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 * level,const char * fmt,...)65 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
66 ...)
67 {
68 struct va_format vaf;
69 va_list args;
70
71 va_start(args, fmt);
72 vaf.fmt = fmt;
73 vaf.va = &args;
74 if (sb)
75 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
76 else
77 printk("%sNILFS: %pV\n", level, &vaf);
78 va_end(args);
79 }
80
nilfs_set_error(struct super_block * sb)81 static void nilfs_set_error(struct super_block *sb)
82 {
83 struct the_nilfs *nilfs = sb->s_fs_info;
84 struct nilfs_super_block **sbp;
85
86 down_write(&nilfs->ns_sem);
87 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
88 nilfs->ns_mount_state |= NILFS_ERROR_FS;
89 sbp = nilfs_prepare_super(sb, 0);
90 if (likely(sbp)) {
91 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
92 if (sbp[1])
93 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
94 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
95 }
96 }
97 up_write(&nilfs->ns_sem);
98 }
99
100 /**
101 * __nilfs_error() - report failure condition on a filesystem
102 *
103 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
104 * reporting an error message. This function should be called when
105 * NILFS detects incoherences or defects of meta data on disk.
106 *
107 * This implements the body of nilfs_error() macro. Normally,
108 * nilfs_error() should be used. As for sustainable errors such as a
109 * single-shot I/O error, nilfs_err() should be used instead.
110 *
111 * Callers should not add a trailing newline since this will do it.
112 */
__nilfs_error(struct super_block * sb,const char * function,const char * fmt,...)113 void __nilfs_error(struct super_block *sb, const char *function,
114 const char *fmt, ...)
115 {
116 struct the_nilfs *nilfs = sb->s_fs_info;
117 struct va_format vaf;
118 va_list args;
119
120 va_start(args, fmt);
121
122 vaf.fmt = fmt;
123 vaf.va = &args;
124
125 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
126 sb->s_id, function, &vaf);
127
128 va_end(args);
129
130 if (!sb_rdonly(sb)) {
131 nilfs_set_error(sb);
132
133 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
134 printk(KERN_CRIT "Remounting filesystem read-only\n");
135 sb->s_flags |= SB_RDONLY;
136 }
137 }
138
139 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
140 panic("NILFS (device %s): panic forced after error\n",
141 sb->s_id);
142 }
143
nilfs_alloc_inode(struct super_block * sb)144 struct inode *nilfs_alloc_inode(struct super_block *sb)
145 {
146 struct nilfs_inode_info *ii;
147
148 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
149 if (!ii)
150 return NULL;
151 ii->i_bh = NULL;
152 ii->i_state = 0;
153 ii->i_cno = 0;
154 ii->i_assoc_inode = NULL;
155 ii->i_bmap = &ii->i_bmap_data;
156 return &ii->vfs_inode;
157 }
158
nilfs_free_inode(struct inode * inode)159 static void nilfs_free_inode(struct inode *inode)
160 {
161 if (nilfs_is_metadata_file_inode(inode))
162 nilfs_mdt_destroy(inode);
163
164 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
165 }
166
nilfs_sync_super(struct super_block * sb,int flag)167 static int nilfs_sync_super(struct super_block *sb, int flag)
168 {
169 struct the_nilfs *nilfs = sb->s_fs_info;
170 int err;
171
172 retry:
173 set_buffer_dirty(nilfs->ns_sbh[0]);
174 if (nilfs_test_opt(nilfs, BARRIER)) {
175 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
176 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
177 } else {
178 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
179 }
180
181 if (unlikely(err)) {
182 nilfs_err(sb, "unable to write superblock: err=%d", err);
183 if (err == -EIO && nilfs->ns_sbh[1]) {
184 /*
185 * sbp[0] points to newer log than sbp[1],
186 * so copy sbp[0] to sbp[1] to take over sbp[0].
187 */
188 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
189 nilfs->ns_sbsize);
190 nilfs_fall_back_super_block(nilfs);
191 goto retry;
192 }
193 } else {
194 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
195
196 nilfs->ns_sbwcount++;
197
198 /*
199 * The latest segment becomes trailable from the position
200 * written in superblock.
201 */
202 clear_nilfs_discontinued(nilfs);
203
204 /* update GC protection for recent segments */
205 if (nilfs->ns_sbh[1]) {
206 if (flag == NILFS_SB_COMMIT_ALL) {
207 set_buffer_dirty(nilfs->ns_sbh[1]);
208 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
209 goto out;
210 }
211 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
212 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
213 sbp = nilfs->ns_sbp[1];
214 }
215
216 spin_lock(&nilfs->ns_last_segment_lock);
217 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
218 spin_unlock(&nilfs->ns_last_segment_lock);
219 }
220 out:
221 return err;
222 }
223
nilfs_set_log_cursor(struct nilfs_super_block * sbp,struct the_nilfs * nilfs)224 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
225 struct the_nilfs *nilfs)
226 {
227 sector_t nfreeblocks;
228
229 /* nilfs->ns_sem must be locked by the caller. */
230 nilfs_count_free_blocks(nilfs, &nfreeblocks);
231 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
232
233 spin_lock(&nilfs->ns_last_segment_lock);
234 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
235 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
236 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
237 spin_unlock(&nilfs->ns_last_segment_lock);
238 }
239
nilfs_prepare_super(struct super_block * sb,int flip)240 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
241 int flip)
242 {
243 struct the_nilfs *nilfs = sb->s_fs_info;
244 struct nilfs_super_block **sbp = nilfs->ns_sbp;
245
246 /* nilfs->ns_sem must be locked by the caller. */
247 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
248 if (sbp[1] &&
249 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
250 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
251 } else {
252 nilfs_crit(sb, "superblock broke");
253 return NULL;
254 }
255 } else if (sbp[1] &&
256 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
257 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
258 }
259
260 if (flip && sbp[1])
261 nilfs_swap_super_block(nilfs);
262
263 return sbp;
264 }
265
nilfs_commit_super(struct super_block * sb,int flag)266 int nilfs_commit_super(struct super_block *sb, int flag)
267 {
268 struct the_nilfs *nilfs = sb->s_fs_info;
269 struct nilfs_super_block **sbp = nilfs->ns_sbp;
270 time64_t t;
271
272 /* nilfs->ns_sem must be locked by the caller. */
273 t = ktime_get_real_seconds();
274 nilfs->ns_sbwtime = t;
275 sbp[0]->s_wtime = cpu_to_le64(t);
276 sbp[0]->s_sum = 0;
277 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
278 (unsigned char *)sbp[0],
279 nilfs->ns_sbsize));
280 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
281 sbp[1]->s_wtime = sbp[0]->s_wtime;
282 sbp[1]->s_sum = 0;
283 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
284 (unsigned char *)sbp[1],
285 nilfs->ns_sbsize));
286 }
287 clear_nilfs_sb_dirty(nilfs);
288 nilfs->ns_flushed_device = 1;
289 /* make sure store to ns_flushed_device cannot be reordered */
290 smp_wmb();
291 return nilfs_sync_super(sb, flag);
292 }
293
294 /**
295 * nilfs_cleanup_super() - write filesystem state for cleanup
296 * @sb: super block instance to be unmounted or degraded to read-only
297 *
298 * This function restores state flags in the on-disk super block.
299 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
300 * filesystem was not clean previously.
301 */
nilfs_cleanup_super(struct super_block * sb)302 int nilfs_cleanup_super(struct super_block *sb)
303 {
304 struct the_nilfs *nilfs = sb->s_fs_info;
305 struct nilfs_super_block **sbp;
306 int flag = NILFS_SB_COMMIT;
307 int ret = -EIO;
308
309 sbp = nilfs_prepare_super(sb, 0);
310 if (sbp) {
311 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
312 nilfs_set_log_cursor(sbp[0], nilfs);
313 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
314 /*
315 * make the "clean" flag also to the opposite
316 * super block if both super blocks point to
317 * the same checkpoint.
318 */
319 sbp[1]->s_state = sbp[0]->s_state;
320 flag = NILFS_SB_COMMIT_ALL;
321 }
322 ret = nilfs_commit_super(sb, flag);
323 }
324 return ret;
325 }
326
327 /**
328 * nilfs_move_2nd_super - relocate secondary super block
329 * @sb: super block instance
330 * @sb2off: new offset of the secondary super block (in bytes)
331 */
nilfs_move_2nd_super(struct super_block * sb,loff_t sb2off)332 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
333 {
334 struct the_nilfs *nilfs = sb->s_fs_info;
335 struct buffer_head *nsbh;
336 struct nilfs_super_block *nsbp;
337 sector_t blocknr, newblocknr;
338 unsigned long offset;
339 int sb2i; /* array index of the secondary superblock */
340 int ret = 0;
341
342 /* nilfs->ns_sem must be locked by the caller. */
343 if (nilfs->ns_sbh[1] &&
344 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
345 sb2i = 1;
346 blocknr = nilfs->ns_sbh[1]->b_blocknr;
347 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
348 sb2i = 0;
349 blocknr = nilfs->ns_sbh[0]->b_blocknr;
350 } else {
351 sb2i = -1;
352 blocknr = 0;
353 }
354 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
355 goto out; /* super block location is unchanged */
356
357 /* Get new super block buffer */
358 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
359 offset = sb2off & (nilfs->ns_blocksize - 1);
360 nsbh = sb_getblk(sb, newblocknr);
361 if (!nsbh) {
362 nilfs_warn(sb,
363 "unable to move secondary superblock to block %llu",
364 (unsigned long long)newblocknr);
365 ret = -EIO;
366 goto out;
367 }
368 nsbp = (void *)nsbh->b_data + offset;
369
370 lock_buffer(nsbh);
371 if (sb2i >= 0) {
372 /*
373 * The position of the second superblock only changes by 4KiB,
374 * which is larger than the maximum superblock data size
375 * (= 1KiB), so there is no need to use memmove() to allow
376 * overlap between source and destination.
377 */
378 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
379
380 /*
381 * Zero fill after copy to avoid overwriting in case of move
382 * within the same block.
383 */
384 memset(nsbh->b_data, 0, offset);
385 memset((void *)nsbp + nilfs->ns_sbsize, 0,
386 nsbh->b_size - offset - nilfs->ns_sbsize);
387 } else {
388 memset(nsbh->b_data, 0, nsbh->b_size);
389 }
390 set_buffer_uptodate(nsbh);
391 unlock_buffer(nsbh);
392
393 if (sb2i >= 0) {
394 brelse(nilfs->ns_sbh[sb2i]);
395 nilfs->ns_sbh[sb2i] = nsbh;
396 nilfs->ns_sbp[sb2i] = nsbp;
397 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
398 /* secondary super block will be restored to index 1 */
399 nilfs->ns_sbh[1] = nsbh;
400 nilfs->ns_sbp[1] = nsbp;
401 } else {
402 brelse(nsbh);
403 }
404 out:
405 return ret;
406 }
407
408 /**
409 * nilfs_resize_fs - resize the filesystem
410 * @sb: super block instance
411 * @newsize: new size of the filesystem (in bytes)
412 */
nilfs_resize_fs(struct super_block * sb,__u64 newsize)413 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
414 {
415 struct the_nilfs *nilfs = sb->s_fs_info;
416 struct nilfs_super_block **sbp;
417 __u64 devsize, newnsegs;
418 loff_t sb2off;
419 int ret;
420
421 ret = -ERANGE;
422 devsize = i_size_read(sb->s_bdev->bd_inode);
423 if (newsize > devsize)
424 goto out;
425
426 /*
427 * Prevent underflow in second superblock position calculation.
428 * The exact minimum size check is done in nilfs_sufile_resize().
429 */
430 if (newsize < 4096) {
431 ret = -ENOSPC;
432 goto out;
433 }
434
435 /*
436 * Write lock is required to protect some functions depending
437 * on the number of segments, the number of reserved segments,
438 * and so forth.
439 */
440 down_write(&nilfs->ns_segctor_sem);
441
442 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
443 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
444 do_div(newnsegs, nilfs->ns_blocks_per_segment);
445
446 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
447 up_write(&nilfs->ns_segctor_sem);
448 if (ret < 0)
449 goto out;
450
451 ret = nilfs_construct_segment(sb);
452 if (ret < 0)
453 goto out;
454
455 down_write(&nilfs->ns_sem);
456 nilfs_move_2nd_super(sb, sb2off);
457 ret = -EIO;
458 sbp = nilfs_prepare_super(sb, 0);
459 if (likely(sbp)) {
460 nilfs_set_log_cursor(sbp[0], nilfs);
461 /*
462 * Drop NILFS_RESIZE_FS flag for compatibility with
463 * mount-time resize which may be implemented in a
464 * future release.
465 */
466 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
467 ~NILFS_RESIZE_FS);
468 sbp[0]->s_dev_size = cpu_to_le64(newsize);
469 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
470 if (sbp[1])
471 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
472 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
473 }
474 up_write(&nilfs->ns_sem);
475
476 /*
477 * Reset the range of allocatable segments last. This order
478 * is important in the case of expansion because the secondary
479 * superblock must be protected from log write until migration
480 * completes.
481 */
482 if (!ret)
483 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
484 out:
485 return ret;
486 }
487
nilfs_put_super(struct super_block * sb)488 static void nilfs_put_super(struct super_block *sb)
489 {
490 struct the_nilfs *nilfs = sb->s_fs_info;
491
492 nilfs_detach_log_writer(sb);
493
494 if (!sb_rdonly(sb)) {
495 down_write(&nilfs->ns_sem);
496 nilfs_cleanup_super(sb);
497 up_write(&nilfs->ns_sem);
498 }
499
500 nilfs_sysfs_delete_device_group(nilfs);
501 iput(nilfs->ns_sufile);
502 iput(nilfs->ns_cpfile);
503 iput(nilfs->ns_dat);
504
505 destroy_nilfs(nilfs);
506 sb->s_fs_info = NULL;
507 }
508
nilfs_sync_fs(struct super_block * sb,int wait)509 static int nilfs_sync_fs(struct super_block *sb, int wait)
510 {
511 struct the_nilfs *nilfs = sb->s_fs_info;
512 struct nilfs_super_block **sbp;
513 int err = 0;
514
515 /* This function is called when super block should be written back */
516 if (wait)
517 err = nilfs_construct_segment(sb);
518
519 down_write(&nilfs->ns_sem);
520 if (nilfs_sb_dirty(nilfs)) {
521 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
522 if (likely(sbp)) {
523 nilfs_set_log_cursor(sbp[0], nilfs);
524 nilfs_commit_super(sb, NILFS_SB_COMMIT);
525 }
526 }
527 up_write(&nilfs->ns_sem);
528
529 if (!err)
530 err = nilfs_flush_device(nilfs);
531
532 return err;
533 }
534
nilfs_attach_checkpoint(struct super_block * sb,__u64 cno,int curr_mnt,struct nilfs_root ** rootp)535 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
536 struct nilfs_root **rootp)
537 {
538 struct the_nilfs *nilfs = sb->s_fs_info;
539 struct nilfs_root *root;
540 struct nilfs_checkpoint *raw_cp;
541 struct buffer_head *bh_cp;
542 int err = -ENOMEM;
543
544 root = nilfs_find_or_create_root(
545 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
546 if (!root)
547 return err;
548
549 if (root->ifile)
550 goto reuse; /* already attached checkpoint */
551
552 down_read(&nilfs->ns_segctor_sem);
553 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
554 &bh_cp);
555 up_read(&nilfs->ns_segctor_sem);
556 if (unlikely(err)) {
557 if (err == -ENOENT || err == -EINVAL) {
558 nilfs_err(sb,
559 "Invalid checkpoint (checkpoint number=%llu)",
560 (unsigned long long)cno);
561 err = -EINVAL;
562 }
563 goto failed;
564 }
565
566 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
567 &raw_cp->cp_ifile_inode, &root->ifile);
568 if (err)
569 goto failed_bh;
570
571 atomic64_set(&root->inodes_count,
572 le64_to_cpu(raw_cp->cp_inodes_count));
573 atomic64_set(&root->blocks_count,
574 le64_to_cpu(raw_cp->cp_blocks_count));
575
576 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
577
578 reuse:
579 *rootp = root;
580 return 0;
581
582 failed_bh:
583 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
584 failed:
585 nilfs_put_root(root);
586
587 return err;
588 }
589
nilfs_freeze(struct super_block * sb)590 static int nilfs_freeze(struct super_block *sb)
591 {
592 struct the_nilfs *nilfs = sb->s_fs_info;
593 int err;
594
595 if (sb_rdonly(sb))
596 return 0;
597
598 /* Mark super block clean */
599 down_write(&nilfs->ns_sem);
600 err = nilfs_cleanup_super(sb);
601 up_write(&nilfs->ns_sem);
602 return err;
603 }
604
nilfs_unfreeze(struct super_block * sb)605 static int nilfs_unfreeze(struct super_block *sb)
606 {
607 struct the_nilfs *nilfs = sb->s_fs_info;
608
609 if (sb_rdonly(sb))
610 return 0;
611
612 down_write(&nilfs->ns_sem);
613 nilfs_setup_super(sb, false);
614 up_write(&nilfs->ns_sem);
615 return 0;
616 }
617
nilfs_statfs(struct dentry * dentry,struct kstatfs * buf)618 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
619 {
620 struct super_block *sb = dentry->d_sb;
621 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
622 struct the_nilfs *nilfs = root->nilfs;
623 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
624 unsigned long long blocks;
625 unsigned long overhead;
626 unsigned long nrsvblocks;
627 sector_t nfreeblocks;
628 u64 nmaxinodes, nfreeinodes;
629 int err;
630
631 /*
632 * Compute all of the segment blocks
633 *
634 * The blocks before first segment and after last segment
635 * are excluded.
636 */
637 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
638 - nilfs->ns_first_data_block;
639 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
640
641 /*
642 * Compute the overhead
643 *
644 * When distributing meta data blocks outside segment structure,
645 * We must count them as the overhead.
646 */
647 overhead = 0;
648
649 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
650 if (unlikely(err))
651 return err;
652
653 err = nilfs_ifile_count_free_inodes(root->ifile,
654 &nmaxinodes, &nfreeinodes);
655 if (unlikely(err)) {
656 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
657 if (err == -ERANGE) {
658 /*
659 * If nilfs_palloc_count_max_entries() returns
660 * -ERANGE error code then we simply treat
661 * curent inodes count as maximum possible and
662 * zero as free inodes value.
663 */
664 nmaxinodes = atomic64_read(&root->inodes_count);
665 nfreeinodes = 0;
666 err = 0;
667 } else
668 return err;
669 }
670
671 buf->f_type = NILFS_SUPER_MAGIC;
672 buf->f_bsize = sb->s_blocksize;
673 buf->f_blocks = blocks - overhead;
674 buf->f_bfree = nfreeblocks;
675 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
676 (buf->f_bfree - nrsvblocks) : 0;
677 buf->f_files = nmaxinodes;
678 buf->f_ffree = nfreeinodes;
679 buf->f_namelen = NILFS_NAME_LEN;
680 buf->f_fsid.val[0] = (u32)id;
681 buf->f_fsid.val[1] = (u32)(id >> 32);
682
683 return 0;
684 }
685
nilfs_show_options(struct seq_file * seq,struct dentry * dentry)686 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
687 {
688 struct super_block *sb = dentry->d_sb;
689 struct the_nilfs *nilfs = sb->s_fs_info;
690 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
691
692 if (!nilfs_test_opt(nilfs, BARRIER))
693 seq_puts(seq, ",nobarrier");
694 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
695 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
696 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
697 seq_puts(seq, ",errors=panic");
698 if (nilfs_test_opt(nilfs, ERRORS_CONT))
699 seq_puts(seq, ",errors=continue");
700 if (nilfs_test_opt(nilfs, STRICT_ORDER))
701 seq_puts(seq, ",order=strict");
702 if (nilfs_test_opt(nilfs, NORECOVERY))
703 seq_puts(seq, ",norecovery");
704 if (nilfs_test_opt(nilfs, DISCARD))
705 seq_puts(seq, ",discard");
706
707 return 0;
708 }
709
710 static const struct super_operations nilfs_sops = {
711 .alloc_inode = nilfs_alloc_inode,
712 .free_inode = nilfs_free_inode,
713 .dirty_inode = nilfs_dirty_inode,
714 .evict_inode = nilfs_evict_inode,
715 .put_super = nilfs_put_super,
716 .sync_fs = nilfs_sync_fs,
717 .freeze_fs = nilfs_freeze,
718 .unfreeze_fs = nilfs_unfreeze,
719 .statfs = nilfs_statfs,
720 .remount_fs = nilfs_remount,
721 .show_options = nilfs_show_options
722 };
723
724 enum {
725 Opt_err_cont, Opt_err_panic, Opt_err_ro,
726 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
727 Opt_discard, Opt_nodiscard, Opt_err,
728 };
729
730 static match_table_t tokens = {
731 {Opt_err_cont, "errors=continue"},
732 {Opt_err_panic, "errors=panic"},
733 {Opt_err_ro, "errors=remount-ro"},
734 {Opt_barrier, "barrier"},
735 {Opt_nobarrier, "nobarrier"},
736 {Opt_snapshot, "cp=%u"},
737 {Opt_order, "order=%s"},
738 {Opt_norecovery, "norecovery"},
739 {Opt_discard, "discard"},
740 {Opt_nodiscard, "nodiscard"},
741 {Opt_err, NULL}
742 };
743
parse_options(char * options,struct super_block * sb,int is_remount)744 static int parse_options(char *options, struct super_block *sb, int is_remount)
745 {
746 struct the_nilfs *nilfs = sb->s_fs_info;
747 char *p;
748 substring_t args[MAX_OPT_ARGS];
749
750 if (!options)
751 return 1;
752
753 while ((p = strsep(&options, ",")) != NULL) {
754 int token;
755
756 if (!*p)
757 continue;
758
759 token = match_token(p, tokens, args);
760 switch (token) {
761 case Opt_barrier:
762 nilfs_set_opt(nilfs, BARRIER);
763 break;
764 case Opt_nobarrier:
765 nilfs_clear_opt(nilfs, BARRIER);
766 break;
767 case Opt_order:
768 if (strcmp(args[0].from, "relaxed") == 0)
769 /* Ordered data semantics */
770 nilfs_clear_opt(nilfs, STRICT_ORDER);
771 else if (strcmp(args[0].from, "strict") == 0)
772 /* Strict in-order semantics */
773 nilfs_set_opt(nilfs, STRICT_ORDER);
774 else
775 return 0;
776 break;
777 case Opt_err_panic:
778 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
779 break;
780 case Opt_err_ro:
781 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
782 break;
783 case Opt_err_cont:
784 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
785 break;
786 case Opt_snapshot:
787 if (is_remount) {
788 nilfs_err(sb,
789 "\"%s\" option is invalid for remount",
790 p);
791 return 0;
792 }
793 break;
794 case Opt_norecovery:
795 nilfs_set_opt(nilfs, NORECOVERY);
796 break;
797 case Opt_discard:
798 nilfs_set_opt(nilfs, DISCARD);
799 break;
800 case Opt_nodiscard:
801 nilfs_clear_opt(nilfs, DISCARD);
802 break;
803 default:
804 nilfs_err(sb, "unrecognized mount option \"%s\"", p);
805 return 0;
806 }
807 }
808 return 1;
809 }
810
811 static inline void
nilfs_set_default_options(struct super_block * sb,struct nilfs_super_block * sbp)812 nilfs_set_default_options(struct super_block *sb,
813 struct nilfs_super_block *sbp)
814 {
815 struct the_nilfs *nilfs = sb->s_fs_info;
816
817 nilfs->ns_mount_opt =
818 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
819 }
820
nilfs_setup_super(struct super_block * sb,int is_mount)821 static int nilfs_setup_super(struct super_block *sb, int is_mount)
822 {
823 struct the_nilfs *nilfs = sb->s_fs_info;
824 struct nilfs_super_block **sbp;
825 int max_mnt_count;
826 int mnt_count;
827
828 /* nilfs->ns_sem must be locked by the caller. */
829 sbp = nilfs_prepare_super(sb, 0);
830 if (!sbp)
831 return -EIO;
832
833 if (!is_mount)
834 goto skip_mount_setup;
835
836 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
837 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
838
839 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
840 nilfs_warn(sb, "mounting fs with errors");
841 #if 0
842 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
843 nilfs_warn(sb, "maximal mount count reached");
844 #endif
845 }
846 if (!max_mnt_count)
847 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
848
849 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
850 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
851
852 skip_mount_setup:
853 sbp[0]->s_state =
854 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
855 /* synchronize sbp[1] with sbp[0] */
856 if (sbp[1])
857 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
858 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
859 }
860
nilfs_read_super_block(struct super_block * sb,u64 pos,int blocksize,struct buffer_head ** pbh)861 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
862 u64 pos, int blocksize,
863 struct buffer_head **pbh)
864 {
865 unsigned long long sb_index = pos;
866 unsigned long offset;
867
868 offset = do_div(sb_index, blocksize);
869 *pbh = sb_bread(sb, sb_index);
870 if (!*pbh)
871 return NULL;
872 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
873 }
874
nilfs_store_magic_and_option(struct super_block * sb,struct nilfs_super_block * sbp,char * data)875 int nilfs_store_magic_and_option(struct super_block *sb,
876 struct nilfs_super_block *sbp,
877 char *data)
878 {
879 struct the_nilfs *nilfs = sb->s_fs_info;
880
881 sb->s_magic = le16_to_cpu(sbp->s_magic);
882
883 /* FS independent flags */
884 #ifdef NILFS_ATIME_DISABLE
885 sb->s_flags |= SB_NOATIME;
886 #endif
887
888 nilfs_set_default_options(sb, sbp);
889
890 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
891 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
892 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
893 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
894
895 return !parse_options(data, sb, 0) ? -EINVAL : 0;
896 }
897
nilfs_check_feature_compatibility(struct super_block * sb,struct nilfs_super_block * sbp)898 int nilfs_check_feature_compatibility(struct super_block *sb,
899 struct nilfs_super_block *sbp)
900 {
901 __u64 features;
902
903 features = le64_to_cpu(sbp->s_feature_incompat) &
904 ~NILFS_FEATURE_INCOMPAT_SUPP;
905 if (features) {
906 nilfs_err(sb,
907 "couldn't mount because of unsupported optional features (%llx)",
908 (unsigned long long)features);
909 return -EINVAL;
910 }
911 features = le64_to_cpu(sbp->s_feature_compat_ro) &
912 ~NILFS_FEATURE_COMPAT_RO_SUPP;
913 if (!sb_rdonly(sb) && features) {
914 nilfs_err(sb,
915 "couldn't mount RDWR because of unsupported optional features (%llx)",
916 (unsigned long long)features);
917 return -EINVAL;
918 }
919 return 0;
920 }
921
nilfs_get_root_dentry(struct super_block * sb,struct nilfs_root * root,struct dentry ** root_dentry)922 static int nilfs_get_root_dentry(struct super_block *sb,
923 struct nilfs_root *root,
924 struct dentry **root_dentry)
925 {
926 struct inode *inode;
927 struct dentry *dentry;
928 int ret = 0;
929
930 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
931 if (IS_ERR(inode)) {
932 ret = PTR_ERR(inode);
933 nilfs_err(sb, "error %d getting root inode", ret);
934 goto out;
935 }
936 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
937 iput(inode);
938 nilfs_err(sb, "corrupt root inode");
939 ret = -EINVAL;
940 goto out;
941 }
942
943 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
944 dentry = d_find_alias(inode);
945 if (!dentry) {
946 dentry = d_make_root(inode);
947 if (!dentry) {
948 ret = -ENOMEM;
949 goto failed_dentry;
950 }
951 } else {
952 iput(inode);
953 }
954 } else {
955 dentry = d_obtain_root(inode);
956 if (IS_ERR(dentry)) {
957 ret = PTR_ERR(dentry);
958 goto failed_dentry;
959 }
960 }
961 *root_dentry = dentry;
962 out:
963 return ret;
964
965 failed_dentry:
966 nilfs_err(sb, "error %d getting root dentry", ret);
967 goto out;
968 }
969
nilfs_attach_snapshot(struct super_block * s,__u64 cno,struct dentry ** root_dentry)970 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
971 struct dentry **root_dentry)
972 {
973 struct the_nilfs *nilfs = s->s_fs_info;
974 struct nilfs_root *root;
975 int ret;
976
977 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
978
979 down_read(&nilfs->ns_segctor_sem);
980 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
981 up_read(&nilfs->ns_segctor_sem);
982 if (ret < 0) {
983 ret = (ret == -ENOENT) ? -EINVAL : ret;
984 goto out;
985 } else if (!ret) {
986 nilfs_err(s,
987 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
988 (unsigned long long)cno);
989 ret = -EINVAL;
990 goto out;
991 }
992
993 ret = nilfs_attach_checkpoint(s, cno, false, &root);
994 if (ret) {
995 nilfs_err(s,
996 "error %d while loading snapshot (checkpoint number=%llu)",
997 ret, (unsigned long long)cno);
998 goto out;
999 }
1000 ret = nilfs_get_root_dentry(s, root, root_dentry);
1001 nilfs_put_root(root);
1002 out:
1003 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1004 return ret;
1005 }
1006
1007 /**
1008 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1009 * @root_dentry: root dentry of the tree to be shrunk
1010 *
1011 * This function returns true if the tree was in-use.
1012 */
nilfs_tree_is_busy(struct dentry * root_dentry)1013 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1014 {
1015 shrink_dcache_parent(root_dentry);
1016 return d_count(root_dentry) > 1;
1017 }
1018
nilfs_checkpoint_is_mounted(struct super_block * sb,__u64 cno)1019 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1020 {
1021 struct the_nilfs *nilfs = sb->s_fs_info;
1022 struct nilfs_root *root;
1023 struct inode *inode;
1024 struct dentry *dentry;
1025 int ret;
1026
1027 if (cno > nilfs->ns_cno)
1028 return false;
1029
1030 if (cno >= nilfs_last_cno(nilfs))
1031 return true; /* protect recent checkpoints */
1032
1033 ret = false;
1034 root = nilfs_lookup_root(nilfs, cno);
1035 if (root) {
1036 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1037 if (inode) {
1038 dentry = d_find_alias(inode);
1039 if (dentry) {
1040 ret = nilfs_tree_is_busy(dentry);
1041 dput(dentry);
1042 }
1043 iput(inode);
1044 }
1045 nilfs_put_root(root);
1046 }
1047 return ret;
1048 }
1049
1050 /**
1051 * nilfs_fill_super() - initialize a super block instance
1052 * @sb: super_block
1053 * @data: mount options
1054 * @silent: silent mode flag
1055 *
1056 * This function is called exclusively by nilfs->ns_mount_mutex.
1057 * So, the recovery process is protected from other simultaneous mounts.
1058 */
1059 static int
nilfs_fill_super(struct super_block * sb,void * data,int silent)1060 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1061 {
1062 struct the_nilfs *nilfs;
1063 struct nilfs_root *fsroot;
1064 __u64 cno;
1065 int err;
1066
1067 nilfs = alloc_nilfs(sb);
1068 if (!nilfs)
1069 return -ENOMEM;
1070
1071 sb->s_fs_info = nilfs;
1072
1073 err = init_nilfs(nilfs, sb, (char *)data);
1074 if (err)
1075 goto failed_nilfs;
1076
1077 sb->s_op = &nilfs_sops;
1078 sb->s_export_op = &nilfs_export_ops;
1079 sb->s_root = NULL;
1080 sb->s_time_gran = 1;
1081 sb->s_max_links = NILFS_LINK_MAX;
1082
1083 sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1084
1085 err = load_nilfs(nilfs, sb);
1086 if (err)
1087 goto failed_nilfs;
1088
1089 cno = nilfs_last_cno(nilfs);
1090 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1091 if (err) {
1092 nilfs_err(sb,
1093 "error %d while loading last checkpoint (checkpoint number=%llu)",
1094 err, (unsigned long long)cno);
1095 goto failed_unload;
1096 }
1097
1098 if (!sb_rdonly(sb)) {
1099 err = nilfs_attach_log_writer(sb, fsroot);
1100 if (err)
1101 goto failed_checkpoint;
1102 }
1103
1104 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1105 if (err)
1106 goto failed_segctor;
1107
1108 nilfs_put_root(fsroot);
1109
1110 if (!sb_rdonly(sb)) {
1111 down_write(&nilfs->ns_sem);
1112 nilfs_setup_super(sb, true);
1113 up_write(&nilfs->ns_sem);
1114 }
1115
1116 return 0;
1117
1118 failed_segctor:
1119 nilfs_detach_log_writer(sb);
1120
1121 failed_checkpoint:
1122 nilfs_put_root(fsroot);
1123
1124 failed_unload:
1125 nilfs_sysfs_delete_device_group(nilfs);
1126 iput(nilfs->ns_sufile);
1127 iput(nilfs->ns_cpfile);
1128 iput(nilfs->ns_dat);
1129
1130 failed_nilfs:
1131 destroy_nilfs(nilfs);
1132 return err;
1133 }
1134
nilfs_remount(struct super_block * sb,int * flags,char * data)1135 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1136 {
1137 struct the_nilfs *nilfs = sb->s_fs_info;
1138 unsigned long old_sb_flags;
1139 unsigned long old_mount_opt;
1140 int err;
1141
1142 sync_filesystem(sb);
1143 old_sb_flags = sb->s_flags;
1144 old_mount_opt = nilfs->ns_mount_opt;
1145
1146 if (!parse_options(data, sb, 1)) {
1147 err = -EINVAL;
1148 goto restore_opts;
1149 }
1150 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1151
1152 err = -EINVAL;
1153
1154 if (!nilfs_valid_fs(nilfs)) {
1155 nilfs_warn(sb,
1156 "couldn't remount because the filesystem is in an incomplete recovery state");
1157 goto restore_opts;
1158 }
1159
1160 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1161 goto out;
1162 if (*flags & SB_RDONLY) {
1163 sb->s_flags |= SB_RDONLY;
1164
1165 /*
1166 * Remounting a valid RW partition RDONLY, so set
1167 * the RDONLY flag and then mark the partition as valid again.
1168 */
1169 down_write(&nilfs->ns_sem);
1170 nilfs_cleanup_super(sb);
1171 up_write(&nilfs->ns_sem);
1172 } else {
1173 __u64 features;
1174 struct nilfs_root *root;
1175
1176 /*
1177 * Mounting a RDONLY partition read-write, so reread and
1178 * store the current valid flag. (It may have been changed
1179 * by fsck since we originally mounted the partition.)
1180 */
1181 down_read(&nilfs->ns_sem);
1182 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1183 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1184 up_read(&nilfs->ns_sem);
1185 if (features) {
1186 nilfs_warn(sb,
1187 "couldn't remount RDWR because of unsupported optional features (%llx)",
1188 (unsigned long long)features);
1189 err = -EROFS;
1190 goto restore_opts;
1191 }
1192
1193 sb->s_flags &= ~SB_RDONLY;
1194
1195 root = NILFS_I(d_inode(sb->s_root))->i_root;
1196 err = nilfs_attach_log_writer(sb, root);
1197 if (err)
1198 goto restore_opts;
1199
1200 down_write(&nilfs->ns_sem);
1201 nilfs_setup_super(sb, true);
1202 up_write(&nilfs->ns_sem);
1203 }
1204 out:
1205 return 0;
1206
1207 restore_opts:
1208 sb->s_flags = old_sb_flags;
1209 nilfs->ns_mount_opt = old_mount_opt;
1210 return err;
1211 }
1212
1213 struct nilfs_super_data {
1214 struct block_device *bdev;
1215 __u64 cno;
1216 int flags;
1217 };
1218
nilfs_parse_snapshot_option(const char * option,const substring_t * arg,struct nilfs_super_data * sd)1219 static int nilfs_parse_snapshot_option(const char *option,
1220 const substring_t *arg,
1221 struct nilfs_super_data *sd)
1222 {
1223 unsigned long long val;
1224 const char *msg = NULL;
1225 int err;
1226
1227 if (!(sd->flags & SB_RDONLY)) {
1228 msg = "read-only option is not specified";
1229 goto parse_error;
1230 }
1231
1232 err = kstrtoull(arg->from, 0, &val);
1233 if (err) {
1234 if (err == -ERANGE)
1235 msg = "too large checkpoint number";
1236 else
1237 msg = "malformed argument";
1238 goto parse_error;
1239 } else if (val == 0) {
1240 msg = "invalid checkpoint number 0";
1241 goto parse_error;
1242 }
1243 sd->cno = val;
1244 return 0;
1245
1246 parse_error:
1247 nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1248 return 1;
1249 }
1250
1251 /**
1252 * nilfs_identify - pre-read mount options needed to identify mount instance
1253 * @data: mount options
1254 * @sd: nilfs_super_data
1255 */
nilfs_identify(char * data,struct nilfs_super_data * sd)1256 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1257 {
1258 char *p, *options = data;
1259 substring_t args[MAX_OPT_ARGS];
1260 int token;
1261 int ret = 0;
1262
1263 do {
1264 p = strsep(&options, ",");
1265 if (p != NULL && *p) {
1266 token = match_token(p, tokens, args);
1267 if (token == Opt_snapshot)
1268 ret = nilfs_parse_snapshot_option(p, &args[0],
1269 sd);
1270 }
1271 if (!options)
1272 break;
1273 BUG_ON(options == data);
1274 *(options - 1) = ',';
1275 } while (!ret);
1276 return ret;
1277 }
1278
nilfs_set_bdev_super(struct super_block * s,void * data)1279 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1280 {
1281 s->s_bdev = data;
1282 s->s_dev = s->s_bdev->bd_dev;
1283 return 0;
1284 }
1285
nilfs_test_bdev_super(struct super_block * s,void * data)1286 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1287 {
1288 return (void *)s->s_bdev == data;
1289 }
1290
1291 static struct dentry *
nilfs_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1292 nilfs_mount(struct file_system_type *fs_type, int flags,
1293 const char *dev_name, void *data)
1294 {
1295 struct nilfs_super_data sd;
1296 struct super_block *s;
1297 fmode_t mode = FMODE_READ | FMODE_EXCL;
1298 struct dentry *root_dentry;
1299 int err, s_new = false;
1300
1301 if (!(flags & SB_RDONLY))
1302 mode |= FMODE_WRITE;
1303
1304 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1305 if (IS_ERR(sd.bdev))
1306 return ERR_CAST(sd.bdev);
1307
1308 sd.cno = 0;
1309 sd.flags = flags;
1310 if (nilfs_identify((char *)data, &sd)) {
1311 err = -EINVAL;
1312 goto failed;
1313 }
1314
1315 /*
1316 * once the super is inserted into the list by sget, s_umount
1317 * will protect the lockfs code from trying to start a snapshot
1318 * while we are mounting
1319 */
1320 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1321 if (sd.bdev->bd_fsfreeze_count > 0) {
1322 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1323 err = -EBUSY;
1324 goto failed;
1325 }
1326 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1327 sd.bdev);
1328 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1329 if (IS_ERR(s)) {
1330 err = PTR_ERR(s);
1331 goto failed;
1332 }
1333
1334 if (!s->s_root) {
1335 s_new = true;
1336
1337 /* New superblock instance created */
1338 s->s_mode = mode;
1339 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1340 sb_set_blocksize(s, block_size(sd.bdev));
1341
1342 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1343 if (err)
1344 goto failed_super;
1345
1346 s->s_flags |= SB_ACTIVE;
1347 } else if (!sd.cno) {
1348 if (nilfs_tree_is_busy(s->s_root)) {
1349 if ((flags ^ s->s_flags) & SB_RDONLY) {
1350 nilfs_err(s,
1351 "the device already has a %s mount.",
1352 sb_rdonly(s) ? "read-only" : "read/write");
1353 err = -EBUSY;
1354 goto failed_super;
1355 }
1356 } else {
1357 /*
1358 * Try remount to setup mount states if the current
1359 * tree is not mounted and only snapshots use this sb.
1360 */
1361 err = nilfs_remount(s, &flags, data);
1362 if (err)
1363 goto failed_super;
1364 }
1365 }
1366
1367 if (sd.cno) {
1368 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1369 if (err)
1370 goto failed_super;
1371 } else {
1372 root_dentry = dget(s->s_root);
1373 }
1374
1375 if (!s_new)
1376 blkdev_put(sd.bdev, mode);
1377
1378 return root_dentry;
1379
1380 failed_super:
1381 deactivate_locked_super(s);
1382
1383 failed:
1384 if (!s_new)
1385 blkdev_put(sd.bdev, mode);
1386 return ERR_PTR(err);
1387 }
1388
1389 struct file_system_type nilfs_fs_type = {
1390 .owner = THIS_MODULE,
1391 .name = "nilfs2",
1392 .mount = nilfs_mount,
1393 .kill_sb = kill_block_super,
1394 .fs_flags = FS_REQUIRES_DEV,
1395 };
1396 MODULE_ALIAS_FS("nilfs2");
1397
nilfs_inode_init_once(void * obj)1398 static void nilfs_inode_init_once(void *obj)
1399 {
1400 struct nilfs_inode_info *ii = obj;
1401
1402 INIT_LIST_HEAD(&ii->i_dirty);
1403 #ifdef CONFIG_NILFS_XATTR
1404 init_rwsem(&ii->xattr_sem);
1405 #endif
1406 inode_init_once(&ii->vfs_inode);
1407 }
1408
nilfs_segbuf_init_once(void * obj)1409 static void nilfs_segbuf_init_once(void *obj)
1410 {
1411 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1412 }
1413
nilfs_destroy_cachep(void)1414 static void nilfs_destroy_cachep(void)
1415 {
1416 /*
1417 * Make sure all delayed rcu free inodes are flushed before we
1418 * destroy cache.
1419 */
1420 rcu_barrier();
1421
1422 kmem_cache_destroy(nilfs_inode_cachep);
1423 kmem_cache_destroy(nilfs_transaction_cachep);
1424 kmem_cache_destroy(nilfs_segbuf_cachep);
1425 kmem_cache_destroy(nilfs_btree_path_cache);
1426 }
1427
nilfs_init_cachep(void)1428 static int __init nilfs_init_cachep(void)
1429 {
1430 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1431 sizeof(struct nilfs_inode_info), 0,
1432 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1433 nilfs_inode_init_once);
1434 if (!nilfs_inode_cachep)
1435 goto fail;
1436
1437 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1438 sizeof(struct nilfs_transaction_info), 0,
1439 SLAB_RECLAIM_ACCOUNT, NULL);
1440 if (!nilfs_transaction_cachep)
1441 goto fail;
1442
1443 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1444 sizeof(struct nilfs_segment_buffer), 0,
1445 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1446 if (!nilfs_segbuf_cachep)
1447 goto fail;
1448
1449 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1450 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1451 0, 0, NULL);
1452 if (!nilfs_btree_path_cache)
1453 goto fail;
1454
1455 return 0;
1456
1457 fail:
1458 nilfs_destroy_cachep();
1459 return -ENOMEM;
1460 }
1461
init_nilfs_fs(void)1462 static int __init init_nilfs_fs(void)
1463 {
1464 int err;
1465
1466 err = nilfs_init_cachep();
1467 if (err)
1468 goto fail;
1469
1470 err = nilfs_sysfs_init();
1471 if (err)
1472 goto free_cachep;
1473
1474 err = register_filesystem(&nilfs_fs_type);
1475 if (err)
1476 goto deinit_sysfs_entry;
1477
1478 printk(KERN_INFO "NILFS version 2 loaded\n");
1479 return 0;
1480
1481 deinit_sysfs_entry:
1482 nilfs_sysfs_exit();
1483 free_cachep:
1484 nilfs_destroy_cachep();
1485 fail:
1486 return err;
1487 }
1488
exit_nilfs_fs(void)1489 static void __exit exit_nilfs_fs(void)
1490 {
1491 nilfs_destroy_cachep();
1492 nilfs_sysfs_exit();
1493 unregister_filesystem(&nilfs_fs_type);
1494 }
1495
1496 module_init(init_nilfs_fs)
1497 module_exit(exit_nilfs_fs)
1498