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
52 MODULE_AUTHOR("NTT Corp.");
53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
54 "(NILFS)");
55 MODULE_LICENSE("GPL");
56
57 static struct kmem_cache *nilfs_inode_cachep;
58 struct kmem_cache *nilfs_transaction_cachep;
59 struct kmem_cache *nilfs_segbuf_cachep;
60 struct kmem_cache *nilfs_btree_path_cache;
61
62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
64
__nilfs_msg(struct super_block * sb,const char * fmt,...)65 void __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
nilfs_set_error(struct super_block * sb)87 static 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 */
__nilfs_error(struct super_block * sb,const char * function,const char * fmt,...)119 void __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
nilfs_alloc_inode(struct super_block * sb)150 struct 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 ii->i_assoc_inode = NULL;
161 ii->i_bmap = &ii->i_bmap_data;
162 return &ii->vfs_inode;
163 }
164
nilfs_free_inode(struct inode * inode)165 static void nilfs_free_inode(struct inode *inode)
166 {
167 if (nilfs_is_metadata_file_inode(inode))
168 nilfs_mdt_destroy(inode);
169
170 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
171 }
172
nilfs_sync_super(struct super_block * sb,int flag)173 static int nilfs_sync_super(struct super_block *sb, int flag)
174 {
175 struct the_nilfs *nilfs = sb->s_fs_info;
176 int err;
177
178 retry:
179 set_buffer_dirty(nilfs->ns_sbh[0]);
180 if (nilfs_test_opt(nilfs, BARRIER)) {
181 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
182 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
183 } else {
184 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
185 }
186
187 if (unlikely(err)) {
188 nilfs_err(sb, "unable to write superblock: err=%d", err);
189 if (err == -EIO && nilfs->ns_sbh[1]) {
190 /*
191 * sbp[0] points to newer log than sbp[1],
192 * so copy sbp[0] to sbp[1] to take over sbp[0].
193 */
194 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
195 nilfs->ns_sbsize);
196 nilfs_fall_back_super_block(nilfs);
197 goto retry;
198 }
199 } else {
200 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
201
202 nilfs->ns_sbwcount++;
203
204 /*
205 * The latest segment becomes trailable from the position
206 * written in superblock.
207 */
208 clear_nilfs_discontinued(nilfs);
209
210 /* update GC protection for recent segments */
211 if (nilfs->ns_sbh[1]) {
212 if (flag == NILFS_SB_COMMIT_ALL) {
213 set_buffer_dirty(nilfs->ns_sbh[1]);
214 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
215 goto out;
216 }
217 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
218 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
219 sbp = nilfs->ns_sbp[1];
220 }
221
222 spin_lock(&nilfs->ns_last_segment_lock);
223 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
224 spin_unlock(&nilfs->ns_last_segment_lock);
225 }
226 out:
227 return err;
228 }
229
nilfs_set_log_cursor(struct nilfs_super_block * sbp,struct the_nilfs * nilfs)230 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
231 struct the_nilfs *nilfs)
232 {
233 sector_t nfreeblocks;
234
235 /* nilfs->ns_sem must be locked by the caller. */
236 nilfs_count_free_blocks(nilfs, &nfreeblocks);
237 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
238
239 spin_lock(&nilfs->ns_last_segment_lock);
240 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
241 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
242 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
243 spin_unlock(&nilfs->ns_last_segment_lock);
244 }
245
nilfs_prepare_super(struct super_block * sb,int flip)246 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
247 int flip)
248 {
249 struct the_nilfs *nilfs = sb->s_fs_info;
250 struct nilfs_super_block **sbp = nilfs->ns_sbp;
251
252 /* nilfs->ns_sem must be locked by the caller. */
253 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
254 if (sbp[1] &&
255 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
256 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
257 } else {
258 nilfs_crit(sb, "superblock broke");
259 return NULL;
260 }
261 } else if (sbp[1] &&
262 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
263 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
264 }
265
266 if (flip && sbp[1])
267 nilfs_swap_super_block(nilfs);
268
269 return sbp;
270 }
271
nilfs_commit_super(struct super_block * sb,int flag)272 int nilfs_commit_super(struct super_block *sb, int flag)
273 {
274 struct the_nilfs *nilfs = sb->s_fs_info;
275 struct nilfs_super_block **sbp = nilfs->ns_sbp;
276 time64_t t;
277
278 /* nilfs->ns_sem must be locked by the caller. */
279 t = ktime_get_real_seconds();
280 nilfs->ns_sbwtime = t;
281 sbp[0]->s_wtime = cpu_to_le64(t);
282 sbp[0]->s_sum = 0;
283 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
284 (unsigned char *)sbp[0],
285 nilfs->ns_sbsize));
286 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
287 sbp[1]->s_wtime = sbp[0]->s_wtime;
288 sbp[1]->s_sum = 0;
289 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
290 (unsigned char *)sbp[1],
291 nilfs->ns_sbsize));
292 }
293 clear_nilfs_sb_dirty(nilfs);
294 nilfs->ns_flushed_device = 1;
295 /* make sure store to ns_flushed_device cannot be reordered */
296 smp_wmb();
297 return nilfs_sync_super(sb, flag);
298 }
299
300 /**
301 * nilfs_cleanup_super() - write filesystem state for cleanup
302 * @sb: super block instance to be unmounted or degraded to read-only
303 *
304 * This function restores state flags in the on-disk super block.
305 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
306 * filesystem was not clean previously.
307 */
nilfs_cleanup_super(struct super_block * sb)308 int nilfs_cleanup_super(struct super_block *sb)
309 {
310 struct the_nilfs *nilfs = sb->s_fs_info;
311 struct nilfs_super_block **sbp;
312 int flag = NILFS_SB_COMMIT;
313 int ret = -EIO;
314
315 sbp = nilfs_prepare_super(sb, 0);
316 if (sbp) {
317 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
318 nilfs_set_log_cursor(sbp[0], nilfs);
319 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
320 /*
321 * make the "clean" flag also to the opposite
322 * super block if both super blocks point to
323 * the same checkpoint.
324 */
325 sbp[1]->s_state = sbp[0]->s_state;
326 flag = NILFS_SB_COMMIT_ALL;
327 }
328 ret = nilfs_commit_super(sb, flag);
329 }
330 return ret;
331 }
332
333 /**
334 * nilfs_move_2nd_super - relocate secondary super block
335 * @sb: super block instance
336 * @sb2off: new offset of the secondary super block (in bytes)
337 */
nilfs_move_2nd_super(struct super_block * sb,loff_t sb2off)338 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
339 {
340 struct the_nilfs *nilfs = sb->s_fs_info;
341 struct buffer_head *nsbh;
342 struct nilfs_super_block *nsbp;
343 sector_t blocknr, newblocknr;
344 unsigned long offset;
345 int sb2i; /* array index of the secondary superblock */
346 int ret = 0;
347
348 /* nilfs->ns_sem must be locked by the caller. */
349 if (nilfs->ns_sbh[1] &&
350 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
351 sb2i = 1;
352 blocknr = nilfs->ns_sbh[1]->b_blocknr;
353 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
354 sb2i = 0;
355 blocknr = nilfs->ns_sbh[0]->b_blocknr;
356 } else {
357 sb2i = -1;
358 blocknr = 0;
359 }
360 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
361 goto out; /* super block location is unchanged */
362
363 /* Get new super block buffer */
364 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
365 offset = sb2off & (nilfs->ns_blocksize - 1);
366 nsbh = sb_getblk(sb, newblocknr);
367 if (!nsbh) {
368 nilfs_warn(sb,
369 "unable to move secondary superblock to block %llu",
370 (unsigned long long)newblocknr);
371 ret = -EIO;
372 goto out;
373 }
374 nsbp = (void *)nsbh->b_data + offset;
375 memset(nsbp, 0, nilfs->ns_blocksize);
376
377 if (sb2i >= 0) {
378 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
379 brelse(nilfs->ns_sbh[sb2i]);
380 nilfs->ns_sbh[sb2i] = nsbh;
381 nilfs->ns_sbp[sb2i] = nsbp;
382 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
383 /* secondary super block will be restored to index 1 */
384 nilfs->ns_sbh[1] = nsbh;
385 nilfs->ns_sbp[1] = nsbp;
386 } else {
387 brelse(nsbh);
388 }
389 out:
390 return ret;
391 }
392
393 /**
394 * nilfs_resize_fs - resize the filesystem
395 * @sb: super block instance
396 * @newsize: new size of the filesystem (in bytes)
397 */
nilfs_resize_fs(struct super_block * sb,__u64 newsize)398 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
399 {
400 struct the_nilfs *nilfs = sb->s_fs_info;
401 struct nilfs_super_block **sbp;
402 __u64 devsize, newnsegs;
403 loff_t sb2off;
404 int ret;
405
406 ret = -ERANGE;
407 devsize = i_size_read(sb->s_bdev->bd_inode);
408 if (newsize > devsize)
409 goto out;
410
411 /*
412 * Prevent underflow in second superblock position calculation.
413 * The exact minimum size check is done in nilfs_sufile_resize().
414 */
415 if (newsize < 4096) {
416 ret = -ENOSPC;
417 goto out;
418 }
419
420 /*
421 * Write lock is required to protect some functions depending
422 * on the number of segments, the number of reserved segments,
423 * and so forth.
424 */
425 down_write(&nilfs->ns_segctor_sem);
426
427 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
428 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
429 do_div(newnsegs, nilfs->ns_blocks_per_segment);
430
431 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
432 up_write(&nilfs->ns_segctor_sem);
433 if (ret < 0)
434 goto out;
435
436 ret = nilfs_construct_segment(sb);
437 if (ret < 0)
438 goto out;
439
440 down_write(&nilfs->ns_sem);
441 nilfs_move_2nd_super(sb, sb2off);
442 ret = -EIO;
443 sbp = nilfs_prepare_super(sb, 0);
444 if (likely(sbp)) {
445 nilfs_set_log_cursor(sbp[0], nilfs);
446 /*
447 * Drop NILFS_RESIZE_FS flag for compatibility with
448 * mount-time resize which may be implemented in a
449 * future release.
450 */
451 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
452 ~NILFS_RESIZE_FS);
453 sbp[0]->s_dev_size = cpu_to_le64(newsize);
454 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
455 if (sbp[1])
456 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
457 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
458 }
459 up_write(&nilfs->ns_sem);
460
461 /*
462 * Reset the range of allocatable segments last. This order
463 * is important in the case of expansion because the secondary
464 * superblock must be protected from log write until migration
465 * completes.
466 */
467 if (!ret)
468 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
469 out:
470 return ret;
471 }
472
nilfs_put_super(struct super_block * sb)473 static void nilfs_put_super(struct super_block *sb)
474 {
475 struct the_nilfs *nilfs = sb->s_fs_info;
476
477 nilfs_detach_log_writer(sb);
478
479 if (!sb_rdonly(sb)) {
480 down_write(&nilfs->ns_sem);
481 nilfs_cleanup_super(sb);
482 up_write(&nilfs->ns_sem);
483 }
484
485 nilfs_sysfs_delete_device_group(nilfs);
486 iput(nilfs->ns_sufile);
487 iput(nilfs->ns_cpfile);
488 iput(nilfs->ns_dat);
489
490 destroy_nilfs(nilfs);
491 sb->s_fs_info = NULL;
492 }
493
nilfs_sync_fs(struct super_block * sb,int wait)494 static int nilfs_sync_fs(struct super_block *sb, int wait)
495 {
496 struct the_nilfs *nilfs = sb->s_fs_info;
497 struct nilfs_super_block **sbp;
498 int err = 0;
499
500 /* This function is called when super block should be written back */
501 if (wait)
502 err = nilfs_construct_segment(sb);
503
504 down_write(&nilfs->ns_sem);
505 if (nilfs_sb_dirty(nilfs)) {
506 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
507 if (likely(sbp)) {
508 nilfs_set_log_cursor(sbp[0], nilfs);
509 nilfs_commit_super(sb, NILFS_SB_COMMIT);
510 }
511 }
512 up_write(&nilfs->ns_sem);
513
514 if (!err)
515 err = nilfs_flush_device(nilfs);
516
517 return err;
518 }
519
nilfs_attach_checkpoint(struct super_block * sb,__u64 cno,int curr_mnt,struct nilfs_root ** rootp)520 int 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 nilfs_err(sb,
544 "Invalid checkpoint (checkpoint number=%llu)",
545 (unsigned long long)cno);
546 err = -EINVAL;
547 }
548 goto failed;
549 }
550
551 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
552 &raw_cp->cp_ifile_inode, &root->ifile);
553 if (err)
554 goto failed_bh;
555
556 atomic64_set(&root->inodes_count,
557 le64_to_cpu(raw_cp->cp_inodes_count));
558 atomic64_set(&root->blocks_count,
559 le64_to_cpu(raw_cp->cp_blocks_count));
560
561 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
562
563 reuse:
564 *rootp = root;
565 return 0;
566
567 failed_bh:
568 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
569 failed:
570 nilfs_put_root(root);
571
572 return err;
573 }
574
nilfs_freeze(struct super_block * sb)575 static int nilfs_freeze(struct super_block *sb)
576 {
577 struct the_nilfs *nilfs = sb->s_fs_info;
578 int err;
579
580 if (sb_rdonly(sb))
581 return 0;
582
583 /* Mark super block clean */
584 down_write(&nilfs->ns_sem);
585 err = nilfs_cleanup_super(sb);
586 up_write(&nilfs->ns_sem);
587 return err;
588 }
589
nilfs_unfreeze(struct super_block * sb)590 static int nilfs_unfreeze(struct super_block *sb)
591 {
592 struct the_nilfs *nilfs = sb->s_fs_info;
593
594 if (sb_rdonly(sb))
595 return 0;
596
597 down_write(&nilfs->ns_sem);
598 nilfs_setup_super(sb, false);
599 up_write(&nilfs->ns_sem);
600 return 0;
601 }
602
nilfs_statfs(struct dentry * dentry,struct kstatfs * buf)603 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
604 {
605 struct super_block *sb = dentry->d_sb;
606 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
607 struct the_nilfs *nilfs = root->nilfs;
608 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
609 unsigned long long blocks;
610 unsigned long overhead;
611 unsigned long nrsvblocks;
612 sector_t nfreeblocks;
613 u64 nmaxinodes, nfreeinodes;
614 int err;
615
616 /*
617 * Compute all of the segment blocks
618 *
619 * The blocks before first segment and after last segment
620 * are excluded.
621 */
622 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
623 - nilfs->ns_first_data_block;
624 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
625
626 /*
627 * Compute the overhead
628 *
629 * When distributing meta data blocks outside segment structure,
630 * We must count them as the overhead.
631 */
632 overhead = 0;
633
634 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
635 if (unlikely(err))
636 return err;
637
638 err = nilfs_ifile_count_free_inodes(root->ifile,
639 &nmaxinodes, &nfreeinodes);
640 if (unlikely(err)) {
641 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
642 if (err == -ERANGE) {
643 /*
644 * If nilfs_palloc_count_max_entries() returns
645 * -ERANGE error code then we simply treat
646 * curent inodes count as maximum possible and
647 * zero as free inodes value.
648 */
649 nmaxinodes = atomic64_read(&root->inodes_count);
650 nfreeinodes = 0;
651 err = 0;
652 } else
653 return err;
654 }
655
656 buf->f_type = NILFS_SUPER_MAGIC;
657 buf->f_bsize = sb->s_blocksize;
658 buf->f_blocks = blocks - overhead;
659 buf->f_bfree = nfreeblocks;
660 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
661 (buf->f_bfree - nrsvblocks) : 0;
662 buf->f_files = nmaxinodes;
663 buf->f_ffree = nfreeinodes;
664 buf->f_namelen = NILFS_NAME_LEN;
665 buf->f_fsid = u64_to_fsid(id);
666
667 return 0;
668 }
669
nilfs_show_options(struct seq_file * seq,struct dentry * dentry)670 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
671 {
672 struct super_block *sb = dentry->d_sb;
673 struct the_nilfs *nilfs = sb->s_fs_info;
674 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
675
676 if (!nilfs_test_opt(nilfs, BARRIER))
677 seq_puts(seq, ",nobarrier");
678 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
679 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
680 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
681 seq_puts(seq, ",errors=panic");
682 if (nilfs_test_opt(nilfs, ERRORS_CONT))
683 seq_puts(seq, ",errors=continue");
684 if (nilfs_test_opt(nilfs, STRICT_ORDER))
685 seq_puts(seq, ",order=strict");
686 if (nilfs_test_opt(nilfs, NORECOVERY))
687 seq_puts(seq, ",norecovery");
688 if (nilfs_test_opt(nilfs, DISCARD))
689 seq_puts(seq, ",discard");
690
691 return 0;
692 }
693
694 static const struct super_operations nilfs_sops = {
695 .alloc_inode = nilfs_alloc_inode,
696 .free_inode = nilfs_free_inode,
697 .dirty_inode = nilfs_dirty_inode,
698 .evict_inode = nilfs_evict_inode,
699 .put_super = nilfs_put_super,
700 .sync_fs = nilfs_sync_fs,
701 .freeze_fs = nilfs_freeze,
702 .unfreeze_fs = nilfs_unfreeze,
703 .statfs = nilfs_statfs,
704 .remount_fs = nilfs_remount,
705 .show_options = nilfs_show_options
706 };
707
708 enum {
709 Opt_err_cont, Opt_err_panic, Opt_err_ro,
710 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
711 Opt_discard, Opt_nodiscard, Opt_err,
712 };
713
714 static match_table_t tokens = {
715 {Opt_err_cont, "errors=continue"},
716 {Opt_err_panic, "errors=panic"},
717 {Opt_err_ro, "errors=remount-ro"},
718 {Opt_barrier, "barrier"},
719 {Opt_nobarrier, "nobarrier"},
720 {Opt_snapshot, "cp=%u"},
721 {Opt_order, "order=%s"},
722 {Opt_norecovery, "norecovery"},
723 {Opt_discard, "discard"},
724 {Opt_nodiscard, "nodiscard"},
725 {Opt_err, NULL}
726 };
727
parse_options(char * options,struct super_block * sb,int is_remount)728 static int parse_options(char *options, struct super_block *sb, int is_remount)
729 {
730 struct the_nilfs *nilfs = sb->s_fs_info;
731 char *p;
732 substring_t args[MAX_OPT_ARGS];
733
734 if (!options)
735 return 1;
736
737 while ((p = strsep(&options, ",")) != NULL) {
738 int token;
739
740 if (!*p)
741 continue;
742
743 token = match_token(p, tokens, args);
744 switch (token) {
745 case Opt_barrier:
746 nilfs_set_opt(nilfs, BARRIER);
747 break;
748 case Opt_nobarrier:
749 nilfs_clear_opt(nilfs, BARRIER);
750 break;
751 case Opt_order:
752 if (strcmp(args[0].from, "relaxed") == 0)
753 /* Ordered data semantics */
754 nilfs_clear_opt(nilfs, STRICT_ORDER);
755 else if (strcmp(args[0].from, "strict") == 0)
756 /* Strict in-order semantics */
757 nilfs_set_opt(nilfs, STRICT_ORDER);
758 else
759 return 0;
760 break;
761 case Opt_err_panic:
762 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
763 break;
764 case Opt_err_ro:
765 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
766 break;
767 case Opt_err_cont:
768 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
769 break;
770 case Opt_snapshot:
771 if (is_remount) {
772 nilfs_err(sb,
773 "\"%s\" option is invalid for remount",
774 p);
775 return 0;
776 }
777 break;
778 case Opt_norecovery:
779 nilfs_set_opt(nilfs, NORECOVERY);
780 break;
781 case Opt_discard:
782 nilfs_set_opt(nilfs, DISCARD);
783 break;
784 case Opt_nodiscard:
785 nilfs_clear_opt(nilfs, DISCARD);
786 break;
787 default:
788 nilfs_err(sb, "unrecognized mount option \"%s\"", p);
789 return 0;
790 }
791 }
792 return 1;
793 }
794
795 static inline void
nilfs_set_default_options(struct super_block * sb,struct nilfs_super_block * sbp)796 nilfs_set_default_options(struct super_block *sb,
797 struct nilfs_super_block *sbp)
798 {
799 struct the_nilfs *nilfs = sb->s_fs_info;
800
801 nilfs->ns_mount_opt =
802 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
803 }
804
nilfs_setup_super(struct super_block * sb,int is_mount)805 static int nilfs_setup_super(struct super_block *sb, int is_mount)
806 {
807 struct the_nilfs *nilfs = sb->s_fs_info;
808 struct nilfs_super_block **sbp;
809 int max_mnt_count;
810 int mnt_count;
811
812 /* nilfs->ns_sem must be locked by the caller. */
813 sbp = nilfs_prepare_super(sb, 0);
814 if (!sbp)
815 return -EIO;
816
817 if (!is_mount)
818 goto skip_mount_setup;
819
820 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
821 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
822
823 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
824 nilfs_warn(sb, "mounting fs with errors");
825 #if 0
826 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
827 nilfs_warn(sb, "maximal mount count reached");
828 #endif
829 }
830 if (!max_mnt_count)
831 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
832
833 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
834 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
835
836 skip_mount_setup:
837 sbp[0]->s_state =
838 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
839 /* synchronize sbp[1] with sbp[0] */
840 if (sbp[1])
841 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
842 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
843 }
844
nilfs_read_super_block(struct super_block * sb,u64 pos,int blocksize,struct buffer_head ** pbh)845 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
846 u64 pos, int blocksize,
847 struct buffer_head **pbh)
848 {
849 unsigned long long sb_index = pos;
850 unsigned long offset;
851
852 offset = do_div(sb_index, blocksize);
853 *pbh = sb_bread(sb, sb_index);
854 if (!*pbh)
855 return NULL;
856 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
857 }
858
nilfs_store_magic_and_option(struct super_block * sb,struct nilfs_super_block * sbp,char * data)859 int nilfs_store_magic_and_option(struct super_block *sb,
860 struct nilfs_super_block *sbp,
861 char *data)
862 {
863 struct the_nilfs *nilfs = sb->s_fs_info;
864
865 sb->s_magic = le16_to_cpu(sbp->s_magic);
866
867 /* FS independent flags */
868 #ifdef NILFS_ATIME_DISABLE
869 sb->s_flags |= SB_NOATIME;
870 #endif
871
872 nilfs_set_default_options(sb, sbp);
873
874 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
875 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
876 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
877 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
878
879 return !parse_options(data, sb, 0) ? -EINVAL : 0;
880 }
881
nilfs_check_feature_compatibility(struct super_block * sb,struct nilfs_super_block * sbp)882 int nilfs_check_feature_compatibility(struct super_block *sb,
883 struct nilfs_super_block *sbp)
884 {
885 __u64 features;
886
887 features = le64_to_cpu(sbp->s_feature_incompat) &
888 ~NILFS_FEATURE_INCOMPAT_SUPP;
889 if (features) {
890 nilfs_err(sb,
891 "couldn't mount because of unsupported optional features (%llx)",
892 (unsigned long long)features);
893 return -EINVAL;
894 }
895 features = le64_to_cpu(sbp->s_feature_compat_ro) &
896 ~NILFS_FEATURE_COMPAT_RO_SUPP;
897 if (!sb_rdonly(sb) && features) {
898 nilfs_err(sb,
899 "couldn't mount RDWR because of unsupported optional features (%llx)",
900 (unsigned long long)features);
901 return -EINVAL;
902 }
903 return 0;
904 }
905
nilfs_get_root_dentry(struct super_block * sb,struct nilfs_root * root,struct dentry ** root_dentry)906 static int nilfs_get_root_dentry(struct super_block *sb,
907 struct nilfs_root *root,
908 struct dentry **root_dentry)
909 {
910 struct inode *inode;
911 struct dentry *dentry;
912 int ret = 0;
913
914 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
915 if (IS_ERR(inode)) {
916 ret = PTR_ERR(inode);
917 nilfs_err(sb, "error %d getting root inode", ret);
918 goto out;
919 }
920 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
921 iput(inode);
922 nilfs_err(sb, "corrupt root inode");
923 ret = -EINVAL;
924 goto out;
925 }
926
927 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
928 dentry = d_find_alias(inode);
929 if (!dentry) {
930 dentry = d_make_root(inode);
931 if (!dentry) {
932 ret = -ENOMEM;
933 goto failed_dentry;
934 }
935 } else {
936 iput(inode);
937 }
938 } else {
939 dentry = d_obtain_root(inode);
940 if (IS_ERR(dentry)) {
941 ret = PTR_ERR(dentry);
942 goto failed_dentry;
943 }
944 }
945 *root_dentry = dentry;
946 out:
947 return ret;
948
949 failed_dentry:
950 nilfs_err(sb, "error %d getting root dentry", ret);
951 goto out;
952 }
953
nilfs_attach_snapshot(struct super_block * s,__u64 cno,struct dentry ** root_dentry)954 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
955 struct dentry **root_dentry)
956 {
957 struct the_nilfs *nilfs = s->s_fs_info;
958 struct nilfs_root *root;
959 int ret;
960
961 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
962
963 down_read(&nilfs->ns_segctor_sem);
964 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
965 up_read(&nilfs->ns_segctor_sem);
966 if (ret < 0) {
967 ret = (ret == -ENOENT) ? -EINVAL : ret;
968 goto out;
969 } else if (!ret) {
970 nilfs_err(s,
971 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
972 (unsigned long long)cno);
973 ret = -EINVAL;
974 goto out;
975 }
976
977 ret = nilfs_attach_checkpoint(s, cno, false, &root);
978 if (ret) {
979 nilfs_err(s,
980 "error %d while loading snapshot (checkpoint number=%llu)",
981 ret, (unsigned long long)cno);
982 goto out;
983 }
984 ret = nilfs_get_root_dentry(s, root, root_dentry);
985 nilfs_put_root(root);
986 out:
987 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
988 return ret;
989 }
990
991 /**
992 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
993 * @root_dentry: root dentry of the tree to be shrunk
994 *
995 * This function returns true if the tree was in-use.
996 */
nilfs_tree_is_busy(struct dentry * root_dentry)997 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
998 {
999 shrink_dcache_parent(root_dentry);
1000 return d_count(root_dentry) > 1;
1001 }
1002
nilfs_checkpoint_is_mounted(struct super_block * sb,__u64 cno)1003 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1004 {
1005 struct the_nilfs *nilfs = sb->s_fs_info;
1006 struct nilfs_root *root;
1007 struct inode *inode;
1008 struct dentry *dentry;
1009 int ret;
1010
1011 if (cno > nilfs->ns_cno)
1012 return false;
1013
1014 if (cno >= nilfs_last_cno(nilfs))
1015 return true; /* protect recent checkpoints */
1016
1017 ret = false;
1018 root = nilfs_lookup_root(nilfs, cno);
1019 if (root) {
1020 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1021 if (inode) {
1022 dentry = d_find_alias(inode);
1023 if (dentry) {
1024 ret = nilfs_tree_is_busy(dentry);
1025 dput(dentry);
1026 }
1027 iput(inode);
1028 }
1029 nilfs_put_root(root);
1030 }
1031 return ret;
1032 }
1033
1034 /**
1035 * nilfs_fill_super() - initialize a super block instance
1036 * @sb: super_block
1037 * @data: mount options
1038 * @silent: silent mode flag
1039 *
1040 * This function is called exclusively by nilfs->ns_mount_mutex.
1041 * So, the recovery process is protected from other simultaneous mounts.
1042 */
1043 static int
nilfs_fill_super(struct super_block * sb,void * data,int silent)1044 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1045 {
1046 struct the_nilfs *nilfs;
1047 struct nilfs_root *fsroot;
1048 __u64 cno;
1049 int err;
1050
1051 nilfs = alloc_nilfs(sb);
1052 if (!nilfs)
1053 return -ENOMEM;
1054
1055 sb->s_fs_info = nilfs;
1056
1057 err = init_nilfs(nilfs, sb, (char *)data);
1058 if (err)
1059 goto failed_nilfs;
1060
1061 sb->s_op = &nilfs_sops;
1062 sb->s_export_op = &nilfs_export_ops;
1063 sb->s_root = NULL;
1064 sb->s_time_gran = 1;
1065 sb->s_max_links = NILFS_LINK_MAX;
1066
1067 sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
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 nilfs_err(sb,
1077 "error %d while loading last checkpoint (checkpoint number=%llu)",
1078 err, (unsigned long long)cno);
1079 goto failed_unload;
1080 }
1081
1082 if (!sb_rdonly(sb)) {
1083 err = nilfs_attach_log_writer(sb, fsroot);
1084 if (err)
1085 goto failed_checkpoint;
1086 }
1087
1088 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1089 if (err)
1090 goto failed_segctor;
1091
1092 nilfs_put_root(fsroot);
1093
1094 if (!sb_rdonly(sb)) {
1095 down_write(&nilfs->ns_sem);
1096 nilfs_setup_super(sb, true);
1097 up_write(&nilfs->ns_sem);
1098 }
1099
1100 return 0;
1101
1102 failed_segctor:
1103 nilfs_detach_log_writer(sb);
1104
1105 failed_checkpoint:
1106 nilfs_put_root(fsroot);
1107
1108 failed_unload:
1109 nilfs_sysfs_delete_device_group(nilfs);
1110 iput(nilfs->ns_sufile);
1111 iput(nilfs->ns_cpfile);
1112 iput(nilfs->ns_dat);
1113
1114 failed_nilfs:
1115 destroy_nilfs(nilfs);
1116 return err;
1117 }
1118
nilfs_remount(struct super_block * sb,int * flags,char * data)1119 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1120 {
1121 struct the_nilfs *nilfs = sb->s_fs_info;
1122 unsigned long old_sb_flags;
1123 unsigned long old_mount_opt;
1124 int err;
1125
1126 sync_filesystem(sb);
1127 old_sb_flags = sb->s_flags;
1128 old_mount_opt = nilfs->ns_mount_opt;
1129
1130 if (!parse_options(data, sb, 1)) {
1131 err = -EINVAL;
1132 goto restore_opts;
1133 }
1134 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1135
1136 err = -EINVAL;
1137
1138 if (!nilfs_valid_fs(nilfs)) {
1139 nilfs_warn(sb,
1140 "couldn't remount because the filesystem is in an incomplete recovery state");
1141 goto restore_opts;
1142 }
1143
1144 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1145 goto out;
1146 if (*flags & SB_RDONLY) {
1147 sb->s_flags |= SB_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 nilfs_warn(sb,
1171 "couldn't remount RDWR because of unsupported optional features (%llx)",
1172 (unsigned long long)features);
1173 err = -EROFS;
1174 goto restore_opts;
1175 }
1176
1177 sb->s_flags &= ~SB_RDONLY;
1178
1179 root = NILFS_I(d_inode(sb->s_root))->i_root;
1180 err = nilfs_attach_log_writer(sb, root);
1181 if (err)
1182 goto restore_opts;
1183
1184 down_write(&nilfs->ns_sem);
1185 nilfs_setup_super(sb, true);
1186 up_write(&nilfs->ns_sem);
1187 }
1188 out:
1189 return 0;
1190
1191 restore_opts:
1192 sb->s_flags = old_sb_flags;
1193 nilfs->ns_mount_opt = old_mount_opt;
1194 return err;
1195 }
1196
1197 struct nilfs_super_data {
1198 struct block_device *bdev;
1199 __u64 cno;
1200 int flags;
1201 };
1202
nilfs_parse_snapshot_option(const char * option,const substring_t * arg,struct nilfs_super_data * sd)1203 static int nilfs_parse_snapshot_option(const char *option,
1204 const substring_t *arg,
1205 struct nilfs_super_data *sd)
1206 {
1207 unsigned long long val;
1208 const char *msg = NULL;
1209 int err;
1210
1211 if (!(sd->flags & SB_RDONLY)) {
1212 msg = "read-only option is not specified";
1213 goto parse_error;
1214 }
1215
1216 err = kstrtoull(arg->from, 0, &val);
1217 if (err) {
1218 if (err == -ERANGE)
1219 msg = "too large checkpoint number";
1220 else
1221 msg = "malformed argument";
1222 goto parse_error;
1223 } else if (val == 0) {
1224 msg = "invalid checkpoint number 0";
1225 goto parse_error;
1226 }
1227 sd->cno = val;
1228 return 0;
1229
1230 parse_error:
1231 nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1232 return 1;
1233 }
1234
1235 /**
1236 * nilfs_identify - pre-read mount options needed to identify mount instance
1237 * @data: mount options
1238 * @sd: nilfs_super_data
1239 */
nilfs_identify(char * data,struct nilfs_super_data * sd)1240 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1241 {
1242 char *p, *options = data;
1243 substring_t args[MAX_OPT_ARGS];
1244 int token;
1245 int ret = 0;
1246
1247 do {
1248 p = strsep(&options, ",");
1249 if (p != NULL && *p) {
1250 token = match_token(p, tokens, args);
1251 if (token == Opt_snapshot)
1252 ret = nilfs_parse_snapshot_option(p, &args[0],
1253 sd);
1254 }
1255 if (!options)
1256 break;
1257 BUG_ON(options == data);
1258 *(options - 1) = ',';
1259 } while (!ret);
1260 return ret;
1261 }
1262
nilfs_set_bdev_super(struct super_block * s,void * data)1263 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1264 {
1265 s->s_bdev = data;
1266 s->s_dev = s->s_bdev->bd_dev;
1267 return 0;
1268 }
1269
nilfs_test_bdev_super(struct super_block * s,void * data)1270 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1271 {
1272 return (void *)s->s_bdev == data;
1273 }
1274
1275 static struct dentry *
nilfs_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1276 nilfs_mount(struct file_system_type *fs_type, int flags,
1277 const char *dev_name, void *data)
1278 {
1279 struct nilfs_super_data sd;
1280 struct super_block *s;
1281 fmode_t mode = FMODE_READ | FMODE_EXCL;
1282 struct dentry *root_dentry;
1283 int err, s_new = false;
1284
1285 if (!(flags & SB_RDONLY))
1286 mode |= FMODE_WRITE;
1287
1288 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1289 if (IS_ERR(sd.bdev))
1290 return ERR_CAST(sd.bdev);
1291
1292 sd.cno = 0;
1293 sd.flags = flags;
1294 if (nilfs_identify((char *)data, &sd)) {
1295 err = -EINVAL;
1296 goto failed;
1297 }
1298
1299 /*
1300 * once the super is inserted into the list by sget, s_umount
1301 * will protect the lockfs code from trying to start a snapshot
1302 * while we are mounting
1303 */
1304 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1305 if (sd.bdev->bd_fsfreeze_count > 0) {
1306 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1307 err = -EBUSY;
1308 goto failed;
1309 }
1310 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1311 sd.bdev);
1312 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1313 if (IS_ERR(s)) {
1314 err = PTR_ERR(s);
1315 goto failed;
1316 }
1317
1318 if (!s->s_root) {
1319 s_new = true;
1320
1321 /* New superblock instance created */
1322 s->s_mode = mode;
1323 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1324 sb_set_blocksize(s, block_size(sd.bdev));
1325
1326 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1327 if (err)
1328 goto failed_super;
1329
1330 s->s_flags |= SB_ACTIVE;
1331 } else if (!sd.cno) {
1332 if (nilfs_tree_is_busy(s->s_root)) {
1333 if ((flags ^ s->s_flags) & SB_RDONLY) {
1334 nilfs_err(s,
1335 "the device already has a %s mount.",
1336 sb_rdonly(s) ? "read-only" : "read/write");
1337 err = -EBUSY;
1338 goto failed_super;
1339 }
1340 } else {
1341 /*
1342 * Try remount to setup mount states if the current
1343 * tree is not mounted and only snapshots use this sb.
1344 */
1345 err = nilfs_remount(s, &flags, data);
1346 if (err)
1347 goto failed_super;
1348 }
1349 }
1350
1351 if (sd.cno) {
1352 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1353 if (err)
1354 goto failed_super;
1355 } else {
1356 root_dentry = dget(s->s_root);
1357 }
1358
1359 if (!s_new)
1360 blkdev_put(sd.bdev, mode);
1361
1362 return root_dentry;
1363
1364 failed_super:
1365 deactivate_locked_super(s);
1366
1367 failed:
1368 if (!s_new)
1369 blkdev_put(sd.bdev, mode);
1370 return ERR_PTR(err);
1371 }
1372
1373 struct file_system_type nilfs_fs_type = {
1374 .owner = THIS_MODULE,
1375 .name = "nilfs2",
1376 .mount = nilfs_mount,
1377 .kill_sb = kill_block_super,
1378 .fs_flags = FS_REQUIRES_DEV,
1379 };
1380 MODULE_ALIAS_FS("nilfs2");
1381
nilfs_inode_init_once(void * obj)1382 static void nilfs_inode_init_once(void *obj)
1383 {
1384 struct nilfs_inode_info *ii = obj;
1385
1386 INIT_LIST_HEAD(&ii->i_dirty);
1387 #ifdef CONFIG_NILFS_XATTR
1388 init_rwsem(&ii->xattr_sem);
1389 #endif
1390 inode_init_once(&ii->vfs_inode);
1391 }
1392
nilfs_segbuf_init_once(void * obj)1393 static void nilfs_segbuf_init_once(void *obj)
1394 {
1395 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1396 }
1397
nilfs_destroy_cachep(void)1398 static void nilfs_destroy_cachep(void)
1399 {
1400 /*
1401 * Make sure all delayed rcu free inodes are flushed before we
1402 * destroy cache.
1403 */
1404 rcu_barrier();
1405
1406 kmem_cache_destroy(nilfs_inode_cachep);
1407 kmem_cache_destroy(nilfs_transaction_cachep);
1408 kmem_cache_destroy(nilfs_segbuf_cachep);
1409 kmem_cache_destroy(nilfs_btree_path_cache);
1410 }
1411
nilfs_init_cachep(void)1412 static int __init nilfs_init_cachep(void)
1413 {
1414 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1415 sizeof(struct nilfs_inode_info), 0,
1416 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1417 nilfs_inode_init_once);
1418 if (!nilfs_inode_cachep)
1419 goto fail;
1420
1421 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1422 sizeof(struct nilfs_transaction_info), 0,
1423 SLAB_RECLAIM_ACCOUNT, NULL);
1424 if (!nilfs_transaction_cachep)
1425 goto fail;
1426
1427 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1428 sizeof(struct nilfs_segment_buffer), 0,
1429 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1430 if (!nilfs_segbuf_cachep)
1431 goto fail;
1432
1433 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1434 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1435 0, 0, NULL);
1436 if (!nilfs_btree_path_cache)
1437 goto fail;
1438
1439 return 0;
1440
1441 fail:
1442 nilfs_destroy_cachep();
1443 return -ENOMEM;
1444 }
1445
init_nilfs_fs(void)1446 static int __init init_nilfs_fs(void)
1447 {
1448 int err;
1449
1450 err = nilfs_init_cachep();
1451 if (err)
1452 goto fail;
1453
1454 err = nilfs_sysfs_init();
1455 if (err)
1456 goto free_cachep;
1457
1458 err = register_filesystem(&nilfs_fs_type);
1459 if (err)
1460 goto deinit_sysfs_entry;
1461
1462 printk(KERN_INFO "NILFS version 2 loaded\n");
1463 return 0;
1464
1465 deinit_sysfs_entry:
1466 nilfs_sysfs_exit();
1467 free_cachep:
1468 nilfs_destroy_cachep();
1469 fail:
1470 return err;
1471 }
1472
exit_nilfs_fs(void)1473 static void __exit exit_nilfs_fs(void)
1474 {
1475 nilfs_destroy_cachep();
1476 nilfs_sysfs_exit();
1477 unregister_filesystem(&nilfs_fs_type);
1478 }
1479
1480 module_init(init_nilfs_fs)
1481 module_exit(exit_nilfs_fs)
1482