?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.36-rc5
[linux-2.6/next.git] / fs / ext3 / super.c
blob5dbf4dba03c4d99240edd76de5e9a0a2bba4bd04
1 /*
2 * linux/fs/ext3/super.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/exportfs.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <linux/seq_file.h>
39 #include <linux/log2.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
48 #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
49 #else
50 #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
51 #endif
52
53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
54 unsigned long journal_devnum);
55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
56 unsigned int);
57 static int ext3_commit_super(struct super_block *sb,
58 struct ext3_super_block *es,
59 int sync);
60 static void ext3_mark_recovery_complete(struct super_block * sb,
61 struct ext3_super_block * es);
62 static void ext3_clear_journal_err(struct super_block * sb,
63 struct ext3_super_block * es);
64 static int ext3_sync_fs(struct super_block *sb, int wait);
65 static const char *ext3_decode_error(struct super_block * sb, int errno,
66 char nbuf[16]);
67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
69 static int ext3_unfreeze(struct super_block *sb);
70 static int ext3_freeze(struct super_block *sb);
71
72 /*
73 * Wrappers for journal_start/end.
74 *
75 * The only special thing we need to do here is to make sure that all
76 * journal_end calls result in the superblock being marked dirty, so
77 * that sync() will call the filesystem's write_super callback if
78 * appropriate.
79 */
80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
81 {
82 journal_t *journal;
83
84 if (sb->s_flags & MS_RDONLY)
85 return ERR_PTR(-EROFS);
86
87 /* Special case here: if the journal has aborted behind our
88 * backs (eg. EIO in the commit thread), then we still need to
89 * take the FS itself readonly cleanly. */
90 journal = EXT3_SB(sb)->s_journal;
91 if (is_journal_aborted(journal)) {
92 ext3_abort(sb, __func__,
93 "Detected aborted journal");
94 return ERR_PTR(-EROFS);
95 }
96
97 return journal_start(journal, nblocks);
98 }
99
100 /*
101 * The only special thing we need to do here is to make sure that all
102 * journal_stop calls result in the superblock being marked dirty, so
103 * that sync() will call the filesystem's write_super callback if
104 * appropriate.
105 */
106 int __ext3_journal_stop(const char *where, handle_t *handle)
107 {
108 struct super_block *sb;
109 int err;
110 int rc;
111
112 sb = handle->h_transaction->t_journal->j_private;
113 err = handle->h_err;
114 rc = journal_stop(handle);
115
116 if (!err)
117 err = rc;
118 if (err)
119 __ext3_std_error(sb, where, err);
120 return err;
121 }
122
123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
124 struct buffer_head *bh, handle_t *handle, int err)
125 {
126 char nbuf[16];
127 const char *errstr = ext3_decode_error(NULL, err, nbuf);
128
129 if (bh)
130 BUFFER_TRACE(bh, "abort");
131
132 if (!handle->h_err)
133 handle->h_err = err;
134
135 if (is_handle_aborted(handle))
136 return;
137
138 printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
139 caller, errstr, err_fn);
140
141 journal_abort_handle(handle);
142 }
143
144 void ext3_msg(struct super_block *sb, const char *prefix,
145 const char *fmt, ...)
146 {
147 va_list args;
148
149 va_start(args, fmt);
150 printk("%sEXT3-fs (%s): ", prefix, sb->s_id);
151 vprintk(fmt, args);
152 printk("\n");
153 va_end(args);
154 }
155
156 /* Deal with the reporting of failure conditions on a filesystem such as
157 * inconsistencies detected or read IO failures.
158 *
159 * On ext2, we can store the error state of the filesystem in the
160 * superblock. That is not possible on ext3, because we may have other
161 * write ordering constraints on the superblock which prevent us from
162 * writing it out straight away; and given that the journal is about to
163 * be aborted, we can't rely on the current, or future, transactions to
164 * write out the superblock safely.
165 *
166 * We'll just use the journal_abort() error code to record an error in
167 * the journal instead. On recovery, the journal will complain about
168 * that error until we've noted it down and cleared it.
169 */
170
171 static void ext3_handle_error(struct super_block *sb)
172 {
173 struct ext3_super_block *es = EXT3_SB(sb)->s_es;
174
175 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
176 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
177
178 if (sb->s_flags & MS_RDONLY)
179 return;
180
181 if (!test_opt (sb, ERRORS_CONT)) {
182 journal_t *journal = EXT3_SB(sb)->s_journal;
183
184 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
185 if (journal)
186 journal_abort(journal, -EIO);
187 }
188 if (test_opt (sb, ERRORS_RO)) {
189 ext3_msg(sb, KERN_CRIT,
190 "error: remounting filesystem read-only");
191 sb->s_flags |= MS_RDONLY;
192 }
193 ext3_commit_super(sb, es, 1);
194 if (test_opt(sb, ERRORS_PANIC))
195 panic("EXT3-fs (%s): panic forced after error\n",
196 sb->s_id);
197 }
198
199 void ext3_error (struct super_block * sb, const char * function,
200 const char * fmt, ...)
201 {
202 va_list args;
203
204 va_start(args, fmt);
205 printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
206 vprintk(fmt, args);
207 printk("\n");
208 va_end(args);
209
210 ext3_handle_error(sb);
211 }
212
213 static const char *ext3_decode_error(struct super_block * sb, int errno,
214 char nbuf[16])
215 {
216 char *errstr = NULL;
217
218 switch (errno) {
219 case -EIO:
220 errstr = "IO failure";
221 break;
222 case -ENOMEM:
223 errstr = "Out of memory";
224 break;
225 case -EROFS:
226 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
227 errstr = "Journal has aborted";
228 else
229 errstr = "Readonly filesystem";
230 break;
231 default:
232 /* If the caller passed in an extra buffer for unknown
233 * errors, textualise them now. Else we just return
234 * NULL. */
235 if (nbuf) {
236 /* Check for truncated error codes... */
237 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
238 errstr = nbuf;
239 }
240 break;
241 }
242
243 return errstr;
244 }
245
246 /* __ext3_std_error decodes expected errors from journaling functions
247 * automatically and invokes the appropriate error response. */
248
249 void __ext3_std_error (struct super_block * sb, const char * function,
250 int errno)
251 {
252 char nbuf[16];
253 const char *errstr;
254
255 /* Special case: if the error is EROFS, and we're not already
256 * inside a transaction, then there's really no point in logging
257 * an error. */
258 if (errno == -EROFS && journal_current_handle() == NULL &&
259 (sb->s_flags & MS_RDONLY))
260 return;
261
262 errstr = ext3_decode_error(sb, errno, nbuf);
263 ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
264
265 ext3_handle_error(sb);
266 }
267
268 /*
269 * ext3_abort is a much stronger failure handler than ext3_error. The
270 * abort function may be used to deal with unrecoverable failures such
271 * as journal IO errors or ENOMEM at a critical moment in log management.
272 *
273 * We unconditionally force the filesystem into an ABORT|READONLY state,
274 * unless the error response on the fs has been set to panic in which
275 * case we take the easy way out and panic immediately.
276 */
277
278 void ext3_abort (struct super_block * sb, const char * function,
279 const char * fmt, ...)
280 {
281 va_list args;
282
283 va_start(args, fmt);
284 printk(KERN_CRIT "EXT3-fs (%s): error: %s: ", sb->s_id, function);
285 vprintk(fmt, args);
286 printk("\n");
287 va_end(args);
288
289 if (test_opt(sb, ERRORS_PANIC))
290 panic("EXT3-fs: panic from previous error\n");
291
292 if (sb->s_flags & MS_RDONLY)
293 return;
294
295 ext3_msg(sb, KERN_CRIT,
296 "error: remounting filesystem read-only");
297 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
298 sb->s_flags |= MS_RDONLY;
299 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
300 if (EXT3_SB(sb)->s_journal)
301 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
302 }
303
304 void ext3_warning (struct super_block * sb, const char * function,
305 const char * fmt, ...)
306 {
307 va_list args;
308
309 va_start(args, fmt);
310 printk(KERN_WARNING "EXT3-fs (%s): warning: %s: ",
311 sb->s_id, function);
312 vprintk(fmt, args);
313 printk("\n");
314 va_end(args);
315 }
316
317 void ext3_update_dynamic_rev(struct super_block *sb)
318 {
319 struct ext3_super_block *es = EXT3_SB(sb)->s_es;
320
321 if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
322 return;
323
324 ext3_msg(sb, KERN_WARNING,
325 "warning: updating to rev %d because of "
326 "new feature flag, running e2fsck is recommended",
327 EXT3_DYNAMIC_REV);
328
329 es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
330 es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
331 es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
332 /* leave es->s_feature_*compat flags alone */
333 /* es->s_uuid will be set by e2fsck if empty */
334
335 /*
336 * The rest of the superblock fields should be zero, and if not it
337 * means they are likely already in use, so leave them alone. We
338 * can leave it up to e2fsck to clean up any inconsistencies there.
339 */
340 }
341
342 /*
343 * Open the external journal device
344 */
345 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
346 {
347 struct block_device *bdev;
348 char b[BDEVNAME_SIZE];
349
350 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
351 if (IS_ERR(bdev))
352 goto fail;
353 return bdev;
354
355 fail:
356 ext3_msg(sb, "error: failed to open journal device %s: %ld",
357 __bdevname(dev, b), PTR_ERR(bdev));
358
359 return NULL;
360 }
361
362 /*
363 * Release the journal device
364 */
365 static int ext3_blkdev_put(struct block_device *bdev)
366 {
367 bd_release(bdev);
368 return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
369 }
370
371 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
372 {
373 struct block_device *bdev;
374 int ret = -ENODEV;
375
376 bdev = sbi->journal_bdev;
377 if (bdev) {
378 ret = ext3_blkdev_put(bdev);
379 sbi->journal_bdev = NULL;
380 }
381 return ret;
382 }
383
384 static inline struct inode *orphan_list_entry(struct list_head *l)
385 {
386 return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
387 }
388
389 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
390 {
391 struct list_head *l;
392
393 ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
394 le32_to_cpu(sbi->s_es->s_last_orphan));
395
396 ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
397 list_for_each(l, &sbi->s_orphan) {
398 struct inode *inode = orphan_list_entry(l);
399 ext3_msg(sb, KERN_ERR, " "
400 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
401 inode->i_sb->s_id, inode->i_ino, inode,
402 inode->i_mode, inode->i_nlink,
403 NEXT_ORPHAN(inode));
404 }
405 }
406
407 static void ext3_put_super (struct super_block * sb)
408 {
409 struct ext3_sb_info *sbi = EXT3_SB(sb);
410 struct ext3_super_block *es = sbi->s_es;
411 int i, err;
412
413 dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
414
415 lock_kernel();
416
417 ext3_xattr_put_super(sb);
418 err = journal_destroy(sbi->s_journal);
419 sbi->s_journal = NULL;
420 if (err < 0)
421 ext3_abort(sb, __func__, "Couldn't clean up the journal");
422
423 if (!(sb->s_flags & MS_RDONLY)) {
424 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
425 es->s_state = cpu_to_le16(sbi->s_mount_state);
426 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
427 mark_buffer_dirty(sbi->s_sbh);
428 ext3_commit_super(sb, es, 1);
429 }
430
431 for (i = 0; i < sbi->s_gdb_count; i++)
432 brelse(sbi->s_group_desc[i]);
433 kfree(sbi->s_group_desc);
434 percpu_counter_destroy(&sbi->s_freeblocks_counter);
435 percpu_counter_destroy(&sbi->s_freeinodes_counter);
436 percpu_counter_destroy(&sbi->s_dirs_counter);
437 brelse(sbi->s_sbh);
438 #ifdef CONFIG_QUOTA
439 for (i = 0; i < MAXQUOTAS; i++)
440 kfree(sbi->s_qf_names[i]);
441 #endif
442
443 /* Debugging code just in case the in-memory inode orphan list
444 * isn't empty. The on-disk one can be non-empty if we've
445 * detected an error and taken the fs readonly, but the
446 * in-memory list had better be clean by this point. */
447 if (!list_empty(&sbi->s_orphan))
448 dump_orphan_list(sb, sbi);
449 J_ASSERT(list_empty(&sbi->s_orphan));
450
451 invalidate_bdev(sb->s_bdev);
452 if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
453 /*
454 * Invalidate the journal device's buffers. We don't want them
455 * floating about in memory - the physical journal device may
456 * hotswapped, and it breaks the `ro-after' testing code.
457 */
458 sync_blockdev(sbi->journal_bdev);
459 invalidate_bdev(sbi->journal_bdev);
460 ext3_blkdev_remove(sbi);
461 }
462 sb->s_fs_info = NULL;
463 kfree(sbi->s_blockgroup_lock);
464 kfree(sbi);
465
466 unlock_kernel();
467 }
468
469 static struct kmem_cache *ext3_inode_cachep;
470
471 /*
472 * Called inside transaction, so use GFP_NOFS
473 */
474 static struct inode *ext3_alloc_inode(struct super_block *sb)
475 {
476 struct ext3_inode_info *ei;
477
478 ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
479 if (!ei)
480 return NULL;
481 ei->i_block_alloc_info = NULL;
482 ei->vfs_inode.i_version = 1;
483 atomic_set(&ei->i_datasync_tid, 0);
484 atomic_set(&ei->i_sync_tid, 0);
485 return &ei->vfs_inode;
486 }
487
488 static void ext3_destroy_inode(struct inode *inode)
489 {
490 if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
491 printk("EXT3 Inode %p: orphan list check failed!\n",
492 EXT3_I(inode));
493 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
494 EXT3_I(inode), sizeof(struct ext3_inode_info),
495 false);
496 dump_stack();
497 }
498 kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
499 }
500
501 static void init_once(void *foo)
502 {
503 struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
504
505 INIT_LIST_HEAD(&ei->i_orphan);
506 #ifdef CONFIG_EXT3_FS_XATTR
507 init_rwsem(&ei->xattr_sem);
508 #endif
509 mutex_init(&ei->truncate_mutex);
510 inode_init_once(&ei->vfs_inode);
511 }
512
513 static int init_inodecache(void)
514 {
515 ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
516 sizeof(struct ext3_inode_info),
517 0, (SLAB_RECLAIM_ACCOUNT|
518 SLAB_MEM_SPREAD),
519 init_once);
520 if (ext3_inode_cachep == NULL)
521 return -ENOMEM;
522 return 0;
523 }
524
525 static void destroy_inodecache(void)
526 {
527 kmem_cache_destroy(ext3_inode_cachep);
528 }
529
530 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
531 {
532 #if defined(CONFIG_QUOTA)
533 struct ext3_sb_info *sbi = EXT3_SB(sb);
534
535 if (sbi->s_jquota_fmt) {
536 char *fmtname = "";
537
538 switch (sbi->s_jquota_fmt) {
539 case QFMT_VFS_OLD:
540 fmtname = "vfsold";
541 break;
542 case QFMT_VFS_V0:
543 fmtname = "vfsv0";
544 break;
545 case QFMT_VFS_V1:
546 fmtname = "vfsv1";
547 break;
548 }
549 seq_printf(seq, ",jqfmt=%s", fmtname);
550 }
551
552 if (sbi->s_qf_names[USRQUOTA])
553 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
554
555 if (sbi->s_qf_names[GRPQUOTA])
556 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
557
558 if (test_opt(sb, USRQUOTA))
559 seq_puts(seq, ",usrquota");
560
561 if (test_opt(sb, GRPQUOTA))
562 seq_puts(seq, ",grpquota");
563 #endif
564 }
565
566 static char *data_mode_string(unsigned long mode)
567 {
568 switch (mode) {
569 case EXT3_MOUNT_JOURNAL_DATA:
570 return "journal";
571 case EXT3_MOUNT_ORDERED_DATA:
572 return "ordered";
573 case EXT3_MOUNT_WRITEBACK_DATA:
574 return "writeback";
575 }
576 return "unknown";
577 }
578
579 /*
580 * Show an option if
581 * - it's set to a non-default value OR
582 * - if the per-sb default is different from the global default
583 */
584 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
585 {
586 struct super_block *sb = vfs->mnt_sb;
587 struct ext3_sb_info *sbi = EXT3_SB(sb);
588 struct ext3_super_block *es = sbi->s_es;
589 unsigned long def_mount_opts;
590
591 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
592
593 if (sbi->s_sb_block != 1)
594 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
595 if (test_opt(sb, MINIX_DF))
596 seq_puts(seq, ",minixdf");
597 if (test_opt(sb, GRPID))
598 seq_puts(seq, ",grpid");
599 if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
600 seq_puts(seq, ",nogrpid");
601 if (sbi->s_resuid != EXT3_DEF_RESUID ||
602 le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
603 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
604 }
605 if (sbi->s_resgid != EXT3_DEF_RESGID ||
606 le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
607 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
608 }
609 if (test_opt(sb, ERRORS_RO)) {
610 int def_errors = le16_to_cpu(es->s_errors);
611
612 if (def_errors == EXT3_ERRORS_PANIC ||
613 def_errors == EXT3_ERRORS_CONTINUE) {
614 seq_puts(seq, ",errors=remount-ro");
615 }
616 }
617 if (test_opt(sb, ERRORS_CONT))
618 seq_puts(seq, ",errors=continue");
619 if (test_opt(sb, ERRORS_PANIC))
620 seq_puts(seq, ",errors=panic");
621 if (test_opt(sb, NO_UID32))
622 seq_puts(seq, ",nouid32");
623 if (test_opt(sb, DEBUG))
624 seq_puts(seq, ",debug");
625 if (test_opt(sb, OLDALLOC))
626 seq_puts(seq, ",oldalloc");
627 #ifdef CONFIG_EXT3_FS_XATTR
628 if (test_opt(sb, XATTR_USER))
629 seq_puts(seq, ",user_xattr");
630 if (!test_opt(sb, XATTR_USER) &&
631 (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
632 seq_puts(seq, ",nouser_xattr");
633 }
634 #endif
635 #ifdef CONFIG_EXT3_FS_POSIX_ACL
636 if (test_opt(sb, POSIX_ACL))
637 seq_puts(seq, ",acl");
638 if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
639 seq_puts(seq, ",noacl");
640 #endif
641 if (!test_opt(sb, RESERVATION))
642 seq_puts(seq, ",noreservation");
643 if (sbi->s_commit_interval) {
644 seq_printf(seq, ",commit=%u",
645 (unsigned) (sbi->s_commit_interval / HZ));
646 }
647
648 /*
649 * Always display barrier state so it's clear what the status is.
650 */
651 seq_puts(seq, ",barrier=");
652 seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
653 seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
654 if (test_opt(sb, DATA_ERR_ABORT))
655 seq_puts(seq, ",data_err=abort");
656
657 if (test_opt(sb, NOLOAD))
658 seq_puts(seq, ",norecovery");
659
660 ext3_show_quota_options(seq, sb);
661
662 return 0;
663 }
664
665
666 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
667 u64 ino, u32 generation)
668 {
669 struct inode *inode;
670
671 if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
672 return ERR_PTR(-ESTALE);
673 if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
674 return ERR_PTR(-ESTALE);
675
676 /* iget isn't really right if the inode is currently unallocated!!
677 *
678 * ext3_read_inode will return a bad_inode if the inode had been
679 * deleted, so we should be safe.
680 *
681 * Currently we don't know the generation for parent directory, so
682 * a generation of 0 means "accept any"
683 */
684 inode = ext3_iget(sb, ino);
685 if (IS_ERR(inode))
686 return ERR_CAST(inode);
687 if (generation && inode->i_generation != generation) {
688 iput(inode);
689 return ERR_PTR(-ESTALE);
690 }
691
692 return inode;
693 }
694
695 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
696 int fh_len, int fh_type)
697 {
698 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
699 ext3_nfs_get_inode);
700 }
701
702 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
703 int fh_len, int fh_type)
704 {
705 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
706 ext3_nfs_get_inode);
707 }
708
709 /*
710 * Try to release metadata pages (indirect blocks, directories) which are
711 * mapped via the block device. Since these pages could have journal heads
712 * which would prevent try_to_free_buffers() from freeing them, we must use
713 * jbd layer's try_to_free_buffers() function to release them.
714 */
715 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
716 gfp_t wait)
717 {
718 journal_t *journal = EXT3_SB(sb)->s_journal;
719
720 WARN_ON(PageChecked(page));
721 if (!page_has_buffers(page))
722 return 0;
723 if (journal)
724 return journal_try_to_free_buffers(journal, page,
725 wait & ~__GFP_WAIT);
726 return try_to_free_buffers(page);
727 }
728
729 #ifdef CONFIG_QUOTA
730 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
731 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
732
733 static int ext3_write_dquot(struct dquot *dquot);
734 static int ext3_acquire_dquot(struct dquot *dquot);
735 static int ext3_release_dquot(struct dquot *dquot);
736 static int ext3_mark_dquot_dirty(struct dquot *dquot);
737 static int ext3_write_info(struct super_block *sb, int type);
738 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
739 char *path);
740 static int ext3_quota_on_mount(struct super_block *sb, int type);
741 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
742 size_t len, loff_t off);
743 static ssize_t ext3_quota_write(struct super_block *sb, int type,
744 const char *data, size_t len, loff_t off);
745
746 static const struct dquot_operations ext3_quota_operations = {
747 .write_dquot = ext3_write_dquot,
748 .acquire_dquot = ext3_acquire_dquot,
749 .release_dquot = ext3_release_dquot,
750 .mark_dirty = ext3_mark_dquot_dirty,
751 .write_info = ext3_write_info,
752 .alloc_dquot = dquot_alloc,
753 .destroy_dquot = dquot_destroy,
754 };
755
756 static const struct quotactl_ops ext3_qctl_operations = {
757 .quota_on = ext3_quota_on,
758 .quota_off = dquot_quota_off,
759 .quota_sync = dquot_quota_sync,
760 .get_info = dquot_get_dqinfo,
761 .set_info = dquot_set_dqinfo,
762 .get_dqblk = dquot_get_dqblk,
763 .set_dqblk = dquot_set_dqblk
764 };
765 #endif
766
767 static const struct super_operations ext3_sops = {
768 .alloc_inode = ext3_alloc_inode,
769 .destroy_inode = ext3_destroy_inode,
770 .write_inode = ext3_write_inode,
771 .dirty_inode = ext3_dirty_inode,
772 .evict_inode = ext3_evict_inode,
773 .put_super = ext3_put_super,
774 .sync_fs = ext3_sync_fs,
775 .freeze_fs = ext3_freeze,
776 .unfreeze_fs = ext3_unfreeze,
777 .statfs = ext3_statfs,
778 .remount_fs = ext3_remount,
779 .show_options = ext3_show_options,
780 #ifdef CONFIG_QUOTA
781 .quota_read = ext3_quota_read,
782 .quota_write = ext3_quota_write,
783 #endif
784 .bdev_try_to_free_page = bdev_try_to_free_page,
785 };
786
787 static const struct export_operations ext3_export_ops = {
788 .fh_to_dentry = ext3_fh_to_dentry,
789 .fh_to_parent = ext3_fh_to_parent,
790 .get_parent = ext3_get_parent,
791 };
792
793 enum {
794 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
795 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
796 Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
797 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
798 Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
799 Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
800 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
801 Opt_data_err_abort, Opt_data_err_ignore,
802 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
803 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
804 Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
805 Opt_resize, Opt_usrquota, Opt_grpquota
806 };
807
808 static const match_table_t tokens = {
809 {Opt_bsd_df, "bsddf"},
810 {Opt_minix_df, "minixdf"},
811 {Opt_grpid, "grpid"},
812 {Opt_grpid, "bsdgroups"},
813 {Opt_nogrpid, "nogrpid"},
814 {Opt_nogrpid, "sysvgroups"},
815 {Opt_resgid, "resgid=%u"},
816 {Opt_resuid, "resuid=%u"},
817 {Opt_sb, "sb=%u"},
818 {Opt_err_cont, "errors=continue"},
819 {Opt_err_panic, "errors=panic"},
820 {Opt_err_ro, "errors=remount-ro"},
821 {Opt_nouid32, "nouid32"},
822 {Opt_nocheck, "nocheck"},
823 {Opt_nocheck, "check=none"},
824 {Opt_debug, "debug"},
825 {Opt_oldalloc, "oldalloc"},
826 {Opt_orlov, "orlov"},
827 {Opt_user_xattr, "user_xattr"},
828 {Opt_nouser_xattr, "nouser_xattr"},
829 {Opt_acl, "acl"},
830 {Opt_noacl, "noacl"},
831 {Opt_reservation, "reservation"},
832 {Opt_noreservation, "noreservation"},
833 {Opt_noload, "noload"},
834 {Opt_noload, "norecovery"},
835 {Opt_nobh, "nobh"},
836 {Opt_bh, "bh"},
837 {Opt_commit, "commit=%u"},
838 {Opt_journal_update, "journal=update"},
839 {Opt_journal_inum, "journal=%u"},
840 {Opt_journal_dev, "journal_dev=%u"},
841 {Opt_abort, "abort"},
842 {Opt_data_journal, "data=journal"},
843 {Opt_data_ordered, "data=ordered"},
844 {Opt_data_writeback, "data=writeback"},
845 {Opt_data_err_abort, "data_err=abort"},
846 {Opt_data_err_ignore, "data_err=ignore"},
847 {Opt_offusrjquota, "usrjquota="},
848 {Opt_usrjquota, "usrjquota=%s"},
849 {Opt_offgrpjquota, "grpjquota="},
850 {Opt_grpjquota, "grpjquota=%s"},
851 {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
852 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
853 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
854 {Opt_grpquota, "grpquota"},
855 {Opt_noquota, "noquota"},
856 {Opt_quota, "quota"},
857 {Opt_usrquota, "usrquota"},
858 {Opt_barrier, "barrier=%u"},
859 {Opt_barrier, "barrier"},
860 {Opt_nobarrier, "nobarrier"},
861 {Opt_resize, "resize"},
862 {Opt_err, NULL},
863 };
864
865 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
866 {
867 ext3_fsblk_t sb_block;
868 char *options = (char *) *data;
869
870 if (!options || strncmp(options, "sb=", 3) != 0)
871 return 1; /* Default location */
872 options += 3;
873 /*todo: use simple_strtoll with >32bit ext3 */
874 sb_block = simple_strtoul(options, &options, 0);
875 if (*options && *options != ',') {
876 ext3_msg(sb, "error: invalid sb specification: %s",
877 (char *) *data);
878 return 1;
879 }
880 if (*options == ',')
881 options++;
882 *data = (void *) options;
883 return sb_block;
884 }
885
886 #ifdef CONFIG_QUOTA
887 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
888 {
889 struct ext3_sb_info *sbi = EXT3_SB(sb);
890 char *qname;
891
892 if (sb_any_quota_loaded(sb) &&
893 !sbi->s_qf_names[qtype]) {
894 ext3_msg(sb, KERN_ERR,
895 "Cannot change journaled "
896 "quota options when quota turned on");
897 return 0;
898 }
899 qname = match_strdup(args);
900 if (!qname) {
901 ext3_msg(sb, KERN_ERR,
902 "Not enough memory for storing quotafile name");
903 return 0;
904 }
905 if (sbi->s_qf_names[qtype] &&
906 strcmp(sbi->s_qf_names[qtype], qname)) {
907 ext3_msg(sb, KERN_ERR,
908 "%s quota file already specified", QTYPE2NAME(qtype));
909 kfree(qname);
910 return 0;
911 }
912 sbi->s_qf_names[qtype] = qname;
913 if (strchr(sbi->s_qf_names[qtype], '/')) {
914 ext3_msg(sb, KERN_ERR,
915 "quotafile must be on filesystem root");
916 kfree(sbi->s_qf_names[qtype]);
917 sbi->s_qf_names[qtype] = NULL;
918 return 0;
919 }
920 set_opt(sbi->s_mount_opt, QUOTA);
921 return 1;
922 }
923
924 static int clear_qf_name(struct super_block *sb, int qtype) {
925
926 struct ext3_sb_info *sbi = EXT3_SB(sb);
927
928 if (sb_any_quota_loaded(sb) &&
929 sbi->s_qf_names[qtype]) {
930 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
931 " when quota turned on");
932 return 0;
933 }
934 /*
935 * The space will be released later when all options are confirmed
936 * to be correct
937 */
938 sbi->s_qf_names[qtype] = NULL;
939 return 1;
940 }
941 #endif
942
943 static int parse_options (char *options, struct super_block *sb,
944 unsigned int *inum, unsigned long *journal_devnum,
945 ext3_fsblk_t *n_blocks_count, int is_remount)
946 {
947 struct ext3_sb_info *sbi = EXT3_SB(sb);
948 char * p;
949 substring_t args[MAX_OPT_ARGS];
950 int data_opt = 0;
951 int option;
952 #ifdef CONFIG_QUOTA
953 int qfmt;
954 #endif
955
956 if (!options)
957 return 1;
958
959 while ((p = strsep (&options, ",")) != NULL) {
960 int token;
961 if (!*p)
962 continue;
963 /*
964 * Initialize args struct so we know whether arg was
965 * found; some options take optional arguments.
966 */
967 args[0].to = args[0].from = 0;
968 token = match_token(p, tokens, args);
969 switch (token) {
970 case Opt_bsd_df:
971 clear_opt (sbi->s_mount_opt, MINIX_DF);
972 break;
973 case Opt_minix_df:
974 set_opt (sbi->s_mount_opt, MINIX_DF);
975 break;
976 case Opt_grpid:
977 set_opt (sbi->s_mount_opt, GRPID);
978 break;
979 case Opt_nogrpid:
980 clear_opt (sbi->s_mount_opt, GRPID);
981 break;
982 case Opt_resuid:
983 if (match_int(&args[0], &option))
984 return 0;
985 sbi->s_resuid = option;
986 break;
987 case Opt_resgid:
988 if (match_int(&args[0], &option))
989 return 0;
990 sbi->s_resgid = option;
991 break;
992 case Opt_sb:
993 /* handled by get_sb_block() instead of here */
994 /* *sb_block = match_int(&args[0]); */
995 break;
996 case Opt_err_panic:
997 clear_opt (sbi->s_mount_opt, ERRORS_CONT);
998 clear_opt (sbi->s_mount_opt, ERRORS_RO);
999 set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1000 break;
1001 case Opt_err_ro:
1002 clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1003 clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1004 set_opt (sbi->s_mount_opt, ERRORS_RO);
1005 break;
1006 case Opt_err_cont:
1007 clear_opt (sbi->s_mount_opt, ERRORS_RO);
1008 clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1009 set_opt (sbi->s_mount_opt, ERRORS_CONT);
1010 break;
1011 case Opt_nouid32:
1012 set_opt (sbi->s_mount_opt, NO_UID32);
1013 break;
1014 case Opt_nocheck:
1015 clear_opt (sbi->s_mount_opt, CHECK);
1016 break;
1017 case Opt_debug:
1018 set_opt (sbi->s_mount_opt, DEBUG);
1019 break;
1020 case Opt_oldalloc:
1021 set_opt (sbi->s_mount_opt, OLDALLOC);
1022 break;
1023 case Opt_orlov:
1024 clear_opt (sbi->s_mount_opt, OLDALLOC);
1025 break;
1026 #ifdef CONFIG_EXT3_FS_XATTR
1027 case Opt_user_xattr:
1028 set_opt (sbi->s_mount_opt, XATTR_USER);
1029 break;
1030 case Opt_nouser_xattr:
1031 clear_opt (sbi->s_mount_opt, XATTR_USER);
1032 break;
1033 #else
1034 case Opt_user_xattr:
1035 case Opt_nouser_xattr:
1036 ext3_msg(sb, KERN_INFO,
1037 "(no)user_xattr options not supported");
1038 break;
1039 #endif
1040 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1041 case Opt_acl:
1042 set_opt(sbi->s_mount_opt, POSIX_ACL);
1043 break;
1044 case Opt_noacl:
1045 clear_opt(sbi->s_mount_opt, POSIX_ACL);
1046 break;
1047 #else
1048 case Opt_acl:
1049 case Opt_noacl:
1050 ext3_msg(sb, KERN_INFO,
1051 "(no)acl options not supported");
1052 break;
1053 #endif
1054 case Opt_reservation:
1055 set_opt(sbi->s_mount_opt, RESERVATION);
1056 break;
1057 case Opt_noreservation:
1058 clear_opt(sbi->s_mount_opt, RESERVATION);
1059 break;
1060 case Opt_journal_update:
1061 /* @@@ FIXME */
1062 /* Eventually we will want to be able to create
1063 a journal file here. For now, only allow the
1064 user to specify an existing inode to be the
1065 journal file. */
1066 if (is_remount) {
1067 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1068 "journal on remount");
1069 return 0;
1070 }
1071 set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1072 break;
1073 case Opt_journal_inum:
1074 if (is_remount) {
1075 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1076 "journal on remount");
1077 return 0;
1078 }
1079 if (match_int(&args[0], &option))
1080 return 0;
1081 *inum = option;
1082 break;
1083 case Opt_journal_dev:
1084 if (is_remount) {
1085 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1086 "journal on remount");
1087 return 0;
1088 }
1089 if (match_int(&args[0], &option))
1090 return 0;
1091 *journal_devnum = option;
1092 break;
1093 case Opt_noload:
1094 set_opt (sbi->s_mount_opt, NOLOAD);
1095 break;
1096 case Opt_commit:
1097 if (match_int(&args[0], &option))
1098 return 0;
1099 if (option < 0)
1100 return 0;
1101 if (option == 0)
1102 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1103 sbi->s_commit_interval = HZ * option;
1104 break;
1105 case Opt_data_journal:
1106 data_opt = EXT3_MOUNT_JOURNAL_DATA;
1107 goto datacheck;
1108 case Opt_data_ordered:
1109 data_opt = EXT3_MOUNT_ORDERED_DATA;
1110 goto datacheck;
1111 case Opt_data_writeback:
1112 data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1113 datacheck:
1114 if (is_remount) {
1115 if (test_opt(sb, DATA_FLAGS) == data_opt)
1116 break;
1117 ext3_msg(sb, KERN_ERR,
1118 "error: cannot change "
1119 "data mode on remount. The filesystem "
1120 "is mounted in data=%s mode and you "
1121 "try to remount it in data=%s mode.",
1122 data_mode_string(test_opt(sb,
1123 DATA_FLAGS)),
1124 data_mode_string(data_opt));
1125 return 0;
1126 } else {
1127 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1128 sbi->s_mount_opt |= data_opt;
1129 }
1130 break;
1131 case Opt_data_err_abort:
1132 set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1133 break;
1134 case Opt_data_err_ignore:
1135 clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1136 break;
1137 #ifdef CONFIG_QUOTA
1138 case Opt_usrjquota:
1139 if (!set_qf_name(sb, USRQUOTA, &args[0]))
1140 return 0;
1141 break;
1142 case Opt_grpjquota:
1143 if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1144 return 0;
1145 break;
1146 case Opt_offusrjquota:
1147 if (!clear_qf_name(sb, USRQUOTA))
1148 return 0;
1149 break;
1150 case Opt_offgrpjquota:
1151 if (!clear_qf_name(sb, GRPQUOTA))
1152 return 0;
1153 break;
1154 case Opt_jqfmt_vfsold:
1155 qfmt = QFMT_VFS_OLD;
1156 goto set_qf_format;
1157 case Opt_jqfmt_vfsv0:
1158 qfmt = QFMT_VFS_V0;
1159 goto set_qf_format;
1160 case Opt_jqfmt_vfsv1:
1161 qfmt = QFMT_VFS_V1;
1162 set_qf_format:
1163 if (sb_any_quota_loaded(sb) &&
1164 sbi->s_jquota_fmt != qfmt) {
1165 ext3_msg(sb, KERN_ERR, "error: cannot change "
1166 "journaled quota options when "
1167 "quota turned on.");
1168 return 0;
1169 }
1170 sbi->s_jquota_fmt = qfmt;
1171 break;
1172 case Opt_quota:
1173 case Opt_usrquota:
1174 set_opt(sbi->s_mount_opt, QUOTA);
1175 set_opt(sbi->s_mount_opt, USRQUOTA);
1176 break;
1177 case Opt_grpquota:
1178 set_opt(sbi->s_mount_opt, QUOTA);
1179 set_opt(sbi->s_mount_opt, GRPQUOTA);
1180 break;
1181 case Opt_noquota:
1182 if (sb_any_quota_loaded(sb)) {
1183 ext3_msg(sb, KERN_ERR, "error: cannot change "
1184 "quota options when quota turned on.");
1185 return 0;
1186 }
1187 clear_opt(sbi->s_mount_opt, QUOTA);
1188 clear_opt(sbi->s_mount_opt, USRQUOTA);
1189 clear_opt(sbi->s_mount_opt, GRPQUOTA);
1190 break;
1191 #else
1192 case Opt_quota:
1193 case Opt_usrquota:
1194 case Opt_grpquota:
1195 ext3_msg(sb, KERN_ERR,
1196 "error: quota options not supported.");
1197 break;
1198 case Opt_usrjquota:
1199 case Opt_grpjquota:
1200 case Opt_offusrjquota:
1201 case Opt_offgrpjquota:
1202 case Opt_jqfmt_vfsold:
1203 case Opt_jqfmt_vfsv0:
1204 case Opt_jqfmt_vfsv1:
1205 ext3_msg(sb, KERN_ERR,
1206 "error: journaled quota options not "
1207 "supported.");
1208 break;
1209 case Opt_noquota:
1210 break;
1211 #endif
1212 case Opt_abort:
1213 set_opt(sbi->s_mount_opt, ABORT);
1214 break;
1215 case Opt_nobarrier:
1216 clear_opt(sbi->s_mount_opt, BARRIER);
1217 break;
1218 case Opt_barrier:
1219 if (args[0].from) {
1220 if (match_int(&args[0], &option))
1221 return 0;
1222 } else
1223 option = 1; /* No argument, default to 1 */
1224 if (option)
1225 set_opt(sbi->s_mount_opt, BARRIER);
1226 else
1227 clear_opt(sbi->s_mount_opt, BARRIER);
1228 break;
1229 case Opt_ignore:
1230 break;
1231 case Opt_resize:
1232 if (!is_remount) {
1233 ext3_msg(sb, KERN_ERR,
1234 "error: resize option only available "
1235 "for remount");
1236 return 0;
1237 }
1238 if (match_int(&args[0], &option) != 0)
1239 return 0;
1240 *n_blocks_count = option;
1241 break;
1242 case Opt_nobh:
1243 ext3_msg(sb, KERN_WARNING,
1244 "warning: ignoring deprecated nobh option");
1245 break;
1246 case Opt_bh:
1247 ext3_msg(sb, KERN_WARNING,
1248 "warning: ignoring deprecated bh option");
1249 break;
1250 default:
1251 ext3_msg(sb, KERN_ERR,
1252 "error: unrecognized mount option \"%s\" "
1253 "or missing value", p);
1254 return 0;
1255 }
1256 }
1257 #ifdef CONFIG_QUOTA
1258 if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1259 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1260 clear_opt(sbi->s_mount_opt, USRQUOTA);
1261 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1262 clear_opt(sbi->s_mount_opt, GRPQUOTA);
1263
1264 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1265 ext3_msg(sb, KERN_ERR, "error: old and new quota "
1266 "format mixing.");
1267 return 0;
1268 }
1269
1270 if (!sbi->s_jquota_fmt) {
1271 ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1272 "not specified.");
1273 return 0;
1274 }
1275 } else {
1276 if (sbi->s_jquota_fmt) {
1277 ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1278 "specified with no journaling "
1279 "enabled.");
1280 return 0;
1281 }
1282 }
1283 #endif
1284 return 1;
1285 }
1286
1287 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1288 int read_only)
1289 {
1290 struct ext3_sb_info *sbi = EXT3_SB(sb);
1291 int res = 0;
1292
1293 if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1294 ext3_msg(sb, KERN_ERR,
1295 "error: revision level too high, "
1296 "forcing read-only mode");
1297 res = MS_RDONLY;
1298 }
1299 if (read_only)
1300 return res;
1301 if (!(sbi->s_mount_state & EXT3_VALID_FS))
1302 ext3_msg(sb, KERN_WARNING,
1303 "warning: mounting unchecked fs, "
1304 "running e2fsck is recommended");
1305 else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1306 ext3_msg(sb, KERN_WARNING,
1307 "warning: mounting fs with errors, "
1308 "running e2fsck is recommended");
1309 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1310 le16_to_cpu(es->s_mnt_count) >=
1311 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1312 ext3_msg(sb, KERN_WARNING,
1313 "warning: maximal mount count reached, "
1314 "running e2fsck is recommended");
1315 else if (le32_to_cpu(es->s_checkinterval) &&
1316 (le32_to_cpu(es->s_lastcheck) +
1317 le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1318 ext3_msg(sb, KERN_WARNING,
1319 "warning: checktime reached, "
1320 "running e2fsck is recommended");
1321 #if 0
1322 /* @@@ We _will_ want to clear the valid bit if we find
1323 inconsistencies, to force a fsck at reboot. But for
1324 a plain journaled filesystem we can keep it set as
1325 valid forever! :) */
1326 es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1327 #endif
1328 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1329 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1330 le16_add_cpu(&es->s_mnt_count, 1);
1331 es->s_mtime = cpu_to_le32(get_seconds());
1332 ext3_update_dynamic_rev(sb);
1333 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1334
1335 ext3_commit_super(sb, es, 1);
1336 if (test_opt(sb, DEBUG))
1337 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1338 "bpg=%lu, ipg=%lu, mo=%04lx]",
1339 sb->s_blocksize,
1340 sbi->s_groups_count,
1341 EXT3_BLOCKS_PER_GROUP(sb),
1342 EXT3_INODES_PER_GROUP(sb),
1343 sbi->s_mount_opt);
1344
1345 if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1346 char b[BDEVNAME_SIZE];
1347 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1348 bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1349 } else {
1350 ext3_msg(sb, KERN_INFO, "using internal journal");
1351 }
1352 return res;
1353 }
1354
1355 /* Called at mount-time, super-block is locked */
1356 static int ext3_check_descriptors(struct super_block *sb)
1357 {
1358 struct ext3_sb_info *sbi = EXT3_SB(sb);
1359 int i;
1360
1361 ext3_debug ("Checking group descriptors");
1362
1363 for (i = 0; i < sbi->s_groups_count; i++) {
1364 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1365 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1366 ext3_fsblk_t last_block;
1367
1368 if (i == sbi->s_groups_count - 1)
1369 last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1370 else
1371 last_block = first_block +
1372 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1373
1374 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1375 le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1376 {
1377 ext3_error (sb, "ext3_check_descriptors",
1378 "Block bitmap for group %d"
1379 " not in group (block %lu)!",
1380 i, (unsigned long)
1381 le32_to_cpu(gdp->bg_block_bitmap));
1382 return 0;
1383 }
1384 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1385 le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1386 {
1387 ext3_error (sb, "ext3_check_descriptors",
1388 "Inode bitmap for group %d"
1389 " not in group (block %lu)!",
1390 i, (unsigned long)
1391 le32_to_cpu(gdp->bg_inode_bitmap));
1392 return 0;
1393 }
1394 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1395 le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1396 last_block)
1397 {
1398 ext3_error (sb, "ext3_check_descriptors",
1399 "Inode table for group %d"
1400 " not in group (block %lu)!",
1401 i, (unsigned long)
1402 le32_to_cpu(gdp->bg_inode_table));
1403 return 0;
1404 }
1405 }
1406
1407 sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1408 sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1409 return 1;
1410 }
1411
1412
1413 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1414 * the superblock) which were deleted from all directories, but held open by
1415 * a process at the time of a crash. We walk the list and try to delete these
1416 * inodes at recovery time (only with a read-write filesystem).
1417 *
1418 * In order to keep the orphan inode chain consistent during traversal (in
1419 * case of crash during recovery), we link each inode into the superblock
1420 * orphan list_head and handle it the same way as an inode deletion during
1421 * normal operation (which journals the operations for us).
1422 *
1423 * We only do an iget() and an iput() on each inode, which is very safe if we
1424 * accidentally point at an in-use or already deleted inode. The worst that
1425 * can happen in this case is that we get a "bit already cleared" message from
1426 * ext3_free_inode(). The only reason we would point at a wrong inode is if
1427 * e2fsck was run on this filesystem, and it must have already done the orphan
1428 * inode cleanup for us, so we can safely abort without any further action.
1429 */
1430 static void ext3_orphan_cleanup (struct super_block * sb,
1431 struct ext3_super_block * es)
1432 {
1433 unsigned int s_flags = sb->s_flags;
1434 int nr_orphans = 0, nr_truncates = 0;
1435 #ifdef CONFIG_QUOTA
1436 int i;
1437 #endif
1438 if (!es->s_last_orphan) {
1439 jbd_debug(4, "no orphan inodes to clean up\n");
1440 return;
1441 }
1442
1443 if (bdev_read_only(sb->s_bdev)) {
1444 ext3_msg(sb, KERN_ERR, "error: write access "
1445 "unavailable, skipping orphan cleanup.");
1446 return;
1447 }
1448
1449 if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1450 if (es->s_last_orphan)
1451 jbd_debug(1, "Errors on filesystem, "
1452 "clearing orphan list.\n");
1453 es->s_last_orphan = 0;
1454 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1455 return;
1456 }
1457
1458 if (s_flags & MS_RDONLY) {
1459 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1460 sb->s_flags &= ~MS_RDONLY;
1461 }
1462 #ifdef CONFIG_QUOTA
1463 /* Needed for iput() to work correctly and not trash data */
1464 sb->s_flags |= MS_ACTIVE;
1465 /* Turn on quotas so that they are updated correctly */
1466 for (i = 0; i < MAXQUOTAS; i++) {
1467 if (EXT3_SB(sb)->s_qf_names[i]) {
1468 int ret = ext3_quota_on_mount(sb, i);
1469 if (ret < 0)
1470 ext3_msg(sb, KERN_ERR,
1471 "error: cannot turn on journaled "
1472 "quota: %d", ret);
1473 }
1474 }
1475 #endif
1476
1477 while (es->s_last_orphan) {
1478 struct inode *inode;
1479
1480 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1481 if (IS_ERR(inode)) {
1482 es->s_last_orphan = 0;
1483 break;
1484 }
1485
1486 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1487 dquot_initialize(inode);
1488 if (inode->i_nlink) {
1489 printk(KERN_DEBUG
1490 "%s: truncating inode %lu to %Ld bytes\n",
1491 __func__, inode->i_ino, inode->i_size);
1492 jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1493 inode->i_ino, inode->i_size);
1494 ext3_truncate(inode);
1495 nr_truncates++;
1496 } else {
1497 printk(KERN_DEBUG
1498 "%s: deleting unreferenced inode %lu\n",
1499 __func__, inode->i_ino);
1500 jbd_debug(2, "deleting unreferenced inode %lu\n",
1501 inode->i_ino);
1502 nr_orphans++;
1503 }
1504 iput(inode); /* The delete magic happens here! */
1505 }
1506
1507 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1508
1509 if (nr_orphans)
1510 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1511 PLURAL(nr_orphans));
1512 if (nr_truncates)
1513 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1514 PLURAL(nr_truncates));
1515 #ifdef CONFIG_QUOTA
1516 /* Turn quotas off */
1517 for (i = 0; i < MAXQUOTAS; i++) {
1518 if (sb_dqopt(sb)->files[i])
1519 dquot_quota_off(sb, i);
1520 }
1521 #endif
1522 sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1523 }
1524
1525 /*
1526 * Maximal file size. There is a direct, and {,double-,triple-}indirect
1527 * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1528 * We need to be 1 filesystem block less than the 2^32 sector limit.
1529 */
1530 static loff_t ext3_max_size(int bits)
1531 {
1532 loff_t res = EXT3_NDIR_BLOCKS;
1533 int meta_blocks;
1534 loff_t upper_limit;
1535
1536 /* This is calculated to be the largest file size for a
1537 * dense, file such that the total number of
1538 * sectors in the file, including data and all indirect blocks,
1539 * does not exceed 2^32 -1
1540 * __u32 i_blocks representing the total number of
1541 * 512 bytes blocks of the file
1542 */
1543 upper_limit = (1LL << 32) - 1;
1544
1545 /* total blocks in file system block size */
1546 upper_limit >>= (bits - 9);
1547
1548
1549 /* indirect blocks */
1550 meta_blocks = 1;
1551 /* double indirect blocks */
1552 meta_blocks += 1 + (1LL << (bits-2));
1553 /* tripple indirect blocks */
1554 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1555
1556 upper_limit -= meta_blocks;
1557 upper_limit <<= bits;
1558
1559 res += 1LL << (bits-2);
1560 res += 1LL << (2*(bits-2));
1561 res += 1LL << (3*(bits-2));
1562 res <<= bits;
1563 if (res > upper_limit)
1564 res = upper_limit;
1565
1566 if (res > MAX_LFS_FILESIZE)
1567 res = MAX_LFS_FILESIZE;
1568
1569 return res;
1570 }
1571
1572 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1573 ext3_fsblk_t logic_sb_block,
1574 int nr)
1575 {
1576 struct ext3_sb_info *sbi = EXT3_SB(sb);
1577 unsigned long bg, first_meta_bg;
1578 int has_super = 0;
1579
1580 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1581
1582 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1583 nr < first_meta_bg)
1584 return (logic_sb_block + nr + 1);
1585 bg = sbi->s_desc_per_block * nr;
1586 if (ext3_bg_has_super(sb, bg))
1587 has_super = 1;
1588 return (has_super + ext3_group_first_block_no(sb, bg));
1589 }
1590
1591
1592 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1593 {
1594 struct buffer_head * bh;
1595 struct ext3_super_block *es = NULL;
1596 struct ext3_sb_info *sbi;
1597 ext3_fsblk_t block;
1598 ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1599 ext3_fsblk_t logic_sb_block;
1600 unsigned long offset = 0;
1601 unsigned int journal_inum = 0;
1602 unsigned long journal_devnum = 0;
1603 unsigned long def_mount_opts;
1604 struct inode *root;
1605 int blocksize;
1606 int hblock;
1607 int db_count;
1608 int i;
1609 int needs_recovery;
1610 int ret = -EINVAL;
1611 __le32 features;
1612 int err;
1613
1614 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1615 if (!sbi)
1616 return -ENOMEM;
1617
1618 sbi->s_blockgroup_lock =
1619 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1620 if (!sbi->s_blockgroup_lock) {
1621 kfree(sbi);
1622 return -ENOMEM;
1623 }
1624 sb->s_fs_info = sbi;
1625 sbi->s_mount_opt = 0;
1626 sbi->s_resuid = EXT3_DEF_RESUID;
1627 sbi->s_resgid = EXT3_DEF_RESGID;
1628 sbi->s_sb_block = sb_block;
1629
1630 unlock_kernel();
1631
1632 blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1633 if (!blocksize) {
1634 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1635 goto out_fail;
1636 }
1637
1638 /*
1639 * The ext3 superblock will not be buffer aligned for other than 1kB
1640 * block sizes. We need to calculate the offset from buffer start.
1641 */
1642 if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1643 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1644 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1645 } else {
1646 logic_sb_block = sb_block;
1647 }
1648
1649 if (!(bh = sb_bread(sb, logic_sb_block))) {
1650 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1651 goto out_fail;
1652 }
1653 /*
1654 * Note: s_es must be initialized as soon as possible because
1655 * some ext3 macro-instructions depend on its value
1656 */
1657 es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1658 sbi->s_es = es;
1659 sb->s_magic = le16_to_cpu(es->s_magic);
1660 if (sb->s_magic != EXT3_SUPER_MAGIC)
1661 goto cantfind_ext3;
1662
1663 /* Set defaults before we parse the mount options */
1664 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1665 if (def_mount_opts & EXT3_DEFM_DEBUG)
1666 set_opt(sbi->s_mount_opt, DEBUG);
1667 if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1668 set_opt(sbi->s_mount_opt, GRPID);
1669 if (def_mount_opts & EXT3_DEFM_UID16)
1670 set_opt(sbi->s_mount_opt, NO_UID32);
1671 #ifdef CONFIG_EXT3_FS_XATTR
1672 if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1673 set_opt(sbi->s_mount_opt, XATTR_USER);
1674 #endif
1675 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1676 if (def_mount_opts & EXT3_DEFM_ACL)
1677 set_opt(sbi->s_mount_opt, POSIX_ACL);
1678 #endif
1679 if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1680 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1681 else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1682 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1683 else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1684 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1685
1686 if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1687 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1688 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1689 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1690 else
1691 set_opt(sbi->s_mount_opt, ERRORS_RO);
1692
1693 sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1694 sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1695
1696 set_opt(sbi->s_mount_opt, RESERVATION);
1697
1698 if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1699 NULL, 0))
1700 goto failed_mount;
1701
1702 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1703 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1704
1705 if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1706 (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1707 EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1708 EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1709 ext3_msg(sb, KERN_WARNING,
1710 "warning: feature flags set on rev 0 fs, "
1711 "running e2fsck is recommended");
1712 /*
1713 * Check feature flags regardless of the revision level, since we
1714 * previously didn't change the revision level when setting the flags,
1715 * so there is a chance incompat flags are set on a rev 0 filesystem.
1716 */
1717 features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1718 if (features) {
1719 ext3_msg(sb, KERN_ERR,
1720 "error: couldn't mount because of unsupported "
1721 "optional features (%x)", le32_to_cpu(features));
1722 goto failed_mount;
1723 }
1724 features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1725 if (!(sb->s_flags & MS_RDONLY) && features) {
1726 ext3_msg(sb, KERN_ERR,
1727 "error: couldn't mount RDWR because of unsupported "
1728 "optional features (%x)", le32_to_cpu(features));
1729 goto failed_mount;
1730 }
1731 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1732
1733 if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1734 blocksize > EXT3_MAX_BLOCK_SIZE) {
1735 ext3_msg(sb, KERN_ERR,
1736 "error: couldn't mount because of unsupported "
1737 "filesystem blocksize %d", blocksize);
1738 goto failed_mount;
1739 }
1740
1741 hblock = bdev_logical_block_size(sb->s_bdev);
1742 if (sb->s_blocksize != blocksize) {
1743 /*
1744 * Make sure the blocksize for the filesystem is larger
1745 * than the hardware sectorsize for the machine.
1746 */
1747 if (blocksize < hblock) {
1748 ext3_msg(sb, KERN_ERR,
1749 "error: fsblocksize %d too small for "
1750 "hardware sectorsize %d", blocksize, hblock);
1751 goto failed_mount;
1752 }
1753
1754 brelse (bh);
1755 if (!sb_set_blocksize(sb, blocksize)) {
1756 ext3_msg(sb, KERN_ERR,
1757 "error: bad blocksize %d", blocksize);
1758 goto out_fail;
1759 }
1760 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1761 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1762 bh = sb_bread(sb, logic_sb_block);
1763 if (!bh) {
1764 ext3_msg(sb, KERN_ERR,
1765 "error: can't read superblock on 2nd try");
1766 goto failed_mount;
1767 }
1768 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1769 sbi->s_es = es;
1770 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1771 ext3_msg(sb, KERN_ERR,
1772 "error: magic mismatch");
1773 goto failed_mount;
1774 }
1775 }
1776
1777 sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1778
1779 if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1780 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1781 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1782 } else {
1783 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1784 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1785 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1786 (!is_power_of_2(sbi->s_inode_size)) ||
1787 (sbi->s_inode_size > blocksize)) {
1788 ext3_msg(sb, KERN_ERR,
1789 "error: unsupported inode size: %d",
1790 sbi->s_inode_size);
1791 goto failed_mount;
1792 }
1793 }
1794 sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1795 le32_to_cpu(es->s_log_frag_size);
1796 if (blocksize != sbi->s_frag_size) {
1797 ext3_msg(sb, KERN_ERR,
1798 "error: fragsize %lu != blocksize %u (unsupported)",
1799 sbi->s_frag_size, blocksize);
1800 goto failed_mount;
1801 }
1802 sbi->s_frags_per_block = 1;
1803 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1804 sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1805 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1806 if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1807 goto cantfind_ext3;
1808 sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1809 if (sbi->s_inodes_per_block == 0)
1810 goto cantfind_ext3;
1811 sbi->s_itb_per_group = sbi->s_inodes_per_group /
1812 sbi->s_inodes_per_block;
1813 sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1814 sbi->s_sbh = bh;
1815 sbi->s_mount_state = le16_to_cpu(es->s_state);
1816 sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1817 sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1818 for (i=0; i < 4; i++)
1819 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1820 sbi->s_def_hash_version = es->s_def_hash_version;
1821 i = le32_to_cpu(es->s_flags);
1822 if (i & EXT2_FLAGS_UNSIGNED_HASH)
1823 sbi->s_hash_unsigned = 3;
1824 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1825 #ifdef __CHAR_UNSIGNED__
1826 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1827 sbi->s_hash_unsigned = 3;
1828 #else
1829 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1830 #endif
1831 }
1832
1833 if (sbi->s_blocks_per_group > blocksize * 8) {
1834 ext3_msg(sb, KERN_ERR,
1835 "#blocks per group too big: %lu",
1836 sbi->s_blocks_per_group);
1837 goto failed_mount;
1838 }
1839 if (sbi->s_frags_per_group > blocksize * 8) {
1840 ext3_msg(sb, KERN_ERR,
1841 "error: #fragments per group too big: %lu",
1842 sbi->s_frags_per_group);
1843 goto failed_mount;
1844 }
1845 if (sbi->s_inodes_per_group > blocksize * 8) {
1846 ext3_msg(sb, KERN_ERR,
1847 "error: #inodes per group too big: %lu",
1848 sbi->s_inodes_per_group);
1849 goto failed_mount;
1850 }
1851
1852 if (le32_to_cpu(es->s_blocks_count) >
1853 (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1854 ext3_msg(sb, KERN_ERR,
1855 "error: filesystem is too large to mount safely");
1856 if (sizeof(sector_t) < 8)
1857 ext3_msg(sb, KERN_ERR,
1858 "error: CONFIG_LBDAF not enabled");
1859 goto failed_mount;
1860 }
1861
1862 if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1863 goto cantfind_ext3;
1864 sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1865 le32_to_cpu(es->s_first_data_block) - 1)
1866 / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1867 db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1868 EXT3_DESC_PER_BLOCK(sb);
1869 sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1870 GFP_KERNEL);
1871 if (sbi->s_group_desc == NULL) {
1872 ext3_msg(sb, KERN_ERR,
1873 "error: not enough memory");
1874 goto failed_mount;
1875 }
1876
1877 bgl_lock_init(sbi->s_blockgroup_lock);
1878
1879 for (i = 0; i < db_count; i++) {
1880 block = descriptor_loc(sb, logic_sb_block, i);
1881 sbi->s_group_desc[i] = sb_bread(sb, block);
1882 if (!sbi->s_group_desc[i]) {
1883 ext3_msg(sb, KERN_ERR,
1884 "error: can't read group descriptor %d", i);
1885 db_count = i;
1886 goto failed_mount2;
1887 }
1888 }
1889 if (!ext3_check_descriptors (sb)) {
1890 ext3_msg(sb, KERN_ERR,
1891 "error: group descriptors corrupted");
1892 goto failed_mount2;
1893 }
1894 sbi->s_gdb_count = db_count;
1895 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1896 spin_lock_init(&sbi->s_next_gen_lock);
1897
1898 /* per fileystem reservation list head & lock */
1899 spin_lock_init(&sbi->s_rsv_window_lock);
1900 sbi->s_rsv_window_root = RB_ROOT;
1901 /* Add a single, static dummy reservation to the start of the
1902 * reservation window list --- it gives us a placeholder for
1903 * append-at-start-of-list which makes the allocation logic
1904 * _much_ simpler. */
1905 sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1906 sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1907 sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1908 sbi->s_rsv_window_head.rsv_goal_size = 0;
1909 ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1910
1911 /*
1912 * set up enough so that it can read an inode
1913 */
1914 sb->s_op = &ext3_sops;
1915 sb->s_export_op = &ext3_export_ops;
1916 sb->s_xattr = ext3_xattr_handlers;
1917 #ifdef CONFIG_QUOTA
1918 sb->s_qcop = &ext3_qctl_operations;
1919 sb->dq_op = &ext3_quota_operations;
1920 #endif
1921 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1922 mutex_init(&sbi->s_orphan_lock);
1923 mutex_init(&sbi->s_resize_lock);
1924
1925 sb->s_root = NULL;
1926
1927 needs_recovery = (es->s_last_orphan != 0 ||
1928 EXT3_HAS_INCOMPAT_FEATURE(sb,
1929 EXT3_FEATURE_INCOMPAT_RECOVER));
1930
1931 /*
1932 * The first inode we look at is the journal inode. Don't try
1933 * root first: it may be modified in the journal!
1934 */
1935 if (!test_opt(sb, NOLOAD) &&
1936 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1937 if (ext3_load_journal(sb, es, journal_devnum))
1938 goto failed_mount2;
1939 } else if (journal_inum) {
1940 if (ext3_create_journal(sb, es, journal_inum))
1941 goto failed_mount2;
1942 } else {
1943 if (!silent)
1944 ext3_msg(sb, KERN_ERR,
1945 "error: no journal found. "
1946 "mounting ext3 over ext2?");
1947 goto failed_mount2;
1948 }
1949 err = percpu_counter_init(&sbi->s_freeblocks_counter,
1950 ext3_count_free_blocks(sb));
1951 if (!err) {
1952 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1953 ext3_count_free_inodes(sb));
1954 }
1955 if (!err) {
1956 err = percpu_counter_init(&sbi->s_dirs_counter,
1957 ext3_count_dirs(sb));
1958 }
1959 if (err) {
1960 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1961 goto failed_mount3;
1962 }
1963
1964 /* We have now updated the journal if required, so we can
1965 * validate the data journaling mode. */
1966 switch (test_opt(sb, DATA_FLAGS)) {
1967 case 0:
1968 /* No mode set, assume a default based on the journal
1969 capabilities: ORDERED_DATA if the journal can
1970 cope, else JOURNAL_DATA */
1971 if (journal_check_available_features
1972 (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1973 set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1974 else
1975 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1976 break;
1977
1978 case EXT3_MOUNT_ORDERED_DATA:
1979 case EXT3_MOUNT_WRITEBACK_DATA:
1980 if (!journal_check_available_features
1981 (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1982 ext3_msg(sb, KERN_ERR,
1983 "error: journal does not support "
1984 "requested data journaling mode");
1985 goto failed_mount3;
1986 }
1987 default:
1988 break;
1989 }
1990
1991 /*
1992 * The journal_load will have done any necessary log recovery,
1993 * so we can safely mount the rest of the filesystem now.
1994 */
1995
1996 root = ext3_iget(sb, EXT3_ROOT_INO);
1997 if (IS_ERR(root)) {
1998 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
1999 ret = PTR_ERR(root);
2000 goto failed_mount3;
2001 }
2002 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2003 iput(root);
2004 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2005 goto failed_mount3;
2006 }
2007 sb->s_root = d_alloc_root(root);
2008 if (!sb->s_root) {
2009 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2010 iput(root);
2011 ret = -ENOMEM;
2012 goto failed_mount3;
2013 }
2014
2015 ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2016
2017 EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2018 ext3_orphan_cleanup(sb, es);
2019 EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2020 if (needs_recovery)
2021 ext3_msg(sb, KERN_INFO, "recovery complete");
2022 ext3_mark_recovery_complete(sb, es);
2023 ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2024 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2025 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2026 "writeback");
2027
2028 lock_kernel();
2029 return 0;
2030
2031 cantfind_ext3:
2032 if (!silent)
2033 ext3_msg(sb, KERN_INFO,
2034 "error: can't find ext3 filesystem on dev %s.",
2035 sb->s_id);
2036 goto failed_mount;
2037
2038 failed_mount3:
2039 percpu_counter_destroy(&sbi->s_freeblocks_counter);
2040 percpu_counter_destroy(&sbi->s_freeinodes_counter);
2041 percpu_counter_destroy(&sbi->s_dirs_counter);
2042 journal_destroy(sbi->s_journal);
2043 failed_mount2:
2044 for (i = 0; i < db_count; i++)
2045 brelse(sbi->s_group_desc[i]);
2046 kfree(sbi->s_group_desc);
2047 failed_mount:
2048 #ifdef CONFIG_QUOTA
2049 for (i = 0; i < MAXQUOTAS; i++)
2050 kfree(sbi->s_qf_names[i]);
2051 #endif
2052 ext3_blkdev_remove(sbi);
2053 brelse(bh);
2054 out_fail:
2055 sb->s_fs_info = NULL;
2056 kfree(sbi->s_blockgroup_lock);
2057 kfree(sbi);
2058 lock_kernel();
2059 return ret;
2060 }
2061
2062 /*
2063 * Setup any per-fs journal parameters now. We'll do this both on
2064 * initial mount, once the journal has been initialised but before we've
2065 * done any recovery; and again on any subsequent remount.
2066 */
2067 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2068 {
2069 struct ext3_sb_info *sbi = EXT3_SB(sb);
2070
2071 if (sbi->s_commit_interval)
2072 journal->j_commit_interval = sbi->s_commit_interval;
2073 /* We could also set up an ext3-specific default for the commit
2074 * interval here, but for now we'll just fall back to the jbd
2075 * default. */
2076
2077 spin_lock(&journal->j_state_lock);
2078 if (test_opt(sb, BARRIER))
2079 journal->j_flags |= JFS_BARRIER;
2080 else
2081 journal->j_flags &= ~JFS_BARRIER;
2082 if (test_opt(sb, DATA_ERR_ABORT))
2083 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2084 else
2085 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2086 spin_unlock(&journal->j_state_lock);
2087 }
2088
2089 static journal_t *ext3_get_journal(struct super_block *sb,
2090 unsigned int journal_inum)
2091 {
2092 struct inode *journal_inode;
2093 journal_t *journal;
2094
2095 /* First, test for the existence of a valid inode on disk. Bad
2096 * things happen if we iget() an unused inode, as the subsequent
2097 * iput() will try to delete it. */
2098
2099 journal_inode = ext3_iget(sb, journal_inum);
2100 if (IS_ERR(journal_inode)) {
2101 ext3_msg(sb, KERN_ERR, "error: no journal found");
2102 return NULL;
2103 }
2104 if (!journal_inode->i_nlink) {
2105 make_bad_inode(journal_inode);
2106 iput(journal_inode);
2107 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2108 return NULL;
2109 }
2110
2111 jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2112 journal_inode, journal_inode->i_size);
2113 if (!S_ISREG(journal_inode->i_mode)) {
2114 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2115 iput(journal_inode);
2116 return NULL;
2117 }
2118
2119 journal = journal_init_inode(journal_inode);
2120 if (!journal) {
2121 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2122 iput(journal_inode);
2123 return NULL;
2124 }
2125 journal->j_private = sb;
2126 ext3_init_journal_params(sb, journal);
2127 return journal;
2128 }
2129
2130 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2131 dev_t j_dev)
2132 {
2133 struct buffer_head * bh;
2134 journal_t *journal;
2135 ext3_fsblk_t start;
2136 ext3_fsblk_t len;
2137 int hblock, blocksize;
2138 ext3_fsblk_t sb_block;
2139 unsigned long offset;
2140 struct ext3_super_block * es;
2141 struct block_device *bdev;
2142
2143 bdev = ext3_blkdev_get(j_dev, sb);
2144 if (bdev == NULL)
2145 return NULL;
2146
2147 if (bd_claim(bdev, sb)) {
2148 ext3_msg(sb, KERN_ERR,
2149 "error: failed to claim external journal device");
2150 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2151 return NULL;
2152 }
2153
2154 blocksize = sb->s_blocksize;
2155 hblock = bdev_logical_block_size(bdev);
2156 if (blocksize < hblock) {
2157 ext3_msg(sb, KERN_ERR,
2158 "error: blocksize too small for journal device");
2159 goto out_bdev;
2160 }
2161
2162 sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2163 offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2164 set_blocksize(bdev, blocksize);
2165 if (!(bh = __bread(bdev, sb_block, blocksize))) {
2166 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2167 "external journal");
2168 goto out_bdev;
2169 }
2170
2171 es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2172 if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2173 !(le32_to_cpu(es->s_feature_incompat) &
2174 EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2175 ext3_msg(sb, KERN_ERR, "error: external journal has "
2176 "bad superblock");
2177 brelse(bh);
2178 goto out_bdev;
2179 }
2180
2181 if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2182 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2183 brelse(bh);
2184 goto out_bdev;
2185 }
2186
2187 len = le32_to_cpu(es->s_blocks_count);
2188 start = sb_block + 1;
2189 brelse(bh); /* we're done with the superblock */
2190
2191 journal = journal_init_dev(bdev, sb->s_bdev,
2192 start, len, blocksize);
2193 if (!journal) {
2194 ext3_msg(sb, KERN_ERR,
2195 "error: failed to create device journal");
2196 goto out_bdev;
2197 }
2198 journal->j_private = sb;
2199 ll_rw_block(READ, 1, &journal->j_sb_buffer);
2200 wait_on_buffer(journal->j_sb_buffer);
2201 if (!buffer_uptodate(journal->j_sb_buffer)) {
2202 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2203 goto out_journal;
2204 }
2205 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2206 ext3_msg(sb, KERN_ERR,
2207 "error: external journal has more than one "
2208 "user (unsupported) - %d",
2209 be32_to_cpu(journal->j_superblock->s_nr_users));
2210 goto out_journal;
2211 }
2212 EXT3_SB(sb)->journal_bdev = bdev;
2213 ext3_init_journal_params(sb, journal);
2214 return journal;
2215 out_journal:
2216 journal_destroy(journal);
2217 out_bdev:
2218 ext3_blkdev_put(bdev);
2219 return NULL;
2220 }
2221
2222 static int ext3_load_journal(struct super_block *sb,
2223 struct ext3_super_block *es,
2224 unsigned long journal_devnum)
2225 {
2226 journal_t *journal;
2227 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2228 dev_t journal_dev;
2229 int err = 0;
2230 int really_read_only;
2231
2232 if (journal_devnum &&
2233 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2234 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2235 "numbers have changed");
2236 journal_dev = new_decode_dev(journal_devnum);
2237 } else
2238 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2239
2240 really_read_only = bdev_read_only(sb->s_bdev);
2241
2242 /*
2243 * Are we loading a blank journal or performing recovery after a
2244 * crash? For recovery, we need to check in advance whether we
2245 * can get read-write access to the device.
2246 */
2247
2248 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2249 if (sb->s_flags & MS_RDONLY) {
2250 ext3_msg(sb, KERN_INFO,
2251 "recovery required on readonly filesystem");
2252 if (really_read_only) {
2253 ext3_msg(sb, KERN_ERR, "error: write access "
2254 "unavailable, cannot proceed");
2255 return -EROFS;
2256 }
2257 ext3_msg(sb, KERN_INFO,
2258 "write access will be enabled during recovery");
2259 }
2260 }
2261
2262 if (journal_inum && journal_dev) {
2263 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2264 "and inode journals");
2265 return -EINVAL;
2266 }
2267
2268 if (journal_inum) {
2269 if (!(journal = ext3_get_journal(sb, journal_inum)))
2270 return -EINVAL;
2271 } else {
2272 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2273 return -EINVAL;
2274 }
2275
2276 if (!(journal->j_flags & JFS_BARRIER))
2277 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2278
2279 if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2280 err = journal_update_format(journal);
2281 if (err) {
2282 ext3_msg(sb, KERN_ERR, "error updating journal");
2283 journal_destroy(journal);
2284 return err;
2285 }
2286 }
2287
2288 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2289 err = journal_wipe(journal, !really_read_only);
2290 if (!err)
2291 err = journal_load(journal);
2292
2293 if (err) {
2294 ext3_msg(sb, KERN_ERR, "error loading journal");
2295 journal_destroy(journal);
2296 return err;
2297 }
2298
2299 EXT3_SB(sb)->s_journal = journal;
2300 ext3_clear_journal_err(sb, es);
2301
2302 if (journal_devnum &&
2303 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2304 es->s_journal_dev = cpu_to_le32(journal_devnum);
2305
2306 /* Make sure we flush the recovery flag to disk. */
2307 ext3_commit_super(sb, es, 1);
2308 }
2309
2310 return 0;
2311 }
2312
2313 static int ext3_create_journal(struct super_block *sb,
2314 struct ext3_super_block *es,
2315 unsigned int journal_inum)
2316 {
2317 journal_t *journal;
2318 int err;
2319
2320 if (sb->s_flags & MS_RDONLY) {
2321 ext3_msg(sb, KERN_ERR,
2322 "error: readonly filesystem when trying to "
2323 "create journal");
2324 return -EROFS;
2325 }
2326
2327 journal = ext3_get_journal(sb, journal_inum);
2328 if (!journal)
2329 return -EINVAL;
2330
2331 ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2332 journal_inum);
2333
2334 err = journal_create(journal);
2335 if (err) {
2336 ext3_msg(sb, KERN_ERR, "error creating journal");
2337 journal_destroy(journal);
2338 return -EIO;
2339 }
2340
2341 EXT3_SB(sb)->s_journal = journal;
2342
2343 ext3_update_dynamic_rev(sb);
2344 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2345 EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2346
2347 es->s_journal_inum = cpu_to_le32(journal_inum);
2348
2349 /* Make sure we flush the recovery flag to disk. */
2350 ext3_commit_super(sb, es, 1);
2351
2352 return 0;
2353 }
2354
2355 static int ext3_commit_super(struct super_block *sb,
2356 struct ext3_super_block *es,
2357 int sync)
2358 {
2359 struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2360 int error = 0;
2361
2362 if (!sbh)
2363 return error;
2364 /*
2365 * If the file system is mounted read-only, don't update the
2366 * superblock write time. This avoids updating the superblock
2367 * write time when we are mounting the root file system
2368 * read/only but we need to replay the journal; at that point,
2369 * for people who are east of GMT and who make their clock
2370 * tick in localtime for Windows bug-for-bug compatibility,
2371 * the clock is set in the future, and this will cause e2fsck
2372 * to complain and force a full file system check.
2373 */
2374 if (!(sb->s_flags & MS_RDONLY))
2375 es->s_wtime = cpu_to_le32(get_seconds());
2376 es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2377 es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2378 BUFFER_TRACE(sbh, "marking dirty");
2379 mark_buffer_dirty(sbh);
2380 if (sync)
2381 error = sync_dirty_buffer(sbh);
2382 return error;
2383 }
2384
2385
2386 /*
2387 * Have we just finished recovery? If so, and if we are mounting (or
2388 * remounting) the filesystem readonly, then we will end up with a
2389 * consistent fs on disk. Record that fact.
2390 */
2391 static void ext3_mark_recovery_complete(struct super_block * sb,
2392 struct ext3_super_block * es)
2393 {
2394 journal_t *journal = EXT3_SB(sb)->s_journal;
2395
2396 journal_lock_updates(journal);
2397 if (journal_flush(journal) < 0)
2398 goto out;
2399
2400 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2401 sb->s_flags & MS_RDONLY) {
2402 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2403 ext3_commit_super(sb, es, 1);
2404 }
2405
2406 out:
2407 journal_unlock_updates(journal);
2408 }
2409
2410 /*
2411 * If we are mounting (or read-write remounting) a filesystem whose journal
2412 * has recorded an error from a previous lifetime, move that error to the
2413 * main filesystem now.
2414 */
2415 static void ext3_clear_journal_err(struct super_block *sb,
2416 struct ext3_super_block *es)
2417 {
2418 journal_t *journal;
2419 int j_errno;
2420 const char *errstr;
2421
2422 journal = EXT3_SB(sb)->s_journal;
2423
2424 /*
2425 * Now check for any error status which may have been recorded in the
2426 * journal by a prior ext3_error() or ext3_abort()
2427 */
2428
2429 j_errno = journal_errno(journal);
2430 if (j_errno) {
2431 char nbuf[16];
2432
2433 errstr = ext3_decode_error(sb, j_errno, nbuf);
2434 ext3_warning(sb, __func__, "Filesystem error recorded "
2435 "from previous mount: %s", errstr);
2436 ext3_warning(sb, __func__, "Marking fs in need of "
2437 "filesystem check.");
2438
2439 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2440 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2441 ext3_commit_super (sb, es, 1);
2442
2443 journal_clear_err(journal);
2444 }
2445 }
2446
2447 /*
2448 * Force the running and committing transactions to commit,
2449 * and wait on the commit.
2450 */
2451 int ext3_force_commit(struct super_block *sb)
2452 {
2453 journal_t *journal;
2454 int ret;
2455
2456 if (sb->s_flags & MS_RDONLY)
2457 return 0;
2458
2459 journal = EXT3_SB(sb)->s_journal;
2460 ret = ext3_journal_force_commit(journal);
2461 return ret;
2462 }
2463
2464 static int ext3_sync_fs(struct super_block *sb, int wait)
2465 {
2466 tid_t target;
2467
2468 if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2469 if (wait)
2470 log_wait_commit(EXT3_SB(sb)->s_journal, target);
2471 }
2472 return 0;
2473 }
2474
2475 /*
2476 * LVM calls this function before a (read-only) snapshot is created. This
2477 * gives us a chance to flush the journal completely and mark the fs clean.
2478 */
2479 static int ext3_freeze(struct super_block *sb)
2480 {
2481 int error = 0;
2482 journal_t *journal;
2483
2484 if (!(sb->s_flags & MS_RDONLY)) {
2485 journal = EXT3_SB(sb)->s_journal;
2486
2487 /* Now we set up the journal barrier. */
2488 journal_lock_updates(journal);
2489
2490 /*
2491 * We don't want to clear needs_recovery flag when we failed
2492 * to flush the journal.
2493 */
2494 error = journal_flush(journal);
2495 if (error < 0)
2496 goto out;
2497
2498 /* Journal blocked and flushed, clear needs_recovery flag. */
2499 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2500 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2501 if (error)
2502 goto out;
2503 }
2504 return 0;
2505
2506 out:
2507 journal_unlock_updates(journal);
2508 return error;
2509 }
2510
2511 /*
2512 * Called by LVM after the snapshot is done. We need to reset the RECOVER
2513 * flag here, even though the filesystem is not technically dirty yet.
2514 */
2515 static int ext3_unfreeze(struct super_block *sb)
2516 {
2517 if (!(sb->s_flags & MS_RDONLY)) {
2518 lock_super(sb);
2519 /* Reser the needs_recovery flag before the fs is unlocked. */
2520 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2521 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2522 unlock_super(sb);
2523 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2524 }
2525 return 0;
2526 }
2527
2528 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2529 {
2530 struct ext3_super_block * es;
2531 struct ext3_sb_info *sbi = EXT3_SB(sb);
2532 ext3_fsblk_t n_blocks_count = 0;
2533 unsigned long old_sb_flags;
2534 struct ext3_mount_options old_opts;
2535 int enable_quota = 0;
2536 int err;
2537 #ifdef CONFIG_QUOTA
2538 int i;
2539 #endif
2540
2541 lock_kernel();
2542
2543 /* Store the original options */
2544 lock_super(sb);
2545 old_sb_flags = sb->s_flags;
2546 old_opts.s_mount_opt = sbi->s_mount_opt;
2547 old_opts.s_resuid = sbi->s_resuid;
2548 old_opts.s_resgid = sbi->s_resgid;
2549 old_opts.s_commit_interval = sbi->s_commit_interval;
2550 #ifdef CONFIG_QUOTA
2551 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2552 for (i = 0; i < MAXQUOTAS; i++)
2553 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2554 #endif
2555
2556 /*
2557 * Allow the "check" option to be passed as a remount option.
2558 */
2559 if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2560 err = -EINVAL;
2561 goto restore_opts;
2562 }
2563
2564 if (test_opt(sb, ABORT))
2565 ext3_abort(sb, __func__, "Abort forced by user");
2566
2567 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2568 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2569
2570 es = sbi->s_es;
2571
2572 ext3_init_journal_params(sb, sbi->s_journal);
2573
2574 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2575 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2576 if (test_opt(sb, ABORT)) {
2577 err = -EROFS;
2578 goto restore_opts;
2579 }
2580
2581 if (*flags & MS_RDONLY) {
2582 err = dquot_suspend(sb, -1);
2583 if (err < 0)
2584 goto restore_opts;
2585
2586 /*
2587 * First of all, the unconditional stuff we have to do
2588 * to disable replay of the journal when we next remount
2589 */
2590 sb->s_flags |= MS_RDONLY;
2591
2592 /*
2593 * OK, test if we are remounting a valid rw partition
2594 * readonly, and if so set the rdonly flag and then
2595 * mark the partition as valid again.
2596 */
2597 if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2598 (sbi->s_mount_state & EXT3_VALID_FS))
2599 es->s_state = cpu_to_le16(sbi->s_mount_state);
2600
2601 ext3_mark_recovery_complete(sb, es);
2602 } else {
2603 __le32 ret;
2604 if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2605 ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2606 ext3_msg(sb, KERN_WARNING,
2607 "warning: couldn't remount RDWR "
2608 "because of unsupported optional "
2609 "features (%x)", le32_to_cpu(ret));
2610 err = -EROFS;
2611 goto restore_opts;
2612 }
2613
2614 /*
2615 * If we have an unprocessed orphan list hanging
2616 * around from a previously readonly bdev mount,
2617 * require a full umount/remount for now.
2618 */
2619 if (es->s_last_orphan) {
2620 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2621 "remount RDWR because of unprocessed "
2622 "orphan inode list. Please "
2623 "umount/remount instead.");
2624 err = -EINVAL;
2625 goto restore_opts;
2626 }
2627
2628 /*
2629 * Mounting a RDONLY partition read-write, so reread
2630 * and store the current valid flag. (It may have
2631 * been changed by e2fsck since we originally mounted
2632 * the partition.)
2633 */
2634 ext3_clear_journal_err(sb, es);
2635 sbi->s_mount_state = le16_to_cpu(es->s_state);
2636 if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2637 goto restore_opts;
2638 if (!ext3_setup_super (sb, es, 0))
2639 sb->s_flags &= ~MS_RDONLY;
2640 enable_quota = 1;
2641 }
2642 }
2643 #ifdef CONFIG_QUOTA
2644 /* Release old quota file names */
2645 for (i = 0; i < MAXQUOTAS; i++)
2646 if (old_opts.s_qf_names[i] &&
2647 old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2648 kfree(old_opts.s_qf_names[i]);
2649 #endif
2650 unlock_super(sb);
2651 unlock_kernel();
2652
2653 if (enable_quota)
2654 dquot_resume(sb, -1);
2655 return 0;
2656 restore_opts:
2657 sb->s_flags = old_sb_flags;
2658 sbi->s_mount_opt = old_opts.s_mount_opt;
2659 sbi->s_resuid = old_opts.s_resuid;
2660 sbi->s_resgid = old_opts.s_resgid;
2661 sbi->s_commit_interval = old_opts.s_commit_interval;
2662 #ifdef CONFIG_QUOTA
2663 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2664 for (i = 0; i < MAXQUOTAS; i++) {
2665 if (sbi->s_qf_names[i] &&
2666 old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2667 kfree(sbi->s_qf_names[i]);
2668 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2669 }
2670 #endif
2671 unlock_super(sb);
2672 unlock_kernel();
2673 return err;
2674 }
2675
2676 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2677 {
2678 struct super_block *sb = dentry->d_sb;
2679 struct ext3_sb_info *sbi = EXT3_SB(sb);
2680 struct ext3_super_block *es = sbi->s_es;
2681 u64 fsid;
2682
2683 if (test_opt(sb, MINIX_DF)) {
2684 sbi->s_overhead_last = 0;
2685 } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2686 unsigned long ngroups = sbi->s_groups_count, i;
2687 ext3_fsblk_t overhead = 0;
2688 smp_rmb();
2689
2690 /*
2691 * Compute the overhead (FS structures). This is constant
2692 * for a given filesystem unless the number of block groups
2693 * changes so we cache the previous value until it does.
2694 */
2695
2696 /*
2697 * All of the blocks before first_data_block are
2698 * overhead
2699 */
2700 overhead = le32_to_cpu(es->s_first_data_block);
2701
2702 /*
2703 * Add the overhead attributed to the superblock and
2704 * block group descriptors. If the sparse superblocks
2705 * feature is turned on, then not all groups have this.
2706 */
2707 for (i = 0; i < ngroups; i++) {
2708 overhead += ext3_bg_has_super(sb, i) +
2709 ext3_bg_num_gdb(sb, i);
2710 cond_resched();
2711 }
2712
2713 /*
2714 * Every block group has an inode bitmap, a block
2715 * bitmap, and an inode table.
2716 */
2717 overhead += ngroups * (2 + sbi->s_itb_per_group);
2718 sbi->s_overhead_last = overhead;
2719 smp_wmb();
2720 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2721 }
2722
2723 buf->f_type = EXT3_SUPER_MAGIC;
2724 buf->f_bsize = sb->s_blocksize;
2725 buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2726 buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2727 buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2728 if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2729 buf->f_bavail = 0;
2730 buf->f_files = le32_to_cpu(es->s_inodes_count);
2731 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2732 buf->f_namelen = EXT3_NAME_LEN;
2733 fsid = le64_to_cpup((void *)es->s_uuid) ^
2734 le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2735 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2736 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2737 return 0;
2738 }
2739
2740 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2741 * is locked for write. Otherwise the are possible deadlocks:
2742 * Process 1 Process 2
2743 * ext3_create() quota_sync()
2744 * journal_start() write_dquot()
2745 * dquot_initialize() down(dqio_mutex)
2746 * down(dqio_mutex) journal_start()
2747 *
2748 */
2749
2750 #ifdef CONFIG_QUOTA
2751
2752 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2753 {
2754 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2755 }
2756
2757 static int ext3_write_dquot(struct dquot *dquot)
2758 {
2759 int ret, err;
2760 handle_t *handle;
2761 struct inode *inode;
2762
2763 inode = dquot_to_inode(dquot);
2764 handle = ext3_journal_start(inode,
2765 EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2766 if (IS_ERR(handle))
2767 return PTR_ERR(handle);
2768 ret = dquot_commit(dquot);
2769 err = ext3_journal_stop(handle);
2770 if (!ret)
2771 ret = err;
2772 return ret;
2773 }
2774
2775 static int ext3_acquire_dquot(struct dquot *dquot)
2776 {
2777 int ret, err;
2778 handle_t *handle;
2779
2780 handle = ext3_journal_start(dquot_to_inode(dquot),
2781 EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2782 if (IS_ERR(handle))
2783 return PTR_ERR(handle);
2784 ret = dquot_acquire(dquot);
2785 err = ext3_journal_stop(handle);
2786 if (!ret)
2787 ret = err;
2788 return ret;
2789 }
2790
2791 static int ext3_release_dquot(struct dquot *dquot)
2792 {
2793 int ret, err;
2794 handle_t *handle;
2795
2796 handle = ext3_journal_start(dquot_to_inode(dquot),
2797 EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2798 if (IS_ERR(handle)) {
2799 /* Release dquot anyway to avoid endless cycle in dqput() */
2800 dquot_release(dquot);
2801 return PTR_ERR(handle);
2802 }
2803 ret = dquot_release(dquot);
2804 err = ext3_journal_stop(handle);
2805 if (!ret)
2806 ret = err;
2807 return ret;
2808 }
2809
2810 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2811 {
2812 /* Are we journaling quotas? */
2813 if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2814 EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2815 dquot_mark_dquot_dirty(dquot);
2816 return ext3_write_dquot(dquot);
2817 } else {
2818 return dquot_mark_dquot_dirty(dquot);
2819 }
2820 }
2821
2822 static int ext3_write_info(struct super_block *sb, int type)
2823 {
2824 int ret, err;
2825 handle_t *handle;
2826
2827 /* Data block + inode block */
2828 handle = ext3_journal_start(sb->s_root->d_inode, 2);
2829 if (IS_ERR(handle))
2830 return PTR_ERR(handle);
2831 ret = dquot_commit_info(sb, type);
2832 err = ext3_journal_stop(handle);
2833 if (!ret)
2834 ret = err;
2835 return ret;
2836 }
2837
2838 /*
2839 * Turn on quotas during mount time - we need to find
2840 * the quota file and such...
2841 */
2842 static int ext3_quota_on_mount(struct super_block *sb, int type)
2843 {
2844 return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2845 EXT3_SB(sb)->s_jquota_fmt, type);
2846 }
2847
2848 /*
2849 * Standard function to be called on quota_on
2850 */
2851 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2852 char *name)
2853 {
2854 int err;
2855 struct path path;
2856
2857 if (!test_opt(sb, QUOTA))
2858 return -EINVAL;
2859
2860 err = kern_path(name, LOOKUP_FOLLOW, &path);
2861 if (err)
2862 return err;
2863
2864 /* Quotafile not on the same filesystem? */
2865 if (path.mnt->mnt_sb != sb) {
2866 path_put(&path);
2867 return -EXDEV;
2868 }
2869 /* Journaling quota? */
2870 if (EXT3_SB(sb)->s_qf_names[type]) {
2871 /* Quotafile not of fs root? */
2872 if (path.dentry->d_parent != sb->s_root)
2873 ext3_msg(sb, KERN_WARNING,
2874 "warning: Quota file not on filesystem root. "
2875 "Journaled quota will not work.");
2876 }
2877
2878 /*
2879 * When we journal data on quota file, we have to flush journal to see
2880 * all updates to the file when we bypass pagecache...
2881 */
2882 if (ext3_should_journal_data(path.dentry->d_inode)) {
2883 /*
2884 * We don't need to lock updates but journal_flush() could
2885 * otherwise be livelocked...
2886 */
2887 journal_lock_updates(EXT3_SB(sb)->s_journal);
2888 err = journal_flush(EXT3_SB(sb)->s_journal);
2889 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2890 if (err) {
2891 path_put(&path);
2892 return err;
2893 }
2894 }
2895
2896 err = dquot_quota_on_path(sb, type, format_id, &path);
2897 path_put(&path);
2898 return err;
2899 }
2900
2901 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2902 * acquiring the locks... As quota files are never truncated and quota code
2903 * itself serializes the operations (and noone else should touch the files)
2904 * we don't have to be afraid of races */
2905 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2906 size_t len, loff_t off)
2907 {
2908 struct inode *inode = sb_dqopt(sb)->files[type];
2909 sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2910 int err = 0;
2911 int offset = off & (sb->s_blocksize - 1);
2912 int tocopy;
2913 size_t toread;
2914 struct buffer_head *bh;
2915 loff_t i_size = i_size_read(inode);
2916
2917 if (off > i_size)
2918 return 0;
2919 if (off+len > i_size)
2920 len = i_size-off;
2921 toread = len;
2922 while (toread > 0) {
2923 tocopy = sb->s_blocksize - offset < toread ?
2924 sb->s_blocksize - offset : toread;
2925 bh = ext3_bread(NULL, inode, blk, 0, &err);
2926 if (err)
2927 return err;
2928 if (!bh) /* A hole? */
2929 memset(data, 0, tocopy);
2930 else
2931 memcpy(data, bh->b_data+offset, tocopy);
2932 brelse(bh);
2933 offset = 0;
2934 toread -= tocopy;
2935 data += tocopy;
2936 blk++;
2937 }
2938 return len;
2939 }
2940
2941 /* Write to quotafile (we know the transaction is already started and has
2942 * enough credits) */
2943 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2944 const char *data, size_t len, loff_t off)
2945 {
2946 struct inode *inode = sb_dqopt(sb)->files[type];
2947 sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2948 int err = 0;
2949 int offset = off & (sb->s_blocksize - 1);
2950 int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2951 struct buffer_head *bh;
2952 handle_t *handle = journal_current_handle();
2953
2954 if (!handle) {
2955 ext3_msg(sb, KERN_WARNING,
2956 "warning: quota write (off=%llu, len=%llu)"
2957 " cancelled because transaction is not started.",
2958 (unsigned long long)off, (unsigned long long)len);
2959 return -EIO;
2960 }
2961
2962 /*
2963 * Since we account only one data block in transaction credits,
2964 * then it is impossible to cross a block boundary.
2965 */
2966 if (sb->s_blocksize - offset < len) {
2967 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
2968 " cancelled because not block aligned",
2969 (unsigned long long)off, (unsigned long long)len);
2970 return -EIO;
2971 }
2972 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2973 bh = ext3_bread(handle, inode, blk, 1, &err);
2974 if (!bh)
2975 goto out;
2976 if (journal_quota) {
2977 err = ext3_journal_get_write_access(handle, bh);
2978 if (err) {
2979 brelse(bh);
2980 goto out;
2981 }
2982 }
2983 lock_buffer(bh);
2984 memcpy(bh->b_data+offset, data, len);
2985 flush_dcache_page(bh->b_page);
2986 unlock_buffer(bh);
2987 if (journal_quota)
2988 err = ext3_journal_dirty_metadata(handle, bh);
2989 else {
2990 /* Always do at least ordered writes for quotas */
2991 err = ext3_journal_dirty_data(handle, bh);
2992 mark_buffer_dirty(bh);
2993 }
2994 brelse(bh);
2995 out:
2996 if (err) {
2997 mutex_unlock(&inode->i_mutex);
2998 return err;
2999 }
3000 if (inode->i_size < off + len) {
3001 i_size_write(inode, off + len);
3002 EXT3_I(inode)->i_disksize = inode->i_size;
3003 }
3004 inode->i_version++;
3005 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3006 ext3_mark_inode_dirty(handle, inode);
3007 mutex_unlock(&inode->i_mutex);
3008 return len;
3009 }
3010
3011 #endif
3012
3013 static int ext3_get_sb(struct file_system_type *fs_type,
3014 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3015 {
3016 return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3017 }
3018
3019 static struct file_system_type ext3_fs_type = {
3020 .owner = THIS_MODULE,
3021 .name = "ext3",
3022 .get_sb = ext3_get_sb,
3023 .kill_sb = kill_block_super,
3024 .fs_flags = FS_REQUIRES_DEV,
3025 };
3026
3027 static int __init init_ext3_fs(void)
3028 {
3029 int err = init_ext3_xattr();
3030 if (err)
3031 return err;
3032 err = init_inodecache();
3033 if (err)
3034 goto out1;
3035 err = register_filesystem(&ext3_fs_type);
3036 if (err)
3037 goto out;
3038 return 0;
3039 out:
3040 destroy_inodecache();
3041 out1:
3042 exit_ext3_xattr();
3043 return err;
3044 }
3045
3046 static void __exit exit_ext3_fs(void)
3047 {
3048 unregister_filesystem(&ext3_fs_type);
3049 destroy_inodecache();
3050 exit_ext3_xattr();
3051 }
3052
3053 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3054 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3055 MODULE_LICENSE("GPL");
3056 module_init(init_ext3_fs)
3057 module_exit(exit_ext3_fs)
linux-next