1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * NILFS module and super block management.
4 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Written by Ryusuke Konishi.
8 */
9 /*
10 * linux/fs/ext2/super.c
11 *
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise Pascal
15 * Universite Pierre et Marie Curie (Paris VI)
16 *
17 * from
18 *
19 * linux/fs/minix/inode.c
20 *
21 * Copyright (C) 1991, 1992 Linus Torvalds
22 *
23 * Big-endian to little-endian byte-swapping/bitmaps by
24 * David S. Miller (davem@caip.rutgers.edu), 1995
25 */
26
27 #include <linux/module.h>
28 #include <linux/string.h>
29 #include <linux/slab.h>
30 #include <linux/init.h>
31 #include <linux/blkdev.h>
32 #include <linux/parser.h>
33 #include <linux/crc32.h>
34 #include <linux/vfs.h>
35 #include <linux/writeback.h>
36 #include <linux/seq_file.h>
37 #include <linux/mount.h>
38 #include <linux/fs_context.h>
39 #include "nilfs.h"
40 #include "export.h"
41 #include "mdt.h"
42 #include "alloc.h"
43 #include "btree.h"
44 #include "btnode.h"
45 #include "page.h"
46 #include "cpfile.h"
47 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
48 #include "ifile.h"
49 #include "dat.h"
50 #include "segment.h"
51 #include "segbuf.h"
52
53 MODULE_AUTHOR("NTT Corp.");
54 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
55 "(NILFS)");
56 MODULE_LICENSE("GPL");
57 MODULE_IMPORT_NS(ANDROID_GKI_VFS_EXPORT_ONLY);
58
59 static struct kmem_cache *nilfs_inode_cachep;
60 struct kmem_cache *nilfs_transaction_cachep;
61 struct kmem_cache *nilfs_segbuf_cachep;
62 struct kmem_cache *nilfs_btree_path_cache;
63
64 static int nilfs_setup_super(struct super_block *sb, int is_mount);
65 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
66
__nilfs_msg(struct super_block * sb,const char * fmt,...)67 void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
68 {
69 struct va_format vaf;
70 va_list args;
71 int level;
72
73 va_start(args, fmt);
74
75 level = printk_get_level(fmt);
76 vaf.fmt = printk_skip_level(fmt);
77 vaf.va = &args;
78
79 if (sb)
80 printk("%c%cNILFS (%s): %pV\n",
81 KERN_SOH_ASCII, level, sb->s_id, &vaf);
82 else
83 printk("%c%cNILFS: %pV\n",
84 KERN_SOH_ASCII, level, &vaf);
85
86 va_end(args);
87 }
88
nilfs_set_error(struct super_block * sb)89 static void nilfs_set_error(struct super_block *sb)
90 {
91 struct the_nilfs *nilfs = sb->s_fs_info;
92 struct nilfs_super_block **sbp;
93
94 down_write(&nilfs->ns_sem);
95 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
96 nilfs->ns_mount_state |= NILFS_ERROR_FS;
97 sbp = nilfs_prepare_super(sb, 0);
98 if (likely(sbp)) {
99 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
100 if (sbp[1])
101 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
102 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
103 }
104 }
105 up_write(&nilfs->ns_sem);
106 }
107
108 /**
109 * __nilfs_error() - report failure condition on a filesystem
110 *
111 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
112 * reporting an error message. This function should be called when
113 * NILFS detects incoherences or defects of meta data on disk.
114 *
115 * This implements the body of nilfs_error() macro. Normally,
116 * nilfs_error() should be used. As for sustainable errors such as a
117 * single-shot I/O error, nilfs_err() should be used instead.
118 *
119 * Callers should not add a trailing newline since this will do it.
120 */
__nilfs_error(struct super_block * sb,const char * function,const char * fmt,...)121 void __nilfs_error(struct super_block *sb, const char *function,
122 const char *fmt, ...)
123 {
124 struct the_nilfs *nilfs = sb->s_fs_info;
125 struct va_format vaf;
126 va_list args;
127
128 va_start(args, fmt);
129
130 vaf.fmt = fmt;
131 vaf.va = &args;
132
133 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
134 sb->s_id, function, &vaf);
135
136 va_end(args);
137
138 if (!sb_rdonly(sb)) {
139 nilfs_set_error(sb);
140
141 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
142 printk(KERN_CRIT "Remounting filesystem read-only\n");
143 sb->s_flags |= SB_RDONLY;
144 }
145 }
146
147 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
148 panic("NILFS (device %s): panic forced after error\n",
149 sb->s_id);
150 }
151
nilfs_alloc_inode(struct super_block * sb)152 struct inode *nilfs_alloc_inode(struct super_block *sb)
153 {
154 struct nilfs_inode_info *ii;
155
156 ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
157 if (!ii)
158 return NULL;
159 ii->i_bh = NULL;
160 ii->i_state = 0;
161 ii->i_cno = 0;
162 ii->i_assoc_inode = NULL;
163 ii->i_bmap = &ii->i_bmap_data;
164 return &ii->vfs_inode;
165 }
166
nilfs_free_inode(struct inode * inode)167 static void nilfs_free_inode(struct inode *inode)
168 {
169 if (nilfs_is_metadata_file_inode(inode))
170 nilfs_mdt_destroy(inode);
171
172 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
173 }
174
nilfs_sync_super(struct super_block * sb,int flag)175 static int nilfs_sync_super(struct super_block *sb, int flag)
176 {
177 struct the_nilfs *nilfs = sb->s_fs_info;
178 int err;
179
180 retry:
181 set_buffer_dirty(nilfs->ns_sbh[0]);
182 if (nilfs_test_opt(nilfs, BARRIER)) {
183 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
184 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
185 } else {
186 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
187 }
188
189 if (unlikely(err)) {
190 nilfs_err(sb, "unable to write superblock: err=%d", err);
191 if (err == -EIO && nilfs->ns_sbh[1]) {
192 /*
193 * sbp[0] points to newer log than sbp[1],
194 * so copy sbp[0] to sbp[1] to take over sbp[0].
195 */
196 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
197 nilfs->ns_sbsize);
198 nilfs_fall_back_super_block(nilfs);
199 goto retry;
200 }
201 } else {
202 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
203
204 nilfs->ns_sbwcount++;
205
206 /*
207 * The latest segment becomes trailable from the position
208 * written in superblock.
209 */
210 clear_nilfs_discontinued(nilfs);
211
212 /* update GC protection for recent segments */
213 if (nilfs->ns_sbh[1]) {
214 if (flag == NILFS_SB_COMMIT_ALL) {
215 set_buffer_dirty(nilfs->ns_sbh[1]);
216 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
217 goto out;
218 }
219 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
220 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
221 sbp = nilfs->ns_sbp[1];
222 }
223
224 spin_lock(&nilfs->ns_last_segment_lock);
225 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
226 spin_unlock(&nilfs->ns_last_segment_lock);
227 }
228 out:
229 return err;
230 }
231
nilfs_set_log_cursor(struct nilfs_super_block * sbp,struct the_nilfs * nilfs)232 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
233 struct the_nilfs *nilfs)
234 {
235 sector_t nfreeblocks;
236
237 /* nilfs->ns_sem must be locked by the caller. */
238 nilfs_count_free_blocks(nilfs, &nfreeblocks);
239 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
240
241 spin_lock(&nilfs->ns_last_segment_lock);
242 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
243 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
244 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
245 spin_unlock(&nilfs->ns_last_segment_lock);
246 }
247
nilfs_prepare_super(struct super_block * sb,int flip)248 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
249 int flip)
250 {
251 struct the_nilfs *nilfs = sb->s_fs_info;
252 struct nilfs_super_block **sbp = nilfs->ns_sbp;
253
254 /* nilfs->ns_sem must be locked by the caller. */
255 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
256 if (sbp[1] &&
257 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
258 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
259 } else {
260 nilfs_crit(sb, "superblock broke");
261 return NULL;
262 }
263 } else if (sbp[1] &&
264 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
265 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
266 }
267
268 if (flip && sbp[1])
269 nilfs_swap_super_block(nilfs);
270
271 return sbp;
272 }
273
nilfs_commit_super(struct super_block * sb,int flag)274 int nilfs_commit_super(struct super_block *sb, int flag)
275 {
276 struct the_nilfs *nilfs = sb->s_fs_info;
277 struct nilfs_super_block **sbp = nilfs->ns_sbp;
278 time64_t t;
279
280 /* nilfs->ns_sem must be locked by the caller. */
281 t = ktime_get_real_seconds();
282 nilfs->ns_sbwtime = t;
283 sbp[0]->s_wtime = cpu_to_le64(t);
284 sbp[0]->s_sum = 0;
285 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
286 (unsigned char *)sbp[0],
287 nilfs->ns_sbsize));
288 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
289 sbp[1]->s_wtime = sbp[0]->s_wtime;
290 sbp[1]->s_sum = 0;
291 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
292 (unsigned char *)sbp[1],
293 nilfs->ns_sbsize));
294 }
295 clear_nilfs_sb_dirty(nilfs);
296 nilfs->ns_flushed_device = 1;
297 /* make sure store to ns_flushed_device cannot be reordered */
298 smp_wmb();
299 return nilfs_sync_super(sb, flag);
300 }
301
302 /**
303 * nilfs_cleanup_super() - write filesystem state for cleanup
304 * @sb: super block instance to be unmounted or degraded to read-only
305 *
306 * This function restores state flags in the on-disk super block.
307 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
308 * filesystem was not clean previously.
309 */
nilfs_cleanup_super(struct super_block * sb)310 int nilfs_cleanup_super(struct super_block *sb)
311 {
312 struct the_nilfs *nilfs = sb->s_fs_info;
313 struct nilfs_super_block **sbp;
314 int flag = NILFS_SB_COMMIT;
315 int ret = -EIO;
316
317 sbp = nilfs_prepare_super(sb, 0);
318 if (sbp) {
319 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
320 nilfs_set_log_cursor(sbp[0], nilfs);
321 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
322 /*
323 * make the "clean" flag also to the opposite
324 * super block if both super blocks point to
325 * the same checkpoint.
326 */
327 sbp[1]->s_state = sbp[0]->s_state;
328 flag = NILFS_SB_COMMIT_ALL;
329 }
330 ret = nilfs_commit_super(sb, flag);
331 }
332 return ret;
333 }
334
335 /**
336 * nilfs_move_2nd_super - relocate secondary super block
337 * @sb: super block instance
338 * @sb2off: new offset of the secondary super block (in bytes)
339 */
nilfs_move_2nd_super(struct super_block * sb,loff_t sb2off)340 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
341 {
342 struct the_nilfs *nilfs = sb->s_fs_info;
343 struct buffer_head *nsbh;
344 struct nilfs_super_block *nsbp;
345 sector_t blocknr, newblocknr;
346 unsigned long offset;
347 int sb2i; /* array index of the secondary superblock */
348 int ret = 0;
349
350 /* nilfs->ns_sem must be locked by the caller. */
351 if (nilfs->ns_sbh[1] &&
352 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
353 sb2i = 1;
354 blocknr = nilfs->ns_sbh[1]->b_blocknr;
355 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
356 sb2i = 0;
357 blocknr = nilfs->ns_sbh[0]->b_blocknr;
358 } else {
359 sb2i = -1;
360 blocknr = 0;
361 }
362 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
363 goto out; /* super block location is unchanged */
364
365 /* Get new super block buffer */
366 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
367 offset = sb2off & (nilfs->ns_blocksize - 1);
368 nsbh = sb_getblk(sb, newblocknr);
369 if (!nsbh) {
370 nilfs_warn(sb,
371 "unable to move secondary superblock to block %llu",
372 (unsigned long long)newblocknr);
373 ret = -EIO;
374 goto out;
375 }
376 nsbp = (void *)nsbh->b_data + offset;
377
378 lock_buffer(nsbh);
379 if (sb2i >= 0) {
380 /*
381 * The position of the second superblock only changes by 4KiB,
382 * which is larger than the maximum superblock data size
383 * (= 1KiB), so there is no need to use memmove() to allow
384 * overlap between source and destination.
385 */
386 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
387
388 /*
389 * Zero fill after copy to avoid overwriting in case of move
390 * within the same block.
391 */
392 memset(nsbh->b_data, 0, offset);
393 memset((void *)nsbp + nilfs->ns_sbsize, 0,
394 nsbh->b_size - offset - nilfs->ns_sbsize);
395 } else {
396 memset(nsbh->b_data, 0, nsbh->b_size);
397 }
398 set_buffer_uptodate(nsbh);
399 unlock_buffer(nsbh);
400
401 if (sb2i >= 0) {
402 brelse(nilfs->ns_sbh[sb2i]);
403 nilfs->ns_sbh[sb2i] = nsbh;
404 nilfs->ns_sbp[sb2i] = nsbp;
405 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
406 /* secondary super block will be restored to index 1 */
407 nilfs->ns_sbh[1] = nsbh;
408 nilfs->ns_sbp[1] = nsbp;
409 } else {
410 brelse(nsbh);
411 }
412 out:
413 return ret;
414 }
415
416 /**
417 * nilfs_resize_fs - resize the filesystem
418 * @sb: super block instance
419 * @newsize: new size of the filesystem (in bytes)
420 */
nilfs_resize_fs(struct super_block * sb,__u64 newsize)421 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
422 {
423 struct the_nilfs *nilfs = sb->s_fs_info;
424 struct nilfs_super_block **sbp;
425 __u64 devsize, newnsegs;
426 loff_t sb2off;
427 int ret;
428
429 ret = -ERANGE;
430 devsize = bdev_nr_bytes(sb->s_bdev);
431 if (newsize > devsize)
432 goto out;
433
434 /*
435 * Prevent underflow in second superblock position calculation.
436 * The exact minimum size check is done in nilfs_sufile_resize().
437 */
438 if (newsize < 4096) {
439 ret = -ENOSPC;
440 goto out;
441 }
442
443 /*
444 * Write lock is required to protect some functions depending
445 * on the number of segments, the number of reserved segments,
446 * and so forth.
447 */
448 down_write(&nilfs->ns_segctor_sem);
449
450 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
451 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
452 do_div(newnsegs, nilfs->ns_blocks_per_segment);
453
454 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
455 up_write(&nilfs->ns_segctor_sem);
456 if (ret < 0)
457 goto out;
458
459 ret = nilfs_construct_segment(sb);
460 if (ret < 0)
461 goto out;
462
463 down_write(&nilfs->ns_sem);
464 nilfs_move_2nd_super(sb, sb2off);
465 ret = -EIO;
466 sbp = nilfs_prepare_super(sb, 0);
467 if (likely(sbp)) {
468 nilfs_set_log_cursor(sbp[0], nilfs);
469 /*
470 * Drop NILFS_RESIZE_FS flag for compatibility with
471 * mount-time resize which may be implemented in a
472 * future release.
473 */
474 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
475 ~NILFS_RESIZE_FS);
476 sbp[0]->s_dev_size = cpu_to_le64(newsize);
477 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
478 if (sbp[1])
479 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
480 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
481 }
482 up_write(&nilfs->ns_sem);
483
484 /*
485 * Reset the range of allocatable segments last. This order
486 * is important in the case of expansion because the secondary
487 * superblock must be protected from log write until migration
488 * completes.
489 */
490 if (!ret)
491 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
492 out:
493 return ret;
494 }
495
nilfs_put_super(struct super_block * sb)496 static void nilfs_put_super(struct super_block *sb)
497 {
498 struct the_nilfs *nilfs = sb->s_fs_info;
499
500 nilfs_detach_log_writer(sb);
501
502 if (!sb_rdonly(sb)) {
503 down_write(&nilfs->ns_sem);
504 nilfs_cleanup_super(sb);
505 up_write(&nilfs->ns_sem);
506 }
507
508 nilfs_sysfs_delete_device_group(nilfs);
509 iput(nilfs->ns_sufile);
510 iput(nilfs->ns_cpfile);
511 iput(nilfs->ns_dat);
512
513 destroy_nilfs(nilfs);
514 sb->s_fs_info = NULL;
515 }
516
nilfs_sync_fs(struct super_block * sb,int wait)517 static int nilfs_sync_fs(struct super_block *sb, int wait)
518 {
519 struct the_nilfs *nilfs = sb->s_fs_info;
520 struct nilfs_super_block **sbp;
521 int err = 0;
522
523 /* This function is called when super block should be written back */
524 if (wait)
525 err = nilfs_construct_segment(sb);
526
527 down_write(&nilfs->ns_sem);
528 if (nilfs_sb_dirty(nilfs)) {
529 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
530 if (likely(sbp)) {
531 nilfs_set_log_cursor(sbp[0], nilfs);
532 nilfs_commit_super(sb, NILFS_SB_COMMIT);
533 }
534 }
535 up_write(&nilfs->ns_sem);
536
537 if (!err)
538 err = nilfs_flush_device(nilfs);
539
540 return err;
541 }
542
nilfs_attach_checkpoint(struct super_block * sb,__u64 cno,int curr_mnt,struct nilfs_root ** rootp)543 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
544 struct nilfs_root **rootp)
545 {
546 struct the_nilfs *nilfs = sb->s_fs_info;
547 struct nilfs_root *root;
548 struct nilfs_checkpoint *raw_cp;
549 struct buffer_head *bh_cp;
550 int err = -ENOMEM;
551
552 root = nilfs_find_or_create_root(
553 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
554 if (!root)
555 return err;
556
557 if (root->ifile)
558 goto reuse; /* already attached checkpoint */
559
560 down_read(&nilfs->ns_segctor_sem);
561 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
562 &bh_cp);
563 up_read(&nilfs->ns_segctor_sem);
564 if (unlikely(err)) {
565 if (err == -ENOENT || err == -EINVAL) {
566 nilfs_err(sb,
567 "Invalid checkpoint (checkpoint number=%llu)",
568 (unsigned long long)cno);
569 err = -EINVAL;
570 }
571 goto failed;
572 }
573
574 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
575 &raw_cp->cp_ifile_inode, &root->ifile);
576 if (err)
577 goto failed_bh;
578
579 atomic64_set(&root->inodes_count,
580 le64_to_cpu(raw_cp->cp_inodes_count));
581 atomic64_set(&root->blocks_count,
582 le64_to_cpu(raw_cp->cp_blocks_count));
583
584 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
585
586 reuse:
587 *rootp = root;
588 return 0;
589
590 failed_bh:
591 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
592 failed:
593 nilfs_put_root(root);
594
595 return err;
596 }
597
nilfs_freeze(struct super_block * sb)598 static int nilfs_freeze(struct super_block *sb)
599 {
600 struct the_nilfs *nilfs = sb->s_fs_info;
601 int err;
602
603 if (sb_rdonly(sb))
604 return 0;
605
606 /* Mark super block clean */
607 down_write(&nilfs->ns_sem);
608 err = nilfs_cleanup_super(sb);
609 up_write(&nilfs->ns_sem);
610 return err;
611 }
612
nilfs_unfreeze(struct super_block * sb)613 static int nilfs_unfreeze(struct super_block *sb)
614 {
615 struct the_nilfs *nilfs = sb->s_fs_info;
616
617 if (sb_rdonly(sb))
618 return 0;
619
620 down_write(&nilfs->ns_sem);
621 nilfs_setup_super(sb, false);
622 up_write(&nilfs->ns_sem);
623 return 0;
624 }
625
nilfs_statfs(struct dentry * dentry,struct kstatfs * buf)626 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
627 {
628 struct super_block *sb = dentry->d_sb;
629 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
630 struct the_nilfs *nilfs = root->nilfs;
631 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
632 unsigned long long blocks;
633 unsigned long overhead;
634 unsigned long nrsvblocks;
635 sector_t nfreeblocks;
636 u64 nmaxinodes, nfreeinodes;
637 int err;
638
639 /*
640 * Compute all of the segment blocks
641 *
642 * The blocks before first segment and after last segment
643 * are excluded.
644 */
645 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
646 - nilfs->ns_first_data_block;
647 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
648
649 /*
650 * Compute the overhead
651 *
652 * When distributing meta data blocks outside segment structure,
653 * We must count them as the overhead.
654 */
655 overhead = 0;
656
657 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
658 if (unlikely(err))
659 return err;
660
661 err = nilfs_ifile_count_free_inodes(root->ifile,
662 &nmaxinodes, &nfreeinodes);
663 if (unlikely(err)) {
664 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
665 if (err == -ERANGE) {
666 /*
667 * If nilfs_palloc_count_max_entries() returns
668 * -ERANGE error code then we simply treat
669 * curent inodes count as maximum possible and
670 * zero as free inodes value.
671 */
672 nmaxinodes = atomic64_read(&root->inodes_count);
673 nfreeinodes = 0;
674 err = 0;
675 } else
676 return err;
677 }
678
679 buf->f_type = NILFS_SUPER_MAGIC;
680 buf->f_bsize = sb->s_blocksize;
681 buf->f_blocks = blocks - overhead;
682 buf->f_bfree = nfreeblocks;
683 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
684 (buf->f_bfree - nrsvblocks) : 0;
685 buf->f_files = nmaxinodes;
686 buf->f_ffree = nfreeinodes;
687 buf->f_namelen = NILFS_NAME_LEN;
688 buf->f_fsid = u64_to_fsid(id);
689
690 return 0;
691 }
692
nilfs_show_options(struct seq_file * seq,struct dentry * dentry)693 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
694 {
695 struct super_block *sb = dentry->d_sb;
696 struct the_nilfs *nilfs = sb->s_fs_info;
697 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
698
699 if (!nilfs_test_opt(nilfs, BARRIER))
700 seq_puts(seq, ",nobarrier");
701 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
702 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
703 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
704 seq_puts(seq, ",errors=panic");
705 if (nilfs_test_opt(nilfs, ERRORS_CONT))
706 seq_puts(seq, ",errors=continue");
707 if (nilfs_test_opt(nilfs, STRICT_ORDER))
708 seq_puts(seq, ",order=strict");
709 if (nilfs_test_opt(nilfs, NORECOVERY))
710 seq_puts(seq, ",norecovery");
711 if (nilfs_test_opt(nilfs, DISCARD))
712 seq_puts(seq, ",discard");
713
714 return 0;
715 }
716
717 static const struct super_operations nilfs_sops = {
718 .alloc_inode = nilfs_alloc_inode,
719 .free_inode = nilfs_free_inode,
720 .dirty_inode = nilfs_dirty_inode,
721 .evict_inode = nilfs_evict_inode,
722 .put_super = nilfs_put_super,
723 .sync_fs = nilfs_sync_fs,
724 .freeze_fs = nilfs_freeze,
725 .unfreeze_fs = nilfs_unfreeze,
726 .statfs = nilfs_statfs,
727 .remount_fs = nilfs_remount,
728 .show_options = nilfs_show_options
729 };
730
731 enum {
732 Opt_err_cont, Opt_err_panic, Opt_err_ro,
733 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
734 Opt_discard, Opt_nodiscard, Opt_err,
735 };
736
737 static match_table_t tokens = {
738 {Opt_err_cont, "errors=continue"},
739 {Opt_err_panic, "errors=panic"},
740 {Opt_err_ro, "errors=remount-ro"},
741 {Opt_barrier, "barrier"},
742 {Opt_nobarrier, "nobarrier"},
743 {Opt_snapshot, "cp=%u"},
744 {Opt_order, "order=%s"},
745 {Opt_norecovery, "norecovery"},
746 {Opt_discard, "discard"},
747 {Opt_nodiscard, "nodiscard"},
748 {Opt_err, NULL}
749 };
750
parse_options(char * options,struct super_block * sb,int is_remount)751 static int parse_options(char *options, struct super_block *sb, int is_remount)
752 {
753 struct the_nilfs *nilfs = sb->s_fs_info;
754 char *p;
755 substring_t args[MAX_OPT_ARGS];
756
757 if (!options)
758 return 1;
759
760 while ((p = strsep(&options, ",")) != NULL) {
761 int token;
762
763 if (!*p)
764 continue;
765
766 token = match_token(p, tokens, args);
767 switch (token) {
768 case Opt_barrier:
769 nilfs_set_opt(nilfs, BARRIER);
770 break;
771 case Opt_nobarrier:
772 nilfs_clear_opt(nilfs, BARRIER);
773 break;
774 case Opt_order:
775 if (strcmp(args[0].from, "relaxed") == 0)
776 /* Ordered data semantics */
777 nilfs_clear_opt(nilfs, STRICT_ORDER);
778 else if (strcmp(args[0].from, "strict") == 0)
779 /* Strict in-order semantics */
780 nilfs_set_opt(nilfs, STRICT_ORDER);
781 else
782 return 0;
783 break;
784 case Opt_err_panic:
785 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
786 break;
787 case Opt_err_ro:
788 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
789 break;
790 case Opt_err_cont:
791 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
792 break;
793 case Opt_snapshot:
794 if (is_remount) {
795 nilfs_err(sb,
796 "\"%s\" option is invalid for remount",
797 p);
798 return 0;
799 }
800 break;
801 case Opt_norecovery:
802 nilfs_set_opt(nilfs, NORECOVERY);
803 break;
804 case Opt_discard:
805 nilfs_set_opt(nilfs, DISCARD);
806 break;
807 case Opt_nodiscard:
808 nilfs_clear_opt(nilfs, DISCARD);
809 break;
810 default:
811 nilfs_err(sb, "unrecognized mount option \"%s\"", p);
812 return 0;
813 }
814 }
815 return 1;
816 }
817
818 static inline void
nilfs_set_default_options(struct super_block * sb,struct nilfs_super_block * sbp)819 nilfs_set_default_options(struct super_block *sb,
820 struct nilfs_super_block *sbp)
821 {
822 struct the_nilfs *nilfs = sb->s_fs_info;
823
824 nilfs->ns_mount_opt =
825 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
826 }
827
nilfs_setup_super(struct super_block * sb,int is_mount)828 static int nilfs_setup_super(struct super_block *sb, int is_mount)
829 {
830 struct the_nilfs *nilfs = sb->s_fs_info;
831 struct nilfs_super_block **sbp;
832 int max_mnt_count;
833 int mnt_count;
834
835 /* nilfs->ns_sem must be locked by the caller. */
836 sbp = nilfs_prepare_super(sb, 0);
837 if (!sbp)
838 return -EIO;
839
840 if (!is_mount)
841 goto skip_mount_setup;
842
843 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
844 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
845
846 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
847 nilfs_warn(sb, "mounting fs with errors");
848 #if 0
849 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
850 nilfs_warn(sb, "maximal mount count reached");
851 #endif
852 }
853 if (!max_mnt_count)
854 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
855
856 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
857 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
858
859 skip_mount_setup:
860 sbp[0]->s_state =
861 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
862 /* synchronize sbp[1] with sbp[0] */
863 if (sbp[1])
864 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
865 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
866 }
867
nilfs_read_super_block(struct super_block * sb,u64 pos,int blocksize,struct buffer_head ** pbh)868 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
869 u64 pos, int blocksize,
870 struct buffer_head **pbh)
871 {
872 unsigned long long sb_index = pos;
873 unsigned long offset;
874
875 offset = do_div(sb_index, blocksize);
876 *pbh = sb_bread(sb, sb_index);
877 if (!*pbh)
878 return NULL;
879 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
880 }
881
nilfs_store_magic_and_option(struct super_block * sb,struct nilfs_super_block * sbp,char * data)882 int nilfs_store_magic_and_option(struct super_block *sb,
883 struct nilfs_super_block *sbp,
884 char *data)
885 {
886 struct the_nilfs *nilfs = sb->s_fs_info;
887
888 sb->s_magic = le16_to_cpu(sbp->s_magic);
889
890 /* FS independent flags */
891 #ifdef NILFS_ATIME_DISABLE
892 sb->s_flags |= SB_NOATIME;
893 #endif
894
895 nilfs_set_default_options(sb, sbp);
896
897 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
898 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
899 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
900 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
901
902 return !parse_options(data, sb, 0) ? -EINVAL : 0;
903 }
904
nilfs_check_feature_compatibility(struct super_block * sb,struct nilfs_super_block * sbp)905 int nilfs_check_feature_compatibility(struct super_block *sb,
906 struct nilfs_super_block *sbp)
907 {
908 __u64 features;
909
910 features = le64_to_cpu(sbp->s_feature_incompat) &
911 ~NILFS_FEATURE_INCOMPAT_SUPP;
912 if (features) {
913 nilfs_err(sb,
914 "couldn't mount because of unsupported optional features (%llx)",
915 (unsigned long long)features);
916 return -EINVAL;
917 }
918 features = le64_to_cpu(sbp->s_feature_compat_ro) &
919 ~NILFS_FEATURE_COMPAT_RO_SUPP;
920 if (!sb_rdonly(sb) && features) {
921 nilfs_err(sb,
922 "couldn't mount RDWR because of unsupported optional features (%llx)",
923 (unsigned long long)features);
924 return -EINVAL;
925 }
926 return 0;
927 }
928
nilfs_get_root_dentry(struct super_block * sb,struct nilfs_root * root,struct dentry ** root_dentry)929 static int nilfs_get_root_dentry(struct super_block *sb,
930 struct nilfs_root *root,
931 struct dentry **root_dentry)
932 {
933 struct inode *inode;
934 struct dentry *dentry;
935 int ret = 0;
936
937 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
938 if (IS_ERR(inode)) {
939 ret = PTR_ERR(inode);
940 nilfs_err(sb, "error %d getting root inode", ret);
941 goto out;
942 }
943 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
944 iput(inode);
945 nilfs_err(sb, "corrupt root inode");
946 ret = -EINVAL;
947 goto out;
948 }
949
950 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
951 dentry = d_find_alias(inode);
952 if (!dentry) {
953 dentry = d_make_root(inode);
954 if (!dentry) {
955 ret = -ENOMEM;
956 goto failed_dentry;
957 }
958 } else {
959 iput(inode);
960 }
961 } else {
962 dentry = d_obtain_root(inode);
963 if (IS_ERR(dentry)) {
964 ret = PTR_ERR(dentry);
965 goto failed_dentry;
966 }
967 }
968 *root_dentry = dentry;
969 out:
970 return ret;
971
972 failed_dentry:
973 nilfs_err(sb, "error %d getting root dentry", ret);
974 goto out;
975 }
976
nilfs_attach_snapshot(struct super_block * s,__u64 cno,struct dentry ** root_dentry)977 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
978 struct dentry **root_dentry)
979 {
980 struct the_nilfs *nilfs = s->s_fs_info;
981 struct nilfs_root *root;
982 int ret;
983
984 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
985
986 down_read(&nilfs->ns_segctor_sem);
987 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
988 up_read(&nilfs->ns_segctor_sem);
989 if (ret < 0) {
990 ret = (ret == -ENOENT) ? -EINVAL : ret;
991 goto out;
992 } else if (!ret) {
993 nilfs_err(s,
994 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
995 (unsigned long long)cno);
996 ret = -EINVAL;
997 goto out;
998 }
999
1000 ret = nilfs_attach_checkpoint(s, cno, false, &root);
1001 if (ret) {
1002 nilfs_err(s,
1003 "error %d while loading snapshot (checkpoint number=%llu)",
1004 ret, (unsigned long long)cno);
1005 goto out;
1006 }
1007 ret = nilfs_get_root_dentry(s, root, root_dentry);
1008 nilfs_put_root(root);
1009 out:
1010 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1011 return ret;
1012 }
1013
1014 /**
1015 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1016 * @root_dentry: root dentry of the tree to be shrunk
1017 *
1018 * This function returns true if the tree was in-use.
1019 */
nilfs_tree_is_busy(struct dentry * root_dentry)1020 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1021 {
1022 shrink_dcache_parent(root_dentry);
1023 return d_count(root_dentry) > 1;
1024 }
1025
nilfs_checkpoint_is_mounted(struct super_block * sb,__u64 cno)1026 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1027 {
1028 struct the_nilfs *nilfs = sb->s_fs_info;
1029 struct nilfs_root *root;
1030 struct inode *inode;
1031 struct dentry *dentry;
1032 int ret;
1033
1034 if (cno > nilfs->ns_cno)
1035 return false;
1036
1037 if (cno >= nilfs_last_cno(nilfs))
1038 return true; /* protect recent checkpoints */
1039
1040 ret = false;
1041 root = nilfs_lookup_root(nilfs, cno);
1042 if (root) {
1043 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1044 if (inode) {
1045 dentry = d_find_alias(inode);
1046 if (dentry) {
1047 ret = nilfs_tree_is_busy(dentry);
1048 dput(dentry);
1049 }
1050 iput(inode);
1051 }
1052 nilfs_put_root(root);
1053 }
1054 return ret;
1055 }
1056
1057 /**
1058 * nilfs_fill_super() - initialize a super block instance
1059 * @sb: super_block
1060 * @data: mount options
1061 * @silent: silent mode flag
1062 *
1063 * This function is called exclusively by nilfs->ns_mount_mutex.
1064 * So, the recovery process is protected from other simultaneous mounts.
1065 */
1066 static int
nilfs_fill_super(struct super_block * sb,void * data,int silent)1067 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1068 {
1069 struct the_nilfs *nilfs;
1070 struct nilfs_root *fsroot;
1071 __u64 cno;
1072 int err;
1073
1074 nilfs = alloc_nilfs(sb);
1075 if (!nilfs)
1076 return -ENOMEM;
1077
1078 sb->s_fs_info = nilfs;
1079
1080 err = init_nilfs(nilfs, sb, (char *)data);
1081 if (err)
1082 goto failed_nilfs;
1083
1084 sb->s_op = &nilfs_sops;
1085 sb->s_export_op = &nilfs_export_ops;
1086 sb->s_root = NULL;
1087 sb->s_time_gran = 1;
1088 sb->s_max_links = NILFS_LINK_MAX;
1089
1090 sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1091
1092 err = load_nilfs(nilfs, sb);
1093 if (err)
1094 goto failed_nilfs;
1095
1096 cno = nilfs_last_cno(nilfs);
1097 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1098 if (err) {
1099 nilfs_err(sb,
1100 "error %d while loading last checkpoint (checkpoint number=%llu)",
1101 err, (unsigned long long)cno);
1102 goto failed_unload;
1103 }
1104
1105 if (!sb_rdonly(sb)) {
1106 err = nilfs_attach_log_writer(sb, fsroot);
1107 if (err)
1108 goto failed_checkpoint;
1109 }
1110
1111 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1112 if (err)
1113 goto failed_segctor;
1114
1115 nilfs_put_root(fsroot);
1116
1117 if (!sb_rdonly(sb)) {
1118 down_write(&nilfs->ns_sem);
1119 nilfs_setup_super(sb, true);
1120 up_write(&nilfs->ns_sem);
1121 }
1122
1123 return 0;
1124
1125 failed_segctor:
1126 nilfs_detach_log_writer(sb);
1127
1128 failed_checkpoint:
1129 nilfs_put_root(fsroot);
1130
1131 failed_unload:
1132 nilfs_sysfs_delete_device_group(nilfs);
1133 iput(nilfs->ns_sufile);
1134 iput(nilfs->ns_cpfile);
1135 iput(nilfs->ns_dat);
1136
1137 failed_nilfs:
1138 destroy_nilfs(nilfs);
1139 return err;
1140 }
1141
nilfs_remount(struct super_block * sb,int * flags,char * data)1142 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1143 {
1144 struct the_nilfs *nilfs = sb->s_fs_info;
1145 unsigned long old_sb_flags;
1146 unsigned long old_mount_opt;
1147 int err;
1148
1149 sync_filesystem(sb);
1150 old_sb_flags = sb->s_flags;
1151 old_mount_opt = nilfs->ns_mount_opt;
1152
1153 if (!parse_options(data, sb, 1)) {
1154 err = -EINVAL;
1155 goto restore_opts;
1156 }
1157 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1158
1159 err = -EINVAL;
1160
1161 if (!nilfs_valid_fs(nilfs)) {
1162 nilfs_warn(sb,
1163 "couldn't remount because the filesystem is in an incomplete recovery state");
1164 goto restore_opts;
1165 }
1166
1167 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1168 goto out;
1169 if (*flags & SB_RDONLY) {
1170 sb->s_flags |= SB_RDONLY;
1171
1172 /*
1173 * Remounting a valid RW partition RDONLY, so set
1174 * the RDONLY flag and then mark the partition as valid again.
1175 */
1176 down_write(&nilfs->ns_sem);
1177 nilfs_cleanup_super(sb);
1178 up_write(&nilfs->ns_sem);
1179 } else {
1180 __u64 features;
1181 struct nilfs_root *root;
1182
1183 /*
1184 * Mounting a RDONLY partition read-write, so reread and
1185 * store the current valid flag. (It may have been changed
1186 * by fsck since we originally mounted the partition.)
1187 */
1188 down_read(&nilfs->ns_sem);
1189 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1190 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1191 up_read(&nilfs->ns_sem);
1192 if (features) {
1193 nilfs_warn(sb,
1194 "couldn't remount RDWR because of unsupported optional features (%llx)",
1195 (unsigned long long)features);
1196 err = -EROFS;
1197 goto restore_opts;
1198 }
1199
1200 sb->s_flags &= ~SB_RDONLY;
1201
1202 root = NILFS_I(d_inode(sb->s_root))->i_root;
1203 err = nilfs_attach_log_writer(sb, root);
1204 if (err)
1205 goto restore_opts;
1206
1207 down_write(&nilfs->ns_sem);
1208 nilfs_setup_super(sb, true);
1209 up_write(&nilfs->ns_sem);
1210 }
1211 out:
1212 return 0;
1213
1214 restore_opts:
1215 sb->s_flags = old_sb_flags;
1216 nilfs->ns_mount_opt = old_mount_opt;
1217 return err;
1218 }
1219
1220 struct nilfs_super_data {
1221 __u64 cno;
1222 int flags;
1223 };
1224
nilfs_parse_snapshot_option(const char * option,const substring_t * arg,struct nilfs_super_data * sd)1225 static int nilfs_parse_snapshot_option(const char *option,
1226 const substring_t *arg,
1227 struct nilfs_super_data *sd)
1228 {
1229 unsigned long long val;
1230 const char *msg = NULL;
1231 int err;
1232
1233 if (!(sd->flags & SB_RDONLY)) {
1234 msg = "read-only option is not specified";
1235 goto parse_error;
1236 }
1237
1238 err = kstrtoull(arg->from, 0, &val);
1239 if (err) {
1240 if (err == -ERANGE)
1241 msg = "too large checkpoint number";
1242 else
1243 msg = "malformed argument";
1244 goto parse_error;
1245 } else if (val == 0) {
1246 msg = "invalid checkpoint number 0";
1247 goto parse_error;
1248 }
1249 sd->cno = val;
1250 return 0;
1251
1252 parse_error:
1253 nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1254 return 1;
1255 }
1256
1257 /**
1258 * nilfs_identify - pre-read mount options needed to identify mount instance
1259 * @data: mount options
1260 * @sd: nilfs_super_data
1261 */
nilfs_identify(char * data,struct nilfs_super_data * sd)1262 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1263 {
1264 char *p, *options = data;
1265 substring_t args[MAX_OPT_ARGS];
1266 int token;
1267 int ret = 0;
1268
1269 do {
1270 p = strsep(&options, ",");
1271 if (p != NULL && *p) {
1272 token = match_token(p, tokens, args);
1273 if (token == Opt_snapshot)
1274 ret = nilfs_parse_snapshot_option(p, &args[0],
1275 sd);
1276 }
1277 if (!options)
1278 break;
1279 BUG_ON(options == data);
1280 *(options - 1) = ',';
1281 } while (!ret);
1282 return ret;
1283 }
1284
nilfs_set_bdev_super(struct super_block * s,void * data)1285 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1286 {
1287 s->s_dev = *(dev_t *)data;
1288 return 0;
1289 }
1290
nilfs_test_bdev_super(struct super_block * s,void * data)1291 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1292 {
1293 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1294 }
1295
1296 static struct dentry *
nilfs_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1297 nilfs_mount(struct file_system_type *fs_type, int flags,
1298 const char *dev_name, void *data)
1299 {
1300 struct nilfs_super_data sd = { .flags = flags };
1301 struct super_block *s;
1302 dev_t dev;
1303 int err;
1304
1305 if (nilfs_identify(data, &sd))
1306 return ERR_PTR(-EINVAL);
1307
1308 err = lookup_bdev(dev_name, &dev);
1309 if (err)
1310 return ERR_PTR(err);
1311
1312 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1313 &dev);
1314 if (IS_ERR(s))
1315 return ERR_CAST(s);
1316
1317 if (!s->s_root) {
1318 /*
1319 * We drop s_umount here because we need to open the bdev and
1320 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
1321 * __invalidate_device()). It is safe because we have active sb
1322 * reference and SB_BORN is not set yet.
1323 */
1324 up_write(&s->s_umount);
1325 err = setup_bdev_super(s, flags, NULL);
1326 down_write(&s->s_umount);
1327 if (!err)
1328 err = nilfs_fill_super(s, data,
1329 flags & SB_SILENT ? 1 : 0);
1330 if (err)
1331 goto failed_super;
1332
1333 s->s_flags |= SB_ACTIVE;
1334 } else if (!sd.cno) {
1335 if (nilfs_tree_is_busy(s->s_root)) {
1336 if ((flags ^ s->s_flags) & SB_RDONLY) {
1337 nilfs_err(s,
1338 "the device already has a %s mount.",
1339 sb_rdonly(s) ? "read-only" : "read/write");
1340 err = -EBUSY;
1341 goto failed_super;
1342 }
1343 } else {
1344 /*
1345 * Try remount to setup mount states if the current
1346 * tree is not mounted and only snapshots use this sb.
1347 */
1348 err = nilfs_remount(s, &flags, data);
1349 if (err)
1350 goto failed_super;
1351 }
1352 }
1353
1354 if (sd.cno) {
1355 struct dentry *root_dentry;
1356
1357 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1358 if (err)
1359 goto failed_super;
1360 return root_dentry;
1361 }
1362
1363 return dget(s->s_root);
1364
1365 failed_super:
1366 deactivate_locked_super(s);
1367 return ERR_PTR(err);
1368 }
1369
1370 struct file_system_type nilfs_fs_type = {
1371 .owner = THIS_MODULE,
1372 .name = "nilfs2",
1373 .mount = nilfs_mount,
1374 .kill_sb = kill_block_super,
1375 .fs_flags = FS_REQUIRES_DEV,
1376 };
1377 MODULE_ALIAS_FS("nilfs2");
1378
nilfs_inode_init_once(void * obj)1379 static void nilfs_inode_init_once(void *obj)
1380 {
1381 struct nilfs_inode_info *ii = obj;
1382
1383 INIT_LIST_HEAD(&ii->i_dirty);
1384 #ifdef CONFIG_NILFS_XATTR
1385 init_rwsem(&ii->xattr_sem);
1386 #endif
1387 inode_init_once(&ii->vfs_inode);
1388 }
1389
nilfs_segbuf_init_once(void * obj)1390 static void nilfs_segbuf_init_once(void *obj)
1391 {
1392 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1393 }
1394
nilfs_destroy_cachep(void)1395 static void nilfs_destroy_cachep(void)
1396 {
1397 /*
1398 * Make sure all delayed rcu free inodes are flushed before we
1399 * destroy cache.
1400 */
1401 rcu_barrier();
1402
1403 kmem_cache_destroy(nilfs_inode_cachep);
1404 kmem_cache_destroy(nilfs_transaction_cachep);
1405 kmem_cache_destroy(nilfs_segbuf_cachep);
1406 kmem_cache_destroy(nilfs_btree_path_cache);
1407 }
1408
nilfs_init_cachep(void)1409 static int __init nilfs_init_cachep(void)
1410 {
1411 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1412 sizeof(struct nilfs_inode_info), 0,
1413 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1414 nilfs_inode_init_once);
1415 if (!nilfs_inode_cachep)
1416 goto fail;
1417
1418 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1419 sizeof(struct nilfs_transaction_info), 0,
1420 SLAB_RECLAIM_ACCOUNT, NULL);
1421 if (!nilfs_transaction_cachep)
1422 goto fail;
1423
1424 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1425 sizeof(struct nilfs_segment_buffer), 0,
1426 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1427 if (!nilfs_segbuf_cachep)
1428 goto fail;
1429
1430 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1431 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1432 0, 0, NULL);
1433 if (!nilfs_btree_path_cache)
1434 goto fail;
1435
1436 return 0;
1437
1438 fail:
1439 nilfs_destroy_cachep();
1440 return -ENOMEM;
1441 }
1442
init_nilfs_fs(void)1443 static int __init init_nilfs_fs(void)
1444 {
1445 int err;
1446
1447 err = nilfs_init_cachep();
1448 if (err)
1449 goto fail;
1450
1451 err = nilfs_sysfs_init();
1452 if (err)
1453 goto free_cachep;
1454
1455 err = register_filesystem(&nilfs_fs_type);
1456 if (err)
1457 goto deinit_sysfs_entry;
1458
1459 printk(KERN_INFO "NILFS version 2 loaded\n");
1460 return 0;
1461
1462 deinit_sysfs_entry:
1463 nilfs_sysfs_exit();
1464 free_cachep:
1465 nilfs_destroy_cachep();
1466 fail:
1467 return err;
1468 }
1469
exit_nilfs_fs(void)1470 static void __exit exit_nilfs_fs(void)
1471 {
1472 nilfs_destroy_cachep();
1473 nilfs_sysfs_exit();
1474 unregister_filesystem(&nilfs_fs_type);
1475 }
1476
1477 module_init(init_nilfs_fs)
1478 module_exit(exit_nilfs_fs)
1479