aboutsummaryrefslogtreecommitdiffstats
path: root/fs/nilfs2/super.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/nilfs2/super.c')
-rw-r--r--fs/nilfs2/super.c1323
1 files changed, 1323 insertions, 0 deletions
diff --git a/fs/nilfs2/super.c b/fs/nilfs2/super.c
new file mode 100644
index 000000000000..e117e1ea9bff
--- /dev/null
+++ b/fs/nilfs2/super.c
@@ -0,0 +1,1323 @@
1/*
2 * super.c - NILFS module and super block management.
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
21 */
22/*
23 * linux/fs/ext2/super.c
24 *
25 * Copyright (C) 1992, 1993, 1994, 1995
26 * Remy Card (card@masi.ibp.fr)
27 * Laboratoire MASI - Institut Blaise Pascal
28 * Universite Pierre et Marie Curie (Paris VI)
29 *
30 * from
31 *
32 * linux/fs/minix/inode.c
33 *
34 * Copyright (C) 1991, 1992 Linus Torvalds
35 *
36 * Big-endian to little-endian byte-swapping/bitmaps by
37 * David S. Miller (davem@caip.rutgers.edu), 1995
38 */
39
40#include <linux/module.h>
41#include <linux/string.h>
42#include <linux/slab.h>
43#include <linux/init.h>
44#include <linux/blkdev.h>
45#include <linux/parser.h>
46#include <linux/random.h>
47#include <linux/crc32.h>
48#include <linux/smp_lock.h>
49#include <linux/vfs.h>
50#include <linux/writeback.h>
51#include <linux/kobject.h>
52#include <linux/exportfs.h>
53#include "nilfs.h"
54#include "mdt.h"
55#include "alloc.h"
56#include "page.h"
57#include "cpfile.h"
58#include "ifile.h"
59#include "dat.h"
60#include "segment.h"
61#include "segbuf.h"
62
63MODULE_AUTHOR("NTT Corp.");
64MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
65 "(NILFS)");
66MODULE_VERSION(NILFS_VERSION);
67MODULE_LICENSE("GPL");
68
69static int nilfs_remount(struct super_block *sb, int *flags, char *data);
70static int test_exclusive_mount(struct file_system_type *fs_type,
71 struct block_device *bdev, int flags);
72
73/**
74 * nilfs_error() - report failure condition on a filesystem
75 *
76 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
77 * reporting an error message. It should be called when NILFS detects
78 * incoherences or defects of meta data on disk. As for sustainable
79 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
80 * function should be used instead.
81 *
82 * The segment constructor must not call this function because it can
83 * kill itself.
84 */
85void nilfs_error(struct super_block *sb, const char *function,
86 const char *fmt, ...)
87{
88 struct nilfs_sb_info *sbi = NILFS_SB(sb);
89 va_list args;
90
91 va_start(args, fmt);
92 printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
93 vprintk(fmt, args);
94 printk("\n");
95 va_end(args);
96
97 if (!(sb->s_flags & MS_RDONLY)) {
98 struct the_nilfs *nilfs = sbi->s_nilfs;
99
100 if (!nilfs_test_opt(sbi, ERRORS_CONT))
101 nilfs_detach_segment_constructor(sbi);
102
103 down_write(&nilfs->ns_sem);
104 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
105 nilfs->ns_mount_state |= NILFS_ERROR_FS;
106 nilfs->ns_sbp[0]->s_state |=
107 cpu_to_le16(NILFS_ERROR_FS);
108 nilfs_commit_super(sbi, 1);
109 }
110 up_write(&nilfs->ns_sem);
111
112 if (nilfs_test_opt(sbi, ERRORS_RO)) {
113 printk(KERN_CRIT "Remounting filesystem read-only\n");
114 sb->s_flags |= MS_RDONLY;
115 }
116 }
117
118 if (nilfs_test_opt(sbi, ERRORS_PANIC))
119 panic("NILFS (device %s): panic forced after error\n",
120 sb->s_id);
121}
122
123void nilfs_warning(struct super_block *sb, const char *function,
124 const char *fmt, ...)
125{
126 va_list args;
127
128 va_start(args, fmt);
129 printk(KERN_WARNING "NILFS warning (device %s): %s: ",
130 sb->s_id, function);
131 vprintk(fmt, args);
132 printk("\n");
133 va_end(args);
134}
135
136static struct kmem_cache *nilfs_inode_cachep;
137
138struct inode *nilfs_alloc_inode(struct super_block *sb)
139{
140 struct nilfs_inode_info *ii;
141
142 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
143 if (!ii)
144 return NULL;
145 ii->i_bh = NULL;
146 ii->i_state = 0;
147 ii->vfs_inode.i_version = 1;
148 nilfs_btnode_cache_init(&ii->i_btnode_cache);
149 return &ii->vfs_inode;
150}
151
152void nilfs_destroy_inode(struct inode *inode)
153{
154 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
155}
156
157static void init_once(void *obj)
158{
159 struct nilfs_inode_info *ii = obj;
160
161 INIT_LIST_HEAD(&ii->i_dirty);
162#ifdef CONFIG_NILFS_XATTR
163 init_rwsem(&ii->xattr_sem);
164#endif
165 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
166 ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
167 inode_init_once(&ii->vfs_inode);
168}
169
170static int nilfs_init_inode_cache(void)
171{
172 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
173 sizeof(struct nilfs_inode_info),
174 0, SLAB_RECLAIM_ACCOUNT,
175 init_once);
176
177 return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
178}
179
180static inline void nilfs_destroy_inode_cache(void)
181{
182 kmem_cache_destroy(nilfs_inode_cachep);
183}
184
185static void nilfs_clear_inode(struct inode *inode)
186{
187 struct nilfs_inode_info *ii = NILFS_I(inode);
188
189#ifdef CONFIG_NILFS_POSIX_ACL
190 if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
191 posix_acl_release(ii->i_acl);
192 ii->i_acl = NILFS_ACL_NOT_CACHED;
193 }
194 if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
195 posix_acl_release(ii->i_default_acl);
196 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
197 }
198#endif
199 /*
200 * Free resources allocated in nilfs_read_inode(), here.
201 */
202 BUG_ON(!list_empty(&ii->i_dirty));
203 brelse(ii->i_bh);
204 ii->i_bh = NULL;
205
206 if (test_bit(NILFS_I_BMAP, &ii->i_state))
207 nilfs_bmap_clear(ii->i_bmap);
208
209 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
210}
211
212static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
213{
214 struct the_nilfs *nilfs = sbi->s_nilfs;
215 int err;
216 int barrier_done = 0;
217
218 if (nilfs_test_opt(sbi, BARRIER)) {
219 set_buffer_ordered(nilfs->ns_sbh[0]);
220 barrier_done = 1;
221 }
222 retry:
223 set_buffer_dirty(nilfs->ns_sbh[0]);
224 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
225 if (err == -EOPNOTSUPP && barrier_done) {
226 nilfs_warning(sbi->s_super, __func__,
227 "barrier-based sync failed. "
228 "disabling barriers\n");
229 nilfs_clear_opt(sbi, BARRIER);
230 barrier_done = 0;
231 clear_buffer_ordered(nilfs->ns_sbh[0]);
232 goto retry;
233 }
234 if (unlikely(err)) {
235 printk(KERN_ERR
236 "NILFS: unable to write superblock (err=%d)\n", err);
237 if (err == -EIO && nilfs->ns_sbh[1]) {
238 nilfs_fall_back_super_block(nilfs);
239 goto retry;
240 }
241 } else {
242 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
243
244 /*
245 * The latest segment becomes trailable from the position
246 * written in superblock.
247 */
248 clear_nilfs_discontinued(nilfs);
249
250 /* update GC protection for recent segments */
251 if (nilfs->ns_sbh[1]) {
252 sbp = NULL;
253 if (dupsb) {
254 set_buffer_dirty(nilfs->ns_sbh[1]);
255 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
256 sbp = nilfs->ns_sbp[1];
257 }
258 }
259 if (sbp) {
260 spin_lock(&nilfs->ns_last_segment_lock);
261 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
262 spin_unlock(&nilfs->ns_last_segment_lock);
263 }
264 }
265
266 return err;
267}
268
269int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
270{
271 struct the_nilfs *nilfs = sbi->s_nilfs;
272 struct nilfs_super_block **sbp = nilfs->ns_sbp;
273 sector_t nfreeblocks;
274 time_t t;
275 int err;
276
277 /* nilfs->sem must be locked by the caller. */
278 if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
279 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
280 nilfs_swap_super_block(nilfs);
281 else {
282 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
283 sbi->s_super->s_id);
284 return -EIO;
285 }
286 }
287 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
288 if (unlikely(err)) {
289 printk(KERN_ERR "NILFS: failed to count free blocks\n");
290 return err;
291 }
292 spin_lock(&nilfs->ns_last_segment_lock);
293 sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
294 sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
295 sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
296 spin_unlock(&nilfs->ns_last_segment_lock);
297
298 t = get_seconds();
299 nilfs->ns_sbwtime[0] = t;
300 sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
301 sbp[0]->s_wtime = cpu_to_le64(t);
302 sbp[0]->s_sum = 0;
303 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
304 (unsigned char *)sbp[0],
305 nilfs->ns_sbsize));
306 if (dupsb && sbp[1]) {
307 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
308 nilfs->ns_sbwtime[1] = t;
309 }
310 sbi->s_super->s_dirt = 0;
311 return nilfs_sync_super(sbi, dupsb);
312}
313
314static void nilfs_put_super(struct super_block *sb)
315{
316 struct nilfs_sb_info *sbi = NILFS_SB(sb);
317 struct the_nilfs *nilfs = sbi->s_nilfs;
318
319 nilfs_detach_segment_constructor(sbi);
320
321 if (!(sb->s_flags & MS_RDONLY)) {
322 down_write(&nilfs->ns_sem);
323 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
324 nilfs_commit_super(sbi, 1);
325 up_write(&nilfs->ns_sem);
326 }
327
328 nilfs_detach_checkpoint(sbi);
329 put_nilfs(sbi->s_nilfs);
330 sbi->s_super = NULL;
331 sb->s_fs_info = NULL;
332 kfree(sbi);
333}
334
335/**
336 * nilfs_write_super - write super block(s) of NILFS
337 * @sb: super_block
338 *
339 * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
340 * clears s_dirt. This function is called in the section protected by
341 * lock_super().
342 *
343 * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
344 * of the struct the_nilfs. Lock order must be as follows:
345 *
346 * 1. lock_super()
347 * 2. down_write(&nilfs->ns_sem)
348 *
349 * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
350 * of the super block (nilfs->ns_sbp[]).
351 *
352 * In most cases, VFS functions call lock_super() before calling these
353 * methods. So we must be careful not to bring on deadlocks when using
354 * lock_super(); see generic_shutdown_super(), write_super(), and so on.
355 *
356 * Note that order of lock_kernel() and lock_super() depends on contexts
357 * of VFS. We should also note that lock_kernel() can be used in its
358 * protective section and only the outermost one has an effect.
359 */
360static void nilfs_write_super(struct super_block *sb)
361{
362 struct nilfs_sb_info *sbi = NILFS_SB(sb);
363 struct the_nilfs *nilfs = sbi->s_nilfs;
364
365 down_write(&nilfs->ns_sem);
366 if (!(sb->s_flags & MS_RDONLY)) {
367 struct nilfs_super_block **sbp = nilfs->ns_sbp;
368 u64 t = get_seconds();
369 int dupsb;
370
371 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
372 t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
373 up_write(&nilfs->ns_sem);
374 return;
375 }
376 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
377 nilfs_commit_super(sbi, dupsb);
378 }
379 sb->s_dirt = 0;
380 up_write(&nilfs->ns_sem);
381}
382
383static int nilfs_sync_fs(struct super_block *sb, int wait)
384{
385 int err = 0;
386
387 /* This function is called when super block should be written back */
388 if (wait)
389 err = nilfs_construct_segment(sb);
390 return err;
391}
392
393int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
394{
395 struct the_nilfs *nilfs = sbi->s_nilfs;
396 struct nilfs_checkpoint *raw_cp;
397 struct buffer_head *bh_cp;
398 int err;
399
400 down_write(&nilfs->ns_sem);
401 list_add(&sbi->s_list, &nilfs->ns_supers);
402 up_write(&nilfs->ns_sem);
403
404 sbi->s_ifile = nilfs_mdt_new(
405 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
406 if (!sbi->s_ifile)
407 return -ENOMEM;
408
409 err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
410 if (unlikely(err))
411 goto failed;
412
413 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
414 &bh_cp);
415 if (unlikely(err)) {
416 if (err == -ENOENT || err == -EINVAL) {
417 printk(KERN_ERR
418 "NILFS: Invalid checkpoint "
419 "(checkpoint number=%llu)\n",
420 (unsigned long long)cno);
421 err = -EINVAL;
422 }
423 goto failed;
424 }
425 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
426 if (unlikely(err))
427 goto failed_bh;
428 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
429 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
430
431 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
432 return 0;
433
434 failed_bh:
435 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
436 failed:
437 nilfs_mdt_destroy(sbi->s_ifile);
438 sbi->s_ifile = NULL;
439
440 down_write(&nilfs->ns_sem);
441 list_del_init(&sbi->s_list);
442 up_write(&nilfs->ns_sem);
443
444 return err;
445}
446
447void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
448{
449 struct the_nilfs *nilfs = sbi->s_nilfs;
450
451 nilfs_mdt_clear(sbi->s_ifile);
452 nilfs_mdt_destroy(sbi->s_ifile);
453 sbi->s_ifile = NULL;
454 down_write(&nilfs->ns_sem);
455 list_del_init(&sbi->s_list);
456 up_write(&nilfs->ns_sem);
457}
458
459static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
460{
461 struct the_nilfs *nilfs = sbi->s_nilfs;
462 int err = 0;
463
464 down_write(&nilfs->ns_sem);
465 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
466 nilfs->ns_mount_state |= NILFS_VALID_FS;
467 err = nilfs_commit_super(sbi, 1);
468 if (likely(!err))
469 printk(KERN_INFO "NILFS: recovery complete.\n");
470 }
471 up_write(&nilfs->ns_sem);
472 return err;
473}
474
475static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
476{
477 struct super_block *sb = dentry->d_sb;
478 struct nilfs_sb_info *sbi = NILFS_SB(sb);
479 unsigned long long blocks;
480 unsigned long overhead;
481 unsigned long nrsvblocks;
482 sector_t nfreeblocks;
483 struct the_nilfs *nilfs = sbi->s_nilfs;
484 int err;
485
486 /*
487 * Compute all of the segment blocks
488 *
489 * The blocks before first segment and after last segment
490 * are excluded.
491 */
492 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
493 - nilfs->ns_first_data_block;
494 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
495
496 /*
497 * Compute the overhead
498 *
499 * When distributing meta data blocks outside semgent structure,
500 * We must count them as the overhead.
501 */
502 overhead = 0;
503
504 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
505 if (unlikely(err))
506 return err;
507
508 buf->f_type = NILFS_SUPER_MAGIC;
509 buf->f_bsize = sb->s_blocksize;
510 buf->f_blocks = blocks - overhead;
511 buf->f_bfree = nfreeblocks;
512 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
513 (buf->f_bfree - nrsvblocks) : 0;
514 buf->f_files = atomic_read(&sbi->s_inodes_count);
515 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
516 buf->f_namelen = NILFS_NAME_LEN;
517 return 0;
518}
519
520static struct super_operations nilfs_sops = {
521 .alloc_inode = nilfs_alloc_inode,
522 .destroy_inode = nilfs_destroy_inode,
523 .dirty_inode = nilfs_dirty_inode,
524 /* .write_inode = nilfs_write_inode, */
525 /* .put_inode = nilfs_put_inode, */
526 /* .drop_inode = nilfs_drop_inode, */
527 .delete_inode = nilfs_delete_inode,
528 .put_super = nilfs_put_super,
529 .write_super = nilfs_write_super,
530 .sync_fs = nilfs_sync_fs,
531 /* .write_super_lockfs */
532 /* .unlockfs */
533 .statfs = nilfs_statfs,
534 .remount_fs = nilfs_remount,
535 .clear_inode = nilfs_clear_inode,
536 /* .umount_begin */
537 /* .show_options */
538};
539
540static struct inode *
541nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
542{
543 struct inode *inode;
544
545 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
546 ino != NILFS_SKETCH_INO)
547 return ERR_PTR(-ESTALE);
548
549 inode = nilfs_iget(sb, ino);
550 if (IS_ERR(inode))
551 return ERR_CAST(inode);
552 if (generation && inode->i_generation != generation) {
553 iput(inode);
554 return ERR_PTR(-ESTALE);
555 }
556
557 return inode;
558}
559
560static struct dentry *
561nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
562 int fh_type)
563{
564 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
565 nilfs_nfs_get_inode);
566}
567
568static struct dentry *
569nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
570 int fh_type)
571{
572 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
573 nilfs_nfs_get_inode);
574}
575
576static struct export_operations nilfs_export_ops = {
577 .fh_to_dentry = nilfs_fh_to_dentry,
578 .fh_to_parent = nilfs_fh_to_parent,
579 .get_parent = nilfs_get_parent,
580};
581
582enum {
583 Opt_err_cont, Opt_err_panic, Opt_err_ro,
584 Opt_barrier, Opt_snapshot, Opt_order,
585 Opt_err,
586};
587
588static match_table_t tokens = {
589 {Opt_err_cont, "errors=continue"},
590 {Opt_err_panic, "errors=panic"},
591 {Opt_err_ro, "errors=remount-ro"},
592 {Opt_barrier, "barrier=%s"},
593 {Opt_snapshot, "cp=%u"},
594 {Opt_order, "order=%s"},
595 {Opt_err, NULL}
596};
597
598static int match_bool(substring_t *s, int *result)
599{
600 int len = s->to - s->from;
601
602 if (strncmp(s->from, "on", len) == 0)
603 *result = 1;
604 else if (strncmp(s->from, "off", len) == 0)
605 *result = 0;
606 else
607 return 1;
608 return 0;
609}
610
611static int parse_options(char *options, struct super_block *sb)
612{
613 struct nilfs_sb_info *sbi = NILFS_SB(sb);
614 char *p;
615 substring_t args[MAX_OPT_ARGS];
616 int option;
617
618 if (!options)
619 return 1;
620
621 while ((p = strsep(&options, ",")) != NULL) {
622 int token;
623 if (!*p)
624 continue;
625
626 token = match_token(p, tokens, args);
627 switch (token) {
628 case Opt_barrier:
629 if (match_bool(&args[0], &option))
630 return 0;
631 if (option)
632 nilfs_set_opt(sbi, BARRIER);
633 else
634 nilfs_clear_opt(sbi, BARRIER);
635 break;
636 case Opt_order:
637 if (strcmp(args[0].from, "relaxed") == 0)
638 /* Ordered data semantics */
639 nilfs_clear_opt(sbi, STRICT_ORDER);
640 else if (strcmp(args[0].from, "strict") == 0)
641 /* Strict in-order semantics */
642 nilfs_set_opt(sbi, STRICT_ORDER);
643 else
644 return 0;
645 break;
646 case Opt_err_panic:
647 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
648 break;
649 case Opt_err_ro:
650 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
651 break;
652 case Opt_err_cont:
653 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
654 break;
655 case Opt_snapshot:
656 if (match_int(&args[0], &option) || option <= 0)
657 return 0;
658 if (!(sb->s_flags & MS_RDONLY))
659 return 0;
660 sbi->s_snapshot_cno = option;
661 nilfs_set_opt(sbi, SNAPSHOT);
662 break;
663 default:
664 printk(KERN_ERR
665 "NILFS: Unrecognized mount option \"%s\"\n", p);
666 return 0;
667 }
668 }
669 return 1;
670}
671
672static inline void
673nilfs_set_default_options(struct nilfs_sb_info *sbi,
674 struct nilfs_super_block *sbp)
675{
676 sbi->s_mount_opt =
677 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
678}
679
680static int nilfs_setup_super(struct nilfs_sb_info *sbi)
681{
682 struct the_nilfs *nilfs = sbi->s_nilfs;
683 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
684 int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
685 int mnt_count = le16_to_cpu(sbp->s_mnt_count);
686
687 /* nilfs->sem must be locked by the caller. */
688 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
689 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
690 } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
691 printk(KERN_WARNING
692 "NILFS warning: mounting fs with errors\n");
693#if 0
694 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
695 printk(KERN_WARNING
696 "NILFS warning: maximal mount count reached\n");
697#endif
698 }
699 if (!max_mnt_count)
700 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
701
702 sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
703 sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
704 sbp->s_mtime = cpu_to_le64(get_seconds());
705 return nilfs_commit_super(sbi, 1);
706}
707
708struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
709 u64 pos, int blocksize,
710 struct buffer_head **pbh)
711{
712 unsigned long long sb_index = pos;
713 unsigned long offset;
714
715 offset = do_div(sb_index, blocksize);
716 *pbh = sb_bread(sb, sb_index);
717 if (!*pbh)
718 return NULL;
719 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
720}
721
722int nilfs_store_magic_and_option(struct super_block *sb,
723 struct nilfs_super_block *sbp,
724 char *data)
725{
726 struct nilfs_sb_info *sbi = NILFS_SB(sb);
727
728 sb->s_magic = le16_to_cpu(sbp->s_magic);
729
730 /* FS independent flags */
731#ifdef NILFS_ATIME_DISABLE
732 sb->s_flags |= MS_NOATIME;
733#endif
734
735 nilfs_set_default_options(sbi, sbp);
736
737 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
738 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
739 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
740 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
741
742 return !parse_options(data, sb) ? -EINVAL : 0 ;
743}
744
745/**
746 * nilfs_fill_super() - initialize a super block instance
747 * @sb: super_block
748 * @data: mount options
749 * @silent: silent mode flag
750 * @nilfs: the_nilfs struct
751 *
752 * This function is called exclusively by bd_mount_mutex.
753 * So, the recovery process is protected from other simultaneous mounts.
754 */
755static int
756nilfs_fill_super(struct super_block *sb, void *data, int silent,
757 struct the_nilfs *nilfs)
758{
759 struct nilfs_sb_info *sbi;
760 struct inode *root;
761 __u64 cno;
762 int err;
763
764 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
765 if (!sbi)
766 return -ENOMEM;
767
768 sb->s_fs_info = sbi;
769
770 get_nilfs(nilfs);
771 sbi->s_nilfs = nilfs;
772 sbi->s_super = sb;
773
774 err = init_nilfs(nilfs, sbi, (char *)data);
775 if (err)
776 goto failed_sbi;
777
778 spin_lock_init(&sbi->s_inode_lock);
779 INIT_LIST_HEAD(&sbi->s_dirty_files);
780 INIT_LIST_HEAD(&sbi->s_list);
781
782 /*
783 * Following initialization is overlapped because
784 * nilfs_sb_info structure has been cleared at the beginning.
785 * But we reserve them to keep our interest and make ready
786 * for the future change.
787 */
788 get_random_bytes(&sbi->s_next_generation,
789 sizeof(sbi->s_next_generation));
790 spin_lock_init(&sbi->s_next_gen_lock);
791
792 sb->s_op = &nilfs_sops;
793 sb->s_export_op = &nilfs_export_ops;
794 sb->s_root = NULL;
795 sb->s_time_gran = 1;
796
797 if (!nilfs_loaded(nilfs)) {
798 err = load_nilfs(nilfs, sbi);
799 if (err)
800 goto failed_sbi;
801 }
802 cno = nilfs_last_cno(nilfs);
803
804 if (sb->s_flags & MS_RDONLY) {
805 if (nilfs_test_opt(sbi, SNAPSHOT)) {
806 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
807 sbi->s_snapshot_cno);
808 if (err < 0)
809 goto failed_sbi;
810 if (!err) {
811 printk(KERN_ERR
812 "NILFS: The specified checkpoint is "
813 "not a snapshot "
814 "(checkpoint number=%llu).\n",
815 (unsigned long long)sbi->s_snapshot_cno);
816 err = -EINVAL;
817 goto failed_sbi;
818 }
819 cno = sbi->s_snapshot_cno;
820 } else
821 /* Read-only mount */
822 sbi->s_snapshot_cno = cno;
823 }
824
825 err = nilfs_attach_checkpoint(sbi, cno);
826 if (err) {
827 printk(KERN_ERR "NILFS: error loading a checkpoint"
828 " (checkpoint number=%llu).\n", (unsigned long long)cno);
829 goto failed_sbi;
830 }
831
832 if (!(sb->s_flags & MS_RDONLY)) {
833 err = nilfs_attach_segment_constructor(sbi);
834 if (err)
835 goto failed_checkpoint;
836 }
837
838 root = nilfs_iget(sb, NILFS_ROOT_INO);
839 if (IS_ERR(root)) {
840 printk(KERN_ERR "NILFS: get root inode failed\n");
841 err = PTR_ERR(root);
842 goto failed_segctor;
843 }
844 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
845 iput(root);
846 printk(KERN_ERR "NILFS: corrupt root inode.\n");
847 err = -EINVAL;
848 goto failed_segctor;
849 }
850 sb->s_root = d_alloc_root(root);
851 if (!sb->s_root) {
852 iput(root);
853 printk(KERN_ERR "NILFS: get root dentry failed\n");
854 err = -ENOMEM;
855 goto failed_segctor;
856 }
857
858 if (!(sb->s_flags & MS_RDONLY)) {
859 down_write(&nilfs->ns_sem);
860 nilfs_setup_super(sbi);
861 up_write(&nilfs->ns_sem);
862 }
863
864 err = nilfs_mark_recovery_complete(sbi);
865 if (unlikely(err)) {
866 printk(KERN_ERR "NILFS: recovery failed.\n");
867 goto failed_root;
868 }
869
870 return 0;
871
872 failed_root:
873 dput(sb->s_root);
874 sb->s_root = NULL;
875
876 failed_segctor:
877 nilfs_detach_segment_constructor(sbi);
878
879 failed_checkpoint:
880 nilfs_detach_checkpoint(sbi);
881
882 failed_sbi:
883 put_nilfs(nilfs);
884 sb->s_fs_info = NULL;
885 kfree(sbi);
886 return err;
887}
888
889static int nilfs_remount(struct super_block *sb, int *flags, char *data)
890{
891 struct nilfs_sb_info *sbi = NILFS_SB(sb);
892 struct nilfs_super_block *sbp;
893 struct the_nilfs *nilfs = sbi->s_nilfs;
894 unsigned long old_sb_flags;
895 struct nilfs_mount_options old_opts;
896 int err;
897
898 old_sb_flags = sb->s_flags;
899 old_opts.mount_opt = sbi->s_mount_opt;
900 old_opts.snapshot_cno = sbi->s_snapshot_cno;
901
902 if (!parse_options(data, sb)) {
903 err = -EINVAL;
904 goto restore_opts;
905 }
906 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
907
908 if ((*flags & MS_RDONLY) &&
909 sbi->s_snapshot_cno != old_opts.snapshot_cno) {
910 printk(KERN_WARNING "NILFS (device %s): couldn't "
911 "remount to a different snapshot. \n",
912 sb->s_id);
913 err = -EINVAL;
914 goto restore_opts;
915 }
916
917 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
918 goto out;
919 if (*flags & MS_RDONLY) {
920 /* Shutting down the segment constructor */
921 nilfs_detach_segment_constructor(sbi);
922 sb->s_flags |= MS_RDONLY;
923
924 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
925 /* nilfs_set_opt(sbi, SNAPSHOT); */
926
927 /*
928 * Remounting a valid RW partition RDONLY, so set
929 * the RDONLY flag and then mark the partition as valid again.
930 */
931 down_write(&nilfs->ns_sem);
932 sbp = nilfs->ns_sbp[0];
933 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
934 (nilfs->ns_mount_state & NILFS_VALID_FS))
935 sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
936 sbp->s_mtime = cpu_to_le64(get_seconds());
937 nilfs_commit_super(sbi, 1);
938 up_write(&nilfs->ns_sem);
939 } else {
940 /*
941 * Mounting a RDONLY partition read-write, so reread and
942 * store the current valid flag. (It may have been changed
943 * by fsck since we originally mounted the partition.)
944 */
945 down(&sb->s_bdev->bd_mount_sem);
946 /* Check existing RW-mount */
947 if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) {
948 printk(KERN_WARNING "NILFS (device %s): couldn't "
949 "remount because a RW-mount exists.\n",
950 sb->s_id);
951 err = -EBUSY;
952 goto rw_remount_failed;
953 }
954 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
955 printk(KERN_WARNING "NILFS (device %s): couldn't "
956 "remount because the current RO-mount is not "
957 "the latest one.\n",
958 sb->s_id);
959 err = -EINVAL;
960 goto rw_remount_failed;
961 }
962 sb->s_flags &= ~MS_RDONLY;
963 nilfs_clear_opt(sbi, SNAPSHOT);
964 sbi->s_snapshot_cno = 0;
965
966 err = nilfs_attach_segment_constructor(sbi);
967 if (err)
968 goto rw_remount_failed;
969
970 down_write(&nilfs->ns_sem);
971 nilfs_setup_super(sbi);
972 up_write(&nilfs->ns_sem);
973
974 up(&sb->s_bdev->bd_mount_sem);
975 }
976 out:
977 return 0;
978
979 rw_remount_failed:
980 up(&sb->s_bdev->bd_mount_sem);
981 restore_opts:
982 sb->s_flags = old_sb_flags;
983 sbi->s_mount_opt = old_opts.mount_opt;
984 sbi->s_snapshot_cno = old_opts.snapshot_cno;
985 return err;
986}
987
988struct nilfs_super_data {
989 struct block_device *bdev;
990 __u64 cno;
991 int flags;
992};
993
994/**
995 * nilfs_identify - pre-read mount options needed to identify mount instance
996 * @data: mount options
997 * @sd: nilfs_super_data
998 */
999static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1000{
1001 char *p, *options = data;
1002 substring_t args[MAX_OPT_ARGS];
1003 int option, token;
1004 int ret = 0;
1005
1006 do {
1007 p = strsep(&options, ",");
1008 if (p != NULL && *p) {
1009 token = match_token(p, tokens, args);
1010 if (token == Opt_snapshot) {
1011 if (!(sd->flags & MS_RDONLY))
1012 ret++;
1013 else {
1014 ret = match_int(&args[0], &option);
1015 if (!ret) {
1016 if (option > 0)
1017 sd->cno = option;
1018 else
1019 ret++;
1020 }
1021 }
1022 }
1023 if (ret)
1024 printk(KERN_ERR
1025 "NILFS: invalid mount option: %s\n", p);
1026 }
1027 if (!options)
1028 break;
1029 BUG_ON(options == data);
1030 *(options - 1) = ',';
1031 } while (!ret);
1032 return ret;
1033}
1034
1035static int nilfs_set_bdev_super(struct super_block *s, void *data)
1036{
1037 struct nilfs_super_data *sd = data;
1038
1039 s->s_bdev = sd->bdev;
1040 s->s_dev = s->s_bdev->bd_dev;
1041 return 0;
1042}
1043
1044static int nilfs_test_bdev_super(struct super_block *s, void *data)
1045{
1046 struct nilfs_super_data *sd = data;
1047
1048 return s->s_bdev == sd->bdev;
1049}
1050
1051static int nilfs_test_bdev_super2(struct super_block *s, void *data)
1052{
1053 struct nilfs_super_data *sd = data;
1054 int ret;
1055
1056 if (s->s_bdev != sd->bdev)
1057 return 0;
1058
1059 if (!((s->s_flags | sd->flags) & MS_RDONLY))
1060 return 1; /* Reuse an old R/W-mode super_block */
1061
1062 if (s->s_flags & sd->flags & MS_RDONLY) {
1063 if (down_read_trylock(&s->s_umount)) {
1064 ret = s->s_root &&
1065 (sd->cno == NILFS_SB(s)->s_snapshot_cno);
1066 up_read(&s->s_umount);
1067 /*
1068 * This path is locked with sb_lock by sget().
1069 * So, drop_super() causes deadlock.
1070 */
1071 return ret;
1072 }
1073 }
1074 return 0;
1075}
1076
1077static int
1078nilfs_get_sb(struct file_system_type *fs_type, int flags,
1079 const char *dev_name, void *data, struct vfsmount *mnt)
1080{
1081 struct nilfs_super_data sd;
1082 struct super_block *s, *s2;
1083 struct the_nilfs *nilfs = NULL;
1084 int err, need_to_close = 1;
1085
1086 sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1087 if (IS_ERR(sd.bdev))
1088 return PTR_ERR(sd.bdev);
1089
1090 /*
1091 * To get mount instance using sget() vfs-routine, NILFS needs
1092 * much more information than normal filesystems to identify mount
1093 * instance. For snapshot mounts, not only a mount type (ro-mount
1094 * or rw-mount) but also a checkpoint number is required.
1095 * The results are passed in sget() using nilfs_super_data.
1096 */
1097 sd.cno = 0;
1098 sd.flags = flags;
1099 if (nilfs_identify((char *)data, &sd)) {
1100 err = -EINVAL;
1101 goto failed;
1102 }
1103
1104 /*
1105 * once the super is inserted into the list by sget, s_umount
1106 * will protect the lockfs code from trying to start a snapshot
1107 * while we are mounting
1108 */
1109 down(&sd.bdev->bd_mount_sem);
1110 if (!sd.cno &&
1111 (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) {
1112 err = (err < 0) ? : -EBUSY;
1113 goto failed_unlock;
1114 }
1115
1116 /*
1117 * Phase-1: search any existent instance and get the_nilfs
1118 */
1119 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1120 if (IS_ERR(s))
1121 goto error_s;
1122
1123 if (!s->s_root) {
1124 err = -ENOMEM;
1125 nilfs = alloc_nilfs(sd.bdev);
1126 if (!nilfs)
1127 goto cancel_new;
1128 } else {
1129 struct nilfs_sb_info *sbi = NILFS_SB(s);
1130
1131 /*
1132 * s_umount protects super_block from unmount process;
1133 * It covers pointers of nilfs_sb_info and the_nilfs.
1134 */
1135 nilfs = sbi->s_nilfs;
1136 get_nilfs(nilfs);
1137 up_write(&s->s_umount);
1138
1139 /*
1140 * Phase-2: search specified snapshot or R/W mode super_block
1141 */
1142 if (!sd.cno)
1143 /* trying to get the latest checkpoint. */
1144 sd.cno = nilfs_last_cno(nilfs);
1145
1146 s2 = sget(fs_type, nilfs_test_bdev_super2,
1147 nilfs_set_bdev_super, &sd);
1148 deactivate_super(s);
1149 /*
1150 * Although deactivate_super() invokes close_bdev_exclusive() at
1151 * kill_block_super(). Here, s is an existent mount; we need
1152 * one more close_bdev_exclusive() call.
1153 */
1154 s = s2;
1155 if (IS_ERR(s))
1156 goto error_s;
1157 }
1158
1159 if (!s->s_root) {
1160 char b[BDEVNAME_SIZE];
1161
1162 s->s_flags = flags;
1163 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1164 sb_set_blocksize(s, block_size(sd.bdev));
1165
1166 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1167 if (err)
1168 goto cancel_new;
1169
1170 s->s_flags |= MS_ACTIVE;
1171 need_to_close = 0;
1172 } else if (!(s->s_flags & MS_RDONLY)) {
1173 err = -EBUSY;
1174 }
1175
1176 up(&sd.bdev->bd_mount_sem);
1177 put_nilfs(nilfs);
1178 if (need_to_close)
1179 close_bdev_exclusive(sd.bdev, flags);
1180 simple_set_mnt(mnt, s);
1181 return 0;
1182
1183 error_s:
1184 up(&sd.bdev->bd_mount_sem);
1185 if (nilfs)
1186 put_nilfs(nilfs);
1187 close_bdev_exclusive(sd.bdev, flags);
1188 return PTR_ERR(s);
1189
1190 failed_unlock:
1191 up(&sd.bdev->bd_mount_sem);
1192 failed:
1193 close_bdev_exclusive(sd.bdev, flags);
1194
1195 return err;
1196
1197 cancel_new:
1198 /* Abandoning the newly allocated superblock */
1199 up(&sd.bdev->bd_mount_sem);
1200 if (nilfs)
1201 put_nilfs(nilfs);
1202 up_write(&s->s_umount);
1203 deactivate_super(s);
1204 /*
1205 * deactivate_super() invokes close_bdev_exclusive().
1206 * We must finish all post-cleaning before this call;
1207 * put_nilfs() and unlocking bd_mount_sem need the block device.
1208 */
1209 return err;
1210}
1211
1212static int nilfs_test_bdev_super3(struct super_block *s, void *data)
1213{
1214 struct nilfs_super_data *sd = data;
1215 int ret;
1216
1217 if (s->s_bdev != sd->bdev)
1218 return 0;
1219 if (down_read_trylock(&s->s_umount)) {
1220 ret = (s->s_flags & MS_RDONLY) && s->s_root &&
1221 nilfs_test_opt(NILFS_SB(s), SNAPSHOT);
1222 up_read(&s->s_umount);
1223 if (ret)
1224 return 0; /* ignore snapshot mounts */
1225 }
1226 return !((sd->flags ^ s->s_flags) & MS_RDONLY);
1227}
1228
1229static int __false_bdev_super(struct super_block *s, void *data)
1230{
1231#if 0 /* XXX: workaround for lock debug. This is not good idea */
1232 up_write(&s->s_umount);
1233#endif
1234 return -EFAULT;
1235}
1236
1237/**
1238 * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not.
1239 * fs_type: filesystem type
1240 * bdev: block device
1241 * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount)
1242 * res: pointer to an integer to store result
1243 *
1244 * This function must be called within a section protected by bd_mount_mutex.
1245 */
1246static int test_exclusive_mount(struct file_system_type *fs_type,
1247 struct block_device *bdev, int flags)
1248{
1249 struct super_block *s;
1250 struct nilfs_super_data sd = { .flags = flags, .bdev = bdev };
1251
1252 s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd);
1253 if (IS_ERR(s)) {
1254 if (PTR_ERR(s) != -EFAULT)
1255 return PTR_ERR(s);
1256 return 0; /* Not found */
1257 }
1258 up_write(&s->s_umount);
1259 deactivate_super(s);
1260 return 1; /* Found */
1261}
1262
1263struct file_system_type nilfs_fs_type = {
1264 .owner = THIS_MODULE,
1265 .name = "nilfs2",
1266 .get_sb = nilfs_get_sb,
1267 .kill_sb = kill_block_super,
1268 .fs_flags = FS_REQUIRES_DEV,
1269};
1270
1271static int __init init_nilfs_fs(void)
1272{
1273 int err;
1274
1275 err = nilfs_init_inode_cache();
1276 if (err)
1277 goto failed;
1278
1279 err = nilfs_init_transaction_cache();
1280 if (err)
1281 goto failed_inode_cache;
1282
1283 err = nilfs_init_segbuf_cache();
1284 if (err)
1285 goto failed_transaction_cache;
1286
1287 err = nilfs_btree_path_cache_init();
1288 if (err)
1289 goto failed_segbuf_cache;
1290
1291 err = register_filesystem(&nilfs_fs_type);
1292 if (err)
1293 goto failed_btree_path_cache;
1294
1295 return 0;
1296
1297 failed_btree_path_cache:
1298 nilfs_btree_path_cache_destroy();
1299
1300 failed_segbuf_cache:
1301 nilfs_destroy_segbuf_cache();
1302
1303 failed_transaction_cache:
1304 nilfs_destroy_transaction_cache();
1305
1306 failed_inode_cache:
1307 nilfs_destroy_inode_cache();
1308
1309 failed:
1310 return err;
1311}
1312
1313static void __exit exit_nilfs_fs(void)
1314{
1315 nilfs_destroy_segbuf_cache();
1316 nilfs_destroy_transaction_cache();
1317 nilfs_destroy_inode_cache();
1318 nilfs_btree_path_cache_destroy();
1319 unregister_filesystem(&nilfs_fs_type);
1320}
1321
1322module_init(init_nilfs_fs)
1323module_exit(exit_nilfs_fs)
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-23 19:49:03 -0500