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",
323 mutex_unlock(&ui->ui_mutex);
324 ubifs_release_dirty_inode_budget(c, ui);
328 static void ubifs_delete_inode(struct inode *inode)
331 struct ubifs_info *c = inode->i_sb->s_fs_info;
332 struct ubifs_inode *ui = ubifs_inode(inode);
336 * Extended attribute inode deletions are fully handled in
337 * 'ubifs_removexattr()'. These inodes are special and have
338 * limited usage, so there is nothing to do here.
342 dbg_gen("inode %lu, mode %#x", inode->i_ino, (int)inode->i_mode);
343 ubifs_assert(!atomic_read(&inode->i_count));
344 ubifs_assert(inode->i_nlink == 0);
346 truncate_inode_pages(&inode->i_data, 0);
347 if (is_bad_inode(inode))
350 ui->ui_size = inode->i_size = 0;
351 err = ubifs_jnl_delete_inode(c, inode);
354 * Worst case we have a lost orphan inode wasting space, so a
355 * simple error message is OK here.
357 ubifs_err("can't delete inode %lu, error %d",
362 ubifs_release_dirty_inode_budget(c, ui);
366 static void ubifs_dirty_inode(struct inode *inode)
368 struct ubifs_inode *ui = ubifs_inode(inode);
370 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
373 dbg_gen("inode %lu", inode->i_ino);
377 static int ubifs_statfs(struct dentry *dentry, struct kstatfs *buf)
379 struct ubifs_info *c = dentry->d_sb->s_fs_info;
380 unsigned long long free;
381 __le32 *uuid = (__le32 *)c->uuid;
383 free = ubifs_get_free_space(c);
384 dbg_gen("free space %lld bytes (%lld blocks)",
385 free, free >> UBIFS_BLOCK_SHIFT);
387 buf->f_type = UBIFS_SUPER_MAGIC;
388 buf->f_bsize = UBIFS_BLOCK_SIZE;
389 buf->f_blocks = c->block_cnt;
390 buf->f_bfree = free >> UBIFS_BLOCK_SHIFT;
391 if (free > c->report_rp_size)
392 buf->f_bavail = (free - c->report_rp_size) >> UBIFS_BLOCK_SHIFT;
397 buf->f_namelen = UBIFS_MAX_NLEN;
398 buf->f_fsid.val[0] = le32_to_cpu(uuid[0]) ^ le32_to_cpu(uuid[2]);
399 buf->f_fsid.val[1] = le32_to_cpu(uuid[1]) ^ le32_to_cpu(uuid[3]);
403 static int ubifs_show_options(struct seq_file *s, struct vfsmount *mnt)
405 struct ubifs_info *c = mnt->mnt_sb->s_fs_info;
407 if (c->mount_opts.unmount_mode == 2)
408 seq_printf(s, ",fast_unmount");
409 else if (c->mount_opts.unmount_mode == 1)
410 seq_printf(s, ",norm_unmount");
412 if (c->mount_opts.bulk_read == 2)
413 seq_printf(s, ",bulk_read");
414 else if (c->mount_opts.bulk_read == 1)
415 seq_printf(s, ",no_bulk_read");
417 if (c->mount_opts.chk_data_crc == 2)
418 seq_printf(s, ",chk_data_crc");
419 else if (c->mount_opts.chk_data_crc == 1)
420 seq_printf(s, ",no_chk_data_crc");
422 if (c->mount_opts.override_compr) {
423 seq_printf(s, ",compr=");
424 seq_printf(s, ubifs_compr_name(c->mount_opts.compr_type));
430 static int ubifs_sync_fs(struct super_block *sb, int wait)
433 struct ubifs_info *c = sb->s_fs_info;
434 struct writeback_control wbc = {
435 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
437 .range_end = LLONG_MAX,
438 .nr_to_write = LONG_MAX,
442 * Note by akpm about WB_SYNC_NONE used above: zero @wait is just an
443 * advisory thing to help the file system shove lots of data into the
444 * queues. If some gets missed then it'll be picked up on the second
445 * '->sync_fs()' call, with non-zero @wait.
448 if (sb->s_flags & MS_RDONLY)
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 * VFS calls '->sync_fs()' before synchronizing all dirty inodes and
463 * pages, so synchronize them first, then commit the journal. Strictly
464 * speaking, it is not necessary to commit the journal here,
465 * synchronizing write-buffers would be enough. But committing makes
466 * UBIFS free space predictions much more accurate, so we want to let
467 * the user be able to get more accurate results of 'statfs()' after
468 * they synchronize the file system.
470 generic_sync_sb_inodes(sb, &wbc);
472 err = ubifs_run_commit(c);
476 return ubi_sync(c->vi.ubi_num);
480 * init_constants_early - initialize UBIFS constants.
481 * @c: UBIFS file-system description object
483 * This function initialize UBIFS constants which do not need the superblock to
484 * be read. It also checks that the UBI volume satisfies basic UBIFS
485 * requirements. Returns zero in case of success and a negative error code in
488 static int init_constants_early(struct ubifs_info *c)
490 if (c->vi.corrupted) {
491 ubifs_warn("UBI volume is corrupted - read-only mode");
496 ubifs_msg("read-only UBI device");
500 if (c->vi.vol_type == UBI_STATIC_VOLUME) {
501 ubifs_msg("static UBI volume - read-only mode");
505 c->leb_cnt = c->vi.size;
506 c->leb_size = c->vi.usable_leb_size;
507 c->half_leb_size = c->leb_size / 2;
508 c->min_io_size = c->di.min_io_size;
509 c->min_io_shift = fls(c->min_io_size) - 1;
511 if (c->leb_size < UBIFS_MIN_LEB_SZ) {
512 ubifs_err("too small LEBs (%d bytes), min. is %d bytes",
513 c->leb_size, UBIFS_MIN_LEB_SZ);
517 if (c->leb_cnt < UBIFS_MIN_LEB_CNT) {
518 ubifs_err("too few LEBs (%d), min. is %d",
519 c->leb_cnt, UBIFS_MIN_LEB_CNT);
523 if (!is_power_of_2(c->min_io_size)) {
524 ubifs_err("bad min. I/O size %d", c->min_io_size);
529 * UBIFS aligns all node to 8-byte boundary, so to make function in
530 * io.c simpler, assume minimum I/O unit size to be 8 bytes if it is
533 if (c->min_io_size < 8) {
538 c->ref_node_alsz = ALIGN(UBIFS_REF_NODE_SZ, c->min_io_size);
539 c->mst_node_alsz = ALIGN(UBIFS_MST_NODE_SZ, c->min_io_size);
542 * Initialize node length ranges which are mostly needed for node
545 c->ranges[UBIFS_PAD_NODE].len = UBIFS_PAD_NODE_SZ;
546 c->ranges[UBIFS_SB_NODE].len = UBIFS_SB_NODE_SZ;
547 c->ranges[UBIFS_MST_NODE].len = UBIFS_MST_NODE_SZ;
548 c->ranges[UBIFS_REF_NODE].len = UBIFS_REF_NODE_SZ;
549 c->ranges[UBIFS_TRUN_NODE].len = UBIFS_TRUN_NODE_SZ;
550 c->ranges[UBIFS_CS_NODE].len = UBIFS_CS_NODE_SZ;
552 c->ranges[UBIFS_INO_NODE].min_len = UBIFS_INO_NODE_SZ;
553 c->ranges[UBIFS_INO_NODE].max_len = UBIFS_MAX_INO_NODE_SZ;
554 c->ranges[UBIFS_ORPH_NODE].min_len =
555 UBIFS_ORPH_NODE_SZ + sizeof(__le64);
556 c->ranges[UBIFS_ORPH_NODE].max_len = c->leb_size;
557 c->ranges[UBIFS_DENT_NODE].min_len = UBIFS_DENT_NODE_SZ;
558 c->ranges[UBIFS_DENT_NODE].max_len = UBIFS_MAX_DENT_NODE_SZ;
559 c->ranges[UBIFS_XENT_NODE].min_len = UBIFS_XENT_NODE_SZ;
560 c->ranges[UBIFS_XENT_NODE].max_len = UBIFS_MAX_XENT_NODE_SZ;
561 c->ranges[UBIFS_DATA_NODE].min_len = UBIFS_DATA_NODE_SZ;
562 c->ranges[UBIFS_DATA_NODE].max_len = UBIFS_MAX_DATA_NODE_SZ;
564 * Minimum indexing node size is amended later when superblock is
565 * read and the key length is known.
567 c->ranges[UBIFS_IDX_NODE].min_len = UBIFS_IDX_NODE_SZ + UBIFS_BRANCH_SZ;
569 * Maximum indexing node size is amended later when superblock is
570 * read and the fanout is known.
572 c->ranges[UBIFS_IDX_NODE].max_len = INT_MAX;
575 * Initialize dead and dark LEB space watermarks.
577 * Dead space is the space which cannot be used. Its watermark is
578 * equivalent to min. I/O unit or minimum node size if it is greater
579 * then min. I/O unit.
581 * Dark space is the space which might be used, or might not, depending
582 * on which node should be written to the LEB. Its watermark is
583 * equivalent to maximum UBIFS node size.
585 c->dead_wm = ALIGN(MIN_WRITE_SZ, c->min_io_size);
586 c->dark_wm = ALIGN(UBIFS_MAX_NODE_SZ, c->min_io_size);
589 * Calculate how many bytes would be wasted at the end of LEB if it was
590 * fully filled with data nodes of maximum size. This is used in
591 * calculations when reporting free space.
593 c->leb_overhead = c->leb_size % UBIFS_MAX_DATA_NODE_SZ;
595 /* Buffer size for bulk-reads */
596 c->max_bu_buf_len = UBIFS_MAX_BULK_READ * UBIFS_MAX_DATA_NODE_SZ;
597 if (c->max_bu_buf_len > c->leb_size)
598 c->max_bu_buf_len = c->leb_size;
603 * bud_wbuf_callback - bud LEB write-buffer synchronization call-back.
604 * @c: UBIFS file-system description object
605 * @lnum: LEB the write-buffer was synchronized to
606 * @free: how many free bytes left in this LEB
607 * @pad: how many bytes were padded
609 * This is a callback function which is called by the I/O unit when the
610 * write-buffer is synchronized. We need this to correctly maintain space
611 * accounting in bud logical eraseblocks. This function returns zero in case of
612 * success and a negative error code in case of failure.
614 * This function actually belongs to the journal, but we keep it here because
615 * we want to keep it static.
617 static int bud_wbuf_callback(struct ubifs_info *c, int lnum, int free, int pad)
619 return ubifs_update_one_lp(c, lnum, free, pad, 0, 0);
623 * init_constants_sb - initialize UBIFS constants.
624 * @c: UBIFS file-system description object
626 * This is a helper function which initializes various UBIFS constants after
627 * the superblock has been read. It also checks various UBIFS parameters and
628 * makes sure they are all right. Returns zero in case of success and a
629 * negative error code in case of failure.
631 static int init_constants_sb(struct ubifs_info *c)
636 c->main_bytes = (long long)c->main_lebs * c->leb_size;
637 c->max_znode_sz = sizeof(struct ubifs_znode) +
638 c->fanout * sizeof(struct ubifs_zbranch);
640 tmp = ubifs_idx_node_sz(c, 1);
641 c->ranges[UBIFS_IDX_NODE].min_len = tmp;
642 c->min_idx_node_sz = ALIGN(tmp, 8);
644 tmp = ubifs_idx_node_sz(c, c->fanout);
645 c->ranges[UBIFS_IDX_NODE].max_len = tmp;
646 c->max_idx_node_sz = ALIGN(tmp, 8);
648 /* Make sure LEB size is large enough to fit full commit */
649 tmp = UBIFS_CS_NODE_SZ + UBIFS_REF_NODE_SZ * c->jhead_cnt;
650 tmp = ALIGN(tmp, c->min_io_size);
651 if (tmp > c->leb_size) {
652 dbg_err("too small LEB size %d, at least %d needed",
658 * Make sure that the log is large enough to fit reference nodes for
659 * all buds plus one reserved LEB.
661 tmp64 = c->max_bud_bytes + c->leb_size - 1;
662 c->max_bud_cnt = div_u64(tmp64, c->leb_size);
663 tmp = (c->ref_node_alsz * c->max_bud_cnt + c->leb_size - 1);
666 if (c->log_lebs < tmp) {
667 dbg_err("too small log %d LEBs, required min. %d LEBs",
673 * When budgeting we assume worst-case scenarios when the pages are not
674 * be compressed and direntries are of the maximum size.
676 * Note, data, which may be stored in inodes is budgeted separately, so
677 * it is not included into 'c->inode_budget'.
679 c->page_budget = UBIFS_MAX_DATA_NODE_SZ * UBIFS_BLOCKS_PER_PAGE;
680 c->inode_budget = UBIFS_INO_NODE_SZ;
681 c->dent_budget = UBIFS_MAX_DENT_NODE_SZ;
684 * When the amount of flash space used by buds becomes
685 * 'c->max_bud_bytes', UBIFS just blocks all writers and starts commit.
686 * The writers are unblocked when the commit is finished. To avoid
687 * writers to be blocked UBIFS initiates background commit in advance,
688 * when number of bud bytes becomes above the limit defined below.
690 c->bg_bud_bytes = (c->max_bud_bytes * 13) >> 4;
693 * Ensure minimum journal size. All the bytes in the journal heads are
694 * considered to be used, when calculating the current journal usage.
695 * Consequently, if the journal is too small, UBIFS will treat it as
698 tmp64 = (long long)(c->jhead_cnt + 1) * c->leb_size + 1;
699 if (c->bg_bud_bytes < tmp64)
700 c->bg_bud_bytes = tmp64;
701 if (c->max_bud_bytes < tmp64 + c->leb_size)
702 c->max_bud_bytes = tmp64 + c->leb_size;
704 err = ubifs_calc_lpt_geom(c);
712 * init_constants_master - initialize UBIFS constants.
713 * @c: UBIFS file-system description object
715 * This is a helper function which initializes various UBIFS constants after
716 * the master node has been read. It also checks various UBIFS parameters and
717 * makes sure they are all right.
719 static void init_constants_master(struct ubifs_info *c)
723 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
726 * Calculate total amount of FS blocks. This number is not used
727 * internally because it does not make much sense for UBIFS, but it is
728 * necessary to report something for the 'statfs()' call.
730 * Subtract the LEB reserved for GC, the LEB which is reserved for
731 * deletions, minimum LEBs for the index, and assume only one journal
734 tmp64 = c->main_lebs - 1 - 1 - MIN_INDEX_LEBS - c->jhead_cnt + 1;
735 tmp64 *= (long long)c->leb_size - c->leb_overhead;
736 tmp64 = ubifs_reported_space(c, tmp64);
737 c->block_cnt = tmp64 >> UBIFS_BLOCK_SHIFT;
741 * take_gc_lnum - reserve GC LEB.
742 * @c: UBIFS file-system description object
744 * This function ensures that the LEB reserved for garbage collection is
745 * unmapped and is marked as "taken" in lprops. We also have to set free space
746 * to LEB size and dirty space to zero, because lprops may contain out-of-date
747 * information if the file-system was un-mounted before it has been committed.
748 * This function returns zero in case of success and a negative error code in
751 static int take_gc_lnum(struct ubifs_info *c)
755 if (c->gc_lnum == -1) {
756 ubifs_err("no LEB for GC");
760 err = ubifs_leb_unmap(c, c->gc_lnum);
764 /* And we have to tell lprops that this LEB is taken */
765 err = ubifs_change_one_lp(c, c->gc_lnum, c->leb_size, 0,
771 * alloc_wbufs - allocate write-buffers.
772 * @c: UBIFS file-system description object
774 * This helper function allocates and initializes UBIFS write-buffers. Returns
775 * zero in case of success and %-ENOMEM in case of failure.
777 static int alloc_wbufs(struct ubifs_info *c)
781 c->jheads = kzalloc(c->jhead_cnt * sizeof(struct ubifs_jhead),
786 /* Initialize journal heads */
787 for (i = 0; i < c->jhead_cnt; i++) {
788 INIT_LIST_HEAD(&c->jheads[i].buds_list);
789 err = ubifs_wbuf_init(c, &c->jheads[i].wbuf);
793 c->jheads[i].wbuf.sync_callback = &bud_wbuf_callback;
794 c->jheads[i].wbuf.jhead = i;
797 c->jheads[BASEHD].wbuf.dtype = UBI_SHORTTERM;
799 * Garbage Collector head likely contains long-term data and
800 * does not need to be synchronized by timer.
802 c->jheads[GCHD].wbuf.dtype = UBI_LONGTERM;
803 c->jheads[GCHD].wbuf.timeout = 0;
809 * free_wbufs - free write-buffers.
810 * @c: UBIFS file-system description object
812 static void free_wbufs(struct ubifs_info *c)
817 for (i = 0; i < c->jhead_cnt; i++) {
818 kfree(c->jheads[i].wbuf.buf);
819 kfree(c->jheads[i].wbuf.inodes);
827 * free_orphans - free orphans.
828 * @c: UBIFS file-system description object
830 static void free_orphans(struct ubifs_info *c)
832 struct ubifs_orphan *orph;
834 while (c->orph_dnext) {
835 orph = c->orph_dnext;
836 c->orph_dnext = orph->dnext;
837 list_del(&orph->list);
841 while (!list_empty(&c->orph_list)) {
842 orph = list_entry(c->orph_list.next, struct ubifs_orphan, list);
843 list_del(&orph->list);
845 dbg_err("orphan list not empty at unmount");
853 * free_buds - free per-bud objects.
854 * @c: UBIFS file-system description object
856 static void free_buds(struct ubifs_info *c)
858 struct rb_node *this = c->buds.rb_node;
859 struct ubifs_bud *bud;
863 this = this->rb_left;
864 else if (this->rb_right)
865 this = this->rb_right;
867 bud = rb_entry(this, struct ubifs_bud, rb);
868 this = rb_parent(this);
870 if (this->rb_left == &bud->rb)
871 this->rb_left = NULL;
873 this->rb_right = NULL;
881 * check_volume_empty - check if the UBI volume is empty.
882 * @c: UBIFS file-system description object
884 * This function checks if the UBIFS volume is empty by looking if its LEBs are
885 * mapped or not. The result of checking is stored in the @c->empty variable.
886 * Returns zero in case of success and a negative error code in case of
889 static int check_volume_empty(struct ubifs_info *c)
894 for (lnum = 0; lnum < c->leb_cnt; lnum++) {
895 err = ubi_is_mapped(c->ubi, lnum);
896 if (unlikely(err < 0))
910 * UBIFS mount options.
912 * Opt_fast_unmount: do not run a journal commit before un-mounting
913 * Opt_norm_unmount: run a journal commit before un-mounting
914 * Opt_bulk_read: enable bulk-reads
915 * Opt_no_bulk_read: disable bulk-reads
916 * Opt_chk_data_crc: check CRCs when reading data nodes
917 * Opt_no_chk_data_crc: do not check CRCs when reading data nodes
918 * Opt_override_compr: override default compressor
919 * Opt_err: just end of array marker
932 static const match_table_t tokens = {
933 {Opt_fast_unmount, "fast_unmount"},
934 {Opt_norm_unmount, "norm_unmount"},
935 {Opt_bulk_read, "bulk_read"},
936 {Opt_no_bulk_read, "no_bulk_read"},
937 {Opt_chk_data_crc, "chk_data_crc"},
938 {Opt_no_chk_data_crc, "no_chk_data_crc"},
939 {Opt_override_compr, "compr=%s"},
944 * ubifs_parse_options - parse mount parameters.
945 * @c: UBIFS file-system description object
946 * @options: parameters to parse
947 * @is_remount: non-zero if this is FS re-mount
949 * This function parses UBIFS mount options and returns zero in case success
950 * and a negative error code in case of failure.
952 static int ubifs_parse_options(struct ubifs_info *c, char *options,
956 substring_t args[MAX_OPT_ARGS];
961 while ((p = strsep(&options, ","))) {
967 token = match_token(p, tokens, args);
969 case Opt_fast_unmount:
970 c->mount_opts.unmount_mode = 2;
973 case Opt_norm_unmount:
974 c->mount_opts.unmount_mode = 1;
978 c->mount_opts.bulk_read = 2;
981 case Opt_no_bulk_read:
982 c->mount_opts.bulk_read = 1;
985 case Opt_chk_data_crc:
986 c->mount_opts.chk_data_crc = 2;
987 c->no_chk_data_crc = 0;
989 case Opt_no_chk_data_crc:
990 c->mount_opts.chk_data_crc = 1;
991 c->no_chk_data_crc = 1;
993 case Opt_override_compr:
995 char *name = match_strdup(&args[0]);
999 if (!strcmp(name, "none"))
1000 c->mount_opts.compr_type = UBIFS_COMPR_NONE;
1001 else if (!strcmp(name, "lzo"))
1002 c->mount_opts.compr_type = UBIFS_COMPR_LZO;
1003 else if (!strcmp(name, "zlib"))
1004 c->mount_opts.compr_type = UBIFS_COMPR_ZLIB;
1006 ubifs_err("unknown compressor \"%s\"", name);
1011 c->mount_opts.override_compr = 1;
1012 c->default_compr = c->mount_opts.compr_type;
1016 ubifs_err("unrecognized mount option \"%s\" "
1017 "or missing value", p);
1026 * destroy_journal - destroy journal data structures.
1027 * @c: UBIFS file-system description object
1029 * This function destroys journal data structures including those that may have
1030 * been created by recovery functions.
1032 static void destroy_journal(struct ubifs_info *c)
1034 while (!list_empty(&c->unclean_leb_list)) {
1035 struct ubifs_unclean_leb *ucleb;
1037 ucleb = list_entry(c->unclean_leb_list.next,
1038 struct ubifs_unclean_leb, list);
1039 list_del(&ucleb->list);
1042 while (!list_empty(&c->old_buds)) {
1043 struct ubifs_bud *bud;
1045 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
1046 list_del(&bud->list);
1049 ubifs_destroy_idx_gc(c);
1050 ubifs_destroy_size_tree(c);
1056 * bu_init - initialize bulk-read information.
1057 * @c: UBIFS file-system description object
1059 static void bu_init(struct ubifs_info *c)
1061 ubifs_assert(c->bulk_read == 1);
1064 return; /* Already initialized */
1067 c->bu.buf = kmalloc(c->max_bu_buf_len, GFP_KERNEL | __GFP_NOWARN);
1069 if (c->max_bu_buf_len > UBIFS_KMALLOC_OK) {
1070 c->max_bu_buf_len = UBIFS_KMALLOC_OK;
1074 /* Just disable bulk-read */
1075 ubifs_warn("Cannot allocate %d bytes of memory for bulk-read, "
1076 "disabling it", c->max_bu_buf_len);
1077 c->mount_opts.bulk_read = 1;
1084 * check_free_space - check if there is enough free space to mount.
1085 * @c: UBIFS file-system description object
1087 * This function makes sure UBIFS has enough free space to be mounted in
1088 * read/write mode. UBIFS must always have some free space to allow deletions.
1090 static int check_free_space(struct ubifs_info *c)
1092 ubifs_assert(c->dark_wm > 0);
1093 if (c->lst.total_free + c->lst.total_dirty < c->dark_wm) {
1094 ubifs_err("insufficient free space to mount in read/write mode");
1098 * We return %-EINVAL instead of %-ENOSPC because it seems to
1099 * be the closest error code mentioned in the mount function
1108 * mount_ubifs - mount UBIFS file-system.
1109 * @c: UBIFS file-system description object
1111 * This function mounts UBIFS file system. Returns zero in case of success and
1112 * a negative error code in case of failure.
1114 * Note, the function does not de-allocate resources it it fails half way
1115 * through, and the caller has to do this instead.
1117 static int mount_ubifs(struct ubifs_info *c)
1119 struct super_block *sb = c->vfs_sb;
1120 int err, mounted_read_only = (sb->s_flags & MS_RDONLY);
1124 err = init_constants_early(c);
1128 err = ubifs_debugging_init(c);
1132 err = check_volume_empty(c);
1136 if (c->empty && (mounted_read_only || c->ro_media)) {
1138 * This UBI volume is empty, and read-only, or the file system
1139 * is mounted read-only - we cannot format it.
1141 ubifs_err("can't format empty UBI volume: read-only %s",
1142 c->ro_media ? "UBI volume" : "mount");
1147 if (c->ro_media && !mounted_read_only) {
1148 ubifs_err("cannot mount read-write - read-only media");
1154 * The requirement for the buffer is that it should fit indexing B-tree
1155 * height amount of integers. We assume the height if the TNC tree will
1159 c->bottom_up_buf = kmalloc(BOTTOM_UP_HEIGHT * sizeof(int), GFP_KERNEL);
1160 if (!c->bottom_up_buf)
1163 c->sbuf = vmalloc(c->leb_size);
1167 if (!mounted_read_only) {
1168 c->ileb_buf = vmalloc(c->leb_size);
1173 if (c->bulk_read == 1)
1177 * We have to check all CRCs, even for data nodes, when we mount the FS
1178 * (specifically, when we are replaying).
1180 c->always_chk_crc = 1;
1182 err = ubifs_read_superblock(c);
1187 * Make sure the compressor which is set as default in the superblock
1188 * or overridden by mount options is actually compiled in.
1190 if (!ubifs_compr_present(c->default_compr)) {
1191 ubifs_err("'compressor \"%s\" is not compiled in",
1192 ubifs_compr_name(c->default_compr));
1196 err = init_constants_sb(c);
1200 sz = ALIGN(c->max_idx_node_sz, c->min_io_size);
1201 sz = ALIGN(sz + c->max_idx_node_sz, c->min_io_size);
1202 c->cbuf = kmalloc(sz, GFP_NOFS);
1208 sprintf(c->bgt_name, BGT_NAME_PATTERN, c->vi.ubi_num, c->vi.vol_id);
1209 if (!mounted_read_only) {
1210 err = alloc_wbufs(c);
1214 /* Create background thread */
1215 c->bgt = kthread_create(ubifs_bg_thread, c, c->bgt_name);
1216 if (IS_ERR(c->bgt)) {
1217 err = PTR_ERR(c->bgt);
1219 ubifs_err("cannot spawn \"%s\", error %d",
1223 wake_up_process(c->bgt);
1226 err = ubifs_read_master(c);
1230 init_constants_master(c);
1232 if ((c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY)) != 0) {
1233 ubifs_msg("recovery needed");
1234 c->need_recovery = 1;
1235 if (!mounted_read_only) {
1236 err = ubifs_recover_inl_heads(c, c->sbuf);
1240 } else if (!mounted_read_only) {
1242 * Set the "dirty" flag so that if we reboot uncleanly we
1243 * will notice this immediately on the next mount.
1245 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1246 err = ubifs_write_master(c);
1251 err = ubifs_lpt_init(c, 1, !mounted_read_only);
1255 err = dbg_check_idx_size(c, c->old_idx_sz);
1259 err = ubifs_replay_journal(c);
1263 err = ubifs_mount_orphans(c, c->need_recovery, mounted_read_only);
1267 if (!mounted_read_only) {
1270 err = check_free_space(c);
1274 /* Check for enough log space */
1275 lnum = c->lhead_lnum + 1;
1276 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1277 lnum = UBIFS_LOG_LNUM;
1278 if (lnum == c->ltail_lnum) {
1279 err = ubifs_consolidate_log(c);
1284 if (c->need_recovery) {
1285 err = ubifs_recover_size(c);
1288 err = ubifs_rcvry_gc_commit(c);
1290 err = take_gc_lnum(c);
1294 err = dbg_check_lprops(c);
1297 } else if (c->need_recovery) {
1298 err = ubifs_recover_size(c);
1303 spin_lock(&ubifs_infos_lock);
1304 list_add_tail(&c->infos_list, &ubifs_infos);
1305 spin_unlock(&ubifs_infos_lock);
1307 if (c->need_recovery) {
1308 if (mounted_read_only)
1309 ubifs_msg("recovery deferred");
1311 c->need_recovery = 0;
1312 ubifs_msg("recovery completed");
1316 err = dbg_debugfs_init_fs(c);
1320 err = dbg_check_filesystem(c);
1324 c->always_chk_crc = 0;
1326 ubifs_msg("mounted UBI device %d, volume %d, name \"%s\"",
1327 c->vi.ubi_num, c->vi.vol_id, c->vi.name);
1328 if (mounted_read_only)
1329 ubifs_msg("mounted read-only");
1330 x = (long long)c->main_lebs * c->leb_size;
1331 ubifs_msg("file system size: %lld bytes (%lld KiB, %lld MiB, %d "
1332 "LEBs)", x, x >> 10, x >> 20, c->main_lebs);
1333 x = (long long)c->log_lebs * c->leb_size + c->max_bud_bytes;
1334 ubifs_msg("journal size: %lld bytes (%lld KiB, %lld MiB, %d "
1335 "LEBs)", x, x >> 10, x >> 20, c->log_lebs + c->max_bud_cnt);
1336 ubifs_msg("media format: %d (latest is %d)",
1337 c->fmt_version, UBIFS_FORMAT_VERSION);
1338 ubifs_msg("default compressor: %s", ubifs_compr_name(c->default_compr));
1339 ubifs_msg("reserved for root: %llu bytes (%llu KiB)",
1340 c->report_rp_size, c->report_rp_size >> 10);
1342 dbg_msg("compiled on: " __DATE__ " at " __TIME__);
1343 dbg_msg("min. I/O unit size: %d bytes", c->min_io_size);
1344 dbg_msg("LEB size: %d bytes (%d KiB)",
1345 c->leb_size, c->leb_size >> 10);
1346 dbg_msg("data journal heads: %d",
1347 c->jhead_cnt - NONDATA_JHEADS_CNT);
1348 dbg_msg("UUID: %02X%02X%02X%02X-%02X%02X"
1349 "-%02X%02X-%02X%02X-%02X%02X%02X%02X%02X%02X",
1350 c->uuid[0], c->uuid[1], c->uuid[2], c->uuid[3],
1351 c->uuid[4], c->uuid[5], c->uuid[6], c->uuid[7],
1352 c->uuid[8], c->uuid[9], c->uuid[10], c->uuid[11],
1353 c->uuid[12], c->uuid[13], c->uuid[14], c->uuid[15]);
1354 dbg_msg("fast unmount: %d", c->fast_unmount);
1355 dbg_msg("big_lpt %d", c->big_lpt);
1356 dbg_msg("log LEBs: %d (%d - %d)",
1357 c->log_lebs, UBIFS_LOG_LNUM, c->log_last);
1358 dbg_msg("LPT area LEBs: %d (%d - %d)",
1359 c->lpt_lebs, c->lpt_first, c->lpt_last);
1360 dbg_msg("orphan area LEBs: %d (%d - %d)",
1361 c->orph_lebs, c->orph_first, c->orph_last);
1362 dbg_msg("main area LEBs: %d (%d - %d)",
1363 c->main_lebs, c->main_first, c->leb_cnt - 1);
1364 dbg_msg("index LEBs: %d", c->lst.idx_lebs);
1365 dbg_msg("total index bytes: %lld (%lld KiB, %lld MiB)",
1366 c->old_idx_sz, c->old_idx_sz >> 10, c->old_idx_sz >> 20);
1367 dbg_msg("key hash type: %d", c->key_hash_type);
1368 dbg_msg("tree fanout: %d", c->fanout);
1369 dbg_msg("reserved GC LEB: %d", c->gc_lnum);
1370 dbg_msg("first main LEB: %d", c->main_first);
1371 dbg_msg("max. znode size %d", c->max_znode_sz);
1372 dbg_msg("max. index node size %d", c->max_idx_node_sz);
1373 dbg_msg("node sizes: data %zu, inode %zu, dentry %zu",
1374 UBIFS_DATA_NODE_SZ, UBIFS_INO_NODE_SZ, UBIFS_DENT_NODE_SZ);
1375 dbg_msg("node sizes: trun %zu, sb %zu, master %zu",
1376 UBIFS_TRUN_NODE_SZ, UBIFS_SB_NODE_SZ, UBIFS_MST_NODE_SZ);
1377 dbg_msg("node sizes: ref %zu, cmt. start %zu, orph %zu",
1378 UBIFS_REF_NODE_SZ, UBIFS_CS_NODE_SZ, UBIFS_ORPH_NODE_SZ);
1379 dbg_msg("max. node sizes: data %zu, inode %zu dentry %zu",
1380 UBIFS_MAX_DATA_NODE_SZ, UBIFS_MAX_INO_NODE_SZ,
1381 UBIFS_MAX_DENT_NODE_SZ);
1382 dbg_msg("dead watermark: %d", c->dead_wm);
1383 dbg_msg("dark watermark: %d", c->dark_wm);
1384 dbg_msg("LEB overhead: %d", c->leb_overhead);
1385 x = (long long)c->main_lebs * c->dark_wm;
1386 dbg_msg("max. dark space: %lld (%lld KiB, %lld MiB)",
1387 x, x >> 10, x >> 20);
1388 dbg_msg("maximum bud bytes: %lld (%lld KiB, %lld MiB)",
1389 c->max_bud_bytes, c->max_bud_bytes >> 10,
1390 c->max_bud_bytes >> 20);
1391 dbg_msg("BG commit bud bytes: %lld (%lld KiB, %lld MiB)",
1392 c->bg_bud_bytes, c->bg_bud_bytes >> 10,
1393 c->bg_bud_bytes >> 20);
1394 dbg_msg("current bud bytes %lld (%lld KiB, %lld MiB)",
1395 c->bud_bytes, c->bud_bytes >> 10, c->bud_bytes >> 20);
1396 dbg_msg("max. seq. number: %llu", c->max_sqnum);
1397 dbg_msg("commit number: %llu", c->cmt_no);
1402 spin_lock(&ubifs_infos_lock);
1403 list_del(&c->infos_list);
1404 spin_unlock(&ubifs_infos_lock);
1410 ubifs_lpt_free(c, 0);
1413 kfree(c->rcvrd_mst_node);
1415 kthread_stop(c->bgt);
1424 kfree(c->bottom_up_buf);
1425 ubifs_debugging_exit(c);
1430 * ubifs_umount - un-mount UBIFS file-system.
1431 * @c: UBIFS file-system description object
1433 * Note, this function is called to free allocated resourced when un-mounting,
1434 * as well as free resources when an error occurred while we were half way
1435 * through mounting (error path cleanup function). So it has to make sure the
1436 * resource was actually allocated before freeing it.
1438 static void ubifs_umount(struct ubifs_info *c)
1440 dbg_gen("un-mounting UBI device %d, volume %d", c->vi.ubi_num,
1443 dbg_debugfs_exit_fs(c);
1444 spin_lock(&ubifs_infos_lock);
1445 list_del(&c->infos_list);
1446 spin_unlock(&ubifs_infos_lock);
1449 kthread_stop(c->bgt);
1454 ubifs_lpt_free(c, 0);
1457 kfree(c->rcvrd_mst_node);
1462 kfree(c->bottom_up_buf);
1463 ubifs_debugging_exit(c);
1467 * ubifs_remount_rw - re-mount in read-write mode.
1468 * @c: UBIFS file-system description object
1470 * UBIFS avoids allocating many unnecessary resources when mounted in read-only
1471 * mode. This function allocates the needed resources and re-mounts UBIFS in
1474 static int ubifs_remount_rw(struct ubifs_info *c)
1481 mutex_lock(&c->umount_mutex);
1482 c->remounting_rw = 1;
1483 c->always_chk_crc = 1;
1485 err = check_free_space(c);
1489 if (c->old_leb_cnt != c->leb_cnt) {
1490 struct ubifs_sb_node *sup;
1492 sup = ubifs_read_sb_node(c);
1497 sup->leb_cnt = cpu_to_le32(c->leb_cnt);
1498 err = ubifs_write_sb_node(c, sup);
1503 if (c->need_recovery) {
1504 ubifs_msg("completing deferred recovery");
1505 err = ubifs_write_rcvrd_mst_node(c);
1508 err = ubifs_recover_size(c);
1511 err = ubifs_clean_lebs(c, c->sbuf);
1514 err = ubifs_recover_inl_heads(c, c->sbuf);
1519 if (!(c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY))) {
1520 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1521 err = ubifs_write_master(c);
1526 c->ileb_buf = vmalloc(c->leb_size);
1532 err = ubifs_lpt_init(c, 0, 1);
1536 err = alloc_wbufs(c);
1540 ubifs_create_buds_lists(c);
1542 /* Create background thread */
1543 c->bgt = kthread_create(ubifs_bg_thread, c, c->bgt_name);
1544 if (IS_ERR(c->bgt)) {
1545 err = PTR_ERR(c->bgt);
1547 ubifs_err("cannot spawn \"%s\", error %d",
1551 wake_up_process(c->bgt);
1553 c->orph_buf = vmalloc(c->leb_size);
1559 /* Check for enough log space */
1560 lnum = c->lhead_lnum + 1;
1561 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1562 lnum = UBIFS_LOG_LNUM;
1563 if (lnum == c->ltail_lnum) {
1564 err = ubifs_consolidate_log(c);
1569 if (c->need_recovery)
1570 err = ubifs_rcvry_gc_commit(c);
1572 err = take_gc_lnum(c);
1576 if (c->need_recovery) {
1577 c->need_recovery = 0;
1578 ubifs_msg("deferred recovery completed");
1581 dbg_gen("re-mounted read-write");
1582 c->vfs_sb->s_flags &= ~MS_RDONLY;
1583 c->remounting_rw = 0;
1584 c->always_chk_crc = 0;
1585 mutex_unlock(&c->umount_mutex);
1592 kthread_stop(c->bgt);
1598 ubifs_lpt_free(c, 1);
1599 c->remounting_rw = 0;
1600 c->always_chk_crc = 0;
1601 mutex_unlock(&c->umount_mutex);
1606 * commit_on_unmount - commit the journal when un-mounting.
1607 * @c: UBIFS file-system description object
1609 * This function is called during un-mounting and re-mounting, and it commits
1610 * the journal unless the "fast unmount" mode is enabled.
1612 static void commit_on_unmount(struct ubifs_info *c)
1614 struct super_block *sb = c->vfs_sb;
1615 long long bud_bytes;
1618 * This function is called before the background thread is stopped, so
1619 * we may race with ongoing commit, which means we have to take
1620 * @c->bud_lock to access @c->bud_bytes.
1622 spin_lock(&c->buds_lock);
1623 bud_bytes = c->bud_bytes;
1624 spin_unlock(&c->buds_lock);
1626 if (!c->fast_unmount && !(sb->s_flags & MS_RDONLY) && bud_bytes)
1627 ubifs_run_commit(c);
1631 * ubifs_remount_ro - re-mount in read-only mode.
1632 * @c: UBIFS file-system description object
1634 * We rely on VFS to have stopped writing. Possibly the background thread could
1635 * be running a commit, however kthread_stop will wait in that case.
1637 static void ubifs_remount_ro(struct ubifs_info *c)
1641 ubifs_assert(!c->need_recovery);
1642 commit_on_unmount(c);
1644 mutex_lock(&c->umount_mutex);
1646 kthread_stop(c->bgt);
1650 for (i = 0; i < c->jhead_cnt; i++) {
1651 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1652 del_timer_sync(&c->jheads[i].wbuf.timer);
1656 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1657 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1658 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1659 err = ubifs_write_master(c);
1661 ubifs_ro_mode(c, err);
1664 ubifs_destroy_idx_gc(c);
1670 ubifs_lpt_free(c, 1);
1671 mutex_unlock(&c->umount_mutex);
1674 static void ubifs_put_super(struct super_block *sb)
1677 struct ubifs_info *c = sb->s_fs_info;
1679 ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
1682 * The following asserts are only valid if there has not been a failure
1683 * of the media. For example, there will be dirty inodes if we failed
1684 * to write them back because of I/O errors.
1686 ubifs_assert(atomic_long_read(&c->dirty_pg_cnt) == 0);
1687 ubifs_assert(c->budg_idx_growth == 0);
1688 ubifs_assert(c->budg_dd_growth == 0);
1689 ubifs_assert(c->budg_data_growth == 0);
1692 * The 'c->umount_lock' prevents races between UBIFS memory shrinker
1693 * and file system un-mount. Namely, it prevents the shrinker from
1694 * picking this superblock for shrinking - it will be just skipped if
1695 * the mutex is locked.
1697 mutex_lock(&c->umount_mutex);
1698 if (!(c->vfs_sb->s_flags & MS_RDONLY)) {
1700 * First of all kill the background thread to make sure it does
1701 * not interfere with un-mounting and freeing resources.
1704 kthread_stop(c->bgt);
1708 /* Synchronize write-buffers */
1710 for (i = 0; i < c->jhead_cnt; i++) {
1711 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1712 del_timer_sync(&c->jheads[i].wbuf.timer);
1716 * On fatal errors c->ro_media is set to 1, in which case we do
1717 * not write the master node.
1721 * We are being cleanly unmounted which means the
1722 * orphans were killed - indicate this in the master
1723 * node. Also save the reserved GC LEB number.
1727 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1728 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1729 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1730 err = ubifs_write_master(c);
1733 * Recovery will attempt to fix the master area
1734 * next mount, so we just print a message and
1735 * continue to unmount normally.
1737 ubifs_err("failed to write master node, "
1743 bdi_destroy(&c->bdi);
1744 ubi_close_volume(c->ubi);
1745 mutex_unlock(&c->umount_mutex);
1749 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
1752 struct ubifs_info *c = sb->s_fs_info;
1754 dbg_gen("old flags %#lx, new flags %#x", sb->s_flags, *flags);
1756 err = ubifs_parse_options(c, data, 1);
1758 ubifs_err("invalid or unknown remount parameter");
1762 if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
1763 err = ubifs_remount_rw(c);
1766 } else if (!(sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY))
1767 ubifs_remount_ro(c);
1769 if (c->bulk_read == 1)
1772 dbg_gen("disable bulk-read");
1780 struct super_operations ubifs_super_operations = {
1781 .alloc_inode = ubifs_alloc_inode,
1782 .destroy_inode = ubifs_destroy_inode,
1783 .put_super = ubifs_put_super,
1784 .write_inode = ubifs_write_inode,
1785 .delete_inode = ubifs_delete_inode,
1786 .statfs = ubifs_statfs,
1787 .dirty_inode = ubifs_dirty_inode,
1788 .remount_fs = ubifs_remount_fs,
1789 .show_options = ubifs_show_options,
1790 .sync_fs = ubifs_sync_fs,
1794 * open_ubi - parse UBI device name string and open the UBI device.
1795 * @name: UBI volume name
1796 * @mode: UBI volume open mode
1798 * There are several ways to specify UBI volumes when mounting UBIFS:
1799 * o ubiX_Y - UBI device number X, volume Y;
1800 * o ubiY - UBI device number 0, volume Y;
1801 * o ubiX:NAME - mount UBI device X, volume with name NAME;
1802 * o ubi:NAME - mount UBI device 0, volume with name NAME.
1804 * Alternative '!' separator may be used instead of ':' (because some shells
1805 * like busybox may interpret ':' as an NFS host name separator). This function
1806 * returns ubi volume object in case of success and a negative error code in
1809 static struct ubi_volume_desc *open_ubi(const char *name, int mode)
1814 if (name[0] != 'u' || name[1] != 'b' || name[2] != 'i')
1815 return ERR_PTR(-EINVAL);
1817 /* ubi:NAME method */
1818 if ((name[3] == ':' || name[3] == '!') && name[4] != '\0')
1819 return ubi_open_volume_nm(0, name + 4, mode);
1821 if (!isdigit(name[3]))
1822 return ERR_PTR(-EINVAL);
1824 dev = simple_strtoul(name + 3, &endptr, 0);
1827 if (*endptr == '\0')
1828 return ubi_open_volume(0, dev, mode);
1831 if (*endptr == '_' && isdigit(endptr[1])) {
1832 vol = simple_strtoul(endptr + 1, &endptr, 0);
1833 if (*endptr != '\0')
1834 return ERR_PTR(-EINVAL);
1835 return ubi_open_volume(dev, vol, mode);
1838 /* ubiX:NAME method */
1839 if ((*endptr == ':' || *endptr == '!') && endptr[1] != '\0')
1840 return ubi_open_volume_nm(dev, ++endptr, mode);
1842 return ERR_PTR(-EINVAL);
1845 static int ubifs_fill_super(struct super_block *sb, void *data, int silent)
1847 struct ubi_volume_desc *ubi = sb->s_fs_info;
1848 struct ubifs_info *c;
1852 c = kzalloc(sizeof(struct ubifs_info), GFP_KERNEL);
1856 spin_lock_init(&c->cnt_lock);
1857 spin_lock_init(&c->cs_lock);
1858 spin_lock_init(&c->buds_lock);
1859 spin_lock_init(&c->space_lock);
1860 spin_lock_init(&c->orphan_lock);
1861 init_rwsem(&c->commit_sem);
1862 mutex_init(&c->lp_mutex);
1863 mutex_init(&c->tnc_mutex);
1864 mutex_init(&c->log_mutex);
1865 mutex_init(&c->mst_mutex);
1866 mutex_init(&c->umount_mutex);
1867 mutex_init(&c->bu_mutex);
1868 init_waitqueue_head(&c->cmt_wq);
1870 c->old_idx = RB_ROOT;
1871 c->size_tree = RB_ROOT;
1872 c->orph_tree = RB_ROOT;
1873 INIT_LIST_HEAD(&c->infos_list);
1874 INIT_LIST_HEAD(&c->idx_gc);
1875 INIT_LIST_HEAD(&c->replay_list);
1876 INIT_LIST_HEAD(&c->replay_buds);
1877 INIT_LIST_HEAD(&c->uncat_list);
1878 INIT_LIST_HEAD(&c->empty_list);
1879 INIT_LIST_HEAD(&c->freeable_list);
1880 INIT_LIST_HEAD(&c->frdi_idx_list);
1881 INIT_LIST_HEAD(&c->unclean_leb_list);
1882 INIT_LIST_HEAD(&c->old_buds);
1883 INIT_LIST_HEAD(&c->orph_list);
1884 INIT_LIST_HEAD(&c->orph_new);
1886 c->highest_inum = UBIFS_FIRST_INO;
1887 c->lhead_lnum = c->ltail_lnum = UBIFS_LOG_LNUM;
1889 ubi_get_volume_info(ubi, &c->vi);
1890 ubi_get_device_info(c->vi.ubi_num, &c->di);
1892 /* Re-open the UBI device in read-write mode */
1893 c->ubi = ubi_open_volume(c->vi.ubi_num, c->vi.vol_id, UBI_READWRITE);
1894 if (IS_ERR(c->ubi)) {
1895 err = PTR_ERR(c->ubi);
1900 * UBIFS provides 'backing_dev_info' in order to disable read-ahead. For
1901 * UBIFS, I/O is not deferred, it is done immediately in readpage,
1902 * which means the user would have to wait not just for their own I/O
1903 * but the read-ahead I/O as well i.e. completely pointless.
1905 * Read-ahead will be disabled because @c->bdi.ra_pages is 0.
1907 c->bdi.capabilities = BDI_CAP_MAP_COPY;
1908 c->bdi.unplug_io_fn = default_unplug_io_fn;
1909 err = bdi_init(&c->bdi);
1913 err = ubifs_parse_options(c, data, 0);
1920 sb->s_magic = UBIFS_SUPER_MAGIC;
1921 sb->s_blocksize = UBIFS_BLOCK_SIZE;
1922 sb->s_blocksize_bits = UBIFS_BLOCK_SHIFT;
1923 sb->s_dev = c->vi.cdev;
1924 sb->s_maxbytes = c->max_inode_sz = key_max_inode_size(c);
1925 if (c->max_inode_sz > MAX_LFS_FILESIZE)
1926 sb->s_maxbytes = c->max_inode_sz = MAX_LFS_FILESIZE;
1927 sb->s_op = &ubifs_super_operations;
1929 mutex_lock(&c->umount_mutex);
1930 err = mount_ubifs(c);
1932 ubifs_assert(err < 0);
1936 /* Read the root inode */
1937 root = ubifs_iget(sb, UBIFS_ROOT_INO);
1939 err = PTR_ERR(root);
1943 sb->s_root = d_alloc_root(root);
1947 mutex_unlock(&c->umount_mutex);
1955 mutex_unlock(&c->umount_mutex);
1957 bdi_destroy(&c->bdi);
1959 ubi_close_volume(c->ubi);
1965 static int sb_test(struct super_block *sb, void *data)
1969 return sb->s_dev == *dev;
1972 static int sb_set(struct super_block *sb, void *data)
1980 static int ubifs_get_sb(struct file_system_type *fs_type, int flags,
1981 const char *name, void *data, struct vfsmount *mnt)
1983 struct ubi_volume_desc *ubi;
1984 struct ubi_volume_info vi;
1985 struct super_block *sb;
1988 dbg_gen("name %s, flags %#x", name, flags);
1991 * Get UBI device number and volume ID. Mount it read-only so far
1992 * because this might be a new mount point, and UBI allows only one
1993 * read-write user at a time.
1995 ubi = open_ubi(name, UBI_READONLY);
1997 ubifs_err("cannot open \"%s\", error %d",
1998 name, (int)PTR_ERR(ubi));
1999 return PTR_ERR(ubi);
2001 ubi_get_volume_info(ubi, &vi);
2003 dbg_gen("opened ubi%d_%d", vi.ubi_num, vi.vol_id);
2005 sb = sget(fs_type, &sb_test, &sb_set, &vi.cdev);
2012 /* A new mount point for already mounted UBIFS */
2013 dbg_gen("this ubi volume is already mounted");
2014 if ((flags ^ sb->s_flags) & MS_RDONLY) {
2019 sb->s_flags = flags;
2021 * Pass 'ubi' to 'fill_super()' in sb->s_fs_info where it is
2024 sb->s_fs_info = ubi;
2025 err = ubifs_fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
2028 /* We do not support atime */
2029 sb->s_flags |= MS_ACTIVE | MS_NOATIME;
2032 /* 'fill_super()' opens ubi again so we must close it here */
2033 ubi_close_volume(ubi);
2035 return simple_set_mnt(mnt, sb);
2038 up_write(&sb->s_umount);
2039 deactivate_super(sb);
2041 ubi_close_volume(ubi);
2045 static void ubifs_kill_sb(struct super_block *sb)
2047 struct ubifs_info *c = sb->s_fs_info;
2050 * We do 'commit_on_unmount()' here instead of 'ubifs_put_super()'
2051 * in order to be outside BKL.
2054 commit_on_unmount(c);
2055 /* The un-mount routine is actually done in put_super() */
2056 generic_shutdown_super(sb);
2059 static struct file_system_type ubifs_fs_type = {
2061 .owner = THIS_MODULE,
2062 .get_sb = ubifs_get_sb,
2063 .kill_sb = ubifs_kill_sb
2067 * Inode slab cache constructor.
2069 static void inode_slab_ctor(void *obj)
2071 struct ubifs_inode *ui = obj;
2072 inode_init_once(&ui->vfs_inode);
2075 static int __init ubifs_init(void)
2079 BUILD_BUG_ON(sizeof(struct ubifs_ch) != 24);
2081 /* Make sure node sizes are 8-byte aligned */
2082 BUILD_BUG_ON(UBIFS_CH_SZ & 7);
2083 BUILD_BUG_ON(UBIFS_INO_NODE_SZ & 7);
2084 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ & 7);
2085 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ & 7);
2086 BUILD_BUG_ON(UBIFS_DATA_NODE_SZ & 7);
2087 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ & 7);
2088 BUILD_BUG_ON(UBIFS_SB_NODE_SZ & 7);
2089 BUILD_BUG_ON(UBIFS_MST_NODE_SZ & 7);
2090 BUILD_BUG_ON(UBIFS_REF_NODE_SZ & 7);
2091 BUILD_BUG_ON(UBIFS_CS_NODE_SZ & 7);
2092 BUILD_BUG_ON(UBIFS_ORPH_NODE_SZ & 7);
2094 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ & 7);
2095 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ & 7);
2096 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ & 7);
2097 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ & 7);
2098 BUILD_BUG_ON(UBIFS_MAX_NODE_SZ & 7);
2099 BUILD_BUG_ON(MIN_WRITE_SZ & 7);
2101 /* Check min. node size */
2102 BUILD_BUG_ON(UBIFS_INO_NODE_SZ < MIN_WRITE_SZ);
2103 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ < MIN_WRITE_SZ);
2104 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ < MIN_WRITE_SZ);
2105 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ < MIN_WRITE_SZ);
2107 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2108 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2109 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ > UBIFS_MAX_NODE_SZ);
2110 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ > UBIFS_MAX_NODE_SZ);
2112 /* Defined node sizes */
2113 BUILD_BUG_ON(UBIFS_SB_NODE_SZ != 4096);
2114 BUILD_BUG_ON(UBIFS_MST_NODE_SZ != 512);
2115 BUILD_BUG_ON(UBIFS_INO_NODE_SZ != 160);
2116 BUILD_BUG_ON(UBIFS_REF_NODE_SZ != 64);
2119 * We use 2 bit wide bit-fields to store compression type, which should
2120 * be amended if more compressors are added. The bit-fields are:
2121 * @compr_type in 'struct ubifs_inode', @default_compr in
2122 * 'struct ubifs_info' and @compr_type in 'struct ubifs_mount_opts'.
2124 BUILD_BUG_ON(UBIFS_COMPR_TYPES_CNT > 4);
2127 * We require that PAGE_CACHE_SIZE is greater-than-or-equal-to
2128 * UBIFS_BLOCK_SIZE. It is assumed that both are powers of 2.
2130 if (PAGE_CACHE_SIZE < UBIFS_BLOCK_SIZE) {
2131 ubifs_err("VFS page cache size is %u bytes, but UBIFS requires"
2132 " at least 4096 bytes",
2133 (unsigned int)PAGE_CACHE_SIZE);
2137 err = register_filesystem(&ubifs_fs_type);
2139 ubifs_err("cannot register file system, error %d", err);
2144 ubifs_inode_slab = kmem_cache_create("ubifs_inode_slab",
2145 sizeof(struct ubifs_inode), 0,
2146 SLAB_MEM_SPREAD | SLAB_RECLAIM_ACCOUNT,
2148 if (!ubifs_inode_slab)
2151 register_shrinker(&ubifs_shrinker_info);
2153 err = ubifs_compressors_init();
2157 err = dbg_debugfs_init();
2164 ubifs_compressors_exit();
2166 unregister_shrinker(&ubifs_shrinker_info);
2167 kmem_cache_destroy(ubifs_inode_slab);
2169 unregister_filesystem(&ubifs_fs_type);
2172 /* late_initcall to let compressors initialize first */
2173 late_initcall(ubifs_init);
2175 static void __exit ubifs_exit(void)
2177 ubifs_assert(list_empty(&ubifs_infos));
2178 ubifs_assert(atomic_long_read(&ubifs_clean_zn_cnt) == 0);
2181 ubifs_compressors_exit();
2182 unregister_shrinker(&ubifs_shrinker_info);
2183 kmem_cache_destroy(ubifs_inode_slab);
2184 unregister_filesystem(&ubifs_fs_type);
2186 module_exit(ubifs_exit);
2188 MODULE_LICENSE("GPL");
2189 MODULE_VERSION(__stringify(UBIFS_VERSION));
2190 MODULE_AUTHOR("Artem Bityutskiy, Adrian Hunter");
2191 MODULE_DESCRIPTION("UBIFS - UBI File System");