr;
76
77 inode->i_flags &= ~(S_IMMUTABLE |
78 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
79
80 if (flags & OCFS2_IMMUTABLE_FL)
81 inode->i_flags |= S_IMMUTABLE;
82
83 if (flags & OCFS2_SYNC_FL)
84 inode->i_flags |= S_SYNC;
85 if (flags & OCFS2_APPEND_FL)
86 inode->i_flags |= S_APPEND;
87 if (flags & OCFS2_NOATIME_FL)
88 inode->i_flags |= S_NOATIME;
89 if (flags & OCFS2_DIRSYNC_FL)
90 inode->i_flags |= S_DIRSYNC;
91}
92
93/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
94void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
95{
96 unsigned int flags = oi->vfs_inode.i_flags;
97
98 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
99 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
100 if (flags & S_SYNC)
101 oi->ip_attr |= OCFS2_SYNC_FL;
102 if (flags & S_APPEND)
103 oi->ip_attr |= OCFS2_APPEND_FL;
104 if (flags & S_IMMUTABLE)
105 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
106 if (flags & S_NOATIME)
107 oi->ip_attr |= OCFS2_NOATIME_FL;
108 if (flags & S_DIRSYNC)
109 oi->ip_attr |= OCFS2_DIRSYNC_FL;
110}
111
112struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
113{
114 struct ocfs2_find_inode_args args;
115
116 args.fi_blkno = blkno;
117 args.fi_flags = 0;
118 args.fi_ino = ino_from_blkno(sb, blkno);
119 args.fi_sysfile_type = 0;
120
121 return ilookup5(sb, blkno, ocfs2_find_actor, &args);
122}
123struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
124 int sysfile_type)
125{
126 int rc = -ESTALE;
127 struct inode *inode = NULL;
128 struct super_block *sb = osb->sb;
129 struct ocfs2_find_inode_args args;
130 journal_t *journal = OCFS2_SB(sb)->journal->j_journal;
131
132 trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
133 sysfile_type);
134
135 /* Ok. By now we've either got the offsets passed to us by the
136 * caller, or we just pulled them off the bh. Lets do some
137 * sanity checks to make sure they're OK. */
138 if (blkno == 0) {
139 inode = ERR_PTR(-EINVAL);
140 mlog_errno(PTR_ERR(inode));
141 goto bail;
142 }
143
144 args.fi_blkno = blkno;
145 args.fi_flags = flags;
146 args.fi_ino = ino_from_blkno(sb, blkno);
147 args.fi_sysfile_type = sysfile_type;
148
149 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
150 ocfs2_init_locked_inode, &args);
151 /* inode was *not* in the inode cache. 2.6.x requires
152 * us to do our own read_inode call and unlock it
153 * afterwards. */
154 if (inode == NULL) {
155 inode = ERR_PTR(-ENOMEM);
156 mlog_errno(PTR_ERR(inode));
157 goto bail;
158 }
159 trace_ocfs2_iget5_locked(inode->i_state);
160 if (inode->i_state & I_NEW) {
161 rc = ocfs2_read_locked_inode(inode, &args);
162 unlock_new_inode(inode);
163 }
164 if (is_bad_inode(inode)) {
165 iput(inode);
166 inode = ERR_PTR(rc);
167 goto bail;
168 }
169
170 /*
171 * Set transaction id's of transactions that have to be committed
172 * to finish f[data]sync. We set them to currently running transaction
173 * as we cannot be sure that the inode or some of its metadata isn't
174 * part of the transaction - the inode could have been reclaimed and
175 * now it is reread from disk.
176 */
177 if (journal) {
178 transaction_t *transaction;
179 tid_t tid;
180 struct ocfs2_inode_info *oi = OCFS2_I(inode);
181
182 read_lock(&journal->j_state_lock);
183 if (journal->j_running_transaction)
184 transaction = journal->j_running_transaction;
185 else
186 transaction = journal->j_committing_transaction;
187 if (transaction)
188 tid = transaction->t_tid;
189 else
190 tid = journal->j_commit_sequence;
191 read_unlock(&journal->j_state_lock);
192 oi->i_sync_tid = tid;
193 oi->i_datasync_tid = tid;
194 }
195
196bail:
197 if (!IS_ERR(inode)) {
198 trace_ocfs2_iget_end(inode,
199 (unsigned long long)OCFS2_I(inode)->ip_blkno);
200 }
201
202 return inode;
203}
204
205
206/*
207 * here's how inodes get read from disk:
208 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
209 * found? : return the in-memory inode
210 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
211 */
212
213static int ocfs2_find_actor(struct inode *inode, void *opaque)
214{
215 struct ocfs2_find_inode_args *args = NULL;
216 struct ocfs2_inode_info *oi = OCFS2_I(inode);
217 int ret = 0;
218
219 args = opaque;
220
221 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
222
223 trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
224
225 if (oi->ip_blkno != args->fi_blkno)
226 goto bail;
227
228 ret = 1;
229bail:
230 return ret;
231}
232
233/*
234 * initialize the new inode, but don't do anything that would cause
235 * us to sleep.
236 * return 0 on success, 1 on failure
237 */
238static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
239{
240 struct ocfs2_find_inode_args *args = opaque;
241 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
242 ocfs2_file_ip_alloc_sem_key;
243
244 inode->i_ino = args->fi_ino;
245 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
246 if (args->fi_sysfile_type != 0)
247 lockdep_set_class(&inode->i_rwsem,
248 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
249 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
250 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
251 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
252 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
253 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
254 &ocfs2_quota_ip_alloc_sem_key);
255 else
256 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
257 &ocfs2_file_ip_alloc_sem_key);
258
259 return 0;
260}
261
262void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
263 int create_ino)
264{
265 struct super_block *sb;
266 struct ocfs2_super *osb;
267 int use_plocks = 1;
268
269 sb = inode->i_sb;
270 osb = OCFS2_SB(sb);
271
272 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
273 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
274 use_plocks = 0;
275
276 /*
277 * These have all been checked by ocfs2_read_inode_block() or set
278 * by ocfs2_mknod_locked(), so a failure is a code bug.
279 */
280 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
281 cannot create a superblock
282 inode today. change if
283 that is needed. */
284 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
285 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
286
287
288 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
289 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
290 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
291
292 inode_set_iversion(inode, 1);
293 inode->i_generation = le32_to_cpu(fe->i_generation);
294 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
295 inode->i_mode = le16_to_cpu(fe->i_mode);
296 i_uid_write(inode, le32_to_cpu(fe->i_uid));
297 i_gid_write(inode, le32_to_cpu(fe->i_gid));
298
299 /* Fast symlinks will have i_size but no allocated clusters. */
300 if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
301 inode->i_blocks = 0;
302 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
303 } else {
304 inode->i_blocks = ocfs2_inode_sector_count(inode);
305 inode->i_mapping->a_ops = &ocfs2_aops;
306 }
307 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
308 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
309 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
310 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
311 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
312 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
313
314 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
315 mlog(ML_ERROR,
316 "ip_blkno %llu != i_blkno %llu!\n",
317 (unsigned long long)OCFS2_I(inode)->ip_blkno,
318 (unsigned long long)le64_to_cpu(fe->i_blkno));
319
320 set_nlink(inode, ocfs2_read_links_count(fe));
321
322 trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
323 le32_to_cpu(fe->i_flags));
324 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
325 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
326 inode->i_flags |= S_NOQUOTA;
327 }
328
329 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
330 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
331 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
332 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
333 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
334 inode->i_flags |= S_NOQUOTA;
335 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
336 /* we can't actually hit this as read_inode can't
337 * handle superblocks today ;-) */
338 BUG();
339 }
340
341 switch (inode->i_mode & S_IFMT) {
342 case S_IFREG:
343 if (use_plocks)
344 inode->i_fop = &ocfs2_fops;
345 else
346 inode->i_fop = &ocfs2_fops_no_plocks;
347 inode->i_op = &ocfs2_file_iops;
348 i_size_write(inode, le64_to_cpu(fe->i_size));
349 break;
350 case S_IFDIR:
351 inode->i_op = &ocfs2_dir_iops;
352 if (use_plocks)
353 inode->i_fop = &ocfs2_dops;
354 else
355 inode->i_fop = &ocfs2_dops_no_plocks;
356 i_size_write(inode, le64_to_cpu(fe->i_size));
357 OCFS2_I(inode)->ip_dir_lock_gen = 1;
358 break;
359 case S_IFLNK:
360 inode->i_op = &ocfs2_symlink_inode_operations;
361 inode_nohighmem(inode);
362 i_size_write(inode, le64_to_cpu(fe->i_size));
363 break;
364 default:
365 inode->i_op = &ocfs2_special_file_iops;
366 init_special_inode(inode, inode->i_mode,
367 inode->i_rdev);
368 break;
369 }
370
371 if (create_ino) {
372 inode->i_ino = ino_from_blkno(inode->i_sb,
373 le64_to_cpu(fe->i_blkno));
374
375 /*
376 * If we ever want to create system files from kernel,
377 * the generation argument to
378 * ocfs2_inode_lock_res_init() will have to change.
379 */
380 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
381
382 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
383 OCFS2_LOCK_TYPE_META, 0, inode);
384
385 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
386 OCFS2_LOCK_TYPE_OPEN, 0, inode);
387 }
388
389 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
390 OCFS2_LOCK_TYPE_RW, inode->i_generation,
391 inode);
392
393 ocfs2_set_inode_flags(inode);
394
395 OCFS2_I(inode)->ip_last_used_slot = 0;
396 OCFS2_I(inode)->ip_last_used_group = 0;
397
398 if (S_ISDIR(inode->i_mode))
399 ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
400 OCFS2_RESV_FLAG_DIR);
401}
402
403static int ocfs2_read_locked_inode(struct inode *inode,
404 struct ocfs2_find_inode_args *args)
405{
406 struct super_block *sb;
407 struct ocfs2_super *osb;
408 struct ocfs2_dinode *fe;
409 struct buffer_head *bh = NULL;
410 int status, can_lock, lock_level = 0;
411 u32 generation = 0;
412
413 status = -EINVAL;
414 sb = inode->i_sb;
415 osb = OCFS2_SB(sb);
416
417 /*
418 * To improve performance of cold-cache inode stats, we take
419 * the cluster lock here if possible.
420 *
421 * Generally, OCFS2 never trusts the contents of an inode
422 * unless it's holding a cluster lock, so taking it here isn't
423 * a correctness issue as much as it is a performance
424 * improvement.
425 *
426 * There are three times when taking the lock is not a good idea:
427 *
428 * 1) During startup, before we have initialized the DLM.
429 *
430 * 2) If we are reading certain system files which never get
431 * cluster locks (local alloc, truncate log).
432 *
433 * 3) If the process doing the iget() is responsible for
434 * orphan dir recovery. We're holding the orphan dir lock and
435 * can get into a deadlock with another process on another
436 * node in ->delete_inode().
437 *
438 * #1 and #2 can be simply solved by never taking the lock
439 * here for system files (which are the only type we read
440 * during mount). It's a heavier approach, but our main
441 * concern is user-accessible files anyway.
442 *
443 * #3 works itself out because we'll eventually take the
444 * cluster lock before trusting anything anyway.
445 */
446 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
447 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
448 && !ocfs2_mount_local(osb);
449
450 trace_ocfs2_read_locked_inode(
451 (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
452
453 /*
454 * To maintain backwards compatibility with older versions of
455 * ocfs2-tools, we still store the generation value for system
456 * files. The only ones that actually matter to userspace are
457 * the journals, but it's easier and inexpensive to just flag
458 * all system files similarly.
459 */
460 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
461 generation = osb->fs_generation;
462
463 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
464 OCFS2_LOCK_TYPE_META,
465 generation, inode);
466
467 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
468 OCFS2_LOCK_TYPE_OPEN,
469 0, inode);
470
471 if (can_lock) {
472 status = ocfs2_open_lock(inode);
473 if (status) {
474 make_bad_inode(inode);
475 mlog_errno(status);
476 return status;
477 }
478 status = ocfs2_inode_lock(inode, NULL, lock_level);
479 if (status) {
480 make_bad_inode(inode);
481 mlog_errno(status);
482 return status;
483 }
484 }
485
486 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
487 status = ocfs2_try_open_lock(inode, 0);
488 if (status) {
489 make_bad_inode(inode);
490 return status;
491 }
492 }
493
494 if (can_lock) {
495 if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
496 status = ocfs2_filecheck_read_inode_block_full(inode,
497 &bh, OCFS2_BH_IGNORE_CACHE, 0);
498 else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
499 status = ocfs2_filecheck_read_inode_block_full(inode,
500 &bh, OCFS2_BH_IGNORE_CACHE, 1);
501 else
502 status = ocfs2_read_inode_block_full(inode,
503 &bh, OCFS2_BH_IGNORE_CACHE);
504 } else {
505 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
506 /*
507 * If buffer is in jbd, then its checksum may not have been
508 * computed as yet.
509 */
510 if (!status && !buffer_jbd(bh)) {
511 if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
512 status = ocfs2_filecheck_validate_inode_block(
513 osb->sb, bh);
514 else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
515 status = ocfs2_filecheck_repair_inode_block(
516 osb->sb, bh);
517 else
518 status = ocfs2_validate_inode_block(
519 osb->sb, bh);
520 }
521 }
522 if (status < 0) {
523 mlog_errno(status);
524 goto bail;
525 }
526
527 status = -EINVAL;
528 fe = (struct ocfs2_dinode *) bh->b_data;
529
530 /*
531 * This is a code bug. Right now the caller needs to
532 * understand whether it is asking for a system file inode or
533 * not so the proper lock names can be built.
534 */
535 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
536 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
537 "Inode %llu: system file state is ambiguous\n",
538 (unsigned long long)args->fi_blkno);
539
540 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
541 S_ISBLK(le16_to_cpu(fe->i_mode)))
542 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
543
544 ocfs2_populate_inode(inode, fe, 0);
545
546 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
547
548 if (buffer_dirty(bh) && !buffer_jbd(bh)) {
549 if (can_lock) {
550 ocfs2_inode_unlock(inode, lock_level);
551 lock_level = 1;
552 ocfs2_inode_lock(inode, NULL, lock_level);
553 }
554 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
555 if (status < 0) {
556 mlog_errno(status);
557 goto bail;
558 }
559 }
560
561 status = 0;
562
563bail:
564 if (can_lock)
565 ocfs2_inode_unlock(inode, lock_level);
566
567 if (status < 0)
568 make_bad_inode(inode);
569
570 brelse(bh);
571
572 return status;
573}
574
575void ocfs2_sync_blockdev(struct super_block *sb)
576{
577 sync_blockdev(sb->s_bdev);
578}
579
580static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
581 struct inode *inode,
582 struct buffer_head *fe_bh)
583{
584 int status = 0;
585 struct ocfs2_dinode *fe;
586 handle_t *handle = NULL;
587
588 fe = (struct ocfs2_dinode *) fe_bh->b_data;
589
590 /*
591 * This check will also skip truncate of inodes with inline
592 * data and fast symlinks.
593 */
594 if (fe->i_clusters) {
595 if (ocfs2_should_order_data(inode))
596 ocfs2_begin_ordered_truncate(inode, 0);
597
598 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
599 if (IS_ERR(handle)) {
600 status = PTR_ERR(handle);
601 handle = NULL;
602 mlog_errno(status);
603 goto out;
604 }
605
606 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
607 fe_bh,
608 OCFS2_JOURNAL_ACCESS_WRITE);
609 if (status < 0) {
610 mlog_errno(status);
611 goto out;
612 }
613
614 i_size_write(inode, 0);
615
616 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
617 if (status < 0) {
618 mlog_errno(status);
619 goto out;
620 }
621
622 ocfs2_commit_trans(osb, handle);
623 handle = NULL;
624
625 status = ocfs2_commit_truncate(osb, inode, fe_bh);
626 if (status < 0)
627 mlog_errno(status);
628 }
629
630out:
631 if (handle)
632 ocfs2_commit_trans(osb, handle);
633 return status;
634}
635
636static int ocfs2_remove_inode(struct inode *inode,
637 struct buffer_head *di_bh,
638 struct inode *orphan_dir_inode,
639 struct buffer_head *orphan_dir_bh)
640{
641 int status;
642 struct inode *inode_alloc_inode = NULL;
643 struct buffer_head *inode_alloc_bh = NULL;
644 handle_t *handle;
645 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
646 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
647
648 inode_alloc_inode =
649 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
650 le16_to_cpu(di->i_suballoc_slot));
651 if (!inode_alloc_inode) {
652 status = -ENOENT;
653 mlog_errno(status);
654 goto bail;
655 }
656
657 inode_lock(inode_alloc_inode);
658 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
659 if (status < 0) {
660 inode_unlock(inode_alloc_inode);
661
662 mlog_errno(status);
663 goto bail;
664 }
665
666 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
667 ocfs2_quota_trans_credits(inode->i_sb));
668 if (IS_ERR(handle)) {
669 status = PTR_ERR(handle);
670 mlog_errno(status);
671 goto bail_unlock;
672 }
673
674 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
675 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
676 orphan_dir_bh, false);
677 if (status < 0) {
678 mlog_errno(status);
679 goto bail_commit;
680 }
681 }
682
683 /* set the inodes dtime */
684 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
685 OCFS2_JOURNAL_ACCESS_WRITE);
686 if (status < 0) {
687 mlog_errno(status);
688 goto bail_commit;
689 }
690
691 di->i_dtime = cpu_to_le64(ktime_get_real_seconds());
692 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
693 ocfs2_journal_dirty(handle, di_bh);
694
695 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
696 dquot_free_inode(inode);
697
698 status = ocfs2_free_dinode(handle, inode_alloc_inode,
699 inode_alloc_bh, di);
700 if (status < 0)
701 mlog_errno(status);
702
703bail_commit:
704 ocfs2_commit_trans(osb, handle);
705bail_unlock:
706 ocfs2_inode_unlock(inode_alloc_inode, 1);
707 inode_unlock(inode_alloc_inode);
708 brelse(inode_alloc_bh);
709bail:
710 iput(inode_alloc_inode);
711
712 return status;
713}
714
715/*
716 * Serialize with orphan dir recovery. If the process doing
717 * recovery on this orphan dir does an iget() with the dir
718 * i_mutex held, we'll deadlock here. Instead we detect this
719 * and exit early - recovery will wipe this inode for us.
720 */
721static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
722 int slot)
723{
724 int ret = 0;
725
726 spin_lock(&osb->osb_lock);
727 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
728 ret = -EDEADLK;
729 goto out;
730 }
731 /* This signals to the orphan recovery process that it should
732 * wait for us to handle the wipe. */
733 osb->osb_orphan_wipes[slot]++;
734out:
735 spin_unlock(&osb->osb_lock);
736 trace_ocfs2_check_orphan_recovery_state(slot, ret);
737 return ret;
738}
739
740static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
741 int slot)
742{
743 spin_lock(&osb->osb_lock);
744 osb->osb_orphan_wipes[slot]--;
745 spin_unlock(&osb->osb_lock);
746
747 wake_up(&osb->osb_wipe_event);
748}
749
750static int ocfs2_wipe_inode(struct inode *inode,
751 struct buffer_head *di_bh)
752{
753 int status, orphaned_slot = -1;
754 struct inode *orphan_dir_inode = NULL;
755 struct buffer_head *orphan_dir_bh = NULL;
756 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
757 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
758
759 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
760 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
761
762 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
763 if (status)
764 return status;
765
766 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
767 ORPHAN_DIR_SYSTEM_INODE,
768 orphaned_slot);
769 if (!orphan_dir_inode) {
770 status = -ENOENT;
771 mlog_errno(status);
772 goto bail;
773 }
774
775 /* Lock the orphan dir. The lock will be held for the entire
776 * delete_inode operation. We do this now to avoid races with
777 * recovery completion on other nodes. */
778 inode_lock(orphan_dir_inode);
779 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
780 if (status < 0) {
781 inode_unlock(orphan_dir_inode);
782
783 mlog_errno(status);
784 goto bail;
785 }
786 }
787
788 /* we do this while holding the orphan dir lock because we
789 * don't want recovery being run from another node to try an
790 * inode delete underneath us -- this will result in two nodes
791 * truncating the same file! */
792 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
793 if (status < 0) {
794 mlog_errno(status);
795 goto bail_unlock_dir;
796 }
797
798 /* Remove any dir index tree */
799 if (S_ISDIR(inode->i_mode)) {
800 status = ocfs2_dx_dir_truncate(inode, di_bh);
801 if (status) {
802 mlog_errno(status);
803 goto bail_unlock_dir;
804 }
805 }
806
807 /*Free extended attribute resources associated with this inode.*/
808 status = ocfs2_xattr_remove(inode, di_bh);
809 if (status < 0) {
810 mlog_errno(status);
811 goto bail_unlock_dir;
812 }
813
814 status = ocfs2_remove_refcount_tree(inode, di_bh);
815 if (status < 0) {
816 mlog_errno(status);
817 goto bail_unlock_dir;
818 }
819
820 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
821 orphan_dir_bh);
822 if (status < 0)
823 mlog_errno(status);
824
825bail_unlock_dir:
826 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
827 return status;
828
829 ocfs2_inode_unlock(orphan_dir_inode, 1);
830 inode_unlock(orphan_dir_inode);
831 brelse(orphan_dir_bh);
832bail:
833 iput(orphan_dir_inode);
834 ocfs2_signal_wipe_completion(osb, orphaned_slot);
835
836 return status;
837}
838
839/* There is a series of simple checks that should be done before a
840 * trylock is even considered. Encapsulate those in this function. */
841static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
842{
843 int ret = 0;
844 struct ocfs2_inode_info *oi = OCFS2_I(inode);
845 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
846
847 trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
848 (unsigned long long)oi->ip_blkno,
849 oi->ip_flags);
850
851 /* We shouldn't be getting here for the root directory
852 * inode.. */
853 if (inode == osb->root_inode) {
854 mlog(ML_ERROR, "Skipping delete of root inode.\n");
855 goto bail;
856 }
857
858 /*
859 * If we're coming from downconvert_thread we can't go into our own
860 * voting [hello, deadlock city!] so we cannot delete the inode. But
861 * since we dropped last inode ref when downconverting dentry lock,
862 * we cannot have the file open and thus the node doing unlink will
863 * take care of deleting the inode.
864 */
865 if (current == osb->dc_task)
866 goto bail;
867
868 spin_lock(&oi->ip_lock);
869 /* OCFS2 *never* deletes system files. This should technically
870 * never get here as system file inodes should always have a
871 * positive link count. */
872 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
873 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
874 (unsigned long long)oi->ip_blkno);
875 goto bail_unlock;
876 }
877
878 ret = 1;
879bail_unlock:
880 spin_unlock(&oi->ip_lock);
881bail:
882 return ret;
883}
884
885/* Query the cluster to determine whether we should wipe an inode from
886 * disk or not.
887 *
888 * Requires the inode to have the cluster lock. */
889static int ocfs2_query_inode_wipe(struct inode *inode,
890 struct buffer_head *di_bh,
891 int *wipe)
892{
893 int status = 0, reason = 0;
894 struct ocfs2_inode_info *oi = OCFS2_I(inode);
895 struct ocfs2_dinode *di;
896
897 *wipe = 0;
898
899 trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
900 inode->i_nlink);
901
902 /* While we were waiting for the cluster lock in
903 * ocfs2_delete_inode, another node might have asked to delete
904 * the inode. Recheck our flags to catch this. */
905 if (!ocfs2_inode_is_valid_to_delete(inode)) {
906 reason = 1;
907 goto bail;
908 }
909
910 /* Now that we have an up to date inode, we can double check
911 * the link count. */
912 if (inode->i_nlink)
913 goto bail;
914
915 /* Do some basic inode verification... */
916 di = (struct ocfs2_dinode *) di_bh->b_data;
917 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
918 !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
919 /*
920 * Inodes in the orphan dir must have ORPHANED_FL. The only
921 * inodes that come back out of the orphan dir are reflink
922 * targets. A reflink target may be moved out of the orphan
923 * dir between the time we scan the directory and the time we
924 * process it. This would lead to HAS_REFCOUNT_FL being set but
925 * ORPHANED_FL not.
926 */
927 if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
928 reason = 2;
929 goto bail;
930 }
931
932 /* for lack of a better error? */
933 status = -EEXIST;
934 mlog(ML_ERROR,
935 "Inode %llu (on-disk %llu) not orphaned! "
936 "Disk flags 0x%x, inode flags 0x%x\n",
937 (unsigned long long)oi->ip_blkno,
938 (unsigned long long)le64_to_cpu(di->i_blkno),
939 le32_to_cpu(di->i_flags), oi->ip_flags);
940 goto bail;
941 }
942
943 /* has someone already deleted us?! baaad... */
944 if (di->i_dtime) {
945 status = -EEXIST;
946 mlog_errno(status);
947 goto bail;
948 }
949
950 /*
951 * This is how ocfs2 determines whether an inode is still live
952 * within the cluster. Every node takes a shared read lock on
953 * the inode open lock in ocfs2_read_locked_inode(). When we
954 * get to ->delete_inode(), each node tries to convert it's
955 * lock to an exclusive. Trylocks are serialized by the inode
956 * meta data lock. If the upconvert succeeds, we know the inode
957 * is no longer live and can be deleted.
958 *
959 * Though we call this with the meta data lock held, the
960 * trylock keeps us from ABBA deadlock.
961 */
962 status = ocfs2_try_open_lock(inode, 1);
963 if (status == -EAGAIN) {
964 status = 0;
965 reason = 3;
966 goto bail;
967 }
968 if (status < 0) {
969 mlog_errno(status);
970 goto bail;
971 }
972
973 *wipe = 1;
974 trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
975
976bail:
977 trace_ocfs2_query_inode_wipe_end(status, reason);
978 return status;
979}
980
981/* Support function for ocfs2_delete_inode. Will help us keep the
982 * inode data in a consistent state for clear_inode. Always truncates
983 * pages, optionally sync's them first. */
984static void ocfs2_cleanup_delete_inode(struct inode *inode,
985 int sync_data)
986{
987 trace_ocfs2_cleanup_delete_inode(
988 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
989 if (sync_data)
990 filemap_write_and_wait(inode->i_mapping);
991 truncate_inode_pages_final(&inode->i_data);
992}
993
994static void ocfs2_delete_inode(struct inode *inode)
995{
996 int wipe, status;
997 sigset_t oldset;
998 struct buffer_head *di_bh = NULL;
999 struct ocfs2_dinode *di = NULL;
1000
1001 trace_ocfs2_delete_inode(inode->i_ino,
1002 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1003 is_bad_inode(inode));
1004
1005 /* When we fail in read_inode() we mark inode as bad. The second test
1006 * catches the case when inode allocation fails before allocating
1007 * a block for inode. */
1008 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
1009 goto bail;
1010
1011 if (!ocfs2_inode_is_valid_to_delete(inode)) {
1012 /* It's probably not necessary to truncate_inode_pages
1013 * here but we do it for safety anyway (it will most
1014 * likely be a no-op anyway) */
1015 ocfs2_cleanup_delete_inode(inode, 0);
1016 goto bail;
1017 }
1018
1019 dquot_initialize(inode);
1020
1021 /* We want to block signals in delete_inode as the lock and
1022 * messaging paths may return us -ERESTARTSYS. Which would
1023 * cause us to exit early, resulting in inodes being orphaned
1024 * forever. */
1025 ocfs2_block_signals(&oldset);
1026
1027 /*
1028 * Synchronize us against ocfs2_get_dentry. We take this in
1029 * shared mode so that all nodes can still concurrently
1030 * process deletes.
1031 */
1032 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1033 if (status < 0) {
1034 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1035 ocfs2_cleanup_delete_inode(inode, 0);
1036 goto bail_unblock;
1037 }
1038 /* Lock down the inode. This gives us an up to date view of
1039 * it's metadata (for verification), and allows us to
1040 * serialize delete_inode on multiple nodes.
1041 *
1042 * Even though we might be doing a truncate, we don't take the
1043 * allocation lock here as it won't be needed - nobody will
1044 * have the file open.
1045 */
1046 status = ocfs2_inode_lock(inode, &di_bh, 1);
1047 if (status < 0) {
1048 if (status != -ENOENT)
1049 mlog_errno(status);
1050 ocfs2_cleanup_delete_inode(inode, 0);
1051 goto bail_unlock_nfs_sync;
1052 }
1053
1054 di = (struct ocfs2_dinode *)di_bh->b_data;
1055 /* Skip inode deletion and wait for dio orphan entry recovered
1056 * first */
1057 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
1058 ocfs2_cleanup_delete_inode(inode, 0);
1059 goto bail_unlock_inode;
1060 }
1061
1062 /* Query the cluster. This will be the final decision made
1063 * before we go ahead and wipe the inode. */
1064 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1065 if (!wipe || status < 0) {
1066 /* Error and remote inode busy both mean we won't be
1067 * removing the inode, so they take almost the same
1068 * path. */
1069 if (status < 0)
1070 mlog_errno(status);
1071
1072 /* Someone in the cluster has disallowed a wipe of
1073 * this inode, or it was never completely
1074 * orphaned. Write out the pages and exit now. */
1075 ocfs2_cleanup_delete_inode(inode, 1);
1076 goto bail_unlock_inode;
1077 }
1078
1079 ocfs2_cleanup_delete_inode(inode, 0);
1080
1081 status = ocfs2_wipe_inode(inode, di_bh);
1082 if (status < 0) {
1083 if (status != -EDEADLK)
1084 mlog_errno(status);
1085 goto bail_unlock_inode;
1086 }
1087
1088 /*
1089 * Mark the inode as successfully deleted.
1090 *
1091 * This is important for ocfs2_clear_inode() as it will check
1092 * this flag and skip any checkpointing work
1093 *
1094 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1095 * the LVB for other nodes.
1096 */
1097 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1098
1099bail_unlock_inode:
1100 ocfs2_inode_unlock(inode, 1);
1101 brelse(di_bh);
1102
1103bail_unlock_nfs_sync:
1104 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1105
1106bail_unblock:
1107 ocfs2_unblock_signals(&oldset);
1108bail:
1109 return;
1110}
1111
1112static void ocfs2_clear_inode(struct inode *inode)
1113{
1114 int status;
1115 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1116 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1117
1118 clear_inode(inode);
1119 trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1120 inode->i_nlink);
1121
1122 mlog_bug_on_msg(osb == NULL,
1123 "Inode=%lu\n", inode->i_ino);
1124
1125 dquot_drop(inode);
1126
1127 /* To preven remote deletes we hold open lock before, now it
1128 * is time to unlock PR and EX open locks. */
1129 ocfs2_open_unlock(inode);
1130
1131 /* Do these before all the other work so that we don't bounce
1132 * the downconvert thread while waiting to destroy the locks. */
1133 ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres);
1134 ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres);
1135 ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres);
1136
1137 ocfs2_resv_discard(&osb->osb_la_resmap,
1138 &oi->ip_la_data_resv);
1139 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1140
1141 /* We very well may get a clear_inode before all an inodes
1142 * metadata has hit disk. Of course, we can't drop any cluster
1143 * locks until the journal has finished with it. The only
1144 * exception here are successfully wiped inodes - their
1145 * metadata can now be considered to be part of the system
1146 * inodes from which it came. */
1147 if (!(oi->ip_flags & OCFS2_INODE_DELETED))
1148 ocfs2_checkpoint_inode(inode);
1149
1150 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1151 "Clear inode of %llu, inode has io markers\n",
1152 (unsigned long long)oi->ip_blkno);
1153 mlog_bug_on_msg(!list_empty(&oi->ip_unwritten_list),
1154 "Clear inode of %llu, inode has unwritten extents\n",
1155 (unsigned long long)oi->ip_blkno);
1156
1157 ocfs2_extent_map_trunc(inode, 0);
1158
1159 status = ocfs2_drop_inode_locks(inode);
1160 if (status < 0)
1161 mlog_errno(status);
1162
1163 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1164 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1165 ocfs2_lock_res_free(&oi->ip_open_lockres);
1166
1167 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1168
1169 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1170 "Clear inode of %llu, inode has %u cache items\n",
1171 (unsigned long long)oi->ip_blkno,
1172 INODE_CACHE(inode)->ci_num_cached);
1173
1174 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1175 "Clear inode of %llu, inode has a bad flag\n",
1176 (unsigned long long)oi->ip_blkno);
1177
1178 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1179 "Clear inode of %llu, inode is locked\n",
1180 (unsigned long long)oi->ip_blkno);
1181
1182 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1183 "Clear inode of %llu, io_mutex is locked\n",
1184 (unsigned long long)oi->ip_blkno);
1185 mutex_unlock(&oi->ip_io_mutex);
1186
1187 /*
1188 * down_trylock() returns 0, down_write_trylock() returns 1
1189 * kernel 1, world 0
1190 */
1191 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1192 "Clear inode of %llu, alloc_sem is locked\n",
1193 (unsigned long long)oi->ip_blkno);
1194 up_write(&oi->ip_alloc_sem);
1195
1196 mlog_bug_on_msg(oi->ip_open_count,
1197 "Clear inode of %llu has open count %d\n",
1198 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1199
1200 /* Clear all other flags. */
1201 oi->ip_flags = 0;
1202 oi->ip_dir_start_lookup = 0;
1203 oi->ip_blkno = 0ULL;
1204
1205 /*
1206 * ip_jinode is used to track txns against this inode. We ensure that
1207 * the journal is flushed before journal shutdown. Thus it is safe to
1208 * have inodes get cleaned up after journal shutdown.
1209 */
1210 jbd2_journal_release_jbd_inode(osb->journal->j_journal,
1211 &oi->ip_jinode);
1212}
1213
1214void ocfs2_evict_inode(struct inode *inode)
1215{
1216 if (!inode->i_nlink ||
1217 (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1218 ocfs2_delete_inode(inode);
1219 } else {
1220 truncate_inode_pages_final(&inode->i_data);
1221 }
1222 ocfs2_clear_inode(inode);
1223}
1224
1225/* Called under inode_lock, with no more references on the
1226 * struct inode, so it's safe here to check the flags field
1227 * and to manipulate i_nlink without any other locks. */
1228int ocfs2_drop_inode(struct inode *inode)
1229{
1230 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1231
1232 trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1233 inode->i_nlink, oi->ip_flags);
1234
1235 assert_spin_locked(&inode->i_lock);
1236 inode->i_state |= I_WILL_FREE;
1237 spin_unlock(&inode->i_lock);
1238 write_inode_now(inode, 1);
1239 spin_lock(&inode->i_lock);
1240 WARN_ON(inode->i_state & I_NEW);
1241 inode->i_state &= ~I_WILL_FREE;
1242
1243 return 1;
1244}
1245
1246/*
1247 * This is called from our getattr.
1248 */
1249int ocfs2_inode_revalidate(struct dentry *dentry)
1250{
1251 struct inode *inode = d_inode(dentry);
1252 int status = 0;
1253
1254 trace_ocfs2_inode_revalidate(inode,
1255 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1256 inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1257
1258 if (!inode) {
1259 status = -ENOENT;
1260 goto bail;
1261 }
1262
1263 spin_lock(&OCFS2_I(inode)->ip_lock);
1264 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1265 spin_unlock(&OCFS2_I(inode)->ip_lock);
1266 status = -ENOENT;
1267 goto bail;
1268 }
1269 spin_unlock(&OCFS2_I(inode)->ip_lock);
1270
1271 /* Let ocfs2_inode_lock do the work of updating our struct
1272 * inode for us. */
1273 status = ocfs2_inode_lock(inode, NULL, 0);
1274 if (status < 0) {
1275 if (status != -ENOENT)
1276 mlog_errno(status);
1277 goto bail;
1278 }
1279 ocfs2_inode_unlock(inode, 0);
1280bail:
1281 return status;
1282}
1283
1284/*
1285 * Updates a disk inode from a
1286 * struct inode.
1287 * Only takes ip_lock.
1288 */
1289int ocfs2_mark_inode_dirty(handle_t *handle,
1290 struct inode *inode,
1291 struct buffer_head *bh)
1292{
1293 int status;
1294 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1295
1296 trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1297
1298 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1299 OCFS2_JOURNAL_ACCESS_WRITE);
1300 if (status < 0) {
1301 mlog_errno(status);
1302 goto leave;
1303 }
1304
1305 spin_lock(&OCFS2_I(inode)->ip_lock);
1306 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1307 ocfs2_get_inode_flags(OCFS2_I(inode));
1308 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1309 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1310 spin_unlock(&OCFS2_I(inode)->ip_lock);
1311
1312 fe->i_size = cpu_to_le64(i_size_read(inode));
1313 ocfs2_set_links_count(fe, inode->i_nlink);
1314 fe->i_uid = cpu_to_le32(i_uid_read(inode));
1315 fe->i_gid = cpu_to_le32(i_gid_read(inode));
1316 fe->i_mode = cpu_to_le16(inode->i_mode);
1317 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1318 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1319 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1320 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1321 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1322 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1323
1324 ocfs2_journal_dirty(handle, bh);
1325 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1326leave:
1327 return status;
1328}
1329
1330/*
1331 *
1332 * Updates a struct inode from a disk inode.
1333 * does no i/o, only takes ip_lock.
1334 */
1335void ocfs2_refresh_inode(struct inode *inode,
1336 struct ocfs2_dinode *fe)
1337{
1338 spin_lock(&OCFS2_I(inode)->ip_lock);
1339
1340 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1341 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1342 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1343 ocfs2_set_inode_flags(inode);
1344 i_size_write(inode, le64_to_cpu(fe->i_size));
1345 set_nlink(inode, ocfs2_read_links_count(fe));
1346 i_uid_write(inode, le32_to_cpu(fe->i_uid));
1347 i_gid_write(inode, le32_to_cpu(fe->i_gid));
1348 inode->i_mode = le16_to_cpu(fe->i_mode);
1349 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1350 inode->i_blocks = 0;
1351 else
1352 inode->i_blocks = ocfs2_inode_sector_count(inode);
1353 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1354 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1355 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1356 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1357 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1358 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1359
1360 spin_unlock(&OCFS2_I(inode)->ip_lock);
1361}
1362
1363int ocfs2_validate_inode_block(struct super_block *sb,
1364 struct buffer_head *bh)
1365{
1366 int rc;
1367 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1368
1369 trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1370
1371 BUG_ON(!buffer_uptodate(bh));
1372
1373 /*
1374 * If the ecc fails, we return the error but otherwise
1375 * leave the filesystem running. We know any error is
1376 * local to this block.
1377 */
1378 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1379 if (rc) {
1380 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1381 (unsigned long long)bh->b_blocknr);
1382 goto bail;
1383 }
1384
1385 /*
1386 * Errors after here are fatal.
1387 */
1388
1389 rc = -EINVAL;
1390
1391 if (!OCFS2_IS_VALID_DINODE(di)) {
1392 rc = ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1393 (unsigned long long)bh->b_blocknr, 7,
1394 di->i_signature);
1395 goto bail;
1396 }
1397
1398 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1399 rc = ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1400 (unsigned long long)bh->b_blocknr,
1401 (unsigned long long)le64_to_cpu(di->i_blkno));
1402 goto bail;
1403 }
1404
1405 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1406 rc = ocfs2_error(sb,
1407 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1408 (unsigned long long)bh->b_blocknr);
1409 goto bail;
1410 }
1411
1412 if (le32_to_cpu(di->i_fs_generation) !=
1413 OCFS2_SB(sb)->fs_generation) {
1414 rc = ocfs2_error(sb,
1415 "Invalid dinode #%llu: fs_generation is %u\n",
1416 (unsigned long long)bh->b_blocknr,
1417 le32_to_cpu(di->i_fs_generation));
1418 goto bail;
1419 }
1420
1421 rc = 0;
1422
1423bail:
1424 return rc;
1425}
1426
1427static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
1428 struct buffer_head *bh)
1429{
1430 int rc = 0;
1431 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1432
1433 trace_ocfs2_filecheck_validate_inode_block(
1434 (unsigned long long)bh->b_blocknr);
1435
1436 BUG_ON(!buffer_uptodate(bh));
1437
1438 /*
1439 * Call ocfs2_validate_meta_ecc() first since it has ecc repair
1440 * function, but we should not return error immediately when ecc
1441 * validation fails, because the reason is quite likely the invalid
1442 * inode number inputed.
1443 */
1444 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1445 if (rc) {
1446 mlog(ML_ERROR,
1447 "Filecheck: checksum failed for dinode %llu\n",
1448 (unsigned long long)bh->b_blocknr);
1449 rc = -OCFS2_FILECHECK_ERR_BLOCKECC;
1450 }
1451
1452 if (!OCFS2_IS_VALID_DINODE(di)) {
1453 mlog(ML_ERROR,
1454 "Filecheck: invalid dinode #%llu: signature = %.*s\n",
1455 (unsigned long long)bh->b_blocknr, 7, di->i_signature);
1456 rc = -OCFS2_FILECHECK_ERR_INVALIDINO;
1457 goto bail;
1458 } else if (rc)
1459 goto bail;
1460
1461 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1462 mlog(ML_ERROR,
1463 "Filecheck: invalid dinode #%llu: i_blkno is %llu\n",
1464 (unsigned long long)bh->b_blocknr,
1465 (unsigned long long)le64_to_cpu(di->i_blkno));
1466 rc = -OCFS2_FILECHECK_ERR_BLOCKNO;
1467 goto bail;
1468 }
1469
1470 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1471 mlog(ML_ERROR,
1472 "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL "
1473 "not set\n",
1474 (unsigned long long)bh->b_blocknr);
1475 rc = -OCFS2_FILECHECK_ERR_VALIDFLAG;
1476 goto bail;
1477 }
1478
1479 if (le32_to_cpu(di->i_fs_generation) !=
1480 OCFS2_SB(sb)->fs_generation) {
1481 mlog(ML_ERROR,
1482 "Filecheck: invalid dinode #%llu: fs_generation is %u\n",
1483 (unsigned long long)bh->b_blocknr,
1484 le32_to_cpu(di->i_fs_generation));
1485 rc = -OCFS2_FILECHECK_ERR_GENERATION;
1486 }
1487
1488bail:
1489 return rc;
1490}
1491
1492static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
1493 struct buffer_head *bh)
1494{
1495 int changed = 0;
1496 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1497
1498 if (!ocfs2_filecheck_validate_inode_block(sb, bh))
1499 return 0;
1500
1501 trace_ocfs2_filecheck_repair_inode_block(
1502 (unsigned long long)bh->b_blocknr);
1503
1504 if (ocfs2_is_hard_readonly(OCFS2_SB(sb)) ||
1505 ocfs2_is_soft_readonly(OCFS2_SB(sb))) {
1506 mlog(ML_ERROR,
1507 "Filecheck: cannot repair dinode #%llu "
1508 "on readonly filesystem\n",
1509 (unsigned long long)bh->b_blocknr);
1510 return -OCFS2_FILECHECK_ERR_READONLY;
1511 }
1512
1513 if (buffer_jbd(bh)) {
1514 mlog(ML_ERROR,
1515 "Filecheck: cannot repair dinode #%llu, "
1516 "its buffer is in jbd\n",
1517 (unsigned long long)bh->b_blocknr);
1518 return -OCFS2_FILECHECK_ERR_INJBD;
1519 }
1520
1521 if (!OCFS2_IS_VALID_DINODE(di)) {
1522 /* Cannot fix invalid inode block */
1523 return -OCFS2_FILECHECK_ERR_INVALIDINO;
1524 }
1525
1526 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1527 /* Cannot just add VALID_FL flag back as a fix,
1528 * need more things to check here.
1529 */
1530 return -OCFS2_FILECHECK_ERR_VALIDFLAG;
1531 }
1532
1533 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1534 di->i_blkno = cpu_to_le64(bh->b_blocknr);
1535 changed = 1;
1536 mlog(ML_ERROR,
1537 "Filecheck: reset dinode #%llu: i_blkno to %llu\n",
1538 (unsigned long long)bh->b_blocknr,
1539 (unsigned long long)le64_to_cpu(di->i_blkno));
1540 }
1541
1542 if (le32_to_cpu(di->i_fs_generation) !=
1543 OCFS2_SB(sb)->fs_generation) {
1544 di->i_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1545 changed = 1;
1546 mlog(ML_ERROR,
1547 "Filecheck: reset dinode #%llu: fs_generation to %u\n",
1548 (unsigned long long)bh->b_blocknr,
1549 le32_to_cpu(di->i_fs_generation));
1550 }
1551
1552 if (changed || ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check)) {
1553 ocfs2_compute_meta_ecc(sb, bh->b_data, &di->i_check);
1554 mark_buffer_dirty(bh);
1555 mlog(ML_ERROR,
1556 "Filecheck: reset dinode #%llu: compute meta ecc\n",
1557 (unsigned long long)bh->b_blocknr);
1558 }
1559
1560 return 0;
1561}
1562
1563static int
1564ocfs2_filecheck_read_inode_block_full(struct inode *inode,
1565 struct buffer_head **bh,
1566 int flags, int type)
1567{
1568 int rc;
1569 struct buffer_head *tmp = *bh;
1570
1571 if (!type) /* Check inode block */
1572 rc = ocfs2_read_blocks(INODE_CACHE(inode),
1573 OCFS2_I(inode)->ip_blkno,
1574 1, &tmp, flags,
1575 ocfs2_filecheck_validate_inode_block);
1576 else /* Repair inode block */
1577 rc = ocfs2_read_blocks(INODE_CACHE(inode),
1578 OCFS2_I(inode)->ip_blkno,
1579 1, &tmp, flags,
1580 ocfs2_filecheck_repair_inode_block);
1581
1582 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1583 if (!rc && !*bh)
1584 *bh = tmp;
1585
1586 return rc;
1587}
1588
1589int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1590 int flags)
1591{
1592 int rc;
1593 struct buffer_head *tmp = *bh;
1594
1595 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1596 1, &tmp, flags, ocfs2_validate_inode_block);
1597
1598 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1599 if (!rc && !*bh)
1600 *bh = tmp;
1601
1602 return rc;
1603}
1604
1605int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1606{
1607 return ocfs2_read_inode_block_full(inode, bh, 0);
1608}
1609
1610
1611static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1612{
1613 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1614
1615 return oi->ip_blkno;
1616}
1617
1618static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1619{
1620 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1621
1622 return oi->vfs_inode.i_sb;
1623}
1624
1625static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1626{
1627 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1628
1629 spin_lock(&oi->ip_lock);
1630}
1631
1632static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1633{
1634 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1635
1636 spin_unlock(&oi->ip_lock);
1637}
1638
1639static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1640{
1641 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1642
1643 mutex_lock(&oi->ip_io_mutex);
1644}
1645
1646static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1647{
1648 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1649
1650 mutex_unlock(&oi->ip_io_mutex);
1651}
1652
1653const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1654 .co_owner = ocfs2_inode_cache_owner,
1655 .co_get_super = ocfs2_inode_cache_get_super,
1656 .co_cache_lock = ocfs2_inode_cache_lock,
1657 .co_cache_unlock = ocfs2_inode_cache_unlock,
1658 .co_io_lock = ocfs2_inode_cache_io_lock,
1659 .co_io_unlock = ocfs2_inode_cache_io_unlock,
1660};
1661
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
59struct 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
67static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
68
69static int ocfs2_read_locked_inode(struct inode *inode,
70 struct ocfs2_find_inode_args *args);
71static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
72static int ocfs2_find_actor(struct inode *inode, void *opaque);
73static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
74 struct inode *inode,
75 struct buffer_head *fe_bh);
76
77void 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 */
98void 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
116struct 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}
127struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
128 int sysfile_type)
129{
130 struct inode *inode = NULL;
131 struct super_block *sb = osb->sb;
132 struct ocfs2_find_inode_args args;
133
134 trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
135 sysfile_type);
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 == NULL) {
157 inode = ERR_PTR(-ENOMEM);
158 mlog_errno(PTR_ERR(inode));
159 goto bail;
160 }
161 trace_ocfs2_iget5_locked(inode->i_state);
162 if (inode->i_state & I_NEW) {
163 ocfs2_read_locked_inode(inode, &args);
164 unlock_new_inode(inode);
165 }
166 if (is_bad_inode(inode)) {
167 iput(inode);
168 inode = ERR_PTR(-ESTALE);
169 goto bail;
170 }
171
172bail:
173 if (!IS_ERR(inode)) {
174 trace_ocfs2_iget_end(inode,
175 (unsigned long long)OCFS2_I(inode)->ip_blkno);
176 }
177
178 return inode;
179}
180
181
182/*
183 * here's how inodes get read from disk:
184 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
185 * found? : return the in-memory inode
186 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
187 */
188
189static int ocfs2_find_actor(struct inode *inode, void *opaque)
190{
191 struct ocfs2_find_inode_args *args = NULL;
192 struct ocfs2_inode_info *oi = OCFS2_I(inode);
193 int ret = 0;
194
195 args = opaque;
196
197 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
198
199 trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
200
201 if (oi->ip_blkno != args->fi_blkno)
202 goto bail;
203
204 ret = 1;
205bail:
206 return ret;
207}
208
209/*
210 * initialize the new inode, but don't do anything that would cause
211 * us to sleep.
212 * return 0 on success, 1 on failure
213 */
214static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
215{
216 struct ocfs2_find_inode_args *args = opaque;
217 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
218 ocfs2_file_ip_alloc_sem_key;
219
220 inode->i_ino = args->fi_ino;
221 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
222 if (args->fi_sysfile_type != 0)
223 lockdep_set_class(&inode->i_mutex,
224 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
225 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
226 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
227 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
228 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
229 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
230 &ocfs2_quota_ip_alloc_sem_key);
231 else
232 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233 &ocfs2_file_ip_alloc_sem_key);
234
235 return 0;
236}
237
238void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
239 int create_ino)
240{
241 struct super_block *sb;
242 struct ocfs2_super *osb;
243 int use_plocks = 1;
244
245 sb = inode->i_sb;
246 osb = OCFS2_SB(sb);
247
248 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
249 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
250 use_plocks = 0;
251
252 /*
253 * These have all been checked by ocfs2_read_inode_block() or set
254 * by ocfs2_mknod_locked(), so a failure is a code bug.
255 */
256 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
257 cannot create a superblock
258 inode today. change if
259 that is needed. */
260 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
261 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
262
263
264 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
265 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
266 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
267
268 inode->i_version = 1;
269 inode->i_generation = le32_to_cpu(fe->i_generation);
270 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
271 inode->i_mode = le16_to_cpu(fe->i_mode);
272 inode->i_uid = le32_to_cpu(fe->i_uid);
273 inode->i_gid = le32_to_cpu(fe->i_gid);
274
275 /* Fast symlinks will have i_size but no allocated clusters. */
276 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
277 inode->i_blocks = 0;
278 else
279 inode->i_blocks = ocfs2_inode_sector_count(inode);
280 inode->i_mapping->a_ops = &ocfs2_aops;
281 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
282 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
283 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
284 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
285 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
286 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
287
288 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
289 mlog(ML_ERROR,
290 "ip_blkno %llu != i_blkno %llu!\n",
291 (unsigned long long)OCFS2_I(inode)->ip_blkno,
292 (unsigned long long)le64_to_cpu(fe->i_blkno));
293
294 inode->i_nlink = ocfs2_read_links_count(fe);
295
296 trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
297 le32_to_cpu(fe->i_flags));
298 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
299 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
300 inode->i_flags |= S_NOQUOTA;
301 }
302
303 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
304 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
305 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
306 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
307 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
308 inode->i_flags |= S_NOQUOTA;
309 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
310 /* we can't actually hit this as read_inode can't
311 * handle superblocks today ;-) */
312 BUG();
313 }
314
315 switch (inode->i_mode & S_IFMT) {
316 case S_IFREG:
317 if (use_plocks)
318 inode->i_fop = &ocfs2_fops;
319 else
320 inode->i_fop = &ocfs2_fops_no_plocks;
321 inode->i_op = &ocfs2_file_iops;
322 i_size_write(inode, le64_to_cpu(fe->i_size));
323 break;
324 case S_IFDIR:
325 inode->i_op = &ocfs2_dir_iops;
326 if (use_plocks)
327 inode->i_fop = &ocfs2_dops;
328 else
329 inode->i_fop = &ocfs2_dops_no_plocks;
330 i_size_write(inode, le64_to_cpu(fe->i_size));
331 OCFS2_I(inode)->ip_dir_lock_gen = 1;
332 break;
333 case S_IFLNK:
334 if (ocfs2_inode_is_fast_symlink(inode))
335 inode->i_op = &ocfs2_fast_symlink_inode_operations;
336 else
337 inode->i_op = &ocfs2_symlink_inode_operations;
338 i_size_write(inode, le64_to_cpu(fe->i_size));
339 break;
340 default:
341 inode->i_op = &ocfs2_special_file_iops;
342 init_special_inode(inode, inode->i_mode,
343 inode->i_rdev);
344 break;
345 }
346
347 if (create_ino) {
348 inode->i_ino = ino_from_blkno(inode->i_sb,
349 le64_to_cpu(fe->i_blkno));
350
351 /*
352 * If we ever want to create system files from kernel,
353 * the generation argument to
354 * ocfs2_inode_lock_res_init() will have to change.
355 */
356 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
357
358 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
359 OCFS2_LOCK_TYPE_META, 0, inode);
360
361 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
362 OCFS2_LOCK_TYPE_OPEN, 0, inode);
363 }
364
365 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
366 OCFS2_LOCK_TYPE_RW, inode->i_generation,
367 inode);
368
369 ocfs2_set_inode_flags(inode);
370
371 OCFS2_I(inode)->ip_last_used_slot = 0;
372 OCFS2_I(inode)->ip_last_used_group = 0;
373
374 if (S_ISDIR(inode->i_mode))
375 ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
376 OCFS2_RESV_FLAG_DIR);
377}
378
379static int ocfs2_read_locked_inode(struct inode *inode,
380 struct ocfs2_find_inode_args *args)
381{
382 struct super_block *sb;
383 struct ocfs2_super *osb;
384 struct ocfs2_dinode *fe;
385 struct buffer_head *bh = NULL;
386 int status, can_lock;
387 u32 generation = 0;
388
389 status = -EINVAL;
390 if (inode == NULL || inode->i_sb == NULL) {
391 mlog(ML_ERROR, "bad inode\n");
392 return status;
393 }
394 sb = inode->i_sb;
395 osb = OCFS2_SB(sb);
396
397 if (!args) {
398 mlog(ML_ERROR, "bad inode args\n");
399 make_bad_inode(inode);
400 return status;
401 }
402
403 /*
404 * To improve performance of cold-cache inode stats, we take
405 * the cluster lock here if possible.
406 *
407 * Generally, OCFS2 never trusts the contents of an inode
408 * unless it's holding a cluster lock, so taking it here isn't
409 * a correctness issue as much as it is a performance
410 * improvement.
411 *
412 * There are three times when taking the lock is not a good idea:
413 *
414 * 1) During startup, before we have initialized the DLM.
415 *
416 * 2) If we are reading certain system files which never get
417 * cluster locks (local alloc, truncate log).
418 *
419 * 3) If the process doing the iget() is responsible for
420 * orphan dir recovery. We're holding the orphan dir lock and
421 * can get into a deadlock with another process on another
422 * node in ->delete_inode().
423 *
424 * #1 and #2 can be simply solved by never taking the lock
425 * here for system files (which are the only type we read
426 * during mount). It's a heavier approach, but our main
427 * concern is user-accessible files anyway.
428 *
429 * #3 works itself out because we'll eventually take the
430 * cluster lock before trusting anything anyway.
431 */
432 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
433 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
434 && !ocfs2_mount_local(osb);
435
436 trace_ocfs2_read_locked_inode(
437 (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
438
439 /*
440 * To maintain backwards compatibility with older versions of
441 * ocfs2-tools, we still store the generation value for system
442 * files. The only ones that actually matter to userspace are
443 * the journals, but it's easier and inexpensive to just flag
444 * all system files similarly.
445 */
446 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
447 generation = osb->fs_generation;
448
449 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
450 OCFS2_LOCK_TYPE_META,
451 generation, inode);
452
453 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
454 OCFS2_LOCK_TYPE_OPEN,
455 0, inode);
456
457 if (can_lock) {
458 status = ocfs2_open_lock(inode);
459 if (status) {
460 make_bad_inode(inode);
461 mlog_errno(status);
462 return status;
463 }
464 status = ocfs2_inode_lock(inode, NULL, 0);
465 if (status) {
466 make_bad_inode(inode);
467 mlog_errno(status);
468 return status;
469 }
470 }
471
472 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
473 status = ocfs2_try_open_lock(inode, 0);
474 if (status) {
475 make_bad_inode(inode);
476 return status;
477 }
478 }
479
480 if (can_lock) {
481 status = ocfs2_read_inode_block_full(inode, &bh,
482 OCFS2_BH_IGNORE_CACHE);
483 } else {
484 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
485 /*
486 * If buffer is in jbd, then its checksum may not have been
487 * computed as yet.
488 */
489 if (!status && !buffer_jbd(bh))
490 status = ocfs2_validate_inode_block(osb->sb, bh);
491 }
492 if (status < 0) {
493 mlog_errno(status);
494 goto bail;
495 }
496
497 status = -EINVAL;
498 fe = (struct ocfs2_dinode *) bh->b_data;
499
500 /*
501 * This is a code bug. Right now the caller needs to
502 * understand whether it is asking for a system file inode or
503 * not so the proper lock names can be built.
504 */
505 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
506 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
507 "Inode %llu: system file state is ambigous\n",
508 (unsigned long long)args->fi_blkno);
509
510 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
511 S_ISBLK(le16_to_cpu(fe->i_mode)))
512 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
513
514 ocfs2_populate_inode(inode, fe, 0);
515
516 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
517
518 status = 0;
519
520bail:
521 if (can_lock)
522 ocfs2_inode_unlock(inode, 0);
523
524 if (status < 0)
525 make_bad_inode(inode);
526
527 if (args && bh)
528 brelse(bh);
529
530 return status;
531}
532
533void ocfs2_sync_blockdev(struct super_block *sb)
534{
535 sync_blockdev(sb->s_bdev);
536}
537
538static 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_dinode *fe;
544 handle_t *handle = NULL;
545
546 fe = (struct ocfs2_dinode *) fe_bh->b_data;
547
548 /*
549 * This check will also skip truncate of inodes with inline
550 * data and fast symlinks.
551 */
552 if (fe->i_clusters) {
553 if (ocfs2_should_order_data(inode))
554 ocfs2_begin_ordered_truncate(inode, 0);
555
556 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
557 if (IS_ERR(handle)) {
558 status = PTR_ERR(handle);
559 handle = NULL;
560 mlog_errno(status);
561 goto out;
562 }
563
564 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
565 fe_bh,
566 OCFS2_JOURNAL_ACCESS_WRITE);
567 if (status < 0) {
568 mlog_errno(status);
569 goto out;
570 }
571
572 i_size_write(inode, 0);
573
574 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
575 if (status < 0) {
576 mlog_errno(status);
577 goto out;
578 }
579
580 ocfs2_commit_trans(osb, handle);
581 handle = NULL;
582
583 status = ocfs2_commit_truncate(osb, inode, fe_bh);
584 if (status < 0) {
585 mlog_errno(status);
586 goto out;
587 }
588 }
589
590out:
591 if (handle)
592 ocfs2_commit_trans(osb, handle);
593 return status;
594}
595
596static int ocfs2_remove_inode(struct inode *inode,
597 struct buffer_head *di_bh,
598 struct inode *orphan_dir_inode,
599 struct buffer_head *orphan_dir_bh)
600{
601 int status;
602 struct inode *inode_alloc_inode = NULL;
603 struct buffer_head *inode_alloc_bh = NULL;
604 handle_t *handle;
605 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
606 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
607
608 inode_alloc_inode =
609 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
610 le16_to_cpu(di->i_suballoc_slot));
611 if (!inode_alloc_inode) {
612 status = -EEXIST;
613 mlog_errno(status);
614 goto bail;
615 }
616
617 mutex_lock(&inode_alloc_inode->i_mutex);
618 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
619 if (status < 0) {
620 mutex_unlock(&inode_alloc_inode->i_mutex);
621
622 mlog_errno(status);
623 goto bail;
624 }
625
626 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
627 ocfs2_quota_trans_credits(inode->i_sb));
628 if (IS_ERR(handle)) {
629 status = PTR_ERR(handle);
630 mlog_errno(status);
631 goto bail_unlock;
632 }
633
634 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
635 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
636 orphan_dir_bh);
637 if (status < 0) {
638 mlog_errno(status);
639 goto bail_commit;
640 }
641 }
642
643 /* set the inodes dtime */
644 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
645 OCFS2_JOURNAL_ACCESS_WRITE);
646 if (status < 0) {
647 mlog_errno(status);
648 goto bail_commit;
649 }
650
651 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
652 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
653 ocfs2_journal_dirty(handle, di_bh);
654
655 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
656 dquot_free_inode(inode);
657
658 status = ocfs2_free_dinode(handle, inode_alloc_inode,
659 inode_alloc_bh, di);
660 if (status < 0)
661 mlog_errno(status);
662
663bail_commit:
664 ocfs2_commit_trans(osb, handle);
665bail_unlock:
666 ocfs2_inode_unlock(inode_alloc_inode, 1);
667 mutex_unlock(&inode_alloc_inode->i_mutex);
668 brelse(inode_alloc_bh);
669bail:
670 iput(inode_alloc_inode);
671
672 return status;
673}
674
675/*
676 * Serialize with orphan dir recovery. If the process doing
677 * recovery on this orphan dir does an iget() with the dir
678 * i_mutex held, we'll deadlock here. Instead we detect this
679 * and exit early - recovery will wipe this inode for us.
680 */
681static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
682 int slot)
683{
684 int ret = 0;
685
686 spin_lock(&osb->osb_lock);
687 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
688 ret = -EDEADLK;
689 goto out;
690 }
691 /* This signals to the orphan recovery process that it should
692 * wait for us to handle the wipe. */
693 osb->osb_orphan_wipes[slot]++;
694out:
695 spin_unlock(&osb->osb_lock);
696 trace_ocfs2_check_orphan_recovery_state(slot, ret);
697 return ret;
698}
699
700static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
701 int slot)
702{
703 spin_lock(&osb->osb_lock);
704 osb->osb_orphan_wipes[slot]--;
705 spin_unlock(&osb->osb_lock);
706
707 wake_up(&osb->osb_wipe_event);
708}
709
710static int ocfs2_wipe_inode(struct inode *inode,
711 struct buffer_head *di_bh)
712{
713 int status, orphaned_slot = -1;
714 struct inode *orphan_dir_inode = NULL;
715 struct buffer_head *orphan_dir_bh = NULL;
716 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
717 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
718
719 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
720 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
721
722 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
723 if (status)
724 return status;
725
726 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
727 ORPHAN_DIR_SYSTEM_INODE,
728 orphaned_slot);
729 if (!orphan_dir_inode) {
730 status = -EEXIST;
731 mlog_errno(status);
732 goto bail;
733 }
734
735 /* Lock the orphan dir. The lock will be held for the entire
736 * delete_inode operation. We do this now to avoid races with
737 * recovery completion on other nodes. */
738 mutex_lock(&orphan_dir_inode->i_mutex);
739 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
740 if (status < 0) {
741 mutex_unlock(&orphan_dir_inode->i_mutex);
742
743 mlog_errno(status);
744 goto bail;
745 }
746 }
747
748 /* we do this while holding the orphan dir lock because we
749 * don't want recovery being run from another node to try an
750 * inode delete underneath us -- this will result in two nodes
751 * truncating the same file! */
752 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
753 if (status < 0) {
754 mlog_errno(status);
755 goto bail_unlock_dir;
756 }
757
758 /* Remove any dir index tree */
759 if (S_ISDIR(inode->i_mode)) {
760 status = ocfs2_dx_dir_truncate(inode, di_bh);
761 if (status) {
762 mlog_errno(status);
763 goto bail_unlock_dir;
764 }
765 }
766
767 /*Free extended attribute resources associated with this inode.*/
768 status = ocfs2_xattr_remove(inode, di_bh);
769 if (status < 0) {
770 mlog_errno(status);
771 goto bail_unlock_dir;
772 }
773
774 status = ocfs2_remove_refcount_tree(inode, di_bh);
775 if (status < 0) {
776 mlog_errno(status);
777 goto bail_unlock_dir;
778 }
779
780 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
781 orphan_dir_bh);
782 if (status < 0)
783 mlog_errno(status);
784
785bail_unlock_dir:
786 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
787 return status;
788
789 ocfs2_inode_unlock(orphan_dir_inode, 1);
790 mutex_unlock(&orphan_dir_inode->i_mutex);
791 brelse(orphan_dir_bh);
792bail:
793 iput(orphan_dir_inode);
794 ocfs2_signal_wipe_completion(osb, orphaned_slot);
795
796 return status;
797}
798
799/* There is a series of simple checks that should be done before a
800 * trylock is even considered. Encapsulate those in this function. */
801static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
802{
803 int ret = 0;
804 struct ocfs2_inode_info *oi = OCFS2_I(inode);
805 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
806
807 trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
808 (unsigned long long)oi->ip_blkno,
809 oi->ip_flags);
810
811 /* We shouldn't be getting here for the root directory
812 * inode.. */
813 if (inode == osb->root_inode) {
814 mlog(ML_ERROR, "Skipping delete of root inode.\n");
815 goto bail;
816 }
817
818 /* If we're coming from downconvert_thread we can't go into our own
819 * voting [hello, deadlock city!], so unforuntately we just
820 * have to skip deleting this guy. That's OK though because
821 * the node who's doing the actual deleting should handle it
822 * anyway. */
823 if (current == osb->dc_task)
824 goto bail;
825
826 spin_lock(&oi->ip_lock);
827 /* OCFS2 *never* deletes system files. This should technically
828 * never get here as system file inodes should always have a
829 * positive link count. */
830 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
831 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
832 (unsigned long long)oi->ip_blkno);
833 goto bail_unlock;
834 }
835
836 /* If we have allowd wipe of this inode for another node, it
837 * will be marked here so we can safely skip it. Recovery will
838 * cleanup any inodes we might inadvertently skip here. */
839 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE)
840 goto bail_unlock;
841
842 ret = 1;
843bail_unlock:
844 spin_unlock(&oi->ip_lock);
845bail:
846 return ret;
847}
848
849/* Query the cluster to determine whether we should wipe an inode from
850 * disk or not.
851 *
852 * Requires the inode to have the cluster lock. */
853static int ocfs2_query_inode_wipe(struct inode *inode,
854 struct buffer_head *di_bh,
855 int *wipe)
856{
857 int status = 0, reason = 0;
858 struct ocfs2_inode_info *oi = OCFS2_I(inode);
859 struct ocfs2_dinode *di;
860
861 *wipe = 0;
862
863 trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
864 inode->i_nlink);
865
866 /* While we were waiting for the cluster lock in
867 * ocfs2_delete_inode, another node might have asked to delete
868 * the inode. Recheck our flags to catch this. */
869 if (!ocfs2_inode_is_valid_to_delete(inode)) {
870 reason = 1;
871 goto bail;
872 }
873
874 /* Now that we have an up to date inode, we can double check
875 * the link count. */
876 if (inode->i_nlink)
877 goto bail;
878
879 /* Do some basic inode verification... */
880 di = (struct ocfs2_dinode *) di_bh->b_data;
881 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
882 !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
883 /*
884 * Inodes in the orphan dir must have ORPHANED_FL. The only
885 * inodes that come back out of the orphan dir are reflink
886 * targets. A reflink target may be moved out of the orphan
887 * dir between the time we scan the directory and the time we
888 * process it. This would lead to HAS_REFCOUNT_FL being set but
889 * ORPHANED_FL not.
890 */
891 if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
892 reason = 2;
893 goto bail;
894 }
895
896 /* for lack of a better error? */
897 status = -EEXIST;
898 mlog(ML_ERROR,
899 "Inode %llu (on-disk %llu) not orphaned! "
900 "Disk flags 0x%x, inode flags 0x%x\n",
901 (unsigned long long)oi->ip_blkno,
902 (unsigned long long)le64_to_cpu(di->i_blkno),
903 le32_to_cpu(di->i_flags), oi->ip_flags);
904 goto bail;
905 }
906
907 /* has someone already deleted us?! baaad... */
908 if (di->i_dtime) {
909 status = -EEXIST;
910 mlog_errno(status);
911 goto bail;
912 }
913
914 /*
915 * This is how ocfs2 determines whether an inode is still live
916 * within the cluster. Every node takes a shared read lock on
917 * the inode open lock in ocfs2_read_locked_inode(). When we
918 * get to ->delete_inode(), each node tries to convert it's
919 * lock to an exclusive. Trylocks are serialized by the inode
920 * meta data lock. If the upconvert succeeds, we know the inode
921 * is no longer live and can be deleted.
922 *
923 * Though we call this with the meta data lock held, the
924 * trylock keeps us from ABBA deadlock.
925 */
926 status = ocfs2_try_open_lock(inode, 1);
927 if (status == -EAGAIN) {
928 status = 0;
929 reason = 3;
930 goto bail;
931 }
932 if (status < 0) {
933 mlog_errno(status);
934 goto bail;
935 }
936
937 *wipe = 1;
938 trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
939
940bail:
941 trace_ocfs2_query_inode_wipe_end(status, reason);
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. */
948static void ocfs2_cleanup_delete_inode(struct inode *inode,
949 int sync_data)
950{
951 trace_ocfs2_cleanup_delete_inode(
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
958static void ocfs2_delete_inode(struct inode *inode)
959{
960 int wipe, status;
961 sigset_t oldset;
962 struct buffer_head *di_bh = NULL;
963
964 trace_ocfs2_delete_inode(inode->i_ino,
965 (unsigned long long)OCFS2_I(inode)->ip_blkno,
966 is_bad_inode(inode));
967
968 /* When we fail in read_inode() we mark inode as bad. The second test
969 * catches the case when inode allocation fails before allocating
970 * a block for inode. */
971 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
972 goto bail;
973
974 dquot_initialize(inode);
975
976 if (!ocfs2_inode_is_valid_to_delete(inode)) {
977 /* It's probably not necessary to truncate_inode_pages
978 * here but we do it for safety anyway (it will most
979 * likely be a no-op anyway) */
980 ocfs2_cleanup_delete_inode(inode, 0);
981 goto bail;
982 }
983
984 /* We want to block signals in delete_inode as the lock and
985 * messaging paths may return us -ERESTARTSYS. Which would
986 * cause us to exit early, resulting in inodes being orphaned
987 * forever. */
988 ocfs2_block_signals(&oldset);
989
990 /*
991 * Synchronize us against ocfs2_get_dentry. We take this in
992 * shared mode so that all nodes can still concurrently
993 * process deletes.
994 */
995 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
996 if (status < 0) {
997 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
998 ocfs2_cleanup_delete_inode(inode, 0);
999 goto bail_unblock;
1000 }
1001 /* Lock down the inode. This gives us an up to date view of
1002 * it's metadata (for verification), and allows us to
1003 * serialize delete_inode on multiple nodes.
1004 *
1005 * Even though we might be doing a truncate, we don't take the
1006 * allocation lock here as it won't be needed - nobody will
1007 * have the file open.
1008 */
1009 status = ocfs2_inode_lock(inode, &di_bh, 1);
1010 if (status < 0) {
1011 if (status != -ENOENT)
1012 mlog_errno(status);
1013 ocfs2_cleanup_delete_inode(inode, 0);
1014 goto bail_unlock_nfs_sync;
1015 }
1016
1017 /* Query the cluster. This will be the final decision made
1018 * before we go ahead and wipe the inode. */
1019 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1020 if (!wipe || status < 0) {
1021 /* Error and remote inode busy both mean we won't be
1022 * removing the inode, so they take almost the same
1023 * path. */
1024 if (status < 0)
1025 mlog_errno(status);
1026
1027 /* Someone in the cluster has disallowed a wipe of
1028 * this inode, or it was never completely
1029 * orphaned. Write out the pages and exit now. */
1030 ocfs2_cleanup_delete_inode(inode, 1);
1031 goto bail_unlock_inode;
1032 }
1033
1034 ocfs2_cleanup_delete_inode(inode, 0);
1035
1036 status = ocfs2_wipe_inode(inode, di_bh);
1037 if (status < 0) {
1038 if (status != -EDEADLK)
1039 mlog_errno(status);
1040 goto bail_unlock_inode;
1041 }
1042
1043 /*
1044 * Mark the inode as successfully deleted.
1045 *
1046 * This is important for ocfs2_clear_inode() as it will check
1047 * this flag and skip any checkpointing work
1048 *
1049 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1050 * the LVB for other nodes.
1051 */
1052 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1053
1054bail_unlock_inode:
1055 ocfs2_inode_unlock(inode, 1);
1056 brelse(di_bh);
1057
1058bail_unlock_nfs_sync:
1059 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1060
1061bail_unblock:
1062 ocfs2_unblock_signals(&oldset);
1063bail:
1064 return;
1065}
1066
1067static void ocfs2_clear_inode(struct inode *inode)
1068{
1069 int status;
1070 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1071
1072 end_writeback(inode);
1073 trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1074 inode->i_nlink);
1075
1076 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1077 "Inode=%lu\n", inode->i_ino);
1078
1079 dquot_drop(inode);
1080
1081 /* To preven remote deletes we hold open lock before, now it
1082 * is time to unlock PR and EX open locks. */
1083 ocfs2_open_unlock(inode);
1084
1085 /* Do these before all the other work so that we don't bounce
1086 * the downconvert thread while waiting to destroy the locks. */
1087 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1088 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1089 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1090
1091 ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1092 &oi->ip_la_data_resv);
1093 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1094
1095 /* We very well may get a clear_inode before all an inodes
1096 * metadata has hit disk. Of course, we can't drop any cluster
1097 * locks until the journal has finished with it. The only
1098 * exception here are successfully wiped inodes - their
1099 * metadata can now be considered to be part of the system
1100 * inodes from which it came. */
1101 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1102 ocfs2_checkpoint_inode(inode);
1103
1104 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1105 "Clear inode of %llu, inode has io markers\n",
1106 (unsigned long long)oi->ip_blkno);
1107
1108 ocfs2_extent_map_trunc(inode, 0);
1109
1110 status = ocfs2_drop_inode_locks(inode);
1111 if (status < 0)
1112 mlog_errno(status);
1113
1114 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1115 ocfs2_lock_res_free(&oi->ip_inode_lockres);
1116 ocfs2_lock_res_free(&oi->ip_open_lockres);
1117
1118 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1119
1120 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1121 "Clear inode of %llu, inode has %u cache items\n",
1122 (unsigned long long)oi->ip_blkno,
1123 INODE_CACHE(inode)->ci_num_cached);
1124
1125 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1126 "Clear inode of %llu, inode has a bad flag\n",
1127 (unsigned long long)oi->ip_blkno);
1128
1129 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1130 "Clear inode of %llu, inode is locked\n",
1131 (unsigned long long)oi->ip_blkno);
1132
1133 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1134 "Clear inode of %llu, io_mutex is locked\n",
1135 (unsigned long long)oi->ip_blkno);
1136 mutex_unlock(&oi->ip_io_mutex);
1137
1138 /*
1139 * down_trylock() returns 0, down_write_trylock() returns 1
1140 * kernel 1, world 0
1141 */
1142 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1143 "Clear inode of %llu, alloc_sem is locked\n",
1144 (unsigned long long)oi->ip_blkno);
1145 up_write(&oi->ip_alloc_sem);
1146
1147 mlog_bug_on_msg(oi->ip_open_count,
1148 "Clear inode of %llu has open count %d\n",
1149 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1150
1151 /* Clear all other flags. */
1152 oi->ip_flags = 0;
1153 oi->ip_dir_start_lookup = 0;
1154 oi->ip_blkno = 0ULL;
1155
1156 /*
1157 * ip_jinode is used to track txns against this inode. We ensure that
1158 * the journal is flushed before journal shutdown. Thus it is safe to
1159 * have inodes get cleaned up after journal shutdown.
1160 */
1161 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1162 &oi->ip_jinode);
1163}
1164
1165void ocfs2_evict_inode(struct inode *inode)
1166{
1167 if (!inode->i_nlink ||
1168 (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1169 ocfs2_delete_inode(inode);
1170 } else {
1171 truncate_inode_pages(&inode->i_data, 0);
1172 }
1173 ocfs2_clear_inode(inode);
1174}
1175
1176/* Called under inode_lock, with no more references on the
1177 * struct inode, so it's safe here to check the flags field
1178 * and to manipulate i_nlink without any other locks. */
1179int ocfs2_drop_inode(struct inode *inode)
1180{
1181 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1182 int res;
1183
1184 trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1185 inode->i_nlink, oi->ip_flags);
1186
1187 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1188 res = 1;
1189 else
1190 res = generic_drop_inode(inode);
1191
1192 return res;
1193}
1194
1195/*
1196 * This is called from our getattr.
1197 */
1198int ocfs2_inode_revalidate(struct dentry *dentry)
1199{
1200 struct inode *inode = dentry->d_inode;
1201 int status = 0;
1202
1203 trace_ocfs2_inode_revalidate(inode,
1204 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1205 inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1206
1207 if (!inode) {
1208 status = -ENOENT;
1209 goto bail;
1210 }
1211
1212 spin_lock(&OCFS2_I(inode)->ip_lock);
1213 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1214 spin_unlock(&OCFS2_I(inode)->ip_lock);
1215 status = -ENOENT;
1216 goto bail;
1217 }
1218 spin_unlock(&OCFS2_I(inode)->ip_lock);
1219
1220 /* Let ocfs2_inode_lock do the work of updating our struct
1221 * inode for us. */
1222 status = ocfs2_inode_lock(inode, NULL, 0);
1223 if (status < 0) {
1224 if (status != -ENOENT)
1225 mlog_errno(status);
1226 goto bail;
1227 }
1228 ocfs2_inode_unlock(inode, 0);
1229bail:
1230 return status;
1231}
1232
1233/*
1234 * Updates a disk inode from a
1235 * struct inode.
1236 * Only takes ip_lock.
1237 */
1238int ocfs2_mark_inode_dirty(handle_t *handle,
1239 struct inode *inode,
1240 struct buffer_head *bh)
1241{
1242 int status;
1243 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1244
1245 trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1246
1247 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1248 OCFS2_JOURNAL_ACCESS_WRITE);
1249 if (status < 0) {
1250 mlog_errno(status);
1251 goto leave;
1252 }
1253
1254 spin_lock(&OCFS2_I(inode)->ip_lock);
1255 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1256 ocfs2_get_inode_flags(OCFS2_I(inode));
1257 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1258 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1259 spin_unlock(&OCFS2_I(inode)->ip_lock);
1260
1261 fe->i_size = cpu_to_le64(i_size_read(inode));
1262 ocfs2_set_links_count(fe, inode->i_nlink);
1263 fe->i_uid = cpu_to_le32(inode->i_uid);
1264 fe->i_gid = cpu_to_le32(inode->i_gid);
1265 fe->i_mode = cpu_to_le16(inode->i_mode);
1266 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1267 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1268 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1269 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1270 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1271 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1272
1273 ocfs2_journal_dirty(handle, bh);
1274leave:
1275 return status;
1276}
1277
1278/*
1279 *
1280 * Updates a struct inode from a disk inode.
1281 * does no i/o, only takes ip_lock.
1282 */
1283void ocfs2_refresh_inode(struct inode *inode,
1284 struct ocfs2_dinode *fe)
1285{
1286 spin_lock(&OCFS2_I(inode)->ip_lock);
1287
1288 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1289 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1290 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1291 ocfs2_set_inode_flags(inode);
1292 i_size_write(inode, le64_to_cpu(fe->i_size));
1293 inode->i_nlink = ocfs2_read_links_count(fe);
1294 inode->i_uid = le32_to_cpu(fe->i_uid);
1295 inode->i_gid = le32_to_cpu(fe->i_gid);
1296 inode->i_mode = le16_to_cpu(fe->i_mode);
1297 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1298 inode->i_blocks = 0;
1299 else
1300 inode->i_blocks = ocfs2_inode_sector_count(inode);
1301 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1302 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1303 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1304 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1305 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1306 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1307
1308 spin_unlock(&OCFS2_I(inode)->ip_lock);
1309}
1310
1311int ocfs2_validate_inode_block(struct super_block *sb,
1312 struct buffer_head *bh)
1313{
1314 int rc;
1315 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1316
1317 trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1318
1319 BUG_ON(!buffer_uptodate(bh));
1320
1321 /*
1322 * If the ecc fails, we return the error but otherwise
1323 * leave the filesystem running. We know any error is
1324 * local to this block.
1325 */
1326 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1327 if (rc) {
1328 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1329 (unsigned long long)bh->b_blocknr);
1330 goto bail;
1331 }
1332
1333 /*
1334 * Errors after here are fatal.
1335 */
1336
1337 rc = -EINVAL;
1338
1339 if (!OCFS2_IS_VALID_DINODE(di)) {
1340 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1341 (unsigned long long)bh->b_blocknr, 7,
1342 di->i_signature);
1343 goto bail;
1344 }
1345
1346 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1347 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1348 (unsigned long long)bh->b_blocknr,
1349 (unsigned long long)le64_to_cpu(di->i_blkno));
1350 goto bail;
1351 }
1352
1353 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1354 ocfs2_error(sb,
1355 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1356 (unsigned long long)bh->b_blocknr);
1357 goto bail;
1358 }
1359
1360 if (le32_to_cpu(di->i_fs_generation) !=
1361 OCFS2_SB(sb)->fs_generation) {
1362 ocfs2_error(sb,
1363 "Invalid dinode #%llu: fs_generation is %u\n",
1364 (unsigned long long)bh->b_blocknr,
1365 le32_to_cpu(di->i_fs_generation));
1366 goto bail;
1367 }
1368
1369 rc = 0;
1370
1371bail:
1372 return rc;
1373}
1374
1375int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1376 int flags)
1377{
1378 int rc;
1379 struct buffer_head *tmp = *bh;
1380
1381 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1382 1, &tmp, flags, ocfs2_validate_inode_block);
1383
1384 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1385 if (!rc && !*bh)
1386 *bh = tmp;
1387
1388 return rc;
1389}
1390
1391int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1392{
1393 return ocfs2_read_inode_block_full(inode, bh, 0);
1394}
1395
1396
1397static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1398{
1399 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1400
1401 return oi->ip_blkno;
1402}
1403
1404static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1405{
1406 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1407
1408 return oi->vfs_inode.i_sb;
1409}
1410
1411static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1412{
1413 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1414
1415 spin_lock(&oi->ip_lock);
1416}
1417
1418static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1419{
1420 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1421
1422 spin_unlock(&oi->ip_lock);
1423}
1424
1425static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1426{
1427 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1428
1429 mutex_lock(&oi->ip_io_mutex);
1430}
1431
1432static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1433{
1434 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1435
1436 mutex_unlock(&oi->ip_io_mutex);
1437}
1438
1439const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1440 .co_owner = ocfs2_inode_cache_owner,
1441 .co_get_super = ocfs2_inode_cache_get_super,
1442 .co_cache_lock = ocfs2_inode_cache_lock,
1443 .co_cache_unlock = ocfs2_inode_cache_unlock,
1444 .co_io_lock = ocfs2_inode_cache_io_lock,
1445 .co_io_unlock = ocfs2_inode_cache_io_unlock,
1446};
1447