2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Artem Bityutskiy (Битюцкий Артём)
24 * This file implements UBIFS initialization and VFS superblock operations. Some
25 * initialization stuff which is rather large and complex is placed at
26 * corresponding subsystems, but most of it is here.
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/module.h>
32 #include <linux/ctype.h>
33 #include <linux/kthread.h>
34 #include <linux/parser.h>
35 #include <linux/seq_file.h>
36 #include <linux/mount.h>
37 #include <linux/math64.h>
38 #include <linux/writeback.h>
42 * Maximum amount of memory we may 'kmalloc()' without worrying that we are
43 * allocating too much.
45 #define UBIFS_KMALLOC_OK (128*1024)
47 /* Slab cache for UBIFS inodes */
48 struct kmem_cache
*ubifs_inode_slab
;
50 /* UBIFS TNC shrinker description */
51 static struct shrinker ubifs_shrinker_info
= {
52 .shrink
= ubifs_shrinker
,
53 .seeks
= DEFAULT_SEEKS
,
57 * validate_inode - validate inode.
58 * @c: UBIFS file-system description object
59 * @inode: the inode to validate
61 * This is a helper function for 'ubifs_iget()' which validates various fields
62 * of a newly built inode to make sure they contain sane values and prevent
63 * possible vulnerabilities. Returns zero if the inode is all right and
64 * a non-zero error code if not.
66 static int validate_inode(struct ubifs_info
*c
, const struct inode
*inode
)
69 const struct ubifs_inode
*ui
= ubifs_inode(inode
);
71 if (inode
->i_size
> c
->max_inode_sz
) {
72 ubifs_err("inode is too large (%lld)",
73 (long long)inode
->i_size
);
77 if (ui
->compr_type
< 0 || ui
->compr_type
>= UBIFS_COMPR_TYPES_CNT
) {
78 ubifs_err("unknown compression type %d", ui
->compr_type
);
82 if (ui
->xattr_names
+ ui
->xattr_cnt
> XATTR_LIST_MAX
)
85 if (ui
->data_len
< 0 || ui
->data_len
> UBIFS_MAX_INO_DATA
)
88 if (ui
->xattr
&& (inode
->i_mode
& S_IFMT
) != S_IFREG
)
91 if (!ubifs_compr_present(ui
->compr_type
)) {
92 ubifs_warn("inode %lu uses '%s' compression, but it was not "
93 "compiled in", inode
->i_ino
,
94 ubifs_compr_name(ui
->compr_type
));
97 err
= dbg_check_dir_size(c
, inode
);
101 struct inode
*ubifs_iget(struct super_block
*sb
, unsigned long inum
)
105 struct ubifs_ino_node
*ino
;
106 struct ubifs_info
*c
= sb
->s_fs_info
;
108 struct ubifs_inode
*ui
;
110 dbg_gen("inode %lu", inum
);
112 inode
= iget_locked(sb
, inum
);
114 return ERR_PTR(-ENOMEM
);
115 if (!(inode
->i_state
& I_NEW
))
117 ui
= ubifs_inode(inode
);
119 ino
= kmalloc(UBIFS_MAX_INO_NODE_SZ
, GFP_NOFS
);
125 ino_key_init(c
, &key
, inode
->i_ino
);
127 err
= ubifs_tnc_lookup(c
, &key
, ino
);
131 inode
->i_flags
|= (S_NOCMTIME
| S_NOATIME
);
132 inode
->i_nlink
= le32_to_cpu(ino
->nlink
);
133 inode
->i_uid
= le32_to_cpu(ino
->uid
);
134 inode
->i_gid
= le32_to_cpu(ino
->gid
);
135 inode
->i_atime
.tv_sec
= (int64_t)le64_to_cpu(ino
->atime_sec
);
136 inode
->i_atime
.tv_nsec
= le32_to_cpu(ino
->atime_nsec
);
137 inode
->i_mtime
.tv_sec
= (int64_t)le64_to_cpu(ino
->mtime_sec
);
138 inode
->i_mtime
.tv_nsec
= le32_to_cpu(ino
->mtime_nsec
);
139 inode
->i_ctime
.tv_sec
= (int64_t)le64_to_cpu(ino
->ctime_sec
);
140 inode
->i_ctime
.tv_nsec
= le32_to_cpu(ino
->ctime_nsec
);
141 inode
->i_mode
= le32_to_cpu(ino
->mode
);
142 inode
->i_size
= le64_to_cpu(ino
->size
);
144 ui
->data_len
= le32_to_cpu(ino
->data_len
);
145 ui
->flags
= le32_to_cpu(ino
->flags
);
146 ui
->compr_type
= le16_to_cpu(ino
->compr_type
);
147 ui
->creat_sqnum
= le64_to_cpu(ino
->creat_sqnum
);
148 ui
->xattr_cnt
= le32_to_cpu(ino
->xattr_cnt
);
149 ui
->xattr_size
= le32_to_cpu(ino
->xattr_size
);
150 ui
->xattr_names
= le32_to_cpu(ino
->xattr_names
);
151 ui
->synced_i_size
= ui
->ui_size
= inode
->i_size
;
153 ui
->xattr
= (ui
->flags
& UBIFS_XATTR_FL
) ? 1 : 0;
155 err
= validate_inode(c
, inode
);
159 /* Disable read-ahead */
160 inode
->i_mapping
->backing_dev_info
= &c
->bdi
;
162 switch (inode
->i_mode
& S_IFMT
) {
164 inode
->i_mapping
->a_ops
= &ubifs_file_address_operations
;
165 inode
->i_op
= &ubifs_file_inode_operations
;
166 inode
->i_fop
= &ubifs_file_operations
;
168 ui
->data
= kmalloc(ui
->data_len
+ 1, GFP_NOFS
);
173 memcpy(ui
->data
, ino
->data
, ui
->data_len
);
174 ((char *)ui
->data
)[ui
->data_len
] = '\0';
175 } else if (ui
->data_len
!= 0) {
181 inode
->i_op
= &ubifs_dir_inode_operations
;
182 inode
->i_fop
= &ubifs_dir_operations
;
183 if (ui
->data_len
!= 0) {
189 inode
->i_op
= &ubifs_symlink_inode_operations
;
190 if (ui
->data_len
<= 0 || ui
->data_len
> UBIFS_MAX_INO_DATA
) {
194 ui
->data
= kmalloc(ui
->data_len
+ 1, GFP_NOFS
);
199 memcpy(ui
->data
, ino
->data
, ui
->data_len
);
200 ((char *)ui
->data
)[ui
->data_len
] = '\0';
206 union ubifs_dev_desc
*dev
;
208 ui
->data
= kmalloc(sizeof(union ubifs_dev_desc
), GFP_NOFS
);
214 dev
= (union ubifs_dev_desc
*)ino
->data
;
215 if (ui
->data_len
== sizeof(dev
->new))
216 rdev
= new_decode_dev(le32_to_cpu(dev
->new));
217 else if (ui
->data_len
== sizeof(dev
->huge
))
218 rdev
= huge_decode_dev(le64_to_cpu(dev
->huge
));
223 memcpy(ui
->data
, ino
->data
, ui
->data_len
);
224 inode
->i_op
= &ubifs_file_inode_operations
;
225 init_special_inode(inode
, inode
->i_mode
, rdev
);
230 inode
->i_op
= &ubifs_file_inode_operations
;
231 init_special_inode(inode
, inode
->i_mode
, 0);
232 if (ui
->data_len
!= 0) {
243 ubifs_set_inode_flags(inode
);
244 unlock_new_inode(inode
);
248 ubifs_err("inode %lu validation failed, error %d", inode
->i_ino
, err
);
249 dbg_dump_node(c
, ino
);
250 dbg_dump_inode(c
, inode
);
255 ubifs_err("failed to read inode %lu, error %d", inode
->i_ino
, err
);
260 static struct inode
*ubifs_alloc_inode(struct super_block
*sb
)
262 struct ubifs_inode
*ui
;
264 ui
= kmem_cache_alloc(ubifs_inode_slab
, GFP_NOFS
);
268 memset((void *)ui
+ sizeof(struct inode
), 0,
269 sizeof(struct ubifs_inode
) - sizeof(struct inode
));
270 mutex_init(&ui
->ui_mutex
);
271 spin_lock_init(&ui
->ui_lock
);
272 return &ui
->vfs_inode
;
275 static void ubifs_destroy_inode(struct inode
*inode
)
277 struct ubifs_inode
*ui
= ubifs_inode(inode
);
280 kmem_cache_free(ubifs_inode_slab
, inode
);
284 * Note, Linux write-back code calls this without 'i_mutex'.
286 static int ubifs_write_inode(struct inode
*inode
, int wait
)
289 struct ubifs_info
*c
= inode
->i_sb
->s_fs_info
;
290 struct ubifs_inode
*ui
= ubifs_inode(inode
);
292 ubifs_assert(!ui
->xattr
);
293 if (is_bad_inode(inode
))
296 mutex_lock(&ui
->ui_mutex
);
298 * Due to races between write-back forced by budgeting
299 * (see 'sync_some_inodes()') and pdflush write-back, the inode may
300 * have already been synchronized, do not do this again. This might
301 * also happen if it was synchronized in an VFS operation, e.g.
305 mutex_unlock(&ui
->ui_mutex
);
310 * As an optimization, do not write orphan inodes to the media just
311 * because this is not needed.
313 dbg_gen("inode %lu, mode %#x, nlink %u",
314 inode
->i_ino
, (int)inode
->i_mode
, inode
->i_nlink
);
315 if (inode
->i_nlink
) {
316 err
= ubifs_jnl_write_inode(c
, inode
);
318 ubifs_err("can't write inode %lu, error %d",
321 err
= dbg_check_inode_size(c
, inode
, ui
->ui_size
);
325 mutex_unlock(&ui
->ui_mutex
);
326 ubifs_release_dirty_inode_budget(c
, ui
);
330 static void ubifs_delete_inode(struct inode
*inode
)
333 struct ubifs_info
*c
= inode
->i_sb
->s_fs_info
;
334 struct ubifs_inode
*ui
= ubifs_inode(inode
);
338 * Extended attribute inode deletions are fully handled in
339 * 'ubifs_removexattr()'. These inodes are special and have
340 * limited usage, so there is nothing to do here.
344 dbg_gen("inode %lu, mode %#x", inode
->i_ino
, (int)inode
->i_mode
);
345 ubifs_assert(!atomic_read(&inode
->i_count
));
346 ubifs_assert(inode
->i_nlink
== 0);
348 truncate_inode_pages(&inode
->i_data
, 0);
349 if (is_bad_inode(inode
))
352 ui
->ui_size
= inode
->i_size
= 0;
353 err
= ubifs_jnl_delete_inode(c
, inode
);
356 * Worst case we have a lost orphan inode wasting space, so a
357 * simple error message is OK here.
359 ubifs_err("can't delete inode %lu, error %d",
364 ubifs_release_dirty_inode_budget(c
, ui
);
366 /* We've deleted something - clean the "no space" flags */
367 c
->nospace
= c
->nospace_rp
= 0;
373 static void ubifs_dirty_inode(struct inode
*inode
)
375 struct ubifs_inode
*ui
= ubifs_inode(inode
);
377 ubifs_assert(mutex_is_locked(&ui
->ui_mutex
));
380 dbg_gen("inode %lu", inode
->i_ino
);
384 static int ubifs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
386 struct ubifs_info
*c
= dentry
->d_sb
->s_fs_info
;
387 unsigned long long free
;
388 __le32
*uuid
= (__le32
*)c
->uuid
;
390 free
= ubifs_get_free_space(c
);
391 dbg_gen("free space %lld bytes (%lld blocks)",
392 free
, free
>> UBIFS_BLOCK_SHIFT
);
394 buf
->f_type
= UBIFS_SUPER_MAGIC
;
395 buf
->f_bsize
= UBIFS_BLOCK_SIZE
;
396 buf
->f_blocks
= c
->block_cnt
;
397 buf
->f_bfree
= free
>> UBIFS_BLOCK_SHIFT
;
398 if (free
> c
->report_rp_size
)
399 buf
->f_bavail
= (free
- c
->report_rp_size
) >> UBIFS_BLOCK_SHIFT
;
404 buf
->f_namelen
= UBIFS_MAX_NLEN
;
405 buf
->f_fsid
.val
[0] = le32_to_cpu(uuid
[0]) ^ le32_to_cpu(uuid
[2]);
406 buf
->f_fsid
.val
[1] = le32_to_cpu(uuid
[1]) ^ le32_to_cpu(uuid
[3]);
407 ubifs_assert(buf
->f_bfree
<= c
->block_cnt
);
411 static int ubifs_show_options(struct seq_file
*s
, struct vfsmount
*mnt
)
413 struct ubifs_info
*c
= mnt
->mnt_sb
->s_fs_info
;
415 if (c
->mount_opts
.unmount_mode
== 2)
416 seq_printf(s
, ",fast_unmount");
417 else if (c
->mount_opts
.unmount_mode
== 1)
418 seq_printf(s
, ",norm_unmount");
420 if (c
->mount_opts
.bulk_read
== 2)
421 seq_printf(s
, ",bulk_read");
422 else if (c
->mount_opts
.bulk_read
== 1)
423 seq_printf(s
, ",no_bulk_read");
425 if (c
->mount_opts
.chk_data_crc
== 2)
426 seq_printf(s
, ",chk_data_crc");
427 else if (c
->mount_opts
.chk_data_crc
== 1)
428 seq_printf(s
, ",no_chk_data_crc");
430 if (c
->mount_opts
.override_compr
) {
431 seq_printf(s
, ",compr=%s",
432 ubifs_compr_name(c
->mount_opts
.compr_type
));
438 static int ubifs_sync_fs(struct super_block
*sb
, int wait
)
441 struct ubifs_info
*c
= sb
->s_fs_info
;
444 * Zero @wait is just an advisory thing to help the file system shove
445 * lots of data into the queues, and there will be the second
446 * '->sync_fs()' call, with non-zero @wait.
452 * Synchronize write buffers, because 'ubifs_run_commit()' does not
453 * do this if it waits for an already running commit.
455 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
456 err
= ubifs_wbuf_sync(&c
->jheads
[i
].wbuf
);
462 * Strictly speaking, it is not necessary to commit the journal here,
463 * synchronizing write-buffers would be enough. But committing makes
464 * UBIFS free space predictions much more accurate, so we want to let
465 * the user be able to get more accurate results of 'statfs()' after
466 * they synchronize the file system.
468 err
= ubifs_run_commit(c
);
472 return ubi_sync(c
->vi
.ubi_num
);
476 * init_constants_early - initialize UBIFS constants.
477 * @c: UBIFS file-system description object
479 * This function initialize UBIFS constants which do not need the superblock to
480 * be read. It also checks that the UBI volume satisfies basic UBIFS
481 * requirements. Returns zero in case of success and a negative error code in
484 static int init_constants_early(struct ubifs_info
*c
)
486 if (c
->vi
.corrupted
) {
487 ubifs_warn("UBI volume is corrupted - read-only mode");
492 ubifs_msg("read-only UBI device");
496 if (c
->vi
.vol_type
== UBI_STATIC_VOLUME
) {
497 ubifs_msg("static UBI volume - read-only mode");
501 c
->leb_cnt
= c
->vi
.size
;
502 c
->leb_size
= c
->vi
.usable_leb_size
;
503 c
->half_leb_size
= c
->leb_size
/ 2;
504 c
->min_io_size
= c
->di
.min_io_size
;
505 c
->min_io_shift
= fls(c
->min_io_size
) - 1;
507 if (c
->leb_size
< UBIFS_MIN_LEB_SZ
) {
508 ubifs_err("too small LEBs (%d bytes), min. is %d bytes",
509 c
->leb_size
, UBIFS_MIN_LEB_SZ
);
513 if (c
->leb_cnt
< UBIFS_MIN_LEB_CNT
) {
514 ubifs_err("too few LEBs (%d), min. is %d",
515 c
->leb_cnt
, UBIFS_MIN_LEB_CNT
);
519 if (!is_power_of_2(c
->min_io_size
)) {
520 ubifs_err("bad min. I/O size %d", c
->min_io_size
);
525 * UBIFS aligns all node to 8-byte boundary, so to make function in
526 * io.c simpler, assume minimum I/O unit size to be 8 bytes if it is
529 if (c
->min_io_size
< 8) {
534 c
->ref_node_alsz
= ALIGN(UBIFS_REF_NODE_SZ
, c
->min_io_size
);
535 c
->mst_node_alsz
= ALIGN(UBIFS_MST_NODE_SZ
, c
->min_io_size
);
538 * Initialize node length ranges which are mostly needed for node
541 c
->ranges
[UBIFS_PAD_NODE
].len
= UBIFS_PAD_NODE_SZ
;
542 c
->ranges
[UBIFS_SB_NODE
].len
= UBIFS_SB_NODE_SZ
;
543 c
->ranges
[UBIFS_MST_NODE
].len
= UBIFS_MST_NODE_SZ
;
544 c
->ranges
[UBIFS_REF_NODE
].len
= UBIFS_REF_NODE_SZ
;
545 c
->ranges
[UBIFS_TRUN_NODE
].len
= UBIFS_TRUN_NODE_SZ
;
546 c
->ranges
[UBIFS_CS_NODE
].len
= UBIFS_CS_NODE_SZ
;
548 c
->ranges
[UBIFS_INO_NODE
].min_len
= UBIFS_INO_NODE_SZ
;
549 c
->ranges
[UBIFS_INO_NODE
].max_len
= UBIFS_MAX_INO_NODE_SZ
;
550 c
->ranges
[UBIFS_ORPH_NODE
].min_len
=
551 UBIFS_ORPH_NODE_SZ
+ sizeof(__le64
);
552 c
->ranges
[UBIFS_ORPH_NODE
].max_len
= c
->leb_size
;
553 c
->ranges
[UBIFS_DENT_NODE
].min_len
= UBIFS_DENT_NODE_SZ
;
554 c
->ranges
[UBIFS_DENT_NODE
].max_len
= UBIFS_MAX_DENT_NODE_SZ
;
555 c
->ranges
[UBIFS_XENT_NODE
].min_len
= UBIFS_XENT_NODE_SZ
;
556 c
->ranges
[UBIFS_XENT_NODE
].max_len
= UBIFS_MAX_XENT_NODE_SZ
;
557 c
->ranges
[UBIFS_DATA_NODE
].min_len
= UBIFS_DATA_NODE_SZ
;
558 c
->ranges
[UBIFS_DATA_NODE
].max_len
= UBIFS_MAX_DATA_NODE_SZ
;
560 * Minimum indexing node size is amended later when superblock is
561 * read and the key length is known.
563 c
->ranges
[UBIFS_IDX_NODE
].min_len
= UBIFS_IDX_NODE_SZ
+ UBIFS_BRANCH_SZ
;
565 * Maximum indexing node size is amended later when superblock is
566 * read and the fanout is known.
568 c
->ranges
[UBIFS_IDX_NODE
].max_len
= INT_MAX
;
571 * Initialize dead and dark LEB space watermarks. See gc.c for comments
572 * about these values.
574 c
->dead_wm
= ALIGN(MIN_WRITE_SZ
, c
->min_io_size
);
575 c
->dark_wm
= ALIGN(UBIFS_MAX_NODE_SZ
, c
->min_io_size
);
578 * Calculate how many bytes would be wasted at the end of LEB if it was
579 * fully filled with data nodes of maximum size. This is used in
580 * calculations when reporting free space.
582 c
->leb_overhead
= c
->leb_size
% UBIFS_MAX_DATA_NODE_SZ
;
584 /* Buffer size for bulk-reads */
585 c
->max_bu_buf_len
= UBIFS_MAX_BULK_READ
* UBIFS_MAX_DATA_NODE_SZ
;
586 if (c
->max_bu_buf_len
> c
->leb_size
)
587 c
->max_bu_buf_len
= c
->leb_size
;
592 * bud_wbuf_callback - bud LEB write-buffer synchronization call-back.
593 * @c: UBIFS file-system description object
594 * @lnum: LEB the write-buffer was synchronized to
595 * @free: how many free bytes left in this LEB
596 * @pad: how many bytes were padded
598 * This is a callback function which is called by the I/O unit when the
599 * write-buffer is synchronized. We need this to correctly maintain space
600 * accounting in bud logical eraseblocks. This function returns zero in case of
601 * success and a negative error code in case of failure.
603 * This function actually belongs to the journal, but we keep it here because
604 * we want to keep it static.
606 static int bud_wbuf_callback(struct ubifs_info
*c
, int lnum
, int free
, int pad
)
608 return ubifs_update_one_lp(c
, lnum
, free
, pad
, 0, 0);
612 * init_constants_sb - initialize UBIFS constants.
613 * @c: UBIFS file-system description object
615 * This is a helper function which initializes various UBIFS constants after
616 * the superblock has been read. It also checks various UBIFS parameters and
617 * makes sure they are all right. Returns zero in case of success and a
618 * negative error code in case of failure.
620 static int init_constants_sb(struct ubifs_info
*c
)
625 c
->main_bytes
= (long long)c
->main_lebs
* c
->leb_size
;
626 c
->max_znode_sz
= sizeof(struct ubifs_znode
) +
627 c
->fanout
* sizeof(struct ubifs_zbranch
);
629 tmp
= ubifs_idx_node_sz(c
, 1);
630 c
->ranges
[UBIFS_IDX_NODE
].min_len
= tmp
;
631 c
->min_idx_node_sz
= ALIGN(tmp
, 8);
633 tmp
= ubifs_idx_node_sz(c
, c
->fanout
);
634 c
->ranges
[UBIFS_IDX_NODE
].max_len
= tmp
;
635 c
->max_idx_node_sz
= ALIGN(tmp
, 8);
637 /* Make sure LEB size is large enough to fit full commit */
638 tmp
= UBIFS_CS_NODE_SZ
+ UBIFS_REF_NODE_SZ
* c
->jhead_cnt
;
639 tmp
= ALIGN(tmp
, c
->min_io_size
);
640 if (tmp
> c
->leb_size
) {
641 dbg_err("too small LEB size %d, at least %d needed",
647 * Make sure that the log is large enough to fit reference nodes for
648 * all buds plus one reserved LEB.
650 tmp64
= c
->max_bud_bytes
+ c
->leb_size
- 1;
651 c
->max_bud_cnt
= div_u64(tmp64
, c
->leb_size
);
652 tmp
= (c
->ref_node_alsz
* c
->max_bud_cnt
+ c
->leb_size
- 1);
655 if (c
->log_lebs
< tmp
) {
656 dbg_err("too small log %d LEBs, required min. %d LEBs",
662 * When budgeting we assume worst-case scenarios when the pages are not
663 * be compressed and direntries are of the maximum size.
665 * Note, data, which may be stored in inodes is budgeted separately, so
666 * it is not included into 'c->inode_budget'.
668 c
->page_budget
= UBIFS_MAX_DATA_NODE_SZ
* UBIFS_BLOCKS_PER_PAGE
;
669 c
->inode_budget
= UBIFS_INO_NODE_SZ
;
670 c
->dent_budget
= UBIFS_MAX_DENT_NODE_SZ
;
673 * When the amount of flash space used by buds becomes
674 * 'c->max_bud_bytes', UBIFS just blocks all writers and starts commit.
675 * The writers are unblocked when the commit is finished. To avoid
676 * writers to be blocked UBIFS initiates background commit in advance,
677 * when number of bud bytes becomes above the limit defined below.
679 c
->bg_bud_bytes
= (c
->max_bud_bytes
* 13) >> 4;
682 * Ensure minimum journal size. All the bytes in the journal heads are
683 * considered to be used, when calculating the current journal usage.
684 * Consequently, if the journal is too small, UBIFS will treat it as
687 tmp64
= (long long)(c
->jhead_cnt
+ 1) * c
->leb_size
+ 1;
688 if (c
->bg_bud_bytes
< tmp64
)
689 c
->bg_bud_bytes
= tmp64
;
690 if (c
->max_bud_bytes
< tmp64
+ c
->leb_size
)
691 c
->max_bud_bytes
= tmp64
+ c
->leb_size
;
693 err
= ubifs_calc_lpt_geom(c
);
697 /* Initialize effective LEB size used in budgeting calculations */
698 c
->idx_leb_size
= c
->leb_size
- c
->max_idx_node_sz
;
703 * init_constants_master - initialize UBIFS constants.
704 * @c: UBIFS file-system description object
706 * This is a helper function which initializes various UBIFS constants after
707 * the master node has been read. It also checks various UBIFS parameters and
708 * makes sure they are all right.
710 static void init_constants_master(struct ubifs_info
*c
)
714 c
->min_idx_lebs
= ubifs_calc_min_idx_lebs(c
);
715 c
->report_rp_size
= ubifs_reported_space(c
, c
->rp_size
);
718 * Calculate total amount of FS blocks. This number is not used
719 * internally because it does not make much sense for UBIFS, but it is
720 * necessary to report something for the 'statfs()' call.
722 * Subtract the LEB reserved for GC, the LEB which is reserved for
723 * deletions, minimum LEBs for the index, and assume only one journal
726 tmp64
= c
->main_lebs
- 1 - 1 - MIN_INDEX_LEBS
- c
->jhead_cnt
+ 1;
727 tmp64
*= (long long)c
->leb_size
- c
->leb_overhead
;
728 tmp64
= ubifs_reported_space(c
, tmp64
);
729 c
->block_cnt
= tmp64
>> UBIFS_BLOCK_SHIFT
;
733 * take_gc_lnum - reserve GC LEB.
734 * @c: UBIFS file-system description object
736 * This function ensures that the LEB reserved for garbage collection is marked
737 * as "taken" in lprops. We also have to set free space to LEB size and dirty
738 * space to zero, because lprops may contain out-of-date information if the
739 * file-system was un-mounted before it has been committed. This function
740 * returns zero in case of success and a negative error code in case of
743 static int take_gc_lnum(struct ubifs_info
*c
)
747 if (c
->gc_lnum
== -1) {
748 ubifs_err("no LEB for GC");
752 /* And we have to tell lprops that this LEB is taken */
753 err
= ubifs_change_one_lp(c
, c
->gc_lnum
, c
->leb_size
, 0,
759 * alloc_wbufs - allocate write-buffers.
760 * @c: UBIFS file-system description object
762 * This helper function allocates and initializes UBIFS write-buffers. Returns
763 * zero in case of success and %-ENOMEM in case of failure.
765 static int alloc_wbufs(struct ubifs_info
*c
)
769 c
->jheads
= kzalloc(c
->jhead_cnt
* sizeof(struct ubifs_jhead
),
774 /* Initialize journal heads */
775 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
776 INIT_LIST_HEAD(&c
->jheads
[i
].buds_list
);
777 err
= ubifs_wbuf_init(c
, &c
->jheads
[i
].wbuf
);
781 c
->jheads
[i
].wbuf
.sync_callback
= &bud_wbuf_callback
;
782 c
->jheads
[i
].wbuf
.jhead
= i
;
785 c
->jheads
[BASEHD
].wbuf
.dtype
= UBI_SHORTTERM
;
787 * Garbage Collector head likely contains long-term data and
788 * does not need to be synchronized by timer.
790 c
->jheads
[GCHD
].wbuf
.dtype
= UBI_LONGTERM
;
791 c
->jheads
[GCHD
].wbuf
.no_timer
= 1;
797 * free_wbufs - free write-buffers.
798 * @c: UBIFS file-system description object
800 static void free_wbufs(struct ubifs_info
*c
)
805 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
806 kfree(c
->jheads
[i
].wbuf
.buf
);
807 kfree(c
->jheads
[i
].wbuf
.inodes
);
815 * free_orphans - free orphans.
816 * @c: UBIFS file-system description object
818 static void free_orphans(struct ubifs_info
*c
)
820 struct ubifs_orphan
*orph
;
822 while (c
->orph_dnext
) {
823 orph
= c
->orph_dnext
;
824 c
->orph_dnext
= orph
->dnext
;
825 list_del(&orph
->list
);
829 while (!list_empty(&c
->orph_list
)) {
830 orph
= list_entry(c
->orph_list
.next
, struct ubifs_orphan
, list
);
831 list_del(&orph
->list
);
833 dbg_err("orphan list not empty at unmount");
841 * free_buds - free per-bud objects.
842 * @c: UBIFS file-system description object
844 static void free_buds(struct ubifs_info
*c
)
846 struct rb_node
*this = c
->buds
.rb_node
;
847 struct ubifs_bud
*bud
;
851 this = this->rb_left
;
852 else if (this->rb_right
)
853 this = this->rb_right
;
855 bud
= rb_entry(this, struct ubifs_bud
, rb
);
856 this = rb_parent(this);
858 if (this->rb_left
== &bud
->rb
)
859 this->rb_left
= NULL
;
861 this->rb_right
= NULL
;
869 * check_volume_empty - check if the UBI volume is empty.
870 * @c: UBIFS file-system description object
872 * This function checks if the UBIFS volume is empty by looking if its LEBs are
873 * mapped or not. The result of checking is stored in the @c->empty variable.
874 * Returns zero in case of success and a negative error code in case of
877 static int check_volume_empty(struct ubifs_info
*c
)
882 for (lnum
= 0; lnum
< c
->leb_cnt
; lnum
++) {
883 err
= ubi_is_mapped(c
->ubi
, lnum
);
884 if (unlikely(err
< 0))
898 * UBIFS mount options.
900 * Opt_fast_unmount: do not run a journal commit before un-mounting
901 * Opt_norm_unmount: run a journal commit before un-mounting
902 * Opt_bulk_read: enable bulk-reads
903 * Opt_no_bulk_read: disable bulk-reads
904 * Opt_chk_data_crc: check CRCs when reading data nodes
905 * Opt_no_chk_data_crc: do not check CRCs when reading data nodes
906 * Opt_override_compr: override default compressor
907 * Opt_err: just end of array marker
920 static const match_table_t tokens
= {
921 {Opt_fast_unmount
, "fast_unmount"},
922 {Opt_norm_unmount
, "norm_unmount"},
923 {Opt_bulk_read
, "bulk_read"},
924 {Opt_no_bulk_read
, "no_bulk_read"},
925 {Opt_chk_data_crc
, "chk_data_crc"},
926 {Opt_no_chk_data_crc
, "no_chk_data_crc"},
927 {Opt_override_compr
, "compr=%s"},
932 * parse_standard_option - parse a standard mount option.
933 * @option: the option to parse
935 * Normally, standard mount options like "sync" are passed to file-systems as
936 * flags. However, when a "rootflags=" kernel boot parameter is used, they may
937 * be present in the options string. This function tries to deal with this
938 * situation and parse standard options. Returns 0 if the option was not
939 * recognized, and the corresponding integer flag if it was.
941 * UBIFS is only interested in the "sync" option, so do not check for anything
944 static int parse_standard_option(const char *option
)
946 ubifs_msg("parse %s", option
);
947 if (!strcmp(option
, "sync"))
948 return MS_SYNCHRONOUS
;
953 * ubifs_parse_options - parse mount parameters.
954 * @c: UBIFS file-system description object
955 * @options: parameters to parse
956 * @is_remount: non-zero if this is FS re-mount
958 * This function parses UBIFS mount options and returns zero in case success
959 * and a negative error code in case of failure.
961 static int ubifs_parse_options(struct ubifs_info
*c
, char *options
,
965 substring_t args
[MAX_OPT_ARGS
];
970 while ((p
= strsep(&options
, ","))) {
976 token
= match_token(p
, tokens
, args
);
979 * %Opt_fast_unmount and %Opt_norm_unmount options are ignored.
980 * We accept them in order to be backward-compatible. But this
981 * should be removed at some point.
983 case Opt_fast_unmount
:
984 c
->mount_opts
.unmount_mode
= 2;
986 case Opt_norm_unmount
:
987 c
->mount_opts
.unmount_mode
= 1;
990 c
->mount_opts
.bulk_read
= 2;
993 case Opt_no_bulk_read
:
994 c
->mount_opts
.bulk_read
= 1;
997 case Opt_chk_data_crc
:
998 c
->mount_opts
.chk_data_crc
= 2;
999 c
->no_chk_data_crc
= 0;
1001 case Opt_no_chk_data_crc
:
1002 c
->mount_opts
.chk_data_crc
= 1;
1003 c
->no_chk_data_crc
= 1;
1005 case Opt_override_compr
:
1007 char *name
= match_strdup(&args
[0]);
1011 if (!strcmp(name
, "none"))
1012 c
->mount_opts
.compr_type
= UBIFS_COMPR_NONE
;
1013 else if (!strcmp(name
, "lzo"))
1014 c
->mount_opts
.compr_type
= UBIFS_COMPR_LZO
;
1015 else if (!strcmp(name
, "zlib"))
1016 c
->mount_opts
.compr_type
= UBIFS_COMPR_ZLIB
;
1018 ubifs_err("unknown compressor \"%s\"", name
);
1023 c
->mount_opts
.override_compr
= 1;
1024 c
->default_compr
= c
->mount_opts
.compr_type
;
1030 struct super_block
*sb
= c
->vfs_sb
;
1032 flag
= parse_standard_option(p
);
1034 ubifs_err("unrecognized mount option \"%s\" "
1035 "or missing value", p
);
1038 sb
->s_flags
|= flag
;
1048 * destroy_journal - destroy journal data structures.
1049 * @c: UBIFS file-system description object
1051 * This function destroys journal data structures including those that may have
1052 * been created by recovery functions.
1054 static void destroy_journal(struct ubifs_info
*c
)
1056 while (!list_empty(&c
->unclean_leb_list
)) {
1057 struct ubifs_unclean_leb
*ucleb
;
1059 ucleb
= list_entry(c
->unclean_leb_list
.next
,
1060 struct ubifs_unclean_leb
, list
);
1061 list_del(&ucleb
->list
);
1064 while (!list_empty(&c
->old_buds
)) {
1065 struct ubifs_bud
*bud
;
1067 bud
= list_entry(c
->old_buds
.next
, struct ubifs_bud
, list
);
1068 list_del(&bud
->list
);
1071 ubifs_destroy_idx_gc(c
);
1072 ubifs_destroy_size_tree(c
);
1078 * bu_init - initialize bulk-read information.
1079 * @c: UBIFS file-system description object
1081 static void bu_init(struct ubifs_info
*c
)
1083 ubifs_assert(c
->bulk_read
== 1);
1086 return; /* Already initialized */
1089 c
->bu
.buf
= kmalloc(c
->max_bu_buf_len
, GFP_KERNEL
| __GFP_NOWARN
);
1091 if (c
->max_bu_buf_len
> UBIFS_KMALLOC_OK
) {
1092 c
->max_bu_buf_len
= UBIFS_KMALLOC_OK
;
1096 /* Just disable bulk-read */
1097 ubifs_warn("Cannot allocate %d bytes of memory for bulk-read, "
1098 "disabling it", c
->max_bu_buf_len
);
1099 c
->mount_opts
.bulk_read
= 1;
1106 * check_free_space - check if there is enough free space to mount.
1107 * @c: UBIFS file-system description object
1109 * This function makes sure UBIFS has enough free space to be mounted in
1110 * read/write mode. UBIFS must always have some free space to allow deletions.
1112 static int check_free_space(struct ubifs_info
*c
)
1114 ubifs_assert(c
->dark_wm
> 0);
1115 if (c
->lst
.total_free
+ c
->lst
.total_dirty
< c
->dark_wm
) {
1116 ubifs_err("insufficient free space to mount in read/write mode");
1125 * mount_ubifs - mount UBIFS file-system.
1126 * @c: UBIFS file-system description object
1128 * This function mounts UBIFS file system. Returns zero in case of success and
1129 * a negative error code in case of failure.
1131 * Note, the function does not de-allocate resources it it fails half way
1132 * through, and the caller has to do this instead.
1134 static int mount_ubifs(struct ubifs_info
*c
)
1136 struct super_block
*sb
= c
->vfs_sb
;
1137 int err
, mounted_read_only
= (sb
->s_flags
& MS_RDONLY
);
1141 err
= init_constants_early(c
);
1145 err
= ubifs_debugging_init(c
);
1149 err
= check_volume_empty(c
);
1153 if (c
->empty
&& (mounted_read_only
|| c
->ro_media
)) {
1155 * This UBI volume is empty, and read-only, or the file system
1156 * is mounted read-only - we cannot format it.
1158 ubifs_err("can't format empty UBI volume: read-only %s",
1159 c
->ro_media
? "UBI volume" : "mount");
1164 if (c
->ro_media
&& !mounted_read_only
) {
1165 ubifs_err("cannot mount read-write - read-only media");
1171 * The requirement for the buffer is that it should fit indexing B-tree
1172 * height amount of integers. We assume the height if the TNC tree will
1176 c
->bottom_up_buf
= kmalloc(BOTTOM_UP_HEIGHT
* sizeof(int), GFP_KERNEL
);
1177 if (!c
->bottom_up_buf
)
1180 c
->sbuf
= vmalloc(c
->leb_size
);
1184 if (!mounted_read_only
) {
1185 c
->ileb_buf
= vmalloc(c
->leb_size
);
1190 if (c
->bulk_read
== 1)
1194 * We have to check all CRCs, even for data nodes, when we mount the FS
1195 * (specifically, when we are replaying).
1197 c
->always_chk_crc
= 1;
1199 err
= ubifs_read_superblock(c
);
1204 * Make sure the compressor which is set as default in the superblock
1205 * or overridden by mount options is actually compiled in.
1207 if (!ubifs_compr_present(c
->default_compr
)) {
1208 ubifs_err("'compressor \"%s\" is not compiled in",
1209 ubifs_compr_name(c
->default_compr
));
1214 err
= init_constants_sb(c
);
1218 sz
= ALIGN(c
->max_idx_node_sz
, c
->min_io_size
);
1219 sz
= ALIGN(sz
+ c
->max_idx_node_sz
, c
->min_io_size
);
1220 c
->cbuf
= kmalloc(sz
, GFP_NOFS
);
1226 sprintf(c
->bgt_name
, BGT_NAME_PATTERN
, c
->vi
.ubi_num
, c
->vi
.vol_id
);
1227 if (!mounted_read_only
) {
1228 err
= alloc_wbufs(c
);
1232 /* Create background thread */
1233 c
->bgt
= kthread_create(ubifs_bg_thread
, c
, "%s", c
->bgt_name
);
1234 if (IS_ERR(c
->bgt
)) {
1235 err
= PTR_ERR(c
->bgt
);
1237 ubifs_err("cannot spawn \"%s\", error %d",
1241 wake_up_process(c
->bgt
);
1244 err
= ubifs_read_master(c
);
1248 init_constants_master(c
);
1250 if ((c
->mst_node
->flags
& cpu_to_le32(UBIFS_MST_DIRTY
)) != 0) {
1251 ubifs_msg("recovery needed");
1252 c
->need_recovery
= 1;
1253 if (!mounted_read_only
) {
1254 err
= ubifs_recover_inl_heads(c
, c
->sbuf
);
1258 } else if (!mounted_read_only
) {
1260 * Set the "dirty" flag so that if we reboot uncleanly we
1261 * will notice this immediately on the next mount.
1263 c
->mst_node
->flags
|= cpu_to_le32(UBIFS_MST_DIRTY
);
1264 err
= ubifs_write_master(c
);
1269 err
= ubifs_lpt_init(c
, 1, !mounted_read_only
);
1273 err
= dbg_check_idx_size(c
, c
->old_idx_sz
);
1277 err
= ubifs_replay_journal(c
);
1281 /* Calculate 'min_idx_lebs' after journal replay */
1282 c
->min_idx_lebs
= ubifs_calc_min_idx_lebs(c
);
1284 err
= ubifs_mount_orphans(c
, c
->need_recovery
, mounted_read_only
);
1288 if (!mounted_read_only
) {
1291 err
= check_free_space(c
);
1295 /* Check for enough log space */
1296 lnum
= c
->lhead_lnum
+ 1;
1297 if (lnum
>= UBIFS_LOG_LNUM
+ c
->log_lebs
)
1298 lnum
= UBIFS_LOG_LNUM
;
1299 if (lnum
== c
->ltail_lnum
) {
1300 err
= ubifs_consolidate_log(c
);
1305 if (c
->need_recovery
) {
1306 err
= ubifs_recover_size(c
);
1309 err
= ubifs_rcvry_gc_commit(c
);
1311 err
= take_gc_lnum(c
);
1316 * GC LEB may contain garbage if there was an unclean
1317 * reboot, and it should be un-mapped.
1319 err
= ubifs_leb_unmap(c
, c
->gc_lnum
);
1324 err
= dbg_check_lprops(c
);
1327 } else if (c
->need_recovery
) {
1328 err
= ubifs_recover_size(c
);
1333 * Even if we mount read-only, we have to set space in GC LEB
1334 * to proper value because this affects UBIFS free space
1335 * reporting. We do not want to have a situation when
1336 * re-mounting from R/O to R/W changes amount of free space.
1338 err
= take_gc_lnum(c
);
1343 spin_lock(&ubifs_infos_lock
);
1344 list_add_tail(&c
->infos_list
, &ubifs_infos
);
1345 spin_unlock(&ubifs_infos_lock
);
1347 if (c
->need_recovery
) {
1348 if (mounted_read_only
)
1349 ubifs_msg("recovery deferred");
1351 c
->need_recovery
= 0;
1352 ubifs_msg("recovery completed");
1354 * GC LEB has to be empty and taken at this point. But
1355 * the journal head LEBs may also be accounted as
1356 * "empty taken" if they are empty.
1358 ubifs_assert(c
->lst
.taken_empty_lebs
> 0);
1361 ubifs_assert(c
->lst
.taken_empty_lebs
> 0);
1363 err
= dbg_check_filesystem(c
);
1367 err
= dbg_debugfs_init_fs(c
);
1371 c
->always_chk_crc
= 0;
1373 ubifs_msg("mounted UBI device %d, volume %d, name \"%s\"",
1374 c
->vi
.ubi_num
, c
->vi
.vol_id
, c
->vi
.name
);
1375 if (mounted_read_only
)
1376 ubifs_msg("mounted read-only");
1377 x
= (long long)c
->main_lebs
* c
->leb_size
;
1378 ubifs_msg("file system size: %lld bytes (%lld KiB, %lld MiB, %d "
1379 "LEBs)", x
, x
>> 10, x
>> 20, c
->main_lebs
);
1380 x
= (long long)c
->log_lebs
* c
->leb_size
+ c
->max_bud_bytes
;
1381 ubifs_msg("journal size: %lld bytes (%lld KiB, %lld MiB, %d "
1382 "LEBs)", x
, x
>> 10, x
>> 20, c
->log_lebs
+ c
->max_bud_cnt
);
1383 ubifs_msg("media format: w%d/r%d (latest is w%d/r%d)",
1384 c
->fmt_version
, c
->ro_compat_version
,
1385 UBIFS_FORMAT_VERSION
, UBIFS_RO_COMPAT_VERSION
);
1386 ubifs_msg("default compressor: %s", ubifs_compr_name(c
->default_compr
));
1387 ubifs_msg("reserved for root: %llu bytes (%llu KiB)",
1388 c
->report_rp_size
, c
->report_rp_size
>> 10);
1390 dbg_msg("compiled on: " __DATE__
" at " __TIME__
);
1391 dbg_msg("min. I/O unit size: %d bytes", c
->min_io_size
);
1392 dbg_msg("LEB size: %d bytes (%d KiB)",
1393 c
->leb_size
, c
->leb_size
>> 10);
1394 dbg_msg("data journal heads: %d",
1395 c
->jhead_cnt
- NONDATA_JHEADS_CNT
);
1396 dbg_msg("UUID: %pUB", c
->uuid
);
1397 dbg_msg("big_lpt %d", c
->big_lpt
);
1398 dbg_msg("log LEBs: %d (%d - %d)",
1399 c
->log_lebs
, UBIFS_LOG_LNUM
, c
->log_last
);
1400 dbg_msg("LPT area LEBs: %d (%d - %d)",
1401 c
->lpt_lebs
, c
->lpt_first
, c
->lpt_last
);
1402 dbg_msg("orphan area LEBs: %d (%d - %d)",
1403 c
->orph_lebs
, c
->orph_first
, c
->orph_last
);
1404 dbg_msg("main area LEBs: %d (%d - %d)",
1405 c
->main_lebs
, c
->main_first
, c
->leb_cnt
- 1);
1406 dbg_msg("index LEBs: %d", c
->lst
.idx_lebs
);
1407 dbg_msg("total index bytes: %lld (%lld KiB, %lld MiB)",
1408 c
->old_idx_sz
, c
->old_idx_sz
>> 10, c
->old_idx_sz
>> 20);
1409 dbg_msg("key hash type: %d", c
->key_hash_type
);
1410 dbg_msg("tree fanout: %d", c
->fanout
);
1411 dbg_msg("reserved GC LEB: %d", c
->gc_lnum
);
1412 dbg_msg("first main LEB: %d", c
->main_first
);
1413 dbg_msg("max. znode size %d", c
->max_znode_sz
);
1414 dbg_msg("max. index node size %d", c
->max_idx_node_sz
);
1415 dbg_msg("node sizes: data %zu, inode %zu, dentry %zu",
1416 UBIFS_DATA_NODE_SZ
, UBIFS_INO_NODE_SZ
, UBIFS_DENT_NODE_SZ
);
1417 dbg_msg("node sizes: trun %zu, sb %zu, master %zu",
1418 UBIFS_TRUN_NODE_SZ
, UBIFS_SB_NODE_SZ
, UBIFS_MST_NODE_SZ
);
1419 dbg_msg("node sizes: ref %zu, cmt. start %zu, orph %zu",
1420 UBIFS_REF_NODE_SZ
, UBIFS_CS_NODE_SZ
, UBIFS_ORPH_NODE_SZ
);
1421 dbg_msg("max. node sizes: data %zu, inode %zu dentry %zu",
1422 UBIFS_MAX_DATA_NODE_SZ
, UBIFS_MAX_INO_NODE_SZ
,
1423 UBIFS_MAX_DENT_NODE_SZ
);
1424 dbg_msg("dead watermark: %d", c
->dead_wm
);
1425 dbg_msg("dark watermark: %d", c
->dark_wm
);
1426 dbg_msg("LEB overhead: %d", c
->leb_overhead
);
1427 x
= (long long)c
->main_lebs
* c
->dark_wm
;
1428 dbg_msg("max. dark space: %lld (%lld KiB, %lld MiB)",
1429 x
, x
>> 10, x
>> 20);
1430 dbg_msg("maximum bud bytes: %lld (%lld KiB, %lld MiB)",
1431 c
->max_bud_bytes
, c
->max_bud_bytes
>> 10,
1432 c
->max_bud_bytes
>> 20);
1433 dbg_msg("BG commit bud bytes: %lld (%lld KiB, %lld MiB)",
1434 c
->bg_bud_bytes
, c
->bg_bud_bytes
>> 10,
1435 c
->bg_bud_bytes
>> 20);
1436 dbg_msg("current bud bytes %lld (%lld KiB, %lld MiB)",
1437 c
->bud_bytes
, c
->bud_bytes
>> 10, c
->bud_bytes
>> 20);
1438 dbg_msg("max. seq. number: %llu", c
->max_sqnum
);
1439 dbg_msg("commit number: %llu", c
->cmt_no
);
1444 spin_lock(&ubifs_infos_lock
);
1445 list_del(&c
->infos_list
);
1446 spin_unlock(&ubifs_infos_lock
);
1452 ubifs_lpt_free(c
, 0);
1455 kfree(c
->rcvrd_mst_node
);
1457 kthread_stop(c
->bgt
);
1466 kfree(c
->bottom_up_buf
);
1467 ubifs_debugging_exit(c
);
1472 * ubifs_umount - un-mount UBIFS file-system.
1473 * @c: UBIFS file-system description object
1475 * Note, this function is called to free allocated resourced when un-mounting,
1476 * as well as free resources when an error occurred while we were half way
1477 * through mounting (error path cleanup function). So it has to make sure the
1478 * resource was actually allocated before freeing it.
1480 static void ubifs_umount(struct ubifs_info
*c
)
1482 dbg_gen("un-mounting UBI device %d, volume %d", c
->vi
.ubi_num
,
1485 dbg_debugfs_exit_fs(c
);
1486 spin_lock(&ubifs_infos_lock
);
1487 list_del(&c
->infos_list
);
1488 spin_unlock(&ubifs_infos_lock
);
1491 kthread_stop(c
->bgt
);
1496 ubifs_lpt_free(c
, 0);
1499 kfree(c
->rcvrd_mst_node
);
1504 kfree(c
->bottom_up_buf
);
1505 ubifs_debugging_exit(c
);
1509 * ubifs_remount_rw - re-mount in read-write mode.
1510 * @c: UBIFS file-system description object
1512 * UBIFS avoids allocating many unnecessary resources when mounted in read-only
1513 * mode. This function allocates the needed resources and re-mounts UBIFS in
1516 static int ubifs_remount_rw(struct ubifs_info
*c
)
1520 if (c
->rw_incompat
) {
1521 ubifs_err("the file-system is not R/W-compatible");
1522 ubifs_msg("on-flash format version is w%d/r%d, but software "
1523 "only supports up to version w%d/r%d", c
->fmt_version
,
1524 c
->ro_compat_version
, UBIFS_FORMAT_VERSION
,
1525 UBIFS_RO_COMPAT_VERSION
);
1529 mutex_lock(&c
->umount_mutex
);
1530 dbg_save_space_info(c
);
1531 c
->remounting_rw
= 1;
1532 c
->always_chk_crc
= 1;
1534 err
= check_free_space(c
);
1538 if (c
->old_leb_cnt
!= c
->leb_cnt
) {
1539 struct ubifs_sb_node
*sup
;
1541 sup
= ubifs_read_sb_node(c
);
1546 sup
->leb_cnt
= cpu_to_le32(c
->leb_cnt
);
1547 err
= ubifs_write_sb_node(c
, sup
);
1552 if (c
->need_recovery
) {
1553 ubifs_msg("completing deferred recovery");
1554 err
= ubifs_write_rcvrd_mst_node(c
);
1557 err
= ubifs_recover_size(c
);
1560 err
= ubifs_clean_lebs(c
, c
->sbuf
);
1563 err
= ubifs_recover_inl_heads(c
, c
->sbuf
);
1567 /* A readonly mount is not allowed to have orphans */
1568 ubifs_assert(c
->tot_orphans
== 0);
1569 err
= ubifs_clear_orphans(c
);
1574 if (!(c
->mst_node
->flags
& cpu_to_le32(UBIFS_MST_DIRTY
))) {
1575 c
->mst_node
->flags
|= cpu_to_le32(UBIFS_MST_DIRTY
);
1576 err
= ubifs_write_master(c
);
1581 c
->ileb_buf
= vmalloc(c
->leb_size
);
1587 err
= ubifs_lpt_init(c
, 0, 1);
1591 err
= alloc_wbufs(c
);
1595 ubifs_create_buds_lists(c
);
1597 /* Create background thread */
1598 c
->bgt
= kthread_create(ubifs_bg_thread
, c
, "%s", c
->bgt_name
);
1599 if (IS_ERR(c
->bgt
)) {
1600 err
= PTR_ERR(c
->bgt
);
1602 ubifs_err("cannot spawn \"%s\", error %d",
1606 wake_up_process(c
->bgt
);
1608 c
->orph_buf
= vmalloc(c
->leb_size
);
1614 /* Check for enough log space */
1615 lnum
= c
->lhead_lnum
+ 1;
1616 if (lnum
>= UBIFS_LOG_LNUM
+ c
->log_lebs
)
1617 lnum
= UBIFS_LOG_LNUM
;
1618 if (lnum
== c
->ltail_lnum
) {
1619 err
= ubifs_consolidate_log(c
);
1624 if (c
->need_recovery
)
1625 err
= ubifs_rcvry_gc_commit(c
);
1627 err
= ubifs_leb_unmap(c
, c
->gc_lnum
);
1631 if (c
->need_recovery
) {
1632 c
->need_recovery
= 0;
1633 ubifs_msg("deferred recovery completed");
1636 dbg_gen("re-mounted read-write");
1637 c
->vfs_sb
->s_flags
&= ~MS_RDONLY
;
1638 c
->remounting_rw
= 0;
1639 c
->always_chk_crc
= 0;
1640 err
= dbg_check_space_info(c
);
1641 mutex_unlock(&c
->umount_mutex
);
1648 kthread_stop(c
->bgt
);
1654 ubifs_lpt_free(c
, 1);
1655 c
->remounting_rw
= 0;
1656 c
->always_chk_crc
= 0;
1657 mutex_unlock(&c
->umount_mutex
);
1662 * ubifs_remount_ro - re-mount in read-only mode.
1663 * @c: UBIFS file-system description object
1665 * We assume VFS has stopped writing. Possibly the background thread could be
1666 * running a commit, however kthread_stop will wait in that case.
1668 static void ubifs_remount_ro(struct ubifs_info
*c
)
1672 ubifs_assert(!c
->need_recovery
);
1673 ubifs_assert(!(c
->vfs_sb
->s_flags
& MS_RDONLY
));
1675 mutex_lock(&c
->umount_mutex
);
1677 kthread_stop(c
->bgt
);
1681 dbg_save_space_info(c
);
1683 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
1684 ubifs_wbuf_sync(&c
->jheads
[i
].wbuf
);
1685 hrtimer_cancel(&c
->jheads
[i
].wbuf
.timer
);
1688 c
->mst_node
->flags
&= ~cpu_to_le32(UBIFS_MST_DIRTY
);
1689 c
->mst_node
->flags
|= cpu_to_le32(UBIFS_MST_NO_ORPHS
);
1690 c
->mst_node
->gc_lnum
= cpu_to_le32(c
->gc_lnum
);
1691 err
= ubifs_write_master(c
);
1693 ubifs_ro_mode(c
, err
);
1700 ubifs_lpt_free(c
, 1);
1701 err
= dbg_check_space_info(c
);
1703 ubifs_ro_mode(c
, err
);
1704 mutex_unlock(&c
->umount_mutex
);
1707 static void ubifs_put_super(struct super_block
*sb
)
1710 struct ubifs_info
*c
= sb
->s_fs_info
;
1712 ubifs_msg("un-mount UBI device %d, volume %d", c
->vi
.ubi_num
,
1716 * The following asserts are only valid if there has not been a failure
1717 * of the media. For example, there will be dirty inodes if we failed
1718 * to write them back because of I/O errors.
1720 ubifs_assert(atomic_long_read(&c
->dirty_pg_cnt
) == 0);
1721 ubifs_assert(c
->budg_idx_growth
== 0);
1722 ubifs_assert(c
->budg_dd_growth
== 0);
1723 ubifs_assert(c
->budg_data_growth
== 0);
1726 * The 'c->umount_lock' prevents races between UBIFS memory shrinker
1727 * and file system un-mount. Namely, it prevents the shrinker from
1728 * picking this superblock for shrinking - it will be just skipped if
1729 * the mutex is locked.
1731 mutex_lock(&c
->umount_mutex
);
1732 if (!(c
->vfs_sb
->s_flags
& MS_RDONLY
)) {
1734 * First of all kill the background thread to make sure it does
1735 * not interfere with un-mounting and freeing resources.
1738 kthread_stop(c
->bgt
);
1742 /* Synchronize write-buffers */
1744 for (i
= 0; i
< c
->jhead_cnt
; i
++)
1745 ubifs_wbuf_sync(&c
->jheads
[i
].wbuf
);
1748 * On fatal errors c->ro_media is set to 1, in which case we do
1749 * not write the master node.
1753 * We are being cleanly unmounted which means the
1754 * orphans were killed - indicate this in the master
1755 * node. Also save the reserved GC LEB number.
1759 c
->mst_node
->flags
&= ~cpu_to_le32(UBIFS_MST_DIRTY
);
1760 c
->mst_node
->flags
|= cpu_to_le32(UBIFS_MST_NO_ORPHS
);
1761 c
->mst_node
->gc_lnum
= cpu_to_le32(c
->gc_lnum
);
1762 err
= ubifs_write_master(c
);
1765 * Recovery will attempt to fix the master area
1766 * next mount, so we just print a message and
1767 * continue to unmount normally.
1769 ubifs_err("failed to write master node, "
1775 bdi_destroy(&c
->bdi
);
1776 ubi_close_volume(c
->ubi
);
1777 mutex_unlock(&c
->umount_mutex
);
1781 static int ubifs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
1784 struct ubifs_info
*c
= sb
->s_fs_info
;
1786 dbg_gen("old flags %#lx, new flags %#x", sb
->s_flags
, *flags
);
1788 err
= ubifs_parse_options(c
, data
, 1);
1790 ubifs_err("invalid or unknown remount parameter");
1794 if ((sb
->s_flags
& MS_RDONLY
) && !(*flags
& MS_RDONLY
)) {
1796 ubifs_msg("cannot re-mount due to prior errors");
1799 err
= ubifs_remount_rw(c
);
1802 } else if (!(sb
->s_flags
& MS_RDONLY
) && (*flags
& MS_RDONLY
)) {
1804 ubifs_msg("cannot re-mount due to prior errors");
1807 ubifs_remount_ro(c
);
1810 if (c
->bulk_read
== 1)
1813 dbg_gen("disable bulk-read");
1818 ubifs_assert(c
->lst
.taken_empty_lebs
> 0);
1822 const struct super_operations ubifs_super_operations
= {
1823 .alloc_inode
= ubifs_alloc_inode
,
1824 .destroy_inode
= ubifs_destroy_inode
,
1825 .put_super
= ubifs_put_super
,
1826 .write_inode
= ubifs_write_inode
,
1827 .delete_inode
= ubifs_delete_inode
,
1828 .statfs
= ubifs_statfs
,
1829 .dirty_inode
= ubifs_dirty_inode
,
1830 .remount_fs
= ubifs_remount_fs
,
1831 .show_options
= ubifs_show_options
,
1832 .sync_fs
= ubifs_sync_fs
,
1836 * open_ubi - parse UBI device name string and open the UBI device.
1837 * @name: UBI volume name
1838 * @mode: UBI volume open mode
1840 * The primary method of mounting UBIFS is by specifying the UBI volume
1841 * character device node path. However, UBIFS may also be mounted withoug any
1842 * character device node using one of the following methods:
1844 * o ubiX_Y - mount UBI device number X, volume Y;
1845 * o ubiY - mount UBI device number 0, volume Y;
1846 * o ubiX:NAME - mount UBI device X, volume with name NAME;
1847 * o ubi:NAME - mount UBI device 0, volume with name NAME.
1849 * Alternative '!' separator may be used instead of ':' (because some shells
1850 * like busybox may interpret ':' as an NFS host name separator). This function
1851 * returns UBI volume description object in case of success and a negative
1852 * error code in case of failure.
1854 static struct ubi_volume_desc
*open_ubi(const char *name
, int mode
)
1856 struct ubi_volume_desc
*ubi
;
1860 /* First, try to open using the device node path method */
1861 ubi
= ubi_open_volume_path(name
, mode
);
1865 /* Try the "nodev" method */
1866 if (name
[0] != 'u' || name
[1] != 'b' || name
[2] != 'i')
1867 return ERR_PTR(-EINVAL
);
1869 /* ubi:NAME method */
1870 if ((name
[3] == ':' || name
[3] == '!') && name
[4] != '\0')
1871 return ubi_open_volume_nm(0, name
+ 4, mode
);
1873 if (!isdigit(name
[3]))
1874 return ERR_PTR(-EINVAL
);
1876 dev
= simple_strtoul(name
+ 3, &endptr
, 0);
1879 if (*endptr
== '\0')
1880 return ubi_open_volume(0, dev
, mode
);
1883 if (*endptr
== '_' && isdigit(endptr
[1])) {
1884 vol
= simple_strtoul(endptr
+ 1, &endptr
, 0);
1885 if (*endptr
!= '\0')
1886 return ERR_PTR(-EINVAL
);
1887 return ubi_open_volume(dev
, vol
, mode
);
1890 /* ubiX:NAME method */
1891 if ((*endptr
== ':' || *endptr
== '!') && endptr
[1] != '\0')
1892 return ubi_open_volume_nm(dev
, ++endptr
, mode
);
1894 return ERR_PTR(-EINVAL
);
1897 static int ubifs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1899 struct ubi_volume_desc
*ubi
= sb
->s_fs_info
;
1900 struct ubifs_info
*c
;
1904 c
= kzalloc(sizeof(struct ubifs_info
), GFP_KERNEL
);
1908 spin_lock_init(&c
->cnt_lock
);
1909 spin_lock_init(&c
->cs_lock
);
1910 spin_lock_init(&c
->buds_lock
);
1911 spin_lock_init(&c
->space_lock
);
1912 spin_lock_init(&c
->orphan_lock
);
1913 init_rwsem(&c
->commit_sem
);
1914 mutex_init(&c
->lp_mutex
);
1915 mutex_init(&c
->tnc_mutex
);
1916 mutex_init(&c
->log_mutex
);
1917 mutex_init(&c
->mst_mutex
);
1918 mutex_init(&c
->umount_mutex
);
1919 mutex_init(&c
->bu_mutex
);
1920 init_waitqueue_head(&c
->cmt_wq
);
1922 c
->old_idx
= RB_ROOT
;
1923 c
->size_tree
= RB_ROOT
;
1924 c
->orph_tree
= RB_ROOT
;
1925 INIT_LIST_HEAD(&c
->infos_list
);
1926 INIT_LIST_HEAD(&c
->idx_gc
);
1927 INIT_LIST_HEAD(&c
->replay_list
);
1928 INIT_LIST_HEAD(&c
->replay_buds
);
1929 INIT_LIST_HEAD(&c
->uncat_list
);
1930 INIT_LIST_HEAD(&c
->empty_list
);
1931 INIT_LIST_HEAD(&c
->freeable_list
);
1932 INIT_LIST_HEAD(&c
->frdi_idx_list
);
1933 INIT_LIST_HEAD(&c
->unclean_leb_list
);
1934 INIT_LIST_HEAD(&c
->old_buds
);
1935 INIT_LIST_HEAD(&c
->orph_list
);
1936 INIT_LIST_HEAD(&c
->orph_new
);
1939 c
->highest_inum
= UBIFS_FIRST_INO
;
1940 c
->lhead_lnum
= c
->ltail_lnum
= UBIFS_LOG_LNUM
;
1942 ubi_get_volume_info(ubi
, &c
->vi
);
1943 ubi_get_device_info(c
->vi
.ubi_num
, &c
->di
);
1945 /* Re-open the UBI device in read-write mode */
1946 c
->ubi
= ubi_open_volume(c
->vi
.ubi_num
, c
->vi
.vol_id
, UBI_READWRITE
);
1947 if (IS_ERR(c
->ubi
)) {
1948 err
= PTR_ERR(c
->ubi
);
1953 * UBIFS provides 'backing_dev_info' in order to disable read-ahead. For
1954 * UBIFS, I/O is not deferred, it is done immediately in readpage,
1955 * which means the user would have to wait not just for their own I/O
1956 * but the read-ahead I/O as well i.e. completely pointless.
1958 * Read-ahead will be disabled because @c->bdi.ra_pages is 0.
1960 c
->bdi
.name
= "ubifs",
1961 c
->bdi
.capabilities
= BDI_CAP_MAP_COPY
;
1962 c
->bdi
.unplug_io_fn
= default_unplug_io_fn
;
1963 err
= bdi_init(&c
->bdi
);
1966 err
= bdi_register(&c
->bdi
, NULL
, "ubifs_%d_%d",
1967 c
->vi
.ubi_num
, c
->vi
.vol_id
);
1971 err
= ubifs_parse_options(c
, data
, 0);
1975 sb
->s_bdi
= &c
->bdi
;
1977 sb
->s_magic
= UBIFS_SUPER_MAGIC
;
1978 sb
->s_blocksize
= UBIFS_BLOCK_SIZE
;
1979 sb
->s_blocksize_bits
= UBIFS_BLOCK_SHIFT
;
1980 sb
->s_maxbytes
= c
->max_inode_sz
= key_max_inode_size(c
);
1981 if (c
->max_inode_sz
> MAX_LFS_FILESIZE
)
1982 sb
->s_maxbytes
= c
->max_inode_sz
= MAX_LFS_FILESIZE
;
1983 sb
->s_op
= &ubifs_super_operations
;
1985 mutex_lock(&c
->umount_mutex
);
1986 err
= mount_ubifs(c
);
1988 ubifs_assert(err
< 0);
1992 /* Read the root inode */
1993 root
= ubifs_iget(sb
, UBIFS_ROOT_INO
);
1995 err
= PTR_ERR(root
);
1999 sb
->s_root
= d_alloc_root(root
);
2003 mutex_unlock(&c
->umount_mutex
);
2011 mutex_unlock(&c
->umount_mutex
);
2013 bdi_destroy(&c
->bdi
);
2015 ubi_close_volume(c
->ubi
);
2021 static int sb_test(struct super_block
*sb
, void *data
)
2024 struct ubifs_info
*c
= sb
->s_fs_info
;
2026 return c
->vi
.cdev
== *dev
;
2029 static int ubifs_get_sb(struct file_system_type
*fs_type
, int flags
,
2030 const char *name
, void *data
, struct vfsmount
*mnt
)
2032 struct ubi_volume_desc
*ubi
;
2033 struct ubi_volume_info vi
;
2034 struct super_block
*sb
;
2037 dbg_gen("name %s, flags %#x", name
, flags
);
2040 * Get UBI device number and volume ID. Mount it read-only so far
2041 * because this might be a new mount point, and UBI allows only one
2042 * read-write user at a time.
2044 ubi
= open_ubi(name
, UBI_READONLY
);
2046 ubifs_err("cannot open \"%s\", error %d",
2047 name
, (int)PTR_ERR(ubi
));
2048 return PTR_ERR(ubi
);
2050 ubi_get_volume_info(ubi
, &vi
);
2052 dbg_gen("opened ubi%d_%d", vi
.ubi_num
, vi
.vol_id
);
2054 sb
= sget(fs_type
, &sb_test
, &set_anon_super
, &vi
.cdev
);
2061 /* A new mount point for already mounted UBIFS */
2062 dbg_gen("this ubi volume is already mounted");
2063 if ((flags
^ sb
->s_flags
) & MS_RDONLY
) {
2068 sb
->s_flags
= flags
;
2070 * Pass 'ubi' to 'fill_super()' in sb->s_fs_info where it is
2073 sb
->s_fs_info
= ubi
;
2074 err
= ubifs_fill_super(sb
, data
, flags
& MS_SILENT
? 1 : 0);
2077 /* We do not support atime */
2078 sb
->s_flags
|= MS_ACTIVE
| MS_NOATIME
;
2081 /* 'fill_super()' opens ubi again so we must close it here */
2082 ubi_close_volume(ubi
);
2084 simple_set_mnt(mnt
, sb
);
2088 deactivate_locked_super(sb
);
2090 ubi_close_volume(ubi
);
2094 static struct file_system_type ubifs_fs_type
= {
2096 .owner
= THIS_MODULE
,
2097 .get_sb
= ubifs_get_sb
,
2098 .kill_sb
= kill_anon_super
,
2102 * Inode slab cache constructor.
2104 static void inode_slab_ctor(void *obj
)
2106 struct ubifs_inode
*ui
= obj
;
2107 inode_init_once(&ui
->vfs_inode
);
2110 static int __init
ubifs_init(void)
2114 BUILD_BUG_ON(sizeof(struct ubifs_ch
) != 24);
2116 /* Make sure node sizes are 8-byte aligned */
2117 BUILD_BUG_ON(UBIFS_CH_SZ
& 7);
2118 BUILD_BUG_ON(UBIFS_INO_NODE_SZ
& 7);
2119 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ
& 7);
2120 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ
& 7);
2121 BUILD_BUG_ON(UBIFS_DATA_NODE_SZ
& 7);
2122 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ
& 7);
2123 BUILD_BUG_ON(UBIFS_SB_NODE_SZ
& 7);
2124 BUILD_BUG_ON(UBIFS_MST_NODE_SZ
& 7);
2125 BUILD_BUG_ON(UBIFS_REF_NODE_SZ
& 7);
2126 BUILD_BUG_ON(UBIFS_CS_NODE_SZ
& 7);
2127 BUILD_BUG_ON(UBIFS_ORPH_NODE_SZ
& 7);
2129 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ
& 7);
2130 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ
& 7);
2131 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ
& 7);
2132 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ
& 7);
2133 BUILD_BUG_ON(UBIFS_MAX_NODE_SZ
& 7);
2134 BUILD_BUG_ON(MIN_WRITE_SZ
& 7);
2136 /* Check min. node size */
2137 BUILD_BUG_ON(UBIFS_INO_NODE_SZ
< MIN_WRITE_SZ
);
2138 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ
< MIN_WRITE_SZ
);
2139 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ
< MIN_WRITE_SZ
);
2140 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ
< MIN_WRITE_SZ
);
2142 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ
> UBIFS_MAX_NODE_SZ
);
2143 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ
> UBIFS_MAX_NODE_SZ
);
2144 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ
> UBIFS_MAX_NODE_SZ
);
2145 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ
> UBIFS_MAX_NODE_SZ
);
2147 /* Defined node sizes */
2148 BUILD_BUG_ON(UBIFS_SB_NODE_SZ
!= 4096);
2149 BUILD_BUG_ON(UBIFS_MST_NODE_SZ
!= 512);
2150 BUILD_BUG_ON(UBIFS_INO_NODE_SZ
!= 160);
2151 BUILD_BUG_ON(UBIFS_REF_NODE_SZ
!= 64);
2154 * We use 2 bit wide bit-fields to store compression type, which should
2155 * be amended if more compressors are added. The bit-fields are:
2156 * @compr_type in 'struct ubifs_inode', @default_compr in
2157 * 'struct ubifs_info' and @compr_type in 'struct ubifs_mount_opts'.
2159 BUILD_BUG_ON(UBIFS_COMPR_TYPES_CNT
> 4);
2162 * We require that PAGE_CACHE_SIZE is greater-than-or-equal-to
2163 * UBIFS_BLOCK_SIZE. It is assumed that both are powers of 2.
2165 if (PAGE_CACHE_SIZE
< UBIFS_BLOCK_SIZE
) {
2166 ubifs_err("VFS page cache size is %u bytes, but UBIFS requires"
2167 " at least 4096 bytes",
2168 (unsigned int)PAGE_CACHE_SIZE
);
2172 err
= register_filesystem(&ubifs_fs_type
);
2174 ubifs_err("cannot register file system, error %d", err
);
2179 ubifs_inode_slab
= kmem_cache_create("ubifs_inode_slab",
2180 sizeof(struct ubifs_inode
), 0,
2181 SLAB_MEM_SPREAD
| SLAB_RECLAIM_ACCOUNT
,
2183 if (!ubifs_inode_slab
)
2186 register_shrinker(&ubifs_shrinker_info
);
2188 err
= ubifs_compressors_init();
2192 err
= dbg_debugfs_init();
2199 ubifs_compressors_exit();
2201 unregister_shrinker(&ubifs_shrinker_info
);
2202 kmem_cache_destroy(ubifs_inode_slab
);
2204 unregister_filesystem(&ubifs_fs_type
);
2207 /* late_initcall to let compressors initialize first */
2208 late_initcall(ubifs_init
);
2210 static void __exit
ubifs_exit(void)
2212 ubifs_assert(list_empty(&ubifs_infos
));
2213 ubifs_assert(atomic_long_read(&ubifs_clean_zn_cnt
) == 0);
2216 ubifs_compressors_exit();
2217 unregister_shrinker(&ubifs_shrinker_info
);
2218 kmem_cache_destroy(ubifs_inode_slab
);
2219 unregister_filesystem(&ubifs_fs_type
);
2221 module_exit(ubifs_exit
);
2223 MODULE_LICENSE("GPL");
2224 MODULE_VERSION(__stringify(UBIFS_VERSION
));
2225 MODULE_AUTHOR("Artem Bityutskiy, Adrian Hunter");
2226 MODULE_DESCRIPTION("UBIFS - UBI File System");