Loading...
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/crc32.h>
33#include <linux/vfs.h>
34#include <linux/writeback.h>
35#include <linux/seq_file.h>
36#include <linux/mount.h>
37#include <linux/fs_context.h>
38#include <linux/fs_parser.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);
64
65void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
66{
67 struct va_format vaf;
68 va_list args;
69 int level;
70
71 va_start(args, fmt);
72
73 level = printk_get_level(fmt);
74 vaf.fmt = printk_skip_level(fmt);
75 vaf.va = &args;
76
77 if (sb)
78 printk("%c%cNILFS (%s): %pV\n",
79 KERN_SOH_ASCII, level, sb->s_id, &vaf);
80 else
81 printk("%c%cNILFS: %pV\n",
82 KERN_SOH_ASCII, level, &vaf);
83
84 va_end(args);
85}
86
87static void nilfs_set_error(struct super_block *sb)
88{
89 struct the_nilfs *nilfs = sb->s_fs_info;
90 struct nilfs_super_block **sbp;
91
92 down_write(&nilfs->ns_sem);
93 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
94 nilfs->ns_mount_state |= NILFS_ERROR_FS;
95 sbp = nilfs_prepare_super(sb, 0);
96 if (likely(sbp)) {
97 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
98 if (sbp[1])
99 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
100 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
101 }
102 }
103 up_write(&nilfs->ns_sem);
104}
105
106/**
107 * __nilfs_error() - report failure condition on a filesystem
108 * @sb: super block instance
109 * @function: name of calling function
110 * @fmt: format string for message to be output
111 * @...: optional arguments to @fmt
112 *
113 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
114 * reporting an error message. This function should be called when
115 * NILFS detects incoherences or defects of meta data on disk.
116 *
117 * This implements the body of nilfs_error() macro. Normally,
118 * nilfs_error() should be used. As for sustainable errors such as a
119 * single-shot I/O error, nilfs_err() should be used instead.
120 *
121 * Callers should not add a trailing newline since this will do it.
122 */
123void __nilfs_error(struct super_block *sb, const char *function,
124 const char *fmt, ...)
125{
126 struct the_nilfs *nilfs = sb->s_fs_info;
127 struct va_format vaf;
128 va_list args;
129
130 va_start(args, fmt);
131
132 vaf.fmt = fmt;
133 vaf.va = &args;
134
135 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
136 sb->s_id, function, &vaf);
137
138 va_end(args);
139
140 if (!sb_rdonly(sb)) {
141 nilfs_set_error(sb);
142
143 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
144 printk(KERN_CRIT "Remounting filesystem read-only\n");
145 sb->s_flags |= SB_RDONLY;
146 }
147 }
148
149 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
150 panic("NILFS (device %s): panic forced after error\n",
151 sb->s_id);
152}
153
154struct inode *nilfs_alloc_inode(struct super_block *sb)
155{
156 struct nilfs_inode_info *ii;
157
158 ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
159 if (!ii)
160 return NULL;
161 ii->i_bh = NULL;
162 ii->i_state = 0;
163 ii->i_type = 0;
164 ii->i_cno = 0;
165 ii->i_assoc_inode = NULL;
166 ii->i_bmap = &ii->i_bmap_data;
167 return &ii->vfs_inode;
168}
169
170static void nilfs_free_inode(struct inode *inode)
171{
172 if (nilfs_is_metadata_file_inode(inode))
173 nilfs_mdt_destroy(inode);
174
175 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
176}
177
178static int nilfs_sync_super(struct super_block *sb, int flag)
179{
180 struct the_nilfs *nilfs = sb->s_fs_info;
181 int err;
182
183 retry:
184 set_buffer_dirty(nilfs->ns_sbh[0]);
185 if (nilfs_test_opt(nilfs, BARRIER)) {
186 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
187 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
188 } else {
189 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
190 }
191
192 if (unlikely(err)) {
193 nilfs_err(sb, "unable to write superblock: err=%d", err);
194 if (err == -EIO && nilfs->ns_sbh[1]) {
195 /*
196 * sbp[0] points to newer log than sbp[1],
197 * so copy sbp[0] to sbp[1] to take over sbp[0].
198 */
199 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
200 nilfs->ns_sbsize);
201 nilfs_fall_back_super_block(nilfs);
202 goto retry;
203 }
204 } else {
205 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
206
207 nilfs->ns_sbwcount++;
208
209 /*
210 * The latest segment becomes trailable from the position
211 * written in superblock.
212 */
213 clear_nilfs_discontinued(nilfs);
214
215 /* update GC protection for recent segments */
216 if (nilfs->ns_sbh[1]) {
217 if (flag == NILFS_SB_COMMIT_ALL) {
218 set_buffer_dirty(nilfs->ns_sbh[1]);
219 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
220 goto out;
221 }
222 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
223 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
224 sbp = nilfs->ns_sbp[1];
225 }
226
227 spin_lock(&nilfs->ns_last_segment_lock);
228 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
229 spin_unlock(&nilfs->ns_last_segment_lock);
230 }
231 out:
232 return err;
233}
234
235void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
236 struct the_nilfs *nilfs)
237{
238 sector_t nfreeblocks;
239
240 /* nilfs->ns_sem must be locked by the caller. */
241 nilfs_count_free_blocks(nilfs, &nfreeblocks);
242 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
243
244 spin_lock(&nilfs->ns_last_segment_lock);
245 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
246 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
247 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
248 spin_unlock(&nilfs->ns_last_segment_lock);
249}
250
251struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
252 int flip)
253{
254 struct the_nilfs *nilfs = sb->s_fs_info;
255 struct nilfs_super_block **sbp = nilfs->ns_sbp;
256
257 /* nilfs->ns_sem must be locked by the caller. */
258 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
259 if (sbp[1] &&
260 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
261 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
262 } else {
263 nilfs_crit(sb, "superblock broke");
264 return NULL;
265 }
266 } else if (sbp[1] &&
267 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
268 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
269 }
270
271 if (flip && sbp[1])
272 nilfs_swap_super_block(nilfs);
273
274 return sbp;
275}
276
277int nilfs_commit_super(struct super_block *sb, int flag)
278{
279 struct the_nilfs *nilfs = sb->s_fs_info;
280 struct nilfs_super_block **sbp = nilfs->ns_sbp;
281 time64_t t;
282
283 /* nilfs->ns_sem must be locked by the caller. */
284 t = ktime_get_real_seconds();
285 nilfs->ns_sbwtime = t;
286 sbp[0]->s_wtime = cpu_to_le64(t);
287 sbp[0]->s_sum = 0;
288 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
289 (unsigned char *)sbp[0],
290 nilfs->ns_sbsize));
291 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
292 sbp[1]->s_wtime = sbp[0]->s_wtime;
293 sbp[1]->s_sum = 0;
294 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
295 (unsigned char *)sbp[1],
296 nilfs->ns_sbsize));
297 }
298 clear_nilfs_sb_dirty(nilfs);
299 nilfs->ns_flushed_device = 1;
300 /* make sure store to ns_flushed_device cannot be reordered */
301 smp_wmb();
302 return nilfs_sync_super(sb, flag);
303}
304
305/**
306 * nilfs_cleanup_super() - write filesystem state for cleanup
307 * @sb: super block instance to be unmounted or degraded to read-only
308 *
309 * This function restores state flags in the on-disk super block.
310 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
311 * filesystem was not clean previously.
312 */
313int nilfs_cleanup_super(struct super_block *sb)
314{
315 struct the_nilfs *nilfs = sb->s_fs_info;
316 struct nilfs_super_block **sbp;
317 int flag = NILFS_SB_COMMIT;
318 int ret = -EIO;
319
320 sbp = nilfs_prepare_super(sb, 0);
321 if (sbp) {
322 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
323 nilfs_set_log_cursor(sbp[0], nilfs);
324 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
325 /*
326 * make the "clean" flag also to the opposite
327 * super block if both super blocks point to
328 * the same checkpoint.
329 */
330 sbp[1]->s_state = sbp[0]->s_state;
331 flag = NILFS_SB_COMMIT_ALL;
332 }
333 ret = nilfs_commit_super(sb, flag);
334 }
335 return ret;
336}
337
338/**
339 * nilfs_move_2nd_super - relocate secondary super block
340 * @sb: super block instance
341 * @sb2off: new offset of the secondary super block (in bytes)
342 */
343static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
344{
345 struct the_nilfs *nilfs = sb->s_fs_info;
346 struct buffer_head *nsbh;
347 struct nilfs_super_block *nsbp;
348 sector_t blocknr, newblocknr;
349 unsigned long offset;
350 int sb2i; /* array index of the secondary superblock */
351 int ret = 0;
352
353 /* nilfs->ns_sem must be locked by the caller. */
354 if (nilfs->ns_sbh[1] &&
355 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
356 sb2i = 1;
357 blocknr = nilfs->ns_sbh[1]->b_blocknr;
358 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
359 sb2i = 0;
360 blocknr = nilfs->ns_sbh[0]->b_blocknr;
361 } else {
362 sb2i = -1;
363 blocknr = 0;
364 }
365 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
366 goto out; /* super block location is unchanged */
367
368 /* Get new super block buffer */
369 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
370 offset = sb2off & (nilfs->ns_blocksize - 1);
371 nsbh = sb_getblk(sb, newblocknr);
372 if (!nsbh) {
373 nilfs_warn(sb,
374 "unable to move secondary superblock to block %llu",
375 (unsigned long long)newblocknr);
376 ret = -EIO;
377 goto out;
378 }
379 nsbp = (void *)nsbh->b_data + offset;
380
381 lock_buffer(nsbh);
382 if (sb2i >= 0) {
383 /*
384 * The position of the second superblock only changes by 4KiB,
385 * which is larger than the maximum superblock data size
386 * (= 1KiB), so there is no need to use memmove() to allow
387 * overlap between source and destination.
388 */
389 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
390
391 /*
392 * Zero fill after copy to avoid overwriting in case of move
393 * within the same block.
394 */
395 memset(nsbh->b_data, 0, offset);
396 memset((void *)nsbp + nilfs->ns_sbsize, 0,
397 nsbh->b_size - offset - nilfs->ns_sbsize);
398 } else {
399 memset(nsbh->b_data, 0, nsbh->b_size);
400 }
401 set_buffer_uptodate(nsbh);
402 unlock_buffer(nsbh);
403
404 if (sb2i >= 0) {
405 brelse(nilfs->ns_sbh[sb2i]);
406 nilfs->ns_sbh[sb2i] = nsbh;
407 nilfs->ns_sbp[sb2i] = nsbp;
408 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
409 /* secondary super block will be restored to index 1 */
410 nilfs->ns_sbh[1] = nsbh;
411 nilfs->ns_sbp[1] = nsbp;
412 } else {
413 brelse(nsbh);
414 }
415out:
416 return ret;
417}
418
419/**
420 * nilfs_resize_fs - resize the filesystem
421 * @sb: super block instance
422 * @newsize: new size of the filesystem (in bytes)
423 */
424int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
425{
426 struct the_nilfs *nilfs = sb->s_fs_info;
427 struct nilfs_super_block **sbp;
428 __u64 devsize, newnsegs;
429 loff_t sb2off;
430 int ret;
431
432 ret = -ERANGE;
433 devsize = bdev_nr_bytes(sb->s_bdev);
434 if (newsize > devsize)
435 goto out;
436
437 /*
438 * Prevent underflow in second superblock position calculation.
439 * The exact minimum size check is done in nilfs_sufile_resize().
440 */
441 if (newsize < 4096) {
442 ret = -ENOSPC;
443 goto out;
444 }
445
446 /*
447 * Write lock is required to protect some functions depending
448 * on the number of segments, the number of reserved segments,
449 * and so forth.
450 */
451 down_write(&nilfs->ns_segctor_sem);
452
453 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
454 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
455 newnsegs = div64_ul(newnsegs, nilfs->ns_blocks_per_segment);
456
457 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
458 up_write(&nilfs->ns_segctor_sem);
459 if (ret < 0)
460 goto out;
461
462 ret = nilfs_construct_segment(sb);
463 if (ret < 0)
464 goto out;
465
466 down_write(&nilfs->ns_sem);
467 nilfs_move_2nd_super(sb, sb2off);
468 ret = -EIO;
469 sbp = nilfs_prepare_super(sb, 0);
470 if (likely(sbp)) {
471 nilfs_set_log_cursor(sbp[0], nilfs);
472 /*
473 * Drop NILFS_RESIZE_FS flag for compatibility with
474 * mount-time resize which may be implemented in a
475 * future release.
476 */
477 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
478 ~NILFS_RESIZE_FS);
479 sbp[0]->s_dev_size = cpu_to_le64(newsize);
480 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
481 if (sbp[1])
482 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
483 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
484 }
485 up_write(&nilfs->ns_sem);
486
487 /*
488 * Reset the range of allocatable segments last. This order
489 * is important in the case of expansion because the secondary
490 * superblock must be protected from log write until migration
491 * completes.
492 */
493 if (!ret)
494 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
495out:
496 return ret;
497}
498
499static void nilfs_put_super(struct super_block *sb)
500{
501 struct the_nilfs *nilfs = sb->s_fs_info;
502
503 nilfs_detach_log_writer(sb);
504
505 if (!sb_rdonly(sb)) {
506 down_write(&nilfs->ns_sem);
507 nilfs_cleanup_super(sb);
508 up_write(&nilfs->ns_sem);
509 }
510
511 nilfs_sysfs_delete_device_group(nilfs);
512 iput(nilfs->ns_sufile);
513 iput(nilfs->ns_cpfile);
514 iput(nilfs->ns_dat);
515
516 destroy_nilfs(nilfs);
517 sb->s_fs_info = NULL;
518}
519
520static int nilfs_sync_fs(struct super_block *sb, int wait)
521{
522 struct the_nilfs *nilfs = sb->s_fs_info;
523 struct nilfs_super_block **sbp;
524 int err = 0;
525
526 /* This function is called when super block should be written back */
527 if (wait)
528 err = nilfs_construct_segment(sb);
529
530 down_write(&nilfs->ns_sem);
531 if (nilfs_sb_dirty(nilfs)) {
532 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
533 if (likely(sbp)) {
534 nilfs_set_log_cursor(sbp[0], nilfs);
535 nilfs_commit_super(sb, NILFS_SB_COMMIT);
536 }
537 }
538 up_write(&nilfs->ns_sem);
539
540 if (!err)
541 err = nilfs_flush_device(nilfs);
542
543 return err;
544}
545
546int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
547 struct nilfs_root **rootp)
548{
549 struct the_nilfs *nilfs = sb->s_fs_info;
550 struct nilfs_root *root;
551 int err = -ENOMEM;
552
553 root = nilfs_find_or_create_root(
554 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
555 if (!root)
556 return err;
557
558 if (root->ifile)
559 goto reuse; /* already attached checkpoint */
560
561 down_read(&nilfs->ns_segctor_sem);
562 err = nilfs_ifile_read(sb, root, cno, nilfs->ns_inode_size);
563 up_read(&nilfs->ns_segctor_sem);
564 if (unlikely(err))
565 goto failed;
566
567 reuse:
568 *rootp = root;
569 return 0;
570
571 failed:
572 if (err == -EINVAL)
573 nilfs_err(sb, "Invalid checkpoint (checkpoint number=%llu)",
574 (unsigned long long)cno);
575 nilfs_put_root(root);
576
577 return err;
578}
579
580static int nilfs_freeze(struct super_block *sb)
581{
582 struct the_nilfs *nilfs = sb->s_fs_info;
583 int err;
584
585 if (sb_rdonly(sb))
586 return 0;
587
588 /* Mark super block clean */
589 down_write(&nilfs->ns_sem);
590 err = nilfs_cleanup_super(sb);
591 up_write(&nilfs->ns_sem);
592 return err;
593}
594
595static int nilfs_unfreeze(struct super_block *sb)
596{
597 struct the_nilfs *nilfs = sb->s_fs_info;
598
599 if (sb_rdonly(sb))
600 return 0;
601
602 down_write(&nilfs->ns_sem);
603 nilfs_setup_super(sb, false);
604 up_write(&nilfs->ns_sem);
605 return 0;
606}
607
608static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
609{
610 struct super_block *sb = dentry->d_sb;
611 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
612 struct the_nilfs *nilfs = root->nilfs;
613 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
614 unsigned long long blocks;
615 unsigned long overhead;
616 unsigned long nrsvblocks;
617 sector_t nfreeblocks;
618 u64 nmaxinodes, nfreeinodes;
619 int err;
620
621 /*
622 * Compute all of the segment blocks
623 *
624 * The blocks before first segment and after last segment
625 * are excluded.
626 */
627 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
628 - nilfs->ns_first_data_block;
629 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
630
631 /*
632 * Compute the overhead
633 *
634 * When distributing meta data blocks outside segment structure,
635 * We must count them as the overhead.
636 */
637 overhead = 0;
638
639 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
640 if (unlikely(err))
641 return err;
642
643 err = nilfs_ifile_count_free_inodes(root->ifile,
644 &nmaxinodes, &nfreeinodes);
645 if (unlikely(err)) {
646 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
647 if (err == -ERANGE) {
648 /*
649 * If nilfs_palloc_count_max_entries() returns
650 * -ERANGE error code then we simply treat
651 * curent inodes count as maximum possible and
652 * zero as free inodes value.
653 */
654 nmaxinodes = atomic64_read(&root->inodes_count);
655 nfreeinodes = 0;
656 err = 0;
657 } else
658 return err;
659 }
660
661 buf->f_type = NILFS_SUPER_MAGIC;
662 buf->f_bsize = sb->s_blocksize;
663 buf->f_blocks = blocks - overhead;
664 buf->f_bfree = nfreeblocks;
665 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
666 (buf->f_bfree - nrsvblocks) : 0;
667 buf->f_files = nmaxinodes;
668 buf->f_ffree = nfreeinodes;
669 buf->f_namelen = NILFS_NAME_LEN;
670 buf->f_fsid = u64_to_fsid(id);
671
672 return 0;
673}
674
675static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
676{
677 struct super_block *sb = dentry->d_sb;
678 struct the_nilfs *nilfs = sb->s_fs_info;
679 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
680
681 if (!nilfs_test_opt(nilfs, BARRIER))
682 seq_puts(seq, ",nobarrier");
683 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
684 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
685 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
686 seq_puts(seq, ",errors=panic");
687 if (nilfs_test_opt(nilfs, ERRORS_CONT))
688 seq_puts(seq, ",errors=continue");
689 if (nilfs_test_opt(nilfs, STRICT_ORDER))
690 seq_puts(seq, ",order=strict");
691 if (nilfs_test_opt(nilfs, NORECOVERY))
692 seq_puts(seq, ",norecovery");
693 if (nilfs_test_opt(nilfs, DISCARD))
694 seq_puts(seq, ",discard");
695
696 return 0;
697}
698
699static const struct super_operations nilfs_sops = {
700 .alloc_inode = nilfs_alloc_inode,
701 .free_inode = nilfs_free_inode,
702 .dirty_inode = nilfs_dirty_inode,
703 .evict_inode = nilfs_evict_inode,
704 .put_super = nilfs_put_super,
705 .sync_fs = nilfs_sync_fs,
706 .freeze_fs = nilfs_freeze,
707 .unfreeze_fs = nilfs_unfreeze,
708 .statfs = nilfs_statfs,
709 .show_options = nilfs_show_options
710};
711
712enum {
713 Opt_err, Opt_barrier, Opt_snapshot, Opt_order, Opt_norecovery,
714 Opt_discard,
715};
716
717static const struct constant_table nilfs_param_err[] = {
718 {"continue", NILFS_MOUNT_ERRORS_CONT},
719 {"panic", NILFS_MOUNT_ERRORS_PANIC},
720 {"remount-ro", NILFS_MOUNT_ERRORS_RO},
721 {}
722};
723
724static const struct fs_parameter_spec nilfs_param_spec[] = {
725 fsparam_enum ("errors", Opt_err, nilfs_param_err),
726 fsparam_flag_no ("barrier", Opt_barrier),
727 fsparam_u64 ("cp", Opt_snapshot),
728 fsparam_string ("order", Opt_order),
729 fsparam_flag ("norecovery", Opt_norecovery),
730 fsparam_flag_no ("discard", Opt_discard),
731 {}
732};
733
734struct nilfs_fs_context {
735 unsigned long ns_mount_opt;
736 __u64 cno;
737};
738
739static int nilfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
740{
741 struct nilfs_fs_context *nilfs = fc->fs_private;
742 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
743 struct fs_parse_result result;
744 int opt;
745
746 opt = fs_parse(fc, nilfs_param_spec, param, &result);
747 if (opt < 0)
748 return opt;
749
750 switch (opt) {
751 case Opt_barrier:
752 if (result.negated)
753 nilfs_clear_opt(nilfs, BARRIER);
754 else
755 nilfs_set_opt(nilfs, BARRIER);
756 break;
757 case Opt_order:
758 if (strcmp(param->string, "relaxed") == 0)
759 /* Ordered data semantics */
760 nilfs_clear_opt(nilfs, STRICT_ORDER);
761 else if (strcmp(param->string, "strict") == 0)
762 /* Strict in-order semantics */
763 nilfs_set_opt(nilfs, STRICT_ORDER);
764 else
765 return -EINVAL;
766 break;
767 case Opt_err:
768 nilfs->ns_mount_opt &= ~NILFS_MOUNT_ERROR_MODE;
769 nilfs->ns_mount_opt |= result.uint_32;
770 break;
771 case Opt_snapshot:
772 if (is_remount) {
773 struct super_block *sb = fc->root->d_sb;
774
775 nilfs_err(sb,
776 "\"%s\" option is invalid for remount",
777 param->key);
778 return -EINVAL;
779 }
780 if (result.uint_64 == 0) {
781 nilfs_err(NULL,
782 "invalid option \"cp=0\": invalid checkpoint number 0");
783 return -EINVAL;
784 }
785 nilfs->cno = result.uint_64;
786 break;
787 case Opt_norecovery:
788 nilfs_set_opt(nilfs, NORECOVERY);
789 break;
790 case Opt_discard:
791 if (result.negated)
792 nilfs_clear_opt(nilfs, DISCARD);
793 else
794 nilfs_set_opt(nilfs, DISCARD);
795 break;
796 default:
797 return -EINVAL;
798 }
799
800 return 0;
801}
802
803static int nilfs_setup_super(struct super_block *sb, int is_mount)
804{
805 struct the_nilfs *nilfs = sb->s_fs_info;
806 struct nilfs_super_block **sbp;
807 int max_mnt_count;
808 int mnt_count;
809
810 /* nilfs->ns_sem must be locked by the caller. */
811 sbp = nilfs_prepare_super(sb, 0);
812 if (!sbp)
813 return -EIO;
814
815 if (!is_mount)
816 goto skip_mount_setup;
817
818 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
819 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
820
821 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
822 nilfs_warn(sb, "mounting fs with errors");
823#if 0
824 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
825 nilfs_warn(sb, "maximal mount count reached");
826#endif
827 }
828 if (!max_mnt_count)
829 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
830
831 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
832 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
833
834skip_mount_setup:
835 sbp[0]->s_state =
836 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
837 /* synchronize sbp[1] with sbp[0] */
838 if (sbp[1])
839 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
840 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
841}
842
843struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
844 u64 pos, int blocksize,
845 struct buffer_head **pbh)
846{
847 unsigned long long sb_index = pos;
848 unsigned long offset;
849
850 offset = do_div(sb_index, blocksize);
851 *pbh = sb_bread(sb, sb_index);
852 if (!*pbh)
853 return NULL;
854 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
855}
856
857int nilfs_store_magic(struct super_block *sb,
858 struct nilfs_super_block *sbp)
859{
860 struct the_nilfs *nilfs = sb->s_fs_info;
861
862 sb->s_magic = le16_to_cpu(sbp->s_magic);
863
864 /* FS independent flags */
865#ifdef NILFS_ATIME_DISABLE
866 sb->s_flags |= SB_NOATIME;
867#endif
868
869 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
870 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
871 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
872 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
873
874 return 0;
875}
876
877int nilfs_check_feature_compatibility(struct super_block *sb,
878 struct nilfs_super_block *sbp)
879{
880 __u64 features;
881
882 features = le64_to_cpu(sbp->s_feature_incompat) &
883 ~NILFS_FEATURE_INCOMPAT_SUPP;
884 if (features) {
885 nilfs_err(sb,
886 "couldn't mount because of unsupported optional features (%llx)",
887 (unsigned long long)features);
888 return -EINVAL;
889 }
890 features = le64_to_cpu(sbp->s_feature_compat_ro) &
891 ~NILFS_FEATURE_COMPAT_RO_SUPP;
892 if (!sb_rdonly(sb) && features) {
893 nilfs_err(sb,
894 "couldn't mount RDWR because of unsupported optional features (%llx)",
895 (unsigned long long)features);
896 return -EINVAL;
897 }
898 return 0;
899}
900
901static int nilfs_get_root_dentry(struct super_block *sb,
902 struct nilfs_root *root,
903 struct dentry **root_dentry)
904{
905 struct inode *inode;
906 struct dentry *dentry;
907 int ret = 0;
908
909 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
910 if (IS_ERR(inode)) {
911 ret = PTR_ERR(inode);
912 nilfs_err(sb, "error %d getting root inode", ret);
913 goto out;
914 }
915 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
916 iput(inode);
917 nilfs_err(sb, "corrupt root inode");
918 ret = -EINVAL;
919 goto out;
920 }
921
922 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
923 dentry = d_find_alias(inode);
924 if (!dentry) {
925 dentry = d_make_root(inode);
926 if (!dentry) {
927 ret = -ENOMEM;
928 goto failed_dentry;
929 }
930 } else {
931 iput(inode);
932 }
933 } else {
934 dentry = d_obtain_root(inode);
935 if (IS_ERR(dentry)) {
936 ret = PTR_ERR(dentry);
937 goto failed_dentry;
938 }
939 }
940 *root_dentry = dentry;
941 out:
942 return ret;
943
944 failed_dentry:
945 nilfs_err(sb, "error %d getting root dentry", ret);
946 goto out;
947}
948
949static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
950 struct dentry **root_dentry)
951{
952 struct the_nilfs *nilfs = s->s_fs_info;
953 struct nilfs_root *root;
954 int ret;
955
956 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
957
958 down_read(&nilfs->ns_segctor_sem);
959 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
960 up_read(&nilfs->ns_segctor_sem);
961 if (ret < 0) {
962 ret = (ret == -ENOENT) ? -EINVAL : ret;
963 goto out;
964 } else if (!ret) {
965 nilfs_err(s,
966 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
967 (unsigned long long)cno);
968 ret = -EINVAL;
969 goto out;
970 }
971
972 ret = nilfs_attach_checkpoint(s, cno, false, &root);
973 if (ret) {
974 nilfs_err(s,
975 "error %d while loading snapshot (checkpoint number=%llu)",
976 ret, (unsigned long long)cno);
977 goto out;
978 }
979 ret = nilfs_get_root_dentry(s, root, root_dentry);
980 nilfs_put_root(root);
981 out:
982 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
983 return ret;
984}
985
986/**
987 * nilfs_tree_is_busy() - 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 bool nilfs_tree_is_busy(struct dentry *root_dentry)
993{
994 shrink_dcache_parent(root_dentry);
995 return d_count(root_dentry) > 1;
996}
997
998int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
999{
1000 struct the_nilfs *nilfs = sb->s_fs_info;
1001 struct nilfs_root *root;
1002 struct inode *inode;
1003 struct dentry *dentry;
1004 int ret;
1005
1006 if (cno > nilfs->ns_cno)
1007 return false;
1008
1009 if (cno >= nilfs_last_cno(nilfs))
1010 return true; /* protect recent checkpoints */
1011
1012 ret = false;
1013 root = nilfs_lookup_root(nilfs, cno);
1014 if (root) {
1015 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1016 if (inode) {
1017 dentry = d_find_alias(inode);
1018 if (dentry) {
1019 ret = nilfs_tree_is_busy(dentry);
1020 dput(dentry);
1021 }
1022 iput(inode);
1023 }
1024 nilfs_put_root(root);
1025 }
1026 return ret;
1027}
1028
1029/**
1030 * nilfs_fill_super() - initialize a super block instance
1031 * @sb: super_block
1032 * @fc: filesystem context
1033 *
1034 * This function is called exclusively by nilfs->ns_mount_mutex.
1035 * So, the recovery process is protected from other simultaneous mounts.
1036 */
1037static int
1038nilfs_fill_super(struct super_block *sb, struct fs_context *fc)
1039{
1040 struct the_nilfs *nilfs;
1041 struct nilfs_root *fsroot;
1042 struct nilfs_fs_context *ctx = fc->fs_private;
1043 __u64 cno;
1044 int err;
1045
1046 nilfs = alloc_nilfs(sb);
1047 if (!nilfs)
1048 return -ENOMEM;
1049
1050 sb->s_fs_info = nilfs;
1051
1052 err = init_nilfs(nilfs, sb);
1053 if (err)
1054 goto failed_nilfs;
1055
1056 /* Copy in parsed mount options */
1057 nilfs->ns_mount_opt = ctx->ns_mount_opt;
1058
1059 sb->s_op = &nilfs_sops;
1060 sb->s_export_op = &nilfs_export_ops;
1061 sb->s_root = NULL;
1062 sb->s_time_gran = 1;
1063 sb->s_max_links = NILFS_LINK_MAX;
1064
1065 sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1066
1067 err = load_nilfs(nilfs, sb);
1068 if (err)
1069 goto failed_nilfs;
1070
1071 super_set_uuid(sb, nilfs->ns_sbp[0]->s_uuid,
1072 sizeof(nilfs->ns_sbp[0]->s_uuid));
1073 super_set_sysfs_name_bdev(sb);
1074
1075 cno = nilfs_last_cno(nilfs);
1076 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1077 if (err) {
1078 nilfs_err(sb,
1079 "error %d while loading last checkpoint (checkpoint number=%llu)",
1080 err, (unsigned long long)cno);
1081 goto failed_unload;
1082 }
1083
1084 if (!sb_rdonly(sb)) {
1085 err = nilfs_attach_log_writer(sb, fsroot);
1086 if (err)
1087 goto failed_checkpoint;
1088 }
1089
1090 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1091 if (err)
1092 goto failed_segctor;
1093
1094 nilfs_put_root(fsroot);
1095
1096 if (!sb_rdonly(sb)) {
1097 down_write(&nilfs->ns_sem);
1098 nilfs_setup_super(sb, true);
1099 up_write(&nilfs->ns_sem);
1100 }
1101
1102 return 0;
1103
1104 failed_segctor:
1105 nilfs_detach_log_writer(sb);
1106
1107 failed_checkpoint:
1108 nilfs_put_root(fsroot);
1109
1110 failed_unload:
1111 nilfs_sysfs_delete_device_group(nilfs);
1112 iput(nilfs->ns_sufile);
1113 iput(nilfs->ns_cpfile);
1114 iput(nilfs->ns_dat);
1115
1116 failed_nilfs:
1117 destroy_nilfs(nilfs);
1118 return err;
1119}
1120
1121static int nilfs_reconfigure(struct fs_context *fc)
1122{
1123 struct nilfs_fs_context *ctx = fc->fs_private;
1124 struct super_block *sb = fc->root->d_sb;
1125 struct the_nilfs *nilfs = sb->s_fs_info;
1126 int err;
1127
1128 sync_filesystem(sb);
1129
1130 err = -EINVAL;
1131
1132 if (!nilfs_valid_fs(nilfs)) {
1133 nilfs_warn(sb,
1134 "couldn't remount because the filesystem is in an incomplete recovery state");
1135 goto ignore_opts;
1136 }
1137 if ((bool)(fc->sb_flags & SB_RDONLY) == sb_rdonly(sb))
1138 goto out;
1139 if (fc->sb_flags & SB_RDONLY) {
1140 sb->s_flags |= SB_RDONLY;
1141
1142 /*
1143 * Remounting a valid RW partition RDONLY, so set
1144 * the RDONLY flag and then mark the partition as valid again.
1145 */
1146 down_write(&nilfs->ns_sem);
1147 nilfs_cleanup_super(sb);
1148 up_write(&nilfs->ns_sem);
1149 } else {
1150 __u64 features;
1151 struct nilfs_root *root;
1152
1153 /*
1154 * Mounting a RDONLY partition read-write, so reread and
1155 * store the current valid flag. (It may have been changed
1156 * by fsck since we originally mounted the partition.)
1157 */
1158 down_read(&nilfs->ns_sem);
1159 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1160 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1161 up_read(&nilfs->ns_sem);
1162 if (features) {
1163 nilfs_warn(sb,
1164 "couldn't remount RDWR because of unsupported optional features (%llx)",
1165 (unsigned long long)features);
1166 err = -EROFS;
1167 goto ignore_opts;
1168 }
1169
1170 sb->s_flags &= ~SB_RDONLY;
1171
1172 root = NILFS_I(d_inode(sb->s_root))->i_root;
1173 err = nilfs_attach_log_writer(sb, root);
1174 if (err) {
1175 sb->s_flags |= SB_RDONLY;
1176 goto ignore_opts;
1177 }
1178
1179 down_write(&nilfs->ns_sem);
1180 nilfs_setup_super(sb, true);
1181 up_write(&nilfs->ns_sem);
1182 }
1183 out:
1184 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1185 /* Copy over parsed remount options */
1186 nilfs->ns_mount_opt = ctx->ns_mount_opt;
1187
1188 return 0;
1189
1190 ignore_opts:
1191 return err;
1192}
1193
1194static int
1195nilfs_get_tree(struct fs_context *fc)
1196{
1197 struct nilfs_fs_context *ctx = fc->fs_private;
1198 struct super_block *s;
1199 dev_t dev;
1200 int err;
1201
1202 if (ctx->cno && !(fc->sb_flags & SB_RDONLY)) {
1203 nilfs_err(NULL,
1204 "invalid option \"cp=%llu\": read-only option is not specified",
1205 ctx->cno);
1206 return -EINVAL;
1207 }
1208
1209 err = lookup_bdev(fc->source, &dev);
1210 if (err)
1211 return err;
1212
1213 s = sget_dev(fc, dev);
1214 if (IS_ERR(s))
1215 return PTR_ERR(s);
1216
1217 if (!s->s_root) {
1218 err = setup_bdev_super(s, fc->sb_flags, fc);
1219 if (!err)
1220 err = nilfs_fill_super(s, fc);
1221 if (err)
1222 goto failed_super;
1223
1224 s->s_flags |= SB_ACTIVE;
1225 } else if (!ctx->cno) {
1226 if (nilfs_tree_is_busy(s->s_root)) {
1227 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1228 nilfs_err(s,
1229 "the device already has a %s mount.",
1230 sb_rdonly(s) ? "read-only" : "read/write");
1231 err = -EBUSY;
1232 goto failed_super;
1233 }
1234 } else {
1235 /*
1236 * Try reconfigure to setup mount states if the current
1237 * tree is not mounted and only snapshots use this sb.
1238 *
1239 * Since nilfs_reconfigure() requires fc->root to be
1240 * set, set it first and release it on failure.
1241 */
1242 fc->root = dget(s->s_root);
1243 err = nilfs_reconfigure(fc);
1244 if (err) {
1245 dput(fc->root);
1246 fc->root = NULL; /* prevent double release */
1247 goto failed_super;
1248 }
1249 return 0;
1250 }
1251 }
1252
1253 if (ctx->cno) {
1254 struct dentry *root_dentry;
1255
1256 err = nilfs_attach_snapshot(s, ctx->cno, &root_dentry);
1257 if (err)
1258 goto failed_super;
1259 fc->root = root_dentry;
1260 return 0;
1261 }
1262
1263 fc->root = dget(s->s_root);
1264 return 0;
1265
1266 failed_super:
1267 deactivate_locked_super(s);
1268 return err;
1269}
1270
1271static void nilfs_free_fc(struct fs_context *fc)
1272{
1273 kfree(fc->fs_private);
1274}
1275
1276static const struct fs_context_operations nilfs_context_ops = {
1277 .parse_param = nilfs_parse_param,
1278 .get_tree = nilfs_get_tree,
1279 .reconfigure = nilfs_reconfigure,
1280 .free = nilfs_free_fc,
1281};
1282
1283static int nilfs_init_fs_context(struct fs_context *fc)
1284{
1285 struct nilfs_fs_context *ctx;
1286
1287 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1288 if (!ctx)
1289 return -ENOMEM;
1290
1291 ctx->ns_mount_opt = NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
1292 fc->fs_private = ctx;
1293 fc->ops = &nilfs_context_ops;
1294
1295 return 0;
1296}
1297
1298struct file_system_type nilfs_fs_type = {
1299 .owner = THIS_MODULE,
1300 .name = "nilfs2",
1301 .kill_sb = kill_block_super,
1302 .fs_flags = FS_REQUIRES_DEV,
1303 .init_fs_context = nilfs_init_fs_context,
1304 .parameters = nilfs_param_spec,
1305};
1306MODULE_ALIAS_FS("nilfs2");
1307
1308static void nilfs_inode_init_once(void *obj)
1309{
1310 struct nilfs_inode_info *ii = obj;
1311
1312 INIT_LIST_HEAD(&ii->i_dirty);
1313#ifdef CONFIG_NILFS_XATTR
1314 init_rwsem(&ii->xattr_sem);
1315#endif
1316 inode_init_once(&ii->vfs_inode);
1317}
1318
1319static void nilfs_segbuf_init_once(void *obj)
1320{
1321 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1322}
1323
1324static void nilfs_destroy_cachep(void)
1325{
1326 /*
1327 * Make sure all delayed rcu free inodes are flushed before we
1328 * destroy cache.
1329 */
1330 rcu_barrier();
1331
1332 kmem_cache_destroy(nilfs_inode_cachep);
1333 kmem_cache_destroy(nilfs_transaction_cachep);
1334 kmem_cache_destroy(nilfs_segbuf_cachep);
1335 kmem_cache_destroy(nilfs_btree_path_cache);
1336}
1337
1338static int __init nilfs_init_cachep(void)
1339{
1340 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1341 sizeof(struct nilfs_inode_info), 0,
1342 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1343 nilfs_inode_init_once);
1344 if (!nilfs_inode_cachep)
1345 goto fail;
1346
1347 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1348 sizeof(struct nilfs_transaction_info), 0,
1349 SLAB_RECLAIM_ACCOUNT, NULL);
1350 if (!nilfs_transaction_cachep)
1351 goto fail;
1352
1353 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1354 sizeof(struct nilfs_segment_buffer), 0,
1355 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1356 if (!nilfs_segbuf_cachep)
1357 goto fail;
1358
1359 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1360 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1361 0, 0, NULL);
1362 if (!nilfs_btree_path_cache)
1363 goto fail;
1364
1365 return 0;
1366
1367fail:
1368 nilfs_destroy_cachep();
1369 return -ENOMEM;
1370}
1371
1372static int __init init_nilfs_fs(void)
1373{
1374 int err;
1375
1376 err = nilfs_init_cachep();
1377 if (err)
1378 goto fail;
1379
1380 err = nilfs_sysfs_init();
1381 if (err)
1382 goto free_cachep;
1383
1384 err = register_filesystem(&nilfs_fs_type);
1385 if (err)
1386 goto deinit_sysfs_entry;
1387
1388 printk(KERN_INFO "NILFS version 2 loaded\n");
1389 return 0;
1390
1391deinit_sysfs_entry:
1392 nilfs_sysfs_exit();
1393free_cachep:
1394 nilfs_destroy_cachep();
1395fail:
1396 return err;
1397}
1398
1399static void __exit exit_nilfs_fs(void)
1400{
1401 nilfs_destroy_cachep();
1402 nilfs_sysfs_exit();
1403 unregister_filesystem(&nilfs_fs_type);
1404}
1405
1406module_init(init_nilfs_fs)
1407module_exit(exit_nilfs_fs)
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * super.c - 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 "nilfs.h"
39#include "export.h"
40#include "mdt.h"
41#include "alloc.h"
42#include "btree.h"
43#include "btnode.h"
44#include "page.h"
45#include "cpfile.h"
46#include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
47#include "ifile.h"
48#include "dat.h"
49#include "segment.h"
50#include "segbuf.h"
51
52MODULE_AUTHOR("NTT Corp.");
53MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
54 "(NILFS)");
55MODULE_LICENSE("GPL");
56
57static struct kmem_cache *nilfs_inode_cachep;
58struct kmem_cache *nilfs_transaction_cachep;
59struct kmem_cache *nilfs_segbuf_cachep;
60struct kmem_cache *nilfs_btree_path_cache;
61
62static int nilfs_setup_super(struct super_block *sb, int is_mount);
63static int nilfs_remount(struct super_block *sb, int *flags, char *data);
64
65void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
66{
67 struct va_format vaf;
68 va_list args;
69 int level;
70
71 va_start(args, fmt);
72
73 level = printk_get_level(fmt);
74 vaf.fmt = printk_skip_level(fmt);
75 vaf.va = &args;
76
77 if (sb)
78 printk("%c%cNILFS (%s): %pV\n",
79 KERN_SOH_ASCII, level, sb->s_id, &vaf);
80 else
81 printk("%c%cNILFS: %pV\n",
82 KERN_SOH_ASCII, level, &vaf);
83
84 va_end(args);
85}
86
87static void nilfs_set_error(struct super_block *sb)
88{
89 struct the_nilfs *nilfs = sb->s_fs_info;
90 struct nilfs_super_block **sbp;
91
92 down_write(&nilfs->ns_sem);
93 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
94 nilfs->ns_mount_state |= NILFS_ERROR_FS;
95 sbp = nilfs_prepare_super(sb, 0);
96 if (likely(sbp)) {
97 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
98 if (sbp[1])
99 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
100 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
101 }
102 }
103 up_write(&nilfs->ns_sem);
104}
105
106/**
107 * __nilfs_error() - report failure condition on a filesystem
108 *
109 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
110 * reporting an error message. This function should be called when
111 * NILFS detects incoherences or defects of meta data on disk.
112 *
113 * This implements the body of nilfs_error() macro. Normally,
114 * nilfs_error() should be used. As for sustainable errors such as a
115 * single-shot I/O error, nilfs_err() should be used instead.
116 *
117 * Callers should not add a trailing newline since this will do it.
118 */
119void __nilfs_error(struct super_block *sb, const char *function,
120 const char *fmt, ...)
121{
122 struct the_nilfs *nilfs = sb->s_fs_info;
123 struct va_format vaf;
124 va_list args;
125
126 va_start(args, fmt);
127
128 vaf.fmt = fmt;
129 vaf.va = &args;
130
131 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
132 sb->s_id, function, &vaf);
133
134 va_end(args);
135
136 if (!sb_rdonly(sb)) {
137 nilfs_set_error(sb);
138
139 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
140 printk(KERN_CRIT "Remounting filesystem read-only\n");
141 sb->s_flags |= SB_RDONLY;
142 }
143 }
144
145 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
146 panic("NILFS (device %s): panic forced after error\n",
147 sb->s_id);
148}
149
150struct inode *nilfs_alloc_inode(struct super_block *sb)
151{
152 struct nilfs_inode_info *ii;
153
154 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
155 if (!ii)
156 return NULL;
157 ii->i_bh = NULL;
158 ii->i_state = 0;
159 ii->i_cno = 0;
160 nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
161 return &ii->vfs_inode;
162}
163
164static void nilfs_free_inode(struct inode *inode)
165{
166 if (nilfs_is_metadata_file_inode(inode))
167 nilfs_mdt_destroy(inode);
168
169 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
170}
171
172static int nilfs_sync_super(struct super_block *sb, int flag)
173{
174 struct the_nilfs *nilfs = sb->s_fs_info;
175 int err;
176
177 retry:
178 set_buffer_dirty(nilfs->ns_sbh[0]);
179 if (nilfs_test_opt(nilfs, BARRIER)) {
180 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
181 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
182 } else {
183 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
184 }
185
186 if (unlikely(err)) {
187 nilfs_err(sb, "unable to write superblock: err=%d", err);
188 if (err == -EIO && nilfs->ns_sbh[1]) {
189 /*
190 * sbp[0] points to newer log than sbp[1],
191 * so copy sbp[0] to sbp[1] to take over sbp[0].
192 */
193 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
194 nilfs->ns_sbsize);
195 nilfs_fall_back_super_block(nilfs);
196 goto retry;
197 }
198 } else {
199 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
200
201 nilfs->ns_sbwcount++;
202
203 /*
204 * The latest segment becomes trailable from the position
205 * written in superblock.
206 */
207 clear_nilfs_discontinued(nilfs);
208
209 /* update GC protection for recent segments */
210 if (nilfs->ns_sbh[1]) {
211 if (flag == NILFS_SB_COMMIT_ALL) {
212 set_buffer_dirty(nilfs->ns_sbh[1]);
213 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
214 goto out;
215 }
216 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
217 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
218 sbp = nilfs->ns_sbp[1];
219 }
220
221 spin_lock(&nilfs->ns_last_segment_lock);
222 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
223 spin_unlock(&nilfs->ns_last_segment_lock);
224 }
225 out:
226 return err;
227}
228
229void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
230 struct the_nilfs *nilfs)
231{
232 sector_t nfreeblocks;
233
234 /* nilfs->ns_sem must be locked by the caller. */
235 nilfs_count_free_blocks(nilfs, &nfreeblocks);
236 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
237
238 spin_lock(&nilfs->ns_last_segment_lock);
239 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
240 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
241 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
242 spin_unlock(&nilfs->ns_last_segment_lock);
243}
244
245struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
246 int flip)
247{
248 struct the_nilfs *nilfs = sb->s_fs_info;
249 struct nilfs_super_block **sbp = nilfs->ns_sbp;
250
251 /* nilfs->ns_sem must be locked by the caller. */
252 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
253 if (sbp[1] &&
254 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
255 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
256 } else {
257 nilfs_crit(sb, "superblock broke");
258 return NULL;
259 }
260 } else if (sbp[1] &&
261 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
262 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
263 }
264
265 if (flip && sbp[1])
266 nilfs_swap_super_block(nilfs);
267
268 return sbp;
269}
270
271int nilfs_commit_super(struct super_block *sb, int flag)
272{
273 struct the_nilfs *nilfs = sb->s_fs_info;
274 struct nilfs_super_block **sbp = nilfs->ns_sbp;
275 time64_t t;
276
277 /* nilfs->ns_sem must be locked by the caller. */
278 t = ktime_get_real_seconds();
279 nilfs->ns_sbwtime = t;
280 sbp[0]->s_wtime = cpu_to_le64(t);
281 sbp[0]->s_sum = 0;
282 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
283 (unsigned char *)sbp[0],
284 nilfs->ns_sbsize));
285 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
286 sbp[1]->s_wtime = sbp[0]->s_wtime;
287 sbp[1]->s_sum = 0;
288 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
289 (unsigned char *)sbp[1],
290 nilfs->ns_sbsize));
291 }
292 clear_nilfs_sb_dirty(nilfs);
293 nilfs->ns_flushed_device = 1;
294 /* make sure store to ns_flushed_device cannot be reordered */
295 smp_wmb();
296 return nilfs_sync_super(sb, flag);
297}
298
299/**
300 * nilfs_cleanup_super() - write filesystem state for cleanup
301 * @sb: super block instance to be unmounted or degraded to read-only
302 *
303 * This function restores state flags in the on-disk super block.
304 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
305 * filesystem was not clean previously.
306 */
307int nilfs_cleanup_super(struct super_block *sb)
308{
309 struct the_nilfs *nilfs = sb->s_fs_info;
310 struct nilfs_super_block **sbp;
311 int flag = NILFS_SB_COMMIT;
312 int ret = -EIO;
313
314 sbp = nilfs_prepare_super(sb, 0);
315 if (sbp) {
316 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
317 nilfs_set_log_cursor(sbp[0], nilfs);
318 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
319 /*
320 * make the "clean" flag also to the opposite
321 * super block if both super blocks point to
322 * the same checkpoint.
323 */
324 sbp[1]->s_state = sbp[0]->s_state;
325 flag = NILFS_SB_COMMIT_ALL;
326 }
327 ret = nilfs_commit_super(sb, flag);
328 }
329 return ret;
330}
331
332/**
333 * nilfs_move_2nd_super - relocate secondary super block
334 * @sb: super block instance
335 * @sb2off: new offset of the secondary super block (in bytes)
336 */
337static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
338{
339 struct the_nilfs *nilfs = sb->s_fs_info;
340 struct buffer_head *nsbh;
341 struct nilfs_super_block *nsbp;
342 sector_t blocknr, newblocknr;
343 unsigned long offset;
344 int sb2i; /* array index of the secondary superblock */
345 int ret = 0;
346
347 /* nilfs->ns_sem must be locked by the caller. */
348 if (nilfs->ns_sbh[1] &&
349 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
350 sb2i = 1;
351 blocknr = nilfs->ns_sbh[1]->b_blocknr;
352 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
353 sb2i = 0;
354 blocknr = nilfs->ns_sbh[0]->b_blocknr;
355 } else {
356 sb2i = -1;
357 blocknr = 0;
358 }
359 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
360 goto out; /* super block location is unchanged */
361
362 /* Get new super block buffer */
363 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
364 offset = sb2off & (nilfs->ns_blocksize - 1);
365 nsbh = sb_getblk(sb, newblocknr);
366 if (!nsbh) {
367 nilfs_warn(sb,
368 "unable to move secondary superblock to block %llu",
369 (unsigned long long)newblocknr);
370 ret = -EIO;
371 goto out;
372 }
373 nsbp = (void *)nsbh->b_data + offset;
374 memset(nsbp, 0, nilfs->ns_blocksize);
375
376 if (sb2i >= 0) {
377 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
378 brelse(nilfs->ns_sbh[sb2i]);
379 nilfs->ns_sbh[sb2i] = nsbh;
380 nilfs->ns_sbp[sb2i] = nsbp;
381 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
382 /* secondary super block will be restored to index 1 */
383 nilfs->ns_sbh[1] = nsbh;
384 nilfs->ns_sbp[1] = nsbp;
385 } else {
386 brelse(nsbh);
387 }
388out:
389 return ret;
390}
391
392/**
393 * nilfs_resize_fs - resize the filesystem
394 * @sb: super block instance
395 * @newsize: new size of the filesystem (in bytes)
396 */
397int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
398{
399 struct the_nilfs *nilfs = sb->s_fs_info;
400 struct nilfs_super_block **sbp;
401 __u64 devsize, newnsegs;
402 loff_t sb2off;
403 int ret;
404
405 ret = -ERANGE;
406 devsize = i_size_read(sb->s_bdev->bd_inode);
407 if (newsize > devsize)
408 goto out;
409
410 /*
411 * Write lock is required to protect some functions depending
412 * on the number of segments, the number of reserved segments,
413 * and so forth.
414 */
415 down_write(&nilfs->ns_segctor_sem);
416
417 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
418 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
419 do_div(newnsegs, nilfs->ns_blocks_per_segment);
420
421 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
422 up_write(&nilfs->ns_segctor_sem);
423 if (ret < 0)
424 goto out;
425
426 ret = nilfs_construct_segment(sb);
427 if (ret < 0)
428 goto out;
429
430 down_write(&nilfs->ns_sem);
431 nilfs_move_2nd_super(sb, sb2off);
432 ret = -EIO;
433 sbp = nilfs_prepare_super(sb, 0);
434 if (likely(sbp)) {
435 nilfs_set_log_cursor(sbp[0], nilfs);
436 /*
437 * Drop NILFS_RESIZE_FS flag for compatibility with
438 * mount-time resize which may be implemented in a
439 * future release.
440 */
441 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
442 ~NILFS_RESIZE_FS);
443 sbp[0]->s_dev_size = cpu_to_le64(newsize);
444 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
445 if (sbp[1])
446 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
447 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
448 }
449 up_write(&nilfs->ns_sem);
450
451 /*
452 * Reset the range of allocatable segments last. This order
453 * is important in the case of expansion because the secondary
454 * superblock must be protected from log write until migration
455 * completes.
456 */
457 if (!ret)
458 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
459out:
460 return ret;
461}
462
463static void nilfs_put_super(struct super_block *sb)
464{
465 struct the_nilfs *nilfs = sb->s_fs_info;
466
467 nilfs_detach_log_writer(sb);
468
469 if (!sb_rdonly(sb)) {
470 down_write(&nilfs->ns_sem);
471 nilfs_cleanup_super(sb);
472 up_write(&nilfs->ns_sem);
473 }
474
475 iput(nilfs->ns_sufile);
476 iput(nilfs->ns_cpfile);
477 iput(nilfs->ns_dat);
478
479 destroy_nilfs(nilfs);
480 sb->s_fs_info = NULL;
481}
482
483static int nilfs_sync_fs(struct super_block *sb, int wait)
484{
485 struct the_nilfs *nilfs = sb->s_fs_info;
486 struct nilfs_super_block **sbp;
487 int err = 0;
488
489 /* This function is called when super block should be written back */
490 if (wait)
491 err = nilfs_construct_segment(sb);
492
493 down_write(&nilfs->ns_sem);
494 if (nilfs_sb_dirty(nilfs)) {
495 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
496 if (likely(sbp)) {
497 nilfs_set_log_cursor(sbp[0], nilfs);
498 nilfs_commit_super(sb, NILFS_SB_COMMIT);
499 }
500 }
501 up_write(&nilfs->ns_sem);
502
503 if (!err)
504 err = nilfs_flush_device(nilfs);
505
506 return err;
507}
508
509int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
510 struct nilfs_root **rootp)
511{
512 struct the_nilfs *nilfs = sb->s_fs_info;
513 struct nilfs_root *root;
514 struct nilfs_checkpoint *raw_cp;
515 struct buffer_head *bh_cp;
516 int err = -ENOMEM;
517
518 root = nilfs_find_or_create_root(
519 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
520 if (!root)
521 return err;
522
523 if (root->ifile)
524 goto reuse; /* already attached checkpoint */
525
526 down_read(&nilfs->ns_segctor_sem);
527 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
528 &bh_cp);
529 up_read(&nilfs->ns_segctor_sem);
530 if (unlikely(err)) {
531 if (err == -ENOENT || err == -EINVAL) {
532 nilfs_err(sb,
533 "Invalid checkpoint (checkpoint number=%llu)",
534 (unsigned long long)cno);
535 err = -EINVAL;
536 }
537 goto failed;
538 }
539
540 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
541 &raw_cp->cp_ifile_inode, &root->ifile);
542 if (err)
543 goto failed_bh;
544
545 atomic64_set(&root->inodes_count,
546 le64_to_cpu(raw_cp->cp_inodes_count));
547 atomic64_set(&root->blocks_count,
548 le64_to_cpu(raw_cp->cp_blocks_count));
549
550 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
551
552 reuse:
553 *rootp = root;
554 return 0;
555
556 failed_bh:
557 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
558 failed:
559 nilfs_put_root(root);
560
561 return err;
562}
563
564static int nilfs_freeze(struct super_block *sb)
565{
566 struct the_nilfs *nilfs = sb->s_fs_info;
567 int err;
568
569 if (sb_rdonly(sb))
570 return 0;
571
572 /* Mark super block clean */
573 down_write(&nilfs->ns_sem);
574 err = nilfs_cleanup_super(sb);
575 up_write(&nilfs->ns_sem);
576 return err;
577}
578
579static int nilfs_unfreeze(struct super_block *sb)
580{
581 struct the_nilfs *nilfs = sb->s_fs_info;
582
583 if (sb_rdonly(sb))
584 return 0;
585
586 down_write(&nilfs->ns_sem);
587 nilfs_setup_super(sb, false);
588 up_write(&nilfs->ns_sem);
589 return 0;
590}
591
592static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
593{
594 struct super_block *sb = dentry->d_sb;
595 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
596 struct the_nilfs *nilfs = root->nilfs;
597 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
598 unsigned long long blocks;
599 unsigned long overhead;
600 unsigned long nrsvblocks;
601 sector_t nfreeblocks;
602 u64 nmaxinodes, nfreeinodes;
603 int err;
604
605 /*
606 * Compute all of the segment blocks
607 *
608 * The blocks before first segment and after last segment
609 * are excluded.
610 */
611 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
612 - nilfs->ns_first_data_block;
613 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
614
615 /*
616 * Compute the overhead
617 *
618 * When distributing meta data blocks outside segment structure,
619 * We must count them as the overhead.
620 */
621 overhead = 0;
622
623 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
624 if (unlikely(err))
625 return err;
626
627 err = nilfs_ifile_count_free_inodes(root->ifile,
628 &nmaxinodes, &nfreeinodes);
629 if (unlikely(err)) {
630 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
631 if (err == -ERANGE) {
632 /*
633 * If nilfs_palloc_count_max_entries() returns
634 * -ERANGE error code then we simply treat
635 * curent inodes count as maximum possible and
636 * zero as free inodes value.
637 */
638 nmaxinodes = atomic64_read(&root->inodes_count);
639 nfreeinodes = 0;
640 err = 0;
641 } else
642 return err;
643 }
644
645 buf->f_type = NILFS_SUPER_MAGIC;
646 buf->f_bsize = sb->s_blocksize;
647 buf->f_blocks = blocks - overhead;
648 buf->f_bfree = nfreeblocks;
649 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
650 (buf->f_bfree - nrsvblocks) : 0;
651 buf->f_files = nmaxinodes;
652 buf->f_ffree = nfreeinodes;
653 buf->f_namelen = NILFS_NAME_LEN;
654 buf->f_fsid.val[0] = (u32)id;
655 buf->f_fsid.val[1] = (u32)(id >> 32);
656
657 return 0;
658}
659
660static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
661{
662 struct super_block *sb = dentry->d_sb;
663 struct the_nilfs *nilfs = sb->s_fs_info;
664 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
665
666 if (!nilfs_test_opt(nilfs, BARRIER))
667 seq_puts(seq, ",nobarrier");
668 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
669 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
670 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
671 seq_puts(seq, ",errors=panic");
672 if (nilfs_test_opt(nilfs, ERRORS_CONT))
673 seq_puts(seq, ",errors=continue");
674 if (nilfs_test_opt(nilfs, STRICT_ORDER))
675 seq_puts(seq, ",order=strict");
676 if (nilfs_test_opt(nilfs, NORECOVERY))
677 seq_puts(seq, ",norecovery");
678 if (nilfs_test_opt(nilfs, DISCARD))
679 seq_puts(seq, ",discard");
680
681 return 0;
682}
683
684static const struct super_operations nilfs_sops = {
685 .alloc_inode = nilfs_alloc_inode,
686 .free_inode = nilfs_free_inode,
687 .dirty_inode = nilfs_dirty_inode,
688 .evict_inode = nilfs_evict_inode,
689 .put_super = nilfs_put_super,
690 .sync_fs = nilfs_sync_fs,
691 .freeze_fs = nilfs_freeze,
692 .unfreeze_fs = nilfs_unfreeze,
693 .statfs = nilfs_statfs,
694 .remount_fs = nilfs_remount,
695 .show_options = nilfs_show_options
696};
697
698enum {
699 Opt_err_cont, Opt_err_panic, Opt_err_ro,
700 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
701 Opt_discard, Opt_nodiscard, Opt_err,
702};
703
704static match_table_t tokens = {
705 {Opt_err_cont, "errors=continue"},
706 {Opt_err_panic, "errors=panic"},
707 {Opt_err_ro, "errors=remount-ro"},
708 {Opt_barrier, "barrier"},
709 {Opt_nobarrier, "nobarrier"},
710 {Opt_snapshot, "cp=%u"},
711 {Opt_order, "order=%s"},
712 {Opt_norecovery, "norecovery"},
713 {Opt_discard, "discard"},
714 {Opt_nodiscard, "nodiscard"},
715 {Opt_err, NULL}
716};
717
718static int parse_options(char *options, struct super_block *sb, int is_remount)
719{
720 struct the_nilfs *nilfs = sb->s_fs_info;
721 char *p;
722 substring_t args[MAX_OPT_ARGS];
723
724 if (!options)
725 return 1;
726
727 while ((p = strsep(&options, ",")) != NULL) {
728 int token;
729
730 if (!*p)
731 continue;
732
733 token = match_token(p, tokens, args);
734 switch (token) {
735 case Opt_barrier:
736 nilfs_set_opt(nilfs, BARRIER);
737 break;
738 case Opt_nobarrier:
739 nilfs_clear_opt(nilfs, BARRIER);
740 break;
741 case Opt_order:
742 if (strcmp(args[0].from, "relaxed") == 0)
743 /* Ordered data semantics */
744 nilfs_clear_opt(nilfs, STRICT_ORDER);
745 else if (strcmp(args[0].from, "strict") == 0)
746 /* Strict in-order semantics */
747 nilfs_set_opt(nilfs, STRICT_ORDER);
748 else
749 return 0;
750 break;
751 case Opt_err_panic:
752 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
753 break;
754 case Opt_err_ro:
755 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
756 break;
757 case Opt_err_cont:
758 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
759 break;
760 case Opt_snapshot:
761 if (is_remount) {
762 nilfs_err(sb,
763 "\"%s\" option is invalid for remount",
764 p);
765 return 0;
766 }
767 break;
768 case Opt_norecovery:
769 nilfs_set_opt(nilfs, NORECOVERY);
770 break;
771 case Opt_discard:
772 nilfs_set_opt(nilfs, DISCARD);
773 break;
774 case Opt_nodiscard:
775 nilfs_clear_opt(nilfs, DISCARD);
776 break;
777 default:
778 nilfs_err(sb, "unrecognized mount option \"%s\"", p);
779 return 0;
780 }
781 }
782 return 1;
783}
784
785static inline void
786nilfs_set_default_options(struct super_block *sb,
787 struct nilfs_super_block *sbp)
788{
789 struct the_nilfs *nilfs = sb->s_fs_info;
790
791 nilfs->ns_mount_opt =
792 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
793}
794
795static int nilfs_setup_super(struct super_block *sb, int is_mount)
796{
797 struct the_nilfs *nilfs = sb->s_fs_info;
798 struct nilfs_super_block **sbp;
799 int max_mnt_count;
800 int mnt_count;
801
802 /* nilfs->ns_sem must be locked by the caller. */
803 sbp = nilfs_prepare_super(sb, 0);
804 if (!sbp)
805 return -EIO;
806
807 if (!is_mount)
808 goto skip_mount_setup;
809
810 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
811 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
812
813 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
814 nilfs_warn(sb, "mounting fs with errors");
815#if 0
816 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
817 nilfs_warn(sb, "maximal mount count reached");
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(ktime_get_real_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 |= SB_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 nilfs_err(sb,
881 "couldn't mount because of unsupported optional features (%llx)",
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_rdonly(sb) && features) {
888 nilfs_err(sb,
889 "couldn't mount RDWR because of unsupported optional features (%llx)",
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 ret = PTR_ERR(inode);
907 nilfs_err(sb, "error %d getting root inode", ret);
908 goto out;
909 }
910 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
911 iput(inode);
912 nilfs_err(sb, "corrupt root inode");
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_root(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 nilfs_err(sb, "error %d getting root dentry", ret);
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 nilfs_err(s,
961 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
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 nilfs_err(s,
970 "error %d while loading snapshot (checkpoint number=%llu)",
971 ret, (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
981/**
982 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
983 * @root_dentry: root dentry of the tree to be shrunk
984 *
985 * This function returns true if the tree was in-use.
986 */
987static bool nilfs_tree_is_busy(struct dentry *root_dentry)
988{
989 shrink_dcache_parent(root_dentry);
990 return d_count(root_dentry) > 1;
991}
992
993int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
994{
995 struct the_nilfs *nilfs = sb->s_fs_info;
996 struct nilfs_root *root;
997 struct inode *inode;
998 struct dentry *dentry;
999 int ret;
1000
1001 if (cno > nilfs->ns_cno)
1002 return false;
1003
1004 if (cno >= nilfs_last_cno(nilfs))
1005 return true; /* protect recent checkpoints */
1006
1007 ret = false;
1008 root = nilfs_lookup_root(nilfs, cno);
1009 if (root) {
1010 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1011 if (inode) {
1012 dentry = d_find_alias(inode);
1013 if (dentry) {
1014 ret = nilfs_tree_is_busy(dentry);
1015 dput(dentry);
1016 }
1017 iput(inode);
1018 }
1019 nilfs_put_root(root);
1020 }
1021 return ret;
1022}
1023
1024/**
1025 * nilfs_fill_super() - initialize a super block instance
1026 * @sb: super_block
1027 * @data: mount options
1028 * @silent: silent mode flag
1029 *
1030 * This function is called exclusively by nilfs->ns_mount_mutex.
1031 * So, the recovery process is protected from other simultaneous mounts.
1032 */
1033static int
1034nilfs_fill_super(struct super_block *sb, void *data, int silent)
1035{
1036 struct the_nilfs *nilfs;
1037 struct nilfs_root *fsroot;
1038 __u64 cno;
1039 int err;
1040
1041 nilfs = alloc_nilfs(sb);
1042 if (!nilfs)
1043 return -ENOMEM;
1044
1045 sb->s_fs_info = nilfs;
1046
1047 err = init_nilfs(nilfs, sb, (char *)data);
1048 if (err)
1049 goto failed_nilfs;
1050
1051 sb->s_op = &nilfs_sops;
1052 sb->s_export_op = &nilfs_export_ops;
1053 sb->s_root = NULL;
1054 sb->s_time_gran = 1;
1055 sb->s_max_links = NILFS_LINK_MAX;
1056
1057 sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1058
1059 err = load_nilfs(nilfs, sb);
1060 if (err)
1061 goto failed_nilfs;
1062
1063 cno = nilfs_last_cno(nilfs);
1064 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1065 if (err) {
1066 nilfs_err(sb,
1067 "error %d while loading last checkpoint (checkpoint number=%llu)",
1068 err, (unsigned long long)cno);
1069 goto failed_unload;
1070 }
1071
1072 if (!sb_rdonly(sb)) {
1073 err = nilfs_attach_log_writer(sb, fsroot);
1074 if (err)
1075 goto failed_checkpoint;
1076 }
1077
1078 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1079 if (err)
1080 goto failed_segctor;
1081
1082 nilfs_put_root(fsroot);
1083
1084 if (!sb_rdonly(sb)) {
1085 down_write(&nilfs->ns_sem);
1086 nilfs_setup_super(sb, true);
1087 up_write(&nilfs->ns_sem);
1088 }
1089
1090 return 0;
1091
1092 failed_segctor:
1093 nilfs_detach_log_writer(sb);
1094
1095 failed_checkpoint:
1096 nilfs_put_root(fsroot);
1097
1098 failed_unload:
1099 iput(nilfs->ns_sufile);
1100 iput(nilfs->ns_cpfile);
1101 iput(nilfs->ns_dat);
1102
1103 failed_nilfs:
1104 destroy_nilfs(nilfs);
1105 return err;
1106}
1107
1108static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1109{
1110 struct the_nilfs *nilfs = sb->s_fs_info;
1111 unsigned long old_sb_flags;
1112 unsigned long old_mount_opt;
1113 int err;
1114
1115 sync_filesystem(sb);
1116 old_sb_flags = sb->s_flags;
1117 old_mount_opt = nilfs->ns_mount_opt;
1118
1119 if (!parse_options(data, sb, 1)) {
1120 err = -EINVAL;
1121 goto restore_opts;
1122 }
1123 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1124
1125 err = -EINVAL;
1126
1127 if (!nilfs_valid_fs(nilfs)) {
1128 nilfs_warn(sb,
1129 "couldn't remount because the filesystem is in an incomplete recovery state");
1130 goto restore_opts;
1131 }
1132
1133 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1134 goto out;
1135 if (*flags & SB_RDONLY) {
1136 /* Shutting down log writer */
1137 nilfs_detach_log_writer(sb);
1138 sb->s_flags |= SB_RDONLY;
1139
1140 /*
1141 * Remounting a valid RW partition RDONLY, so set
1142 * the RDONLY flag and then mark the partition as valid again.
1143 */
1144 down_write(&nilfs->ns_sem);
1145 nilfs_cleanup_super(sb);
1146 up_write(&nilfs->ns_sem);
1147 } else {
1148 __u64 features;
1149 struct nilfs_root *root;
1150
1151 /*
1152 * Mounting a RDONLY partition read-write, so reread and
1153 * store the current valid flag. (It may have been changed
1154 * by fsck since we originally mounted the partition.)
1155 */
1156 down_read(&nilfs->ns_sem);
1157 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1158 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1159 up_read(&nilfs->ns_sem);
1160 if (features) {
1161 nilfs_warn(sb,
1162 "couldn't remount RDWR because of unsupported optional features (%llx)",
1163 (unsigned long long)features);
1164 err = -EROFS;
1165 goto restore_opts;
1166 }
1167
1168 sb->s_flags &= ~SB_RDONLY;
1169
1170 root = NILFS_I(d_inode(sb->s_root))->i_root;
1171 err = nilfs_attach_log_writer(sb, root);
1172 if (err)
1173 goto restore_opts;
1174
1175 down_write(&nilfs->ns_sem);
1176 nilfs_setup_super(sb, true);
1177 up_write(&nilfs->ns_sem);
1178 }
1179 out:
1180 return 0;
1181
1182 restore_opts:
1183 sb->s_flags = old_sb_flags;
1184 nilfs->ns_mount_opt = old_mount_opt;
1185 return err;
1186}
1187
1188struct nilfs_super_data {
1189 struct block_device *bdev;
1190 __u64 cno;
1191 int flags;
1192};
1193
1194static int nilfs_parse_snapshot_option(const char *option,
1195 const substring_t *arg,
1196 struct nilfs_super_data *sd)
1197{
1198 unsigned long long val;
1199 const char *msg = NULL;
1200 int err;
1201
1202 if (!(sd->flags & SB_RDONLY)) {
1203 msg = "read-only option is not specified";
1204 goto parse_error;
1205 }
1206
1207 err = kstrtoull(arg->from, 0, &val);
1208 if (err) {
1209 if (err == -ERANGE)
1210 msg = "too large checkpoint number";
1211 else
1212 msg = "malformed argument";
1213 goto parse_error;
1214 } else if (val == 0) {
1215 msg = "invalid checkpoint number 0";
1216 goto parse_error;
1217 }
1218 sd->cno = val;
1219 return 0;
1220
1221parse_error:
1222 nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1223 return 1;
1224}
1225
1226/**
1227 * nilfs_identify - pre-read mount options needed to identify mount instance
1228 * @data: mount options
1229 * @sd: nilfs_super_data
1230 */
1231static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1232{
1233 char *p, *options = data;
1234 substring_t args[MAX_OPT_ARGS];
1235 int token;
1236 int ret = 0;
1237
1238 do {
1239 p = strsep(&options, ",");
1240 if (p != NULL && *p) {
1241 token = match_token(p, tokens, args);
1242 if (token == Opt_snapshot)
1243 ret = nilfs_parse_snapshot_option(p, &args[0],
1244 sd);
1245 }
1246 if (!options)
1247 break;
1248 BUG_ON(options == data);
1249 *(options - 1) = ',';
1250 } while (!ret);
1251 return ret;
1252}
1253
1254static int nilfs_set_bdev_super(struct super_block *s, void *data)
1255{
1256 s->s_bdev = data;
1257 s->s_dev = s->s_bdev->bd_dev;
1258 return 0;
1259}
1260
1261static int nilfs_test_bdev_super(struct super_block *s, void *data)
1262{
1263 return (void *)s->s_bdev == data;
1264}
1265
1266static struct dentry *
1267nilfs_mount(struct file_system_type *fs_type, int flags,
1268 const char *dev_name, void *data)
1269{
1270 struct nilfs_super_data sd;
1271 struct super_block *s;
1272 fmode_t mode = FMODE_READ | FMODE_EXCL;
1273 struct dentry *root_dentry;
1274 int err, s_new = false;
1275
1276 if (!(flags & SB_RDONLY))
1277 mode |= FMODE_WRITE;
1278
1279 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1280 if (IS_ERR(sd.bdev))
1281 return ERR_CAST(sd.bdev);
1282
1283 sd.cno = 0;
1284 sd.flags = flags;
1285 if (nilfs_identify((char *)data, &sd)) {
1286 err = -EINVAL;
1287 goto failed;
1288 }
1289
1290 /*
1291 * once the super is inserted into the list by sget, s_umount
1292 * will protect the lockfs code from trying to start a snapshot
1293 * while we are mounting
1294 */
1295 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1296 if (sd.bdev->bd_fsfreeze_count > 0) {
1297 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1298 err = -EBUSY;
1299 goto failed;
1300 }
1301 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1302 sd.bdev);
1303 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1304 if (IS_ERR(s)) {
1305 err = PTR_ERR(s);
1306 goto failed;
1307 }
1308
1309 if (!s->s_root) {
1310 s_new = true;
1311
1312 /* New superblock instance created */
1313 s->s_mode = mode;
1314 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1315 sb_set_blocksize(s, block_size(sd.bdev));
1316
1317 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1318 if (err)
1319 goto failed_super;
1320
1321 s->s_flags |= SB_ACTIVE;
1322 } else if (!sd.cno) {
1323 if (nilfs_tree_is_busy(s->s_root)) {
1324 if ((flags ^ s->s_flags) & SB_RDONLY) {
1325 nilfs_err(s,
1326 "the device already has a %s mount.",
1327 sb_rdonly(s) ? "read-only" : "read/write");
1328 err = -EBUSY;
1329 goto failed_super;
1330 }
1331 } else {
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};
1371MODULE_ALIAS_FS("nilfs2");
1372
1373static void nilfs_inode_init_once(void *obj)
1374{
1375 struct nilfs_inode_info *ii = obj;
1376
1377 INIT_LIST_HEAD(&ii->i_dirty);
1378#ifdef CONFIG_NILFS_XATTR
1379 init_rwsem(&ii->xattr_sem);
1380#endif
1381 address_space_init_once(&ii->i_btnode_cache);
1382 ii->i_bmap = &ii->i_bmap_data;
1383 inode_init_once(&ii->vfs_inode);
1384}
1385
1386static void nilfs_segbuf_init_once(void *obj)
1387{
1388 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1389}
1390
1391static void nilfs_destroy_cachep(void)
1392{
1393 /*
1394 * Make sure all delayed rcu free inodes are flushed before we
1395 * destroy cache.
1396 */
1397 rcu_barrier();
1398
1399 kmem_cache_destroy(nilfs_inode_cachep);
1400 kmem_cache_destroy(nilfs_transaction_cachep);
1401 kmem_cache_destroy(nilfs_segbuf_cachep);
1402 kmem_cache_destroy(nilfs_btree_path_cache);
1403}
1404
1405static int __init nilfs_init_cachep(void)
1406{
1407 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1408 sizeof(struct nilfs_inode_info), 0,
1409 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1410 nilfs_inode_init_once);
1411 if (!nilfs_inode_cachep)
1412 goto fail;
1413
1414 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1415 sizeof(struct nilfs_transaction_info), 0,
1416 SLAB_RECLAIM_ACCOUNT, NULL);
1417 if (!nilfs_transaction_cachep)
1418 goto fail;
1419
1420 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1421 sizeof(struct nilfs_segment_buffer), 0,
1422 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1423 if (!nilfs_segbuf_cachep)
1424 goto fail;
1425
1426 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1427 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1428 0, 0, NULL);
1429 if (!nilfs_btree_path_cache)
1430 goto fail;
1431
1432 return 0;
1433
1434fail:
1435 nilfs_destroy_cachep();
1436 return -ENOMEM;
1437}
1438
1439static int __init init_nilfs_fs(void)
1440{
1441 int err;
1442
1443 err = nilfs_init_cachep();
1444 if (err)
1445 goto fail;
1446
1447 err = nilfs_sysfs_init();
1448 if (err)
1449 goto free_cachep;
1450
1451 err = register_filesystem(&nilfs_fs_type);
1452 if (err)
1453 goto deinit_sysfs_entry;
1454
1455 printk(KERN_INFO "NILFS version 2 loaded\n");
1456 return 0;
1457
1458deinit_sysfs_entry:
1459 nilfs_sysfs_exit();
1460free_cachep:
1461 nilfs_destroy_cachep();
1462fail:
1463 return err;
1464}
1465
1466static void __exit exit_nilfs_fs(void)
1467{
1468 nilfs_destroy_cachep();
1469 nilfs_sysfs_exit();
1470 unregister_filesystem(&nilfs_fs_type);
1471}
1472
1473module_init(init_nilfs_fs)
1474module_exit(exit_nilfs_fs)