Loading...
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/crc32.h>
47#include <linux/vfs.h>
48#include <linux/writeback.h>
49#include <linux/seq_file.h>
50#include <linux/mount.h>
51#include "nilfs.h"
52#include "export.h"
53#include "mdt.h"
54#include "alloc.h"
55#include "btree.h"
56#include "btnode.h"
57#include "page.h"
58#include "cpfile.h"
59#include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
60#include "ifile.h"
61#include "dat.h"
62#include "segment.h"
63#include "segbuf.h"
64
65MODULE_AUTHOR("NTT Corp.");
66MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
67 "(NILFS)");
68MODULE_LICENSE("GPL");
69
70static struct kmem_cache *nilfs_inode_cachep;
71struct kmem_cache *nilfs_transaction_cachep;
72struct kmem_cache *nilfs_segbuf_cachep;
73struct kmem_cache *nilfs_btree_path_cache;
74
75static int nilfs_setup_super(struct super_block *sb, int is_mount);
76static int nilfs_remount(struct super_block *sb, int *flags, char *data);
77
78static void nilfs_set_error(struct super_block *sb)
79{
80 struct the_nilfs *nilfs = sb->s_fs_info;
81 struct nilfs_super_block **sbp;
82
83 down_write(&nilfs->ns_sem);
84 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
85 nilfs->ns_mount_state |= NILFS_ERROR_FS;
86 sbp = nilfs_prepare_super(sb, 0);
87 if (likely(sbp)) {
88 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
89 if (sbp[1])
90 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
91 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
92 }
93 }
94 up_write(&nilfs->ns_sem);
95}
96
97/**
98 * nilfs_error() - report failure condition on a filesystem
99 *
100 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
101 * reporting an error message. It should be called when NILFS detects
102 * incoherences or defects of meta data on disk. As for sustainable
103 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
104 * function should be used instead.
105 *
106 * The segment constructor must not call this function because it can
107 * kill itself.
108 */
109void nilfs_error(struct super_block *sb, const char *function,
110 const char *fmt, ...)
111{
112 struct the_nilfs *nilfs = sb->s_fs_info;
113 struct va_format vaf;
114 va_list args;
115
116 va_start(args, fmt);
117
118 vaf.fmt = fmt;
119 vaf.va = &args;
120
121 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
122 sb->s_id, function, &vaf);
123
124 va_end(args);
125
126 if (!(sb->s_flags & MS_RDONLY)) {
127 nilfs_set_error(sb);
128
129 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
130 printk(KERN_CRIT "Remounting filesystem read-only\n");
131 sb->s_flags |= MS_RDONLY;
132 }
133 }
134
135 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
136 panic("NILFS (device %s): panic forced after error\n",
137 sb->s_id);
138}
139
140void nilfs_warning(struct super_block *sb, const char *function,
141 const char *fmt, ...)
142{
143 struct va_format vaf;
144 va_list args;
145
146 va_start(args, fmt);
147
148 vaf.fmt = fmt;
149 vaf.va = &args;
150
151 printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
152 sb->s_id, function, &vaf);
153
154 va_end(args);
155}
156
157
158struct inode *nilfs_alloc_inode(struct super_block *sb)
159{
160 struct nilfs_inode_info *ii;
161
162 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
163 if (!ii)
164 return NULL;
165 ii->i_bh = NULL;
166 ii->i_state = 0;
167 ii->i_cno = 0;
168 ii->vfs_inode.i_version = 1;
169 nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode, sb->s_bdi);
170 return &ii->vfs_inode;
171}
172
173static void nilfs_i_callback(struct rcu_head *head)
174{
175 struct inode *inode = container_of(head, struct inode, i_rcu);
176 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
177
178 if (mdi) {
179 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
180 kfree(mdi);
181 }
182 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
183}
184
185void nilfs_destroy_inode(struct inode *inode)
186{
187 call_rcu(&inode->i_rcu, nilfs_i_callback);
188}
189
190static int nilfs_sync_super(struct super_block *sb, int flag)
191{
192 struct the_nilfs *nilfs = sb->s_fs_info;
193 int err;
194
195 retry:
196 set_buffer_dirty(nilfs->ns_sbh[0]);
197 if (nilfs_test_opt(nilfs, BARRIER)) {
198 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
199 WRITE_SYNC | WRITE_FLUSH_FUA);
200 } else {
201 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
202 }
203
204 if (unlikely(err)) {
205 printk(KERN_ERR
206 "NILFS: unable to write superblock (err=%d)\n", err);
207 if (err == -EIO && nilfs->ns_sbh[1]) {
208 /*
209 * sbp[0] points to newer log than sbp[1],
210 * so copy sbp[0] to sbp[1] to take over sbp[0].
211 */
212 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
213 nilfs->ns_sbsize);
214 nilfs_fall_back_super_block(nilfs);
215 goto retry;
216 }
217 } else {
218 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
219
220 nilfs->ns_sbwcount++;
221
222 /*
223 * The latest segment becomes trailable from the position
224 * written in superblock.
225 */
226 clear_nilfs_discontinued(nilfs);
227
228 /* update GC protection for recent segments */
229 if (nilfs->ns_sbh[1]) {
230 if (flag == NILFS_SB_COMMIT_ALL) {
231 set_buffer_dirty(nilfs->ns_sbh[1]);
232 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
233 goto out;
234 }
235 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
236 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
237 sbp = nilfs->ns_sbp[1];
238 }
239
240 spin_lock(&nilfs->ns_last_segment_lock);
241 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
242 spin_unlock(&nilfs->ns_last_segment_lock);
243 }
244 out:
245 return err;
246}
247
248void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
249 struct the_nilfs *nilfs)
250{
251 sector_t nfreeblocks;
252
253 /* nilfs->ns_sem must be locked by the caller. */
254 nilfs_count_free_blocks(nilfs, &nfreeblocks);
255 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
256
257 spin_lock(&nilfs->ns_last_segment_lock);
258 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
259 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
260 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
261 spin_unlock(&nilfs->ns_last_segment_lock);
262}
263
264struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
265 int flip)
266{
267 struct the_nilfs *nilfs = sb->s_fs_info;
268 struct nilfs_super_block **sbp = nilfs->ns_sbp;
269
270 /* nilfs->ns_sem must be locked by the caller. */
271 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
272 if (sbp[1] &&
273 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
274 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
275 } else {
276 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
277 sb->s_id);
278 return NULL;
279 }
280 } else if (sbp[1] &&
281 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
282 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
283 }
284
285 if (flip && sbp[1])
286 nilfs_swap_super_block(nilfs);
287
288 return sbp;
289}
290
291int nilfs_commit_super(struct super_block *sb, int flag)
292{
293 struct the_nilfs *nilfs = sb->s_fs_info;
294 struct nilfs_super_block **sbp = nilfs->ns_sbp;
295 time_t t;
296
297 /* nilfs->ns_sem must be locked by the caller. */
298 t = get_seconds();
299 nilfs->ns_sbwtime = t;
300 sbp[0]->s_wtime = cpu_to_le64(t);
301 sbp[0]->s_sum = 0;
302 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
303 (unsigned char *)sbp[0],
304 nilfs->ns_sbsize));
305 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
306 sbp[1]->s_wtime = sbp[0]->s_wtime;
307 sbp[1]->s_sum = 0;
308 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
309 (unsigned char *)sbp[1],
310 nilfs->ns_sbsize));
311 }
312 clear_nilfs_sb_dirty(nilfs);
313 return nilfs_sync_super(sb, flag);
314}
315
316/**
317 * nilfs_cleanup_super() - write filesystem state for cleanup
318 * @sb: super block instance to be unmounted or degraded to read-only
319 *
320 * This function restores state flags in the on-disk super block.
321 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
322 * filesystem was not clean previously.
323 */
324int nilfs_cleanup_super(struct super_block *sb)
325{
326 struct the_nilfs *nilfs = sb->s_fs_info;
327 struct nilfs_super_block **sbp;
328 int flag = NILFS_SB_COMMIT;
329 int ret = -EIO;
330
331 sbp = nilfs_prepare_super(sb, 0);
332 if (sbp) {
333 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
334 nilfs_set_log_cursor(sbp[0], nilfs);
335 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
336 /*
337 * make the "clean" flag also to the opposite
338 * super block if both super blocks point to
339 * the same checkpoint.
340 */
341 sbp[1]->s_state = sbp[0]->s_state;
342 flag = NILFS_SB_COMMIT_ALL;
343 }
344 ret = nilfs_commit_super(sb, flag);
345 }
346 return ret;
347}
348
349/**
350 * nilfs_move_2nd_super - relocate secondary super block
351 * @sb: super block instance
352 * @sb2off: new offset of the secondary super block (in bytes)
353 */
354static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
355{
356 struct the_nilfs *nilfs = sb->s_fs_info;
357 struct buffer_head *nsbh;
358 struct nilfs_super_block *nsbp;
359 sector_t blocknr, newblocknr;
360 unsigned long offset;
361 int sb2i = -1; /* array index of the secondary superblock */
362 int ret = 0;
363
364 /* nilfs->ns_sem must be locked by the caller. */
365 if (nilfs->ns_sbh[1] &&
366 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
367 sb2i = 1;
368 blocknr = nilfs->ns_sbh[1]->b_blocknr;
369 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
370 sb2i = 0;
371 blocknr = nilfs->ns_sbh[0]->b_blocknr;
372 }
373 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
374 goto out; /* super block location is unchanged */
375
376 /* Get new super block buffer */
377 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
378 offset = sb2off & (nilfs->ns_blocksize - 1);
379 nsbh = sb_getblk(sb, newblocknr);
380 if (!nsbh) {
381 printk(KERN_WARNING
382 "NILFS warning: unable to move secondary superblock "
383 "to block %llu\n", (unsigned long long)newblocknr);
384 ret = -EIO;
385 goto out;
386 }
387 nsbp = (void *)nsbh->b_data + offset;
388 memset(nsbp, 0, nilfs->ns_blocksize);
389
390 if (sb2i >= 0) {
391 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
392 brelse(nilfs->ns_sbh[sb2i]);
393 nilfs->ns_sbh[sb2i] = nsbh;
394 nilfs->ns_sbp[sb2i] = nsbp;
395 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
396 /* secondary super block will be restored to index 1 */
397 nilfs->ns_sbh[1] = nsbh;
398 nilfs->ns_sbp[1] = nsbp;
399 } else {
400 brelse(nsbh);
401 }
402out:
403 return ret;
404}
405
406/**
407 * nilfs_resize_fs - resize the filesystem
408 * @sb: super block instance
409 * @newsize: new size of the filesystem (in bytes)
410 */
411int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
412{
413 struct the_nilfs *nilfs = sb->s_fs_info;
414 struct nilfs_super_block **sbp;
415 __u64 devsize, newnsegs;
416 loff_t sb2off;
417 int ret;
418
419 ret = -ERANGE;
420 devsize = i_size_read(sb->s_bdev->bd_inode);
421 if (newsize > devsize)
422 goto out;
423
424 /*
425 * Write lock is required to protect some functions depending
426 * on the number of segments, the number of reserved segments,
427 * and so forth.
428 */
429 down_write(&nilfs->ns_segctor_sem);
430
431 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
432 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
433 do_div(newnsegs, nilfs->ns_blocks_per_segment);
434
435 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
436 up_write(&nilfs->ns_segctor_sem);
437 if (ret < 0)
438 goto out;
439
440 ret = nilfs_construct_segment(sb);
441 if (ret < 0)
442 goto out;
443
444 down_write(&nilfs->ns_sem);
445 nilfs_move_2nd_super(sb, sb2off);
446 ret = -EIO;
447 sbp = nilfs_prepare_super(sb, 0);
448 if (likely(sbp)) {
449 nilfs_set_log_cursor(sbp[0], nilfs);
450 /*
451 * Drop NILFS_RESIZE_FS flag for compatibility with
452 * mount-time resize which may be implemented in a
453 * future release.
454 */
455 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
456 ~NILFS_RESIZE_FS);
457 sbp[0]->s_dev_size = cpu_to_le64(newsize);
458 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
459 if (sbp[1])
460 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
461 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
462 }
463 up_write(&nilfs->ns_sem);
464
465 /*
466 * Reset the range of allocatable segments last. This order
467 * is important in the case of expansion because the secondary
468 * superblock must be protected from log write until migration
469 * completes.
470 */
471 if (!ret)
472 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
473out:
474 return ret;
475}
476
477static void nilfs_put_super(struct super_block *sb)
478{
479 struct the_nilfs *nilfs = sb->s_fs_info;
480
481 nilfs_detach_log_writer(sb);
482
483 if (!(sb->s_flags & MS_RDONLY)) {
484 down_write(&nilfs->ns_sem);
485 nilfs_cleanup_super(sb);
486 up_write(&nilfs->ns_sem);
487 }
488
489 iput(nilfs->ns_sufile);
490 iput(nilfs->ns_cpfile);
491 iput(nilfs->ns_dat);
492
493 destroy_nilfs(nilfs);
494 sb->s_fs_info = NULL;
495}
496
497static int nilfs_sync_fs(struct super_block *sb, int wait)
498{
499 struct the_nilfs *nilfs = sb->s_fs_info;
500 struct nilfs_super_block **sbp;
501 int err = 0;
502
503 /* This function is called when super block should be written back */
504 if (wait)
505 err = nilfs_construct_segment(sb);
506
507 down_write(&nilfs->ns_sem);
508 if (nilfs_sb_dirty(nilfs)) {
509 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
510 if (likely(sbp)) {
511 nilfs_set_log_cursor(sbp[0], nilfs);
512 nilfs_commit_super(sb, NILFS_SB_COMMIT);
513 }
514 }
515 up_write(&nilfs->ns_sem);
516
517 return err;
518}
519
520int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
521 struct nilfs_root **rootp)
522{
523 struct the_nilfs *nilfs = sb->s_fs_info;
524 struct nilfs_root *root;
525 struct nilfs_checkpoint *raw_cp;
526 struct buffer_head *bh_cp;
527 int err = -ENOMEM;
528
529 root = nilfs_find_or_create_root(
530 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
531 if (!root)
532 return err;
533
534 if (root->ifile)
535 goto reuse; /* already attached checkpoint */
536
537 down_read(&nilfs->ns_segctor_sem);
538 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
539 &bh_cp);
540 up_read(&nilfs->ns_segctor_sem);
541 if (unlikely(err)) {
542 if (err == -ENOENT || err == -EINVAL) {
543 printk(KERN_ERR
544 "NILFS: Invalid checkpoint "
545 "(checkpoint number=%llu)\n",
546 (unsigned long long)cno);
547 err = -EINVAL;
548 }
549 goto failed;
550 }
551
552 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
553 &raw_cp->cp_ifile_inode, &root->ifile);
554 if (err)
555 goto failed_bh;
556
557 atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
558 atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
559
560 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
561
562 reuse:
563 *rootp = root;
564 return 0;
565
566 failed_bh:
567 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
568 failed:
569 nilfs_put_root(root);
570
571 return err;
572}
573
574static int nilfs_freeze(struct super_block *sb)
575{
576 struct the_nilfs *nilfs = sb->s_fs_info;
577 int err;
578
579 if (sb->s_flags & MS_RDONLY)
580 return 0;
581
582 /* Mark super block clean */
583 down_write(&nilfs->ns_sem);
584 err = nilfs_cleanup_super(sb);
585 up_write(&nilfs->ns_sem);
586 return err;
587}
588
589static int nilfs_unfreeze(struct super_block *sb)
590{
591 struct the_nilfs *nilfs = sb->s_fs_info;
592
593 if (sb->s_flags & MS_RDONLY)
594 return 0;
595
596 down_write(&nilfs->ns_sem);
597 nilfs_setup_super(sb, false);
598 up_write(&nilfs->ns_sem);
599 return 0;
600}
601
602static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
603{
604 struct super_block *sb = dentry->d_sb;
605 struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
606 struct the_nilfs *nilfs = root->nilfs;
607 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
608 unsigned long long blocks;
609 unsigned long overhead;
610 unsigned long nrsvblocks;
611 sector_t nfreeblocks;
612 int err;
613
614 /*
615 * Compute all of the segment blocks
616 *
617 * The blocks before first segment and after last segment
618 * are excluded.
619 */
620 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
621 - nilfs->ns_first_data_block;
622 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
623
624 /*
625 * Compute the overhead
626 *
627 * When distributing meta data blocks outside segment structure,
628 * We must count them as the overhead.
629 */
630 overhead = 0;
631
632 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
633 if (unlikely(err))
634 return err;
635
636 buf->f_type = NILFS_SUPER_MAGIC;
637 buf->f_bsize = sb->s_blocksize;
638 buf->f_blocks = blocks - overhead;
639 buf->f_bfree = nfreeblocks;
640 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
641 (buf->f_bfree - nrsvblocks) : 0;
642 buf->f_files = atomic_read(&root->inodes_count);
643 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
644 buf->f_namelen = NILFS_NAME_LEN;
645 buf->f_fsid.val[0] = (u32)id;
646 buf->f_fsid.val[1] = (u32)(id >> 32);
647
648 return 0;
649}
650
651static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
652{
653 struct super_block *sb = dentry->d_sb;
654 struct the_nilfs *nilfs = sb->s_fs_info;
655 struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
656
657 if (!nilfs_test_opt(nilfs, BARRIER))
658 seq_puts(seq, ",nobarrier");
659 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
660 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
661 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
662 seq_puts(seq, ",errors=panic");
663 if (nilfs_test_opt(nilfs, ERRORS_CONT))
664 seq_puts(seq, ",errors=continue");
665 if (nilfs_test_opt(nilfs, STRICT_ORDER))
666 seq_puts(seq, ",order=strict");
667 if (nilfs_test_opt(nilfs, NORECOVERY))
668 seq_puts(seq, ",norecovery");
669 if (nilfs_test_opt(nilfs, DISCARD))
670 seq_puts(seq, ",discard");
671
672 return 0;
673}
674
675static const struct super_operations nilfs_sops = {
676 .alloc_inode = nilfs_alloc_inode,
677 .destroy_inode = nilfs_destroy_inode,
678 .dirty_inode = nilfs_dirty_inode,
679 /* .write_inode = nilfs_write_inode, */
680 /* .put_inode = nilfs_put_inode, */
681 /* .drop_inode = nilfs_drop_inode, */
682 .evict_inode = nilfs_evict_inode,
683 .put_super = nilfs_put_super,
684 /* .write_super = nilfs_write_super, */
685 .sync_fs = nilfs_sync_fs,
686 .freeze_fs = nilfs_freeze,
687 .unfreeze_fs = nilfs_unfreeze,
688 /* .write_super_lockfs */
689 /* .unlockfs */
690 .statfs = nilfs_statfs,
691 .remount_fs = nilfs_remount,
692 /* .umount_begin */
693 .show_options = nilfs_show_options
694};
695
696enum {
697 Opt_err_cont, Opt_err_panic, Opt_err_ro,
698 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
699 Opt_discard, Opt_nodiscard, Opt_err,
700};
701
702static match_table_t tokens = {
703 {Opt_err_cont, "errors=continue"},
704 {Opt_err_panic, "errors=panic"},
705 {Opt_err_ro, "errors=remount-ro"},
706 {Opt_barrier, "barrier"},
707 {Opt_nobarrier, "nobarrier"},
708 {Opt_snapshot, "cp=%u"},
709 {Opt_order, "order=%s"},
710 {Opt_norecovery, "norecovery"},
711 {Opt_discard, "discard"},
712 {Opt_nodiscard, "nodiscard"},
713 {Opt_err, NULL}
714};
715
716static int parse_options(char *options, struct super_block *sb, int is_remount)
717{
718 struct the_nilfs *nilfs = sb->s_fs_info;
719 char *p;
720 substring_t args[MAX_OPT_ARGS];
721
722 if (!options)
723 return 1;
724
725 while ((p = strsep(&options, ",")) != NULL) {
726 int token;
727 if (!*p)
728 continue;
729
730 token = match_token(p, tokens, args);
731 switch (token) {
732 case Opt_barrier:
733 nilfs_set_opt(nilfs, BARRIER);
734 break;
735 case Opt_nobarrier:
736 nilfs_clear_opt(nilfs, BARRIER);
737 break;
738 case Opt_order:
739 if (strcmp(args[0].from, "relaxed") == 0)
740 /* Ordered data semantics */
741 nilfs_clear_opt(nilfs, STRICT_ORDER);
742 else if (strcmp(args[0].from, "strict") == 0)
743 /* Strict in-order semantics */
744 nilfs_set_opt(nilfs, STRICT_ORDER);
745 else
746 return 0;
747 break;
748 case Opt_err_panic:
749 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
750 break;
751 case Opt_err_ro:
752 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
753 break;
754 case Opt_err_cont:
755 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
756 break;
757 case Opt_snapshot:
758 if (is_remount) {
759 printk(KERN_ERR
760 "NILFS: \"%s\" option is invalid "
761 "for remount.\n", p);
762 return 0;
763 }
764 break;
765 case Opt_norecovery:
766 nilfs_set_opt(nilfs, NORECOVERY);
767 break;
768 case Opt_discard:
769 nilfs_set_opt(nilfs, DISCARD);
770 break;
771 case Opt_nodiscard:
772 nilfs_clear_opt(nilfs, DISCARD);
773 break;
774 default:
775 printk(KERN_ERR
776 "NILFS: Unrecognized mount option \"%s\"\n", p);
777 return 0;
778 }
779 }
780 return 1;
781}
782
783static inline void
784nilfs_set_default_options(struct super_block *sb,
785 struct nilfs_super_block *sbp)
786{
787 struct the_nilfs *nilfs = sb->s_fs_info;
788
789 nilfs->ns_mount_opt =
790 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
791}
792
793static int nilfs_setup_super(struct super_block *sb, int is_mount)
794{
795 struct the_nilfs *nilfs = sb->s_fs_info;
796 struct nilfs_super_block **sbp;
797 int max_mnt_count;
798 int mnt_count;
799
800 /* nilfs->ns_sem must be locked by the caller. */
801 sbp = nilfs_prepare_super(sb, 0);
802 if (!sbp)
803 return -EIO;
804
805 if (!is_mount)
806 goto skip_mount_setup;
807
808 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
809 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
810
811 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
812 printk(KERN_WARNING
813 "NILFS warning: mounting fs with errors\n");
814#if 0
815 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
816 printk(KERN_WARNING
817 "NILFS warning: maximal mount count reached\n");
818#endif
819 }
820 if (!max_mnt_count)
821 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
822
823 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
824 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
825
826skip_mount_setup:
827 sbp[0]->s_state =
828 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
829 /* synchronize sbp[1] with sbp[0] */
830 if (sbp[1])
831 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
832 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
833}
834
835struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
836 u64 pos, int blocksize,
837 struct buffer_head **pbh)
838{
839 unsigned long long sb_index = pos;
840 unsigned long offset;
841
842 offset = do_div(sb_index, blocksize);
843 *pbh = sb_bread(sb, sb_index);
844 if (!*pbh)
845 return NULL;
846 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
847}
848
849int nilfs_store_magic_and_option(struct super_block *sb,
850 struct nilfs_super_block *sbp,
851 char *data)
852{
853 struct the_nilfs *nilfs = sb->s_fs_info;
854
855 sb->s_magic = le16_to_cpu(sbp->s_magic);
856
857 /* FS independent flags */
858#ifdef NILFS_ATIME_DISABLE
859 sb->s_flags |= MS_NOATIME;
860#endif
861
862 nilfs_set_default_options(sb, sbp);
863
864 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
865 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
866 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
867 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
868
869 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
870}
871
872int nilfs_check_feature_compatibility(struct super_block *sb,
873 struct nilfs_super_block *sbp)
874{
875 __u64 features;
876
877 features = le64_to_cpu(sbp->s_feature_incompat) &
878 ~NILFS_FEATURE_INCOMPAT_SUPP;
879 if (features) {
880 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
881 "optional features (%llx)\n",
882 (unsigned long long)features);
883 return -EINVAL;
884 }
885 features = le64_to_cpu(sbp->s_feature_compat_ro) &
886 ~NILFS_FEATURE_COMPAT_RO_SUPP;
887 if (!(sb->s_flags & MS_RDONLY) && features) {
888 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
889 "unsupported optional features (%llx)\n",
890 (unsigned long long)features);
891 return -EINVAL;
892 }
893 return 0;
894}
895
896static int nilfs_get_root_dentry(struct super_block *sb,
897 struct nilfs_root *root,
898 struct dentry **root_dentry)
899{
900 struct inode *inode;
901 struct dentry *dentry;
902 int ret = 0;
903
904 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
905 if (IS_ERR(inode)) {
906 printk(KERN_ERR "NILFS: get root inode failed\n");
907 ret = PTR_ERR(inode);
908 goto out;
909 }
910 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
911 iput(inode);
912 printk(KERN_ERR "NILFS: corrupt root inode.\n");
913 ret = -EINVAL;
914 goto out;
915 }
916
917 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
918 dentry = d_find_alias(inode);
919 if (!dentry) {
920 dentry = d_make_root(inode);
921 if (!dentry) {
922 ret = -ENOMEM;
923 goto failed_dentry;
924 }
925 } else {
926 iput(inode);
927 }
928 } else {
929 dentry = d_obtain_alias(inode);
930 if (IS_ERR(dentry)) {
931 ret = PTR_ERR(dentry);
932 goto failed_dentry;
933 }
934 }
935 *root_dentry = dentry;
936 out:
937 return ret;
938
939 failed_dentry:
940 printk(KERN_ERR "NILFS: get root dentry failed\n");
941 goto out;
942}
943
944static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
945 struct dentry **root_dentry)
946{
947 struct the_nilfs *nilfs = s->s_fs_info;
948 struct nilfs_root *root;
949 int ret;
950
951 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
952
953 down_read(&nilfs->ns_segctor_sem);
954 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
955 up_read(&nilfs->ns_segctor_sem);
956 if (ret < 0) {
957 ret = (ret == -ENOENT) ? -EINVAL : ret;
958 goto out;
959 } else if (!ret) {
960 printk(KERN_ERR "NILFS: The specified checkpoint is "
961 "not a snapshot (checkpoint number=%llu).\n",
962 (unsigned long long)cno);
963 ret = -EINVAL;
964 goto out;
965 }
966
967 ret = nilfs_attach_checkpoint(s, cno, false, &root);
968 if (ret) {
969 printk(KERN_ERR "NILFS: error loading snapshot "
970 "(checkpoint number=%llu).\n",
971 (unsigned long long)cno);
972 goto out;
973 }
974 ret = nilfs_get_root_dentry(s, root, root_dentry);
975 nilfs_put_root(root);
976 out:
977 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
978 return ret;
979}
980
981static int nilfs_tree_was_touched(struct dentry *root_dentry)
982{
983 return root_dentry->d_count > 1;
984}
985
986/**
987 * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
988 * @root_dentry: root dentry of the tree to be shrunk
989 *
990 * This function returns true if the tree was in-use.
991 */
992static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
993{
994 if (have_submounts(root_dentry))
995 return true;
996 shrink_dcache_parent(root_dentry);
997 return nilfs_tree_was_touched(root_dentry);
998}
999
1000int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1001{
1002 struct the_nilfs *nilfs = sb->s_fs_info;
1003 struct nilfs_root *root;
1004 struct inode *inode;
1005 struct dentry *dentry;
1006 int ret;
1007
1008 if (cno < 0 || cno > nilfs->ns_cno)
1009 return false;
1010
1011 if (cno >= nilfs_last_cno(nilfs))
1012 return true; /* protect recent checkpoints */
1013
1014 ret = false;
1015 root = nilfs_lookup_root(nilfs, cno);
1016 if (root) {
1017 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1018 if (inode) {
1019 dentry = d_find_alias(inode);
1020 if (dentry) {
1021 if (nilfs_tree_was_touched(dentry))
1022 ret = nilfs_try_to_shrink_tree(dentry);
1023 dput(dentry);
1024 }
1025 iput(inode);
1026 }
1027 nilfs_put_root(root);
1028 }
1029 return ret;
1030}
1031
1032/**
1033 * nilfs_fill_super() - initialize a super block instance
1034 * @sb: super_block
1035 * @data: mount options
1036 * @silent: silent mode flag
1037 *
1038 * This function is called exclusively by nilfs->ns_mount_mutex.
1039 * So, the recovery process is protected from other simultaneous mounts.
1040 */
1041static int
1042nilfs_fill_super(struct super_block *sb, void *data, int silent)
1043{
1044 struct the_nilfs *nilfs;
1045 struct nilfs_root *fsroot;
1046 struct backing_dev_info *bdi;
1047 __u64 cno;
1048 int err;
1049
1050 nilfs = alloc_nilfs(sb->s_bdev);
1051 if (!nilfs)
1052 return -ENOMEM;
1053
1054 sb->s_fs_info = nilfs;
1055
1056 err = init_nilfs(nilfs, sb, (char *)data);
1057 if (err)
1058 goto failed_nilfs;
1059
1060 sb->s_op = &nilfs_sops;
1061 sb->s_export_op = &nilfs_export_ops;
1062 sb->s_root = NULL;
1063 sb->s_time_gran = 1;
1064 sb->s_max_links = NILFS_LINK_MAX;
1065
1066 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1067 sb->s_bdi = bdi ? : &default_backing_dev_info;
1068
1069 err = load_nilfs(nilfs, sb);
1070 if (err)
1071 goto failed_nilfs;
1072
1073 cno = nilfs_last_cno(nilfs);
1074 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1075 if (err) {
1076 printk(KERN_ERR "NILFS: error loading last checkpoint "
1077 "(checkpoint number=%llu).\n", (unsigned long long)cno);
1078 goto failed_unload;
1079 }
1080
1081 if (!(sb->s_flags & MS_RDONLY)) {
1082 err = nilfs_attach_log_writer(sb, fsroot);
1083 if (err)
1084 goto failed_checkpoint;
1085 }
1086
1087 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1088 if (err)
1089 goto failed_segctor;
1090
1091 nilfs_put_root(fsroot);
1092
1093 if (!(sb->s_flags & MS_RDONLY)) {
1094 down_write(&nilfs->ns_sem);
1095 nilfs_setup_super(sb, true);
1096 up_write(&nilfs->ns_sem);
1097 }
1098
1099 return 0;
1100
1101 failed_segctor:
1102 nilfs_detach_log_writer(sb);
1103
1104 failed_checkpoint:
1105 nilfs_put_root(fsroot);
1106
1107 failed_unload:
1108 iput(nilfs->ns_sufile);
1109 iput(nilfs->ns_cpfile);
1110 iput(nilfs->ns_dat);
1111
1112 failed_nilfs:
1113 destroy_nilfs(nilfs);
1114 return err;
1115}
1116
1117static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1118{
1119 struct the_nilfs *nilfs = sb->s_fs_info;
1120 unsigned long old_sb_flags;
1121 unsigned long old_mount_opt;
1122 int err;
1123
1124 old_sb_flags = sb->s_flags;
1125 old_mount_opt = nilfs->ns_mount_opt;
1126
1127 if (!parse_options(data, sb, 1)) {
1128 err = -EINVAL;
1129 goto restore_opts;
1130 }
1131 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1132
1133 err = -EINVAL;
1134
1135 if (!nilfs_valid_fs(nilfs)) {
1136 printk(KERN_WARNING "NILFS (device %s): couldn't "
1137 "remount because the filesystem is in an "
1138 "incomplete recovery state.\n", sb->s_id);
1139 goto restore_opts;
1140 }
1141
1142 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1143 goto out;
1144 if (*flags & MS_RDONLY) {
1145 /* Shutting down log writer */
1146 nilfs_detach_log_writer(sb);
1147 sb->s_flags |= MS_RDONLY;
1148
1149 /*
1150 * Remounting a valid RW partition RDONLY, so set
1151 * the RDONLY flag and then mark the partition as valid again.
1152 */
1153 down_write(&nilfs->ns_sem);
1154 nilfs_cleanup_super(sb);
1155 up_write(&nilfs->ns_sem);
1156 } else {
1157 __u64 features;
1158 struct nilfs_root *root;
1159
1160 /*
1161 * Mounting a RDONLY partition read-write, so reread and
1162 * store the current valid flag. (It may have been changed
1163 * by fsck since we originally mounted the partition.)
1164 */
1165 down_read(&nilfs->ns_sem);
1166 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1167 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1168 up_read(&nilfs->ns_sem);
1169 if (features) {
1170 printk(KERN_WARNING "NILFS (device %s): couldn't "
1171 "remount RDWR because of unsupported optional "
1172 "features (%llx)\n",
1173 sb->s_id, (unsigned long long)features);
1174 err = -EROFS;
1175 goto restore_opts;
1176 }
1177
1178 sb->s_flags &= ~MS_RDONLY;
1179
1180 root = NILFS_I(sb->s_root->d_inode)->i_root;
1181 err = nilfs_attach_log_writer(sb, root);
1182 if (err)
1183 goto restore_opts;
1184
1185 down_write(&nilfs->ns_sem);
1186 nilfs_setup_super(sb, true);
1187 up_write(&nilfs->ns_sem);
1188 }
1189 out:
1190 return 0;
1191
1192 restore_opts:
1193 sb->s_flags = old_sb_flags;
1194 nilfs->ns_mount_opt = old_mount_opt;
1195 return err;
1196}
1197
1198struct nilfs_super_data {
1199 struct block_device *bdev;
1200 __u64 cno;
1201 int flags;
1202};
1203
1204/**
1205 * nilfs_identify - pre-read mount options needed to identify mount instance
1206 * @data: mount options
1207 * @sd: nilfs_super_data
1208 */
1209static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1210{
1211 char *p, *options = data;
1212 substring_t args[MAX_OPT_ARGS];
1213 int token;
1214 int ret = 0;
1215
1216 do {
1217 p = strsep(&options, ",");
1218 if (p != NULL && *p) {
1219 token = match_token(p, tokens, args);
1220 if (token == Opt_snapshot) {
1221 if (!(sd->flags & MS_RDONLY)) {
1222 ret++;
1223 } else {
1224 sd->cno = simple_strtoull(args[0].from,
1225 NULL, 0);
1226 /*
1227 * No need to see the end pointer;
1228 * match_token() has done syntax
1229 * checking.
1230 */
1231 if (sd->cno == 0)
1232 ret++;
1233 }
1234 }
1235 if (ret)
1236 printk(KERN_ERR
1237 "NILFS: invalid mount option: %s\n", p);
1238 }
1239 if (!options)
1240 break;
1241 BUG_ON(options == data);
1242 *(options - 1) = ',';
1243 } while (!ret);
1244 return ret;
1245}
1246
1247static int nilfs_set_bdev_super(struct super_block *s, void *data)
1248{
1249 s->s_bdev = data;
1250 s->s_dev = s->s_bdev->bd_dev;
1251 return 0;
1252}
1253
1254static int nilfs_test_bdev_super(struct super_block *s, void *data)
1255{
1256 return (void *)s->s_bdev == data;
1257}
1258
1259static struct dentry *
1260nilfs_mount(struct file_system_type *fs_type, int flags,
1261 const char *dev_name, void *data)
1262{
1263 struct nilfs_super_data sd;
1264 struct super_block *s;
1265 fmode_t mode = FMODE_READ | FMODE_EXCL;
1266 struct dentry *root_dentry;
1267 int err, s_new = false;
1268
1269 if (!(flags & MS_RDONLY))
1270 mode |= FMODE_WRITE;
1271
1272 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1273 if (IS_ERR(sd.bdev))
1274 return ERR_CAST(sd.bdev);
1275
1276 sd.cno = 0;
1277 sd.flags = flags;
1278 if (nilfs_identify((char *)data, &sd)) {
1279 err = -EINVAL;
1280 goto failed;
1281 }
1282
1283 /*
1284 * once the super is inserted into the list by sget, s_umount
1285 * will protect the lockfs code from trying to start a snapshot
1286 * while we are mounting
1287 */
1288 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1289 if (sd.bdev->bd_fsfreeze_count > 0) {
1290 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1291 err = -EBUSY;
1292 goto failed;
1293 }
1294 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1295 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1296 if (IS_ERR(s)) {
1297 err = PTR_ERR(s);
1298 goto failed;
1299 }
1300
1301 if (!s->s_root) {
1302 char b[BDEVNAME_SIZE];
1303
1304 s_new = true;
1305
1306 /* New superblock instance created */
1307 s->s_flags = flags;
1308 s->s_mode = mode;
1309 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1310 sb_set_blocksize(s, block_size(sd.bdev));
1311
1312 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1313 if (err)
1314 goto failed_super;
1315
1316 s->s_flags |= MS_ACTIVE;
1317 } else if (!sd.cno) {
1318 int busy = false;
1319
1320 if (nilfs_tree_was_touched(s->s_root)) {
1321 busy = nilfs_try_to_shrink_tree(s->s_root);
1322 if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1323 printk(KERN_ERR "NILFS: the device already "
1324 "has a %s mount.\n",
1325 (s->s_flags & MS_RDONLY) ?
1326 "read-only" : "read/write");
1327 err = -EBUSY;
1328 goto failed_super;
1329 }
1330 }
1331 if (!busy) {
1332 /*
1333 * Try remount to setup mount states if the current
1334 * tree is not mounted and only snapshots use this sb.
1335 */
1336 err = nilfs_remount(s, &flags, data);
1337 if (err)
1338 goto failed_super;
1339 }
1340 }
1341
1342 if (sd.cno) {
1343 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1344 if (err)
1345 goto failed_super;
1346 } else {
1347 root_dentry = dget(s->s_root);
1348 }
1349
1350 if (!s_new)
1351 blkdev_put(sd.bdev, mode);
1352
1353 return root_dentry;
1354
1355 failed_super:
1356 deactivate_locked_super(s);
1357
1358 failed:
1359 if (!s_new)
1360 blkdev_put(sd.bdev, mode);
1361 return ERR_PTR(err);
1362}
1363
1364struct file_system_type nilfs_fs_type = {
1365 .owner = THIS_MODULE,
1366 .name = "nilfs2",
1367 .mount = nilfs_mount,
1368 .kill_sb = kill_block_super,
1369 .fs_flags = FS_REQUIRES_DEV,
1370};
1371
1372static void nilfs_inode_init_once(void *obj)
1373{
1374 struct nilfs_inode_info *ii = obj;
1375
1376 INIT_LIST_HEAD(&ii->i_dirty);
1377#ifdef CONFIG_NILFS_XATTR
1378 init_rwsem(&ii->xattr_sem);
1379#endif
1380 address_space_init_once(&ii->i_btnode_cache);
1381 ii->i_bmap = &ii->i_bmap_data;
1382 inode_init_once(&ii->vfs_inode);
1383}
1384
1385static void nilfs_segbuf_init_once(void *obj)
1386{
1387 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1388}
1389
1390static void nilfs_destroy_cachep(void)
1391{
1392 if (nilfs_inode_cachep)
1393 kmem_cache_destroy(nilfs_inode_cachep);
1394 if (nilfs_transaction_cachep)
1395 kmem_cache_destroy(nilfs_transaction_cachep);
1396 if (nilfs_segbuf_cachep)
1397 kmem_cache_destroy(nilfs_segbuf_cachep);
1398 if (nilfs_btree_path_cache)
1399 kmem_cache_destroy(nilfs_btree_path_cache);
1400}
1401
1402static int __init nilfs_init_cachep(void)
1403{
1404 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1405 sizeof(struct nilfs_inode_info), 0,
1406 SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1407 if (!nilfs_inode_cachep)
1408 goto fail;
1409
1410 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1411 sizeof(struct nilfs_transaction_info), 0,
1412 SLAB_RECLAIM_ACCOUNT, NULL);
1413 if (!nilfs_transaction_cachep)
1414 goto fail;
1415
1416 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1417 sizeof(struct nilfs_segment_buffer), 0,
1418 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1419 if (!nilfs_segbuf_cachep)
1420 goto fail;
1421
1422 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1423 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1424 0, 0, NULL);
1425 if (!nilfs_btree_path_cache)
1426 goto fail;
1427
1428 return 0;
1429
1430fail:
1431 nilfs_destroy_cachep();
1432 return -ENOMEM;
1433}
1434
1435static int __init init_nilfs_fs(void)
1436{
1437 int err;
1438
1439 err = nilfs_init_cachep();
1440 if (err)
1441 goto fail;
1442
1443 err = register_filesystem(&nilfs_fs_type);
1444 if (err)
1445 goto free_cachep;
1446
1447 printk(KERN_INFO "NILFS version 2 loaded\n");
1448 return 0;
1449
1450free_cachep:
1451 nilfs_destroy_cachep();
1452fail:
1453 return err;
1454}
1455
1456static void __exit exit_nilfs_fs(void)
1457{
1458 nilfs_destroy_cachep();
1459 unregister_filesystem(&nilfs_fs_type);
1460}
1461
1462module_init(init_nilfs_fs)
1463module_exit(exit_nilfs_fs)
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * NILFS module and super block management.
4 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Written by Ryusuke Konishi.
8 */
9/*
10 * linux/fs/ext2/super.c
11 *
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise Pascal
15 * Universite Pierre et Marie Curie (Paris VI)
16 *
17 * from
18 *
19 * linux/fs/minix/inode.c
20 *
21 * Copyright (C) 1991, 1992 Linus Torvalds
22 *
23 * Big-endian to little-endian byte-swapping/bitmaps by
24 * David S. Miller (davem@caip.rutgers.edu), 1995
25 */
26
27#include <linux/module.h>
28#include <linux/string.h>
29#include <linux/slab.h>
30#include <linux/init.h>
31#include <linux/blkdev.h>
32#include <linux/parser.h>
33#include <linux/crc32.h>
34#include <linux/vfs.h>
35#include <linux/writeback.h>
36#include <linux/seq_file.h>
37#include <linux/mount.h>
38#include <linux/fs_context.h>
39#include "nilfs.h"
40#include "export.h"
41#include "mdt.h"
42#include "alloc.h"
43#include "btree.h"
44#include "btnode.h"
45#include "page.h"
46#include "cpfile.h"
47#include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
48#include "ifile.h"
49#include "dat.h"
50#include "segment.h"
51#include "segbuf.h"
52
53MODULE_AUTHOR("NTT Corp.");
54MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
55 "(NILFS)");
56MODULE_LICENSE("GPL");
57
58static struct kmem_cache *nilfs_inode_cachep;
59struct kmem_cache *nilfs_transaction_cachep;
60struct kmem_cache *nilfs_segbuf_cachep;
61struct kmem_cache *nilfs_btree_path_cache;
62
63static int nilfs_setup_super(struct super_block *sb, int is_mount);
64static int nilfs_remount(struct super_block *sb, int *flags, char *data);
65
66void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
67{
68 struct va_format vaf;
69 va_list args;
70 int level;
71
72 va_start(args, fmt);
73
74 level = printk_get_level(fmt);
75 vaf.fmt = printk_skip_level(fmt);
76 vaf.va = &args;
77
78 if (sb)
79 printk("%c%cNILFS (%s): %pV\n",
80 KERN_SOH_ASCII, level, sb->s_id, &vaf);
81 else
82 printk("%c%cNILFS: %pV\n",
83 KERN_SOH_ASCII, level, &vaf);
84
85 va_end(args);
86}
87
88static void nilfs_set_error(struct super_block *sb)
89{
90 struct the_nilfs *nilfs = sb->s_fs_info;
91 struct nilfs_super_block **sbp;
92
93 down_write(&nilfs->ns_sem);
94 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
95 nilfs->ns_mount_state |= NILFS_ERROR_FS;
96 sbp = nilfs_prepare_super(sb, 0);
97 if (likely(sbp)) {
98 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
99 if (sbp[1])
100 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
101 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
102 }
103 }
104 up_write(&nilfs->ns_sem);
105}
106
107/**
108 * __nilfs_error() - report failure condition on a filesystem
109 *
110 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
111 * reporting an error message. This function should be called when
112 * NILFS detects incoherences or defects of meta data on disk.
113 *
114 * This implements the body of nilfs_error() macro. Normally,
115 * nilfs_error() should be used. As for sustainable errors such as a
116 * single-shot I/O error, nilfs_err() should be used instead.
117 *
118 * Callers should not add a trailing newline since this will do it.
119 */
120void __nilfs_error(struct super_block *sb, const char *function,
121 const char *fmt, ...)
122{
123 struct the_nilfs *nilfs = sb->s_fs_info;
124 struct va_format vaf;
125 va_list args;
126
127 va_start(args, fmt);
128
129 vaf.fmt = fmt;
130 vaf.va = &args;
131
132 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
133 sb->s_id, function, &vaf);
134
135 va_end(args);
136
137 if (!sb_rdonly(sb)) {
138 nilfs_set_error(sb);
139
140 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
141 printk(KERN_CRIT "Remounting filesystem read-only\n");
142 sb->s_flags |= SB_RDONLY;
143 }
144 }
145
146 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
147 panic("NILFS (device %s): panic forced after error\n",
148 sb->s_id);
149}
150
151struct inode *nilfs_alloc_inode(struct super_block *sb)
152{
153 struct nilfs_inode_info *ii;
154
155 ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
156 if (!ii)
157 return NULL;
158 ii->i_bh = NULL;
159 ii->i_state = 0;
160 ii->i_cno = 0;
161 ii->i_assoc_inode = NULL;
162 ii->i_bmap = &ii->i_bmap_data;
163 return &ii->vfs_inode;
164}
165
166static void nilfs_free_inode(struct inode *inode)
167{
168 if (nilfs_is_metadata_file_inode(inode))
169 nilfs_mdt_destroy(inode);
170
171 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
172}
173
174static int nilfs_sync_super(struct super_block *sb, int flag)
175{
176 struct the_nilfs *nilfs = sb->s_fs_info;
177 int err;
178
179 retry:
180 set_buffer_dirty(nilfs->ns_sbh[0]);
181 if (nilfs_test_opt(nilfs, BARRIER)) {
182 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
183 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
184 } else {
185 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
186 }
187
188 if (unlikely(err)) {
189 nilfs_err(sb, "unable to write superblock: err=%d", err);
190 if (err == -EIO && nilfs->ns_sbh[1]) {
191 /*
192 * sbp[0] points to newer log than sbp[1],
193 * so copy sbp[0] to sbp[1] to take over sbp[0].
194 */
195 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
196 nilfs->ns_sbsize);
197 nilfs_fall_back_super_block(nilfs);
198 goto retry;
199 }
200 } else {
201 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
202
203 nilfs->ns_sbwcount++;
204
205 /*
206 * The latest segment becomes trailable from the position
207 * written in superblock.
208 */
209 clear_nilfs_discontinued(nilfs);
210
211 /* update GC protection for recent segments */
212 if (nilfs->ns_sbh[1]) {
213 if (flag == NILFS_SB_COMMIT_ALL) {
214 set_buffer_dirty(nilfs->ns_sbh[1]);
215 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
216 goto out;
217 }
218 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
219 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
220 sbp = nilfs->ns_sbp[1];
221 }
222
223 spin_lock(&nilfs->ns_last_segment_lock);
224 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
225 spin_unlock(&nilfs->ns_last_segment_lock);
226 }
227 out:
228 return err;
229}
230
231void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
232 struct the_nilfs *nilfs)
233{
234 sector_t nfreeblocks;
235
236 /* nilfs->ns_sem must be locked by the caller. */
237 nilfs_count_free_blocks(nilfs, &nfreeblocks);
238 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
239
240 spin_lock(&nilfs->ns_last_segment_lock);
241 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
242 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
243 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
244 spin_unlock(&nilfs->ns_last_segment_lock);
245}
246
247struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
248 int flip)
249{
250 struct the_nilfs *nilfs = sb->s_fs_info;
251 struct nilfs_super_block **sbp = nilfs->ns_sbp;
252
253 /* nilfs->ns_sem must be locked by the caller. */
254 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
255 if (sbp[1] &&
256 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
257 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
258 } else {
259 nilfs_crit(sb, "superblock broke");
260 return NULL;
261 }
262 } else if (sbp[1] &&
263 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
264 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
265 }
266
267 if (flip && sbp[1])
268 nilfs_swap_super_block(nilfs);
269
270 return sbp;
271}
272
273int nilfs_commit_super(struct super_block *sb, int flag)
274{
275 struct the_nilfs *nilfs = sb->s_fs_info;
276 struct nilfs_super_block **sbp = nilfs->ns_sbp;
277 time64_t t;
278
279 /* nilfs->ns_sem must be locked by the caller. */
280 t = ktime_get_real_seconds();
281 nilfs->ns_sbwtime = t;
282 sbp[0]->s_wtime = cpu_to_le64(t);
283 sbp[0]->s_sum = 0;
284 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
285 (unsigned char *)sbp[0],
286 nilfs->ns_sbsize));
287 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
288 sbp[1]->s_wtime = sbp[0]->s_wtime;
289 sbp[1]->s_sum = 0;
290 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
291 (unsigned char *)sbp[1],
292 nilfs->ns_sbsize));
293 }
294 clear_nilfs_sb_dirty(nilfs);
295 nilfs->ns_flushed_device = 1;
296 /* make sure store to ns_flushed_device cannot be reordered */
297 smp_wmb();
298 return nilfs_sync_super(sb, flag);
299}
300
301/**
302 * nilfs_cleanup_super() - write filesystem state for cleanup
303 * @sb: super block instance to be unmounted or degraded to read-only
304 *
305 * This function restores state flags in the on-disk super block.
306 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
307 * filesystem was not clean previously.
308 */
309int nilfs_cleanup_super(struct super_block *sb)
310{
311 struct the_nilfs *nilfs = sb->s_fs_info;
312 struct nilfs_super_block **sbp;
313 int flag = NILFS_SB_COMMIT;
314 int ret = -EIO;
315
316 sbp = nilfs_prepare_super(sb, 0);
317 if (sbp) {
318 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
319 nilfs_set_log_cursor(sbp[0], nilfs);
320 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
321 /*
322 * make the "clean" flag also to the opposite
323 * super block if both super blocks point to
324 * the same checkpoint.
325 */
326 sbp[1]->s_state = sbp[0]->s_state;
327 flag = NILFS_SB_COMMIT_ALL;
328 }
329 ret = nilfs_commit_super(sb, flag);
330 }
331 return ret;
332}
333
334/**
335 * nilfs_move_2nd_super - relocate secondary super block
336 * @sb: super block instance
337 * @sb2off: new offset of the secondary super block (in bytes)
338 */
339static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
340{
341 struct the_nilfs *nilfs = sb->s_fs_info;
342 struct buffer_head *nsbh;
343 struct nilfs_super_block *nsbp;
344 sector_t blocknr, newblocknr;
345 unsigned long offset;
346 int sb2i; /* array index of the secondary superblock */
347 int ret = 0;
348
349 /* nilfs->ns_sem must be locked by the caller. */
350 if (nilfs->ns_sbh[1] &&
351 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
352 sb2i = 1;
353 blocknr = nilfs->ns_sbh[1]->b_blocknr;
354 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
355 sb2i = 0;
356 blocknr = nilfs->ns_sbh[0]->b_blocknr;
357 } else {
358 sb2i = -1;
359 blocknr = 0;
360 }
361 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
362 goto out; /* super block location is unchanged */
363
364 /* Get new super block buffer */
365 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
366 offset = sb2off & (nilfs->ns_blocksize - 1);
367 nsbh = sb_getblk(sb, newblocknr);
368 if (!nsbh) {
369 nilfs_warn(sb,
370 "unable to move secondary superblock to block %llu",
371 (unsigned long long)newblocknr);
372 ret = -EIO;
373 goto out;
374 }
375 nsbp = (void *)nsbh->b_data + offset;
376
377 lock_buffer(nsbh);
378 if (sb2i >= 0) {
379 /*
380 * The position of the second superblock only changes by 4KiB,
381 * which is larger than the maximum superblock data size
382 * (= 1KiB), so there is no need to use memmove() to allow
383 * overlap between source and destination.
384 */
385 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
386
387 /*
388 * Zero fill after copy to avoid overwriting in case of move
389 * within the same block.
390 */
391 memset(nsbh->b_data, 0, offset);
392 memset((void *)nsbp + nilfs->ns_sbsize, 0,
393 nsbh->b_size - offset - nilfs->ns_sbsize);
394 } else {
395 memset(nsbh->b_data, 0, nsbh->b_size);
396 }
397 set_buffer_uptodate(nsbh);
398 unlock_buffer(nsbh);
399
400 if (sb2i >= 0) {
401 brelse(nilfs->ns_sbh[sb2i]);
402 nilfs->ns_sbh[sb2i] = nsbh;
403 nilfs->ns_sbp[sb2i] = nsbp;
404 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
405 /* secondary super block will be restored to index 1 */
406 nilfs->ns_sbh[1] = nsbh;
407 nilfs->ns_sbp[1] = nsbp;
408 } else {
409 brelse(nsbh);
410 }
411out:
412 return ret;
413}
414
415/**
416 * nilfs_resize_fs - resize the filesystem
417 * @sb: super block instance
418 * @newsize: new size of the filesystem (in bytes)
419 */
420int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
421{
422 struct the_nilfs *nilfs = sb->s_fs_info;
423 struct nilfs_super_block **sbp;
424 __u64 devsize, newnsegs;
425 loff_t sb2off;
426 int ret;
427
428 ret = -ERANGE;
429 devsize = bdev_nr_bytes(sb->s_bdev);
430 if (newsize > devsize)
431 goto out;
432
433 /*
434 * Prevent underflow in second superblock position calculation.
435 * The exact minimum size check is done in nilfs_sufile_resize().
436 */
437 if (newsize < 4096) {
438 ret = -ENOSPC;
439 goto out;
440 }
441
442 /*
443 * Write lock is required to protect some functions depending
444 * on the number of segments, the number of reserved segments,
445 * and so forth.
446 */
447 down_write(&nilfs->ns_segctor_sem);
448
449 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
450 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
451 do_div(newnsegs, nilfs->ns_blocks_per_segment);
452
453 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
454 up_write(&nilfs->ns_segctor_sem);
455 if (ret < 0)
456 goto out;
457
458 ret = nilfs_construct_segment(sb);
459 if (ret < 0)
460 goto out;
461
462 down_write(&nilfs->ns_sem);
463 nilfs_move_2nd_super(sb, sb2off);
464 ret = -EIO;
465 sbp = nilfs_prepare_super(sb, 0);
466 if (likely(sbp)) {
467 nilfs_set_log_cursor(sbp[0], nilfs);
468 /*
469 * Drop NILFS_RESIZE_FS flag for compatibility with
470 * mount-time resize which may be implemented in a
471 * future release.
472 */
473 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
474 ~NILFS_RESIZE_FS);
475 sbp[0]->s_dev_size = cpu_to_le64(newsize);
476 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
477 if (sbp[1])
478 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
479 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
480 }
481 up_write(&nilfs->ns_sem);
482
483 /*
484 * Reset the range of allocatable segments last. This order
485 * is important in the case of expansion because the secondary
486 * superblock must be protected from log write until migration
487 * completes.
488 */
489 if (!ret)
490 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
491out:
492 return ret;
493}
494
495static void nilfs_put_super(struct super_block *sb)
496{
497 struct the_nilfs *nilfs = sb->s_fs_info;
498
499 nilfs_detach_log_writer(sb);
500
501 if (!sb_rdonly(sb)) {
502 down_write(&nilfs->ns_sem);
503 nilfs_cleanup_super(sb);
504 up_write(&nilfs->ns_sem);
505 }
506
507 nilfs_sysfs_delete_device_group(nilfs);
508 iput(nilfs->ns_sufile);
509 iput(nilfs->ns_cpfile);
510 iput(nilfs->ns_dat);
511
512 destroy_nilfs(nilfs);
513 sb->s_fs_info = NULL;
514}
515
516static int nilfs_sync_fs(struct super_block *sb, int wait)
517{
518 struct the_nilfs *nilfs = sb->s_fs_info;
519 struct nilfs_super_block **sbp;
520 int err = 0;
521
522 /* This function is called when super block should be written back */
523 if (wait)
524 err = nilfs_construct_segment(sb);
525
526 down_write(&nilfs->ns_sem);
527 if (nilfs_sb_dirty(nilfs)) {
528 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
529 if (likely(sbp)) {
530 nilfs_set_log_cursor(sbp[0], nilfs);
531 nilfs_commit_super(sb, NILFS_SB_COMMIT);
532 }
533 }
534 up_write(&nilfs->ns_sem);
535
536 if (!err)
537 err = nilfs_flush_device(nilfs);
538
539 return err;
540}
541
542int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
543 struct nilfs_root **rootp)
544{
545 struct the_nilfs *nilfs = sb->s_fs_info;
546 struct nilfs_root *root;
547 struct nilfs_checkpoint *raw_cp;
548 struct buffer_head *bh_cp;
549 int err = -ENOMEM;
550
551 root = nilfs_find_or_create_root(
552 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
553 if (!root)
554 return err;
555
556 if (root->ifile)
557 goto reuse; /* already attached checkpoint */
558
559 down_read(&nilfs->ns_segctor_sem);
560 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
561 &bh_cp);
562 up_read(&nilfs->ns_segctor_sem);
563 if (unlikely(err)) {
564 if (err == -ENOENT || err == -EINVAL) {
565 nilfs_err(sb,
566 "Invalid checkpoint (checkpoint number=%llu)",
567 (unsigned long long)cno);
568 err = -EINVAL;
569 }
570 goto failed;
571 }
572
573 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
574 &raw_cp->cp_ifile_inode, &root->ifile);
575 if (err)
576 goto failed_bh;
577
578 atomic64_set(&root->inodes_count,
579 le64_to_cpu(raw_cp->cp_inodes_count));
580 atomic64_set(&root->blocks_count,
581 le64_to_cpu(raw_cp->cp_blocks_count));
582
583 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
584
585 reuse:
586 *rootp = root;
587 return 0;
588
589 failed_bh:
590 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
591 failed:
592 nilfs_put_root(root);
593
594 return err;
595}
596
597static int nilfs_freeze(struct super_block *sb)
598{
599 struct the_nilfs *nilfs = sb->s_fs_info;
600 int err;
601
602 if (sb_rdonly(sb))
603 return 0;
604
605 /* Mark super block clean */
606 down_write(&nilfs->ns_sem);
607 err = nilfs_cleanup_super(sb);
608 up_write(&nilfs->ns_sem);
609 return err;
610}
611
612static int nilfs_unfreeze(struct super_block *sb)
613{
614 struct the_nilfs *nilfs = sb->s_fs_info;
615
616 if (sb_rdonly(sb))
617 return 0;
618
619 down_write(&nilfs->ns_sem);
620 nilfs_setup_super(sb, false);
621 up_write(&nilfs->ns_sem);
622 return 0;
623}
624
625static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
626{
627 struct super_block *sb = dentry->d_sb;
628 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
629 struct the_nilfs *nilfs = root->nilfs;
630 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
631 unsigned long long blocks;
632 unsigned long overhead;
633 unsigned long nrsvblocks;
634 sector_t nfreeblocks;
635 u64 nmaxinodes, nfreeinodes;
636 int err;
637
638 /*
639 * Compute all of the segment blocks
640 *
641 * The blocks before first segment and after last segment
642 * are excluded.
643 */
644 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
645 - nilfs->ns_first_data_block;
646 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
647
648 /*
649 * Compute the overhead
650 *
651 * When distributing meta data blocks outside segment structure,
652 * We must count them as the overhead.
653 */
654 overhead = 0;
655
656 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
657 if (unlikely(err))
658 return err;
659
660 err = nilfs_ifile_count_free_inodes(root->ifile,
661 &nmaxinodes, &nfreeinodes);
662 if (unlikely(err)) {
663 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
664 if (err == -ERANGE) {
665 /*
666 * If nilfs_palloc_count_max_entries() returns
667 * -ERANGE error code then we simply treat
668 * curent inodes count as maximum possible and
669 * zero as free inodes value.
670 */
671 nmaxinodes = atomic64_read(&root->inodes_count);
672 nfreeinodes = 0;
673 err = 0;
674 } else
675 return err;
676 }
677
678 buf->f_type = NILFS_SUPER_MAGIC;
679 buf->f_bsize = sb->s_blocksize;
680 buf->f_blocks = blocks - overhead;
681 buf->f_bfree = nfreeblocks;
682 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
683 (buf->f_bfree - nrsvblocks) : 0;
684 buf->f_files = nmaxinodes;
685 buf->f_ffree = nfreeinodes;
686 buf->f_namelen = NILFS_NAME_LEN;
687 buf->f_fsid = u64_to_fsid(id);
688
689 return 0;
690}
691
692static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
693{
694 struct super_block *sb = dentry->d_sb;
695 struct the_nilfs *nilfs = sb->s_fs_info;
696 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
697
698 if (!nilfs_test_opt(nilfs, BARRIER))
699 seq_puts(seq, ",nobarrier");
700 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
701 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
702 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
703 seq_puts(seq, ",errors=panic");
704 if (nilfs_test_opt(nilfs, ERRORS_CONT))
705 seq_puts(seq, ",errors=continue");
706 if (nilfs_test_opt(nilfs, STRICT_ORDER))
707 seq_puts(seq, ",order=strict");
708 if (nilfs_test_opt(nilfs, NORECOVERY))
709 seq_puts(seq, ",norecovery");
710 if (nilfs_test_opt(nilfs, DISCARD))
711 seq_puts(seq, ",discard");
712
713 return 0;
714}
715
716static const struct super_operations nilfs_sops = {
717 .alloc_inode = nilfs_alloc_inode,
718 .free_inode = nilfs_free_inode,
719 .dirty_inode = nilfs_dirty_inode,
720 .evict_inode = nilfs_evict_inode,
721 .put_super = nilfs_put_super,
722 .sync_fs = nilfs_sync_fs,
723 .freeze_fs = nilfs_freeze,
724 .unfreeze_fs = nilfs_unfreeze,
725 .statfs = nilfs_statfs,
726 .remount_fs = nilfs_remount,
727 .show_options = nilfs_show_options
728};
729
730enum {
731 Opt_err_cont, Opt_err_panic, Opt_err_ro,
732 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
733 Opt_discard, Opt_nodiscard, Opt_err,
734};
735
736static match_table_t tokens = {
737 {Opt_err_cont, "errors=continue"},
738 {Opt_err_panic, "errors=panic"},
739 {Opt_err_ro, "errors=remount-ro"},
740 {Opt_barrier, "barrier"},
741 {Opt_nobarrier, "nobarrier"},
742 {Opt_snapshot, "cp=%u"},
743 {Opt_order, "order=%s"},
744 {Opt_norecovery, "norecovery"},
745 {Opt_discard, "discard"},
746 {Opt_nodiscard, "nodiscard"},
747 {Opt_err, NULL}
748};
749
750static int parse_options(char *options, struct super_block *sb, int is_remount)
751{
752 struct the_nilfs *nilfs = sb->s_fs_info;
753 char *p;
754 substring_t args[MAX_OPT_ARGS];
755
756 if (!options)
757 return 1;
758
759 while ((p = strsep(&options, ",")) != NULL) {
760 int token;
761
762 if (!*p)
763 continue;
764
765 token = match_token(p, tokens, args);
766 switch (token) {
767 case Opt_barrier:
768 nilfs_set_opt(nilfs, BARRIER);
769 break;
770 case Opt_nobarrier:
771 nilfs_clear_opt(nilfs, BARRIER);
772 break;
773 case Opt_order:
774 if (strcmp(args[0].from, "relaxed") == 0)
775 /* Ordered data semantics */
776 nilfs_clear_opt(nilfs, STRICT_ORDER);
777 else if (strcmp(args[0].from, "strict") == 0)
778 /* Strict in-order semantics */
779 nilfs_set_opt(nilfs, STRICT_ORDER);
780 else
781 return 0;
782 break;
783 case Opt_err_panic:
784 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
785 break;
786 case Opt_err_ro:
787 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
788 break;
789 case Opt_err_cont:
790 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
791 break;
792 case Opt_snapshot:
793 if (is_remount) {
794 nilfs_err(sb,
795 "\"%s\" option is invalid for remount",
796 p);
797 return 0;
798 }
799 break;
800 case Opt_norecovery:
801 nilfs_set_opt(nilfs, NORECOVERY);
802 break;
803 case Opt_discard:
804 nilfs_set_opt(nilfs, DISCARD);
805 break;
806 case Opt_nodiscard:
807 nilfs_clear_opt(nilfs, DISCARD);
808 break;
809 default:
810 nilfs_err(sb, "unrecognized mount option \"%s\"", p);
811 return 0;
812 }
813 }
814 return 1;
815}
816
817static inline void
818nilfs_set_default_options(struct super_block *sb,
819 struct nilfs_super_block *sbp)
820{
821 struct the_nilfs *nilfs = sb->s_fs_info;
822
823 nilfs->ns_mount_opt =
824 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
825}
826
827static int nilfs_setup_super(struct super_block *sb, int is_mount)
828{
829 struct the_nilfs *nilfs = sb->s_fs_info;
830 struct nilfs_super_block **sbp;
831 int max_mnt_count;
832 int mnt_count;
833
834 /* nilfs->ns_sem must be locked by the caller. */
835 sbp = nilfs_prepare_super(sb, 0);
836 if (!sbp)
837 return -EIO;
838
839 if (!is_mount)
840 goto skip_mount_setup;
841
842 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
843 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
844
845 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
846 nilfs_warn(sb, "mounting fs with errors");
847#if 0
848 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
849 nilfs_warn(sb, "maximal mount count reached");
850#endif
851 }
852 if (!max_mnt_count)
853 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
854
855 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
856 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
857
858skip_mount_setup:
859 sbp[0]->s_state =
860 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
861 /* synchronize sbp[1] with sbp[0] */
862 if (sbp[1])
863 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
864 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
865}
866
867struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
868 u64 pos, int blocksize,
869 struct buffer_head **pbh)
870{
871 unsigned long long sb_index = pos;
872 unsigned long offset;
873
874 offset = do_div(sb_index, blocksize);
875 *pbh = sb_bread(sb, sb_index);
876 if (!*pbh)
877 return NULL;
878 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
879}
880
881int nilfs_store_magic_and_option(struct super_block *sb,
882 struct nilfs_super_block *sbp,
883 char *data)
884{
885 struct the_nilfs *nilfs = sb->s_fs_info;
886
887 sb->s_magic = le16_to_cpu(sbp->s_magic);
888
889 /* FS independent flags */
890#ifdef NILFS_ATIME_DISABLE
891 sb->s_flags |= SB_NOATIME;
892#endif
893
894 nilfs_set_default_options(sb, sbp);
895
896 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
897 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
898 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
899 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
900
901 return !parse_options(data, sb, 0) ? -EINVAL : 0;
902}
903
904int nilfs_check_feature_compatibility(struct super_block *sb,
905 struct nilfs_super_block *sbp)
906{
907 __u64 features;
908
909 features = le64_to_cpu(sbp->s_feature_incompat) &
910 ~NILFS_FEATURE_INCOMPAT_SUPP;
911 if (features) {
912 nilfs_err(sb,
913 "couldn't mount because of unsupported optional features (%llx)",
914 (unsigned long long)features);
915 return -EINVAL;
916 }
917 features = le64_to_cpu(sbp->s_feature_compat_ro) &
918 ~NILFS_FEATURE_COMPAT_RO_SUPP;
919 if (!sb_rdonly(sb) && features) {
920 nilfs_err(sb,
921 "couldn't mount RDWR because of unsupported optional features (%llx)",
922 (unsigned long long)features);
923 return -EINVAL;
924 }
925 return 0;
926}
927
928static int nilfs_get_root_dentry(struct super_block *sb,
929 struct nilfs_root *root,
930 struct dentry **root_dentry)
931{
932 struct inode *inode;
933 struct dentry *dentry;
934 int ret = 0;
935
936 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
937 if (IS_ERR(inode)) {
938 ret = PTR_ERR(inode);
939 nilfs_err(sb, "error %d getting root inode", ret);
940 goto out;
941 }
942 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
943 iput(inode);
944 nilfs_err(sb, "corrupt root inode");
945 ret = -EINVAL;
946 goto out;
947 }
948
949 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
950 dentry = d_find_alias(inode);
951 if (!dentry) {
952 dentry = d_make_root(inode);
953 if (!dentry) {
954 ret = -ENOMEM;
955 goto failed_dentry;
956 }
957 } else {
958 iput(inode);
959 }
960 } else {
961 dentry = d_obtain_root(inode);
962 if (IS_ERR(dentry)) {
963 ret = PTR_ERR(dentry);
964 goto failed_dentry;
965 }
966 }
967 *root_dentry = dentry;
968 out:
969 return ret;
970
971 failed_dentry:
972 nilfs_err(sb, "error %d getting root dentry", ret);
973 goto out;
974}
975
976static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
977 struct dentry **root_dentry)
978{
979 struct the_nilfs *nilfs = s->s_fs_info;
980 struct nilfs_root *root;
981 int ret;
982
983 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
984
985 down_read(&nilfs->ns_segctor_sem);
986 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
987 up_read(&nilfs->ns_segctor_sem);
988 if (ret < 0) {
989 ret = (ret == -ENOENT) ? -EINVAL : ret;
990 goto out;
991 } else if (!ret) {
992 nilfs_err(s,
993 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
994 (unsigned long long)cno);
995 ret = -EINVAL;
996 goto out;
997 }
998
999 ret = nilfs_attach_checkpoint(s, cno, false, &root);
1000 if (ret) {
1001 nilfs_err(s,
1002 "error %d while loading snapshot (checkpoint number=%llu)",
1003 ret, (unsigned long long)cno);
1004 goto out;
1005 }
1006 ret = nilfs_get_root_dentry(s, root, root_dentry);
1007 nilfs_put_root(root);
1008 out:
1009 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1010 return ret;
1011}
1012
1013/**
1014 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1015 * @root_dentry: root dentry of the tree to be shrunk
1016 *
1017 * This function returns true if the tree was in-use.
1018 */
1019static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1020{
1021 shrink_dcache_parent(root_dentry);
1022 return d_count(root_dentry) > 1;
1023}
1024
1025int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1026{
1027 struct the_nilfs *nilfs = sb->s_fs_info;
1028 struct nilfs_root *root;
1029 struct inode *inode;
1030 struct dentry *dentry;
1031 int ret;
1032
1033 if (cno > nilfs->ns_cno)
1034 return false;
1035
1036 if (cno >= nilfs_last_cno(nilfs))
1037 return true; /* protect recent checkpoints */
1038
1039 ret = false;
1040 root = nilfs_lookup_root(nilfs, cno);
1041 if (root) {
1042 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1043 if (inode) {
1044 dentry = d_find_alias(inode);
1045 if (dentry) {
1046 ret = nilfs_tree_is_busy(dentry);
1047 dput(dentry);
1048 }
1049 iput(inode);
1050 }
1051 nilfs_put_root(root);
1052 }
1053 return ret;
1054}
1055
1056/**
1057 * nilfs_fill_super() - initialize a super block instance
1058 * @sb: super_block
1059 * @data: mount options
1060 * @silent: silent mode flag
1061 *
1062 * This function is called exclusively by nilfs->ns_mount_mutex.
1063 * So, the recovery process is protected from other simultaneous mounts.
1064 */
1065static int
1066nilfs_fill_super(struct super_block *sb, void *data, int silent)
1067{
1068 struct the_nilfs *nilfs;
1069 struct nilfs_root *fsroot;
1070 __u64 cno;
1071 int err;
1072
1073 nilfs = alloc_nilfs(sb);
1074 if (!nilfs)
1075 return -ENOMEM;
1076
1077 sb->s_fs_info = nilfs;
1078
1079 err = init_nilfs(nilfs, sb, (char *)data);
1080 if (err)
1081 goto failed_nilfs;
1082
1083 sb->s_op = &nilfs_sops;
1084 sb->s_export_op = &nilfs_export_ops;
1085 sb->s_root = NULL;
1086 sb->s_time_gran = 1;
1087 sb->s_max_links = NILFS_LINK_MAX;
1088
1089 sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1090
1091 err = load_nilfs(nilfs, sb);
1092 if (err)
1093 goto failed_nilfs;
1094
1095 cno = nilfs_last_cno(nilfs);
1096 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1097 if (err) {
1098 nilfs_err(sb,
1099 "error %d while loading last checkpoint (checkpoint number=%llu)",
1100 err, (unsigned long long)cno);
1101 goto failed_unload;
1102 }
1103
1104 if (!sb_rdonly(sb)) {
1105 err = nilfs_attach_log_writer(sb, fsroot);
1106 if (err)
1107 goto failed_checkpoint;
1108 }
1109
1110 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1111 if (err)
1112 goto failed_segctor;
1113
1114 nilfs_put_root(fsroot);
1115
1116 if (!sb_rdonly(sb)) {
1117 down_write(&nilfs->ns_sem);
1118 nilfs_setup_super(sb, true);
1119 up_write(&nilfs->ns_sem);
1120 }
1121
1122 return 0;
1123
1124 failed_segctor:
1125 nilfs_detach_log_writer(sb);
1126
1127 failed_checkpoint:
1128 nilfs_put_root(fsroot);
1129
1130 failed_unload:
1131 nilfs_sysfs_delete_device_group(nilfs);
1132 iput(nilfs->ns_sufile);
1133 iput(nilfs->ns_cpfile);
1134 iput(nilfs->ns_dat);
1135
1136 failed_nilfs:
1137 destroy_nilfs(nilfs);
1138 return err;
1139}
1140
1141static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1142{
1143 struct the_nilfs *nilfs = sb->s_fs_info;
1144 unsigned long old_sb_flags;
1145 unsigned long old_mount_opt;
1146 int err;
1147
1148 sync_filesystem(sb);
1149 old_sb_flags = sb->s_flags;
1150 old_mount_opt = nilfs->ns_mount_opt;
1151
1152 if (!parse_options(data, sb, 1)) {
1153 err = -EINVAL;
1154 goto restore_opts;
1155 }
1156 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1157
1158 err = -EINVAL;
1159
1160 if (!nilfs_valid_fs(nilfs)) {
1161 nilfs_warn(sb,
1162 "couldn't remount because the filesystem is in an incomplete recovery state");
1163 goto restore_opts;
1164 }
1165
1166 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1167 goto out;
1168 if (*flags & SB_RDONLY) {
1169 sb->s_flags |= SB_RDONLY;
1170
1171 /*
1172 * Remounting a valid RW partition RDONLY, so set
1173 * the RDONLY flag and then mark the partition as valid again.
1174 */
1175 down_write(&nilfs->ns_sem);
1176 nilfs_cleanup_super(sb);
1177 up_write(&nilfs->ns_sem);
1178 } else {
1179 __u64 features;
1180 struct nilfs_root *root;
1181
1182 /*
1183 * Mounting a RDONLY partition read-write, so reread and
1184 * store the current valid flag. (It may have been changed
1185 * by fsck since we originally mounted the partition.)
1186 */
1187 down_read(&nilfs->ns_sem);
1188 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1189 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1190 up_read(&nilfs->ns_sem);
1191 if (features) {
1192 nilfs_warn(sb,
1193 "couldn't remount RDWR because of unsupported optional features (%llx)",
1194 (unsigned long long)features);
1195 err = -EROFS;
1196 goto restore_opts;
1197 }
1198
1199 sb->s_flags &= ~SB_RDONLY;
1200
1201 root = NILFS_I(d_inode(sb->s_root))->i_root;
1202 err = nilfs_attach_log_writer(sb, root);
1203 if (err)
1204 goto restore_opts;
1205
1206 down_write(&nilfs->ns_sem);
1207 nilfs_setup_super(sb, true);
1208 up_write(&nilfs->ns_sem);
1209 }
1210 out:
1211 return 0;
1212
1213 restore_opts:
1214 sb->s_flags = old_sb_flags;
1215 nilfs->ns_mount_opt = old_mount_opt;
1216 return err;
1217}
1218
1219struct nilfs_super_data {
1220 __u64 cno;
1221 int flags;
1222};
1223
1224static int nilfs_parse_snapshot_option(const char *option,
1225 const substring_t *arg,
1226 struct nilfs_super_data *sd)
1227{
1228 unsigned long long val;
1229 const char *msg = NULL;
1230 int err;
1231
1232 if (!(sd->flags & SB_RDONLY)) {
1233 msg = "read-only option is not specified";
1234 goto parse_error;
1235 }
1236
1237 err = kstrtoull(arg->from, 0, &val);
1238 if (err) {
1239 if (err == -ERANGE)
1240 msg = "too large checkpoint number";
1241 else
1242 msg = "malformed argument";
1243 goto parse_error;
1244 } else if (val == 0) {
1245 msg = "invalid checkpoint number 0";
1246 goto parse_error;
1247 }
1248 sd->cno = val;
1249 return 0;
1250
1251parse_error:
1252 nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1253 return 1;
1254}
1255
1256/**
1257 * nilfs_identify - pre-read mount options needed to identify mount instance
1258 * @data: mount options
1259 * @sd: nilfs_super_data
1260 */
1261static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1262{
1263 char *p, *options = data;
1264 substring_t args[MAX_OPT_ARGS];
1265 int token;
1266 int ret = 0;
1267
1268 do {
1269 p = strsep(&options, ",");
1270 if (p != NULL && *p) {
1271 token = match_token(p, tokens, args);
1272 if (token == Opt_snapshot)
1273 ret = nilfs_parse_snapshot_option(p, &args[0],
1274 sd);
1275 }
1276 if (!options)
1277 break;
1278 BUG_ON(options == data);
1279 *(options - 1) = ',';
1280 } while (!ret);
1281 return ret;
1282}
1283
1284static int nilfs_set_bdev_super(struct super_block *s, void *data)
1285{
1286 s->s_dev = *(dev_t *)data;
1287 return 0;
1288}
1289
1290static int nilfs_test_bdev_super(struct super_block *s, void *data)
1291{
1292 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1293}
1294
1295static struct dentry *
1296nilfs_mount(struct file_system_type *fs_type, int flags,
1297 const char *dev_name, void *data)
1298{
1299 struct nilfs_super_data sd = { .flags = flags };
1300 struct super_block *s;
1301 dev_t dev;
1302 int err;
1303
1304 if (nilfs_identify(data, &sd))
1305 return ERR_PTR(-EINVAL);
1306
1307 err = lookup_bdev(dev_name, &dev);
1308 if (err)
1309 return ERR_PTR(err);
1310
1311 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1312 &dev);
1313 if (IS_ERR(s))
1314 return ERR_CAST(s);
1315
1316 if (!s->s_root) {
1317 err = setup_bdev_super(s, flags, NULL);
1318 if (!err)
1319 err = nilfs_fill_super(s, data,
1320 flags & SB_SILENT ? 1 : 0);
1321 if (err)
1322 goto failed_super;
1323
1324 s->s_flags |= SB_ACTIVE;
1325 } else if (!sd.cno) {
1326 if (nilfs_tree_is_busy(s->s_root)) {
1327 if ((flags ^ s->s_flags) & SB_RDONLY) {
1328 nilfs_err(s,
1329 "the device already has a %s mount.",
1330 sb_rdonly(s) ? "read-only" : "read/write");
1331 err = -EBUSY;
1332 goto failed_super;
1333 }
1334 } else {
1335 /*
1336 * Try remount to setup mount states if the current
1337 * tree is not mounted and only snapshots use this sb.
1338 */
1339 err = nilfs_remount(s, &flags, data);
1340 if (err)
1341 goto failed_super;
1342 }
1343 }
1344
1345 if (sd.cno) {
1346 struct dentry *root_dentry;
1347
1348 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1349 if (err)
1350 goto failed_super;
1351 return root_dentry;
1352 }
1353
1354 return dget(s->s_root);
1355
1356 failed_super:
1357 deactivate_locked_super(s);
1358 return ERR_PTR(err);
1359}
1360
1361struct file_system_type nilfs_fs_type = {
1362 .owner = THIS_MODULE,
1363 .name = "nilfs2",
1364 .mount = nilfs_mount,
1365 .kill_sb = kill_block_super,
1366 .fs_flags = FS_REQUIRES_DEV,
1367};
1368MODULE_ALIAS_FS("nilfs2");
1369
1370static void nilfs_inode_init_once(void *obj)
1371{
1372 struct nilfs_inode_info *ii = obj;
1373
1374 INIT_LIST_HEAD(&ii->i_dirty);
1375#ifdef CONFIG_NILFS_XATTR
1376 init_rwsem(&ii->xattr_sem);
1377#endif
1378 inode_init_once(&ii->vfs_inode);
1379}
1380
1381static void nilfs_segbuf_init_once(void *obj)
1382{
1383 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1384}
1385
1386static void nilfs_destroy_cachep(void)
1387{
1388 /*
1389 * Make sure all delayed rcu free inodes are flushed before we
1390 * destroy cache.
1391 */
1392 rcu_barrier();
1393
1394 kmem_cache_destroy(nilfs_inode_cachep);
1395 kmem_cache_destroy(nilfs_transaction_cachep);
1396 kmem_cache_destroy(nilfs_segbuf_cachep);
1397 kmem_cache_destroy(nilfs_btree_path_cache);
1398}
1399
1400static int __init nilfs_init_cachep(void)
1401{
1402 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1403 sizeof(struct nilfs_inode_info), 0,
1404 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1405 nilfs_inode_init_once);
1406 if (!nilfs_inode_cachep)
1407 goto fail;
1408
1409 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1410 sizeof(struct nilfs_transaction_info), 0,
1411 SLAB_RECLAIM_ACCOUNT, NULL);
1412 if (!nilfs_transaction_cachep)
1413 goto fail;
1414
1415 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1416 sizeof(struct nilfs_segment_buffer), 0,
1417 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1418 if (!nilfs_segbuf_cachep)
1419 goto fail;
1420
1421 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1422 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1423 0, 0, NULL);
1424 if (!nilfs_btree_path_cache)
1425 goto fail;
1426
1427 return 0;
1428
1429fail:
1430 nilfs_destroy_cachep();
1431 return -ENOMEM;
1432}
1433
1434static int __init init_nilfs_fs(void)
1435{
1436 int err;
1437
1438 err = nilfs_init_cachep();
1439 if (err)
1440 goto fail;
1441
1442 err = nilfs_sysfs_init();
1443 if (err)
1444 goto free_cachep;
1445
1446 err = register_filesystem(&nilfs_fs_type);
1447 if (err)
1448 goto deinit_sysfs_entry;
1449
1450 printk(KERN_INFO "NILFS version 2 loaded\n");
1451 return 0;
1452
1453deinit_sysfs_entry:
1454 nilfs_sysfs_exit();
1455free_cachep:
1456 nilfs_destroy_cachep();
1457fail:
1458 return err;
1459}
1460
1461static void __exit exit_nilfs_fs(void)
1462{
1463 nilfs_destroy_cachep();
1464 nilfs_sysfs_exit();
1465 unregister_filesystem(&nilfs_fs_type);
1466}
1467
1468module_init(init_nilfs_fs)
1469module_exit(exit_nilfs_fs)