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, struct writeback_control *wbc)
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_evict_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));
347 truncate_inode_pages(&inode->i_data, 0);
352 if (is_bad_inode(inode))
355 ui->ui_size = inode->i_size = 0;
356 err = ubifs_jnl_delete_inode(c, inode);
359 * Worst case we have a lost orphan inode wasting space, so a
360 * simple error message is OK here.
362 ubifs_err("can't delete inode %lu, error %d",
367 ubifs_release_dirty_inode_budget(c, ui);
369 /* We've deleted something - clean the "no space" flags */
370 c->nospace = c->nospace_rp = 0;
374 end_writeback(inode);
377 static void ubifs_dirty_inode(struct inode *inode)
379 struct ubifs_inode *ui = ubifs_inode(inode);
381 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
384 dbg_gen("inode %lu", inode->i_ino);
388 static int ubifs_statfs(struct dentry *dentry, struct kstatfs *buf)
390 struct ubifs_info *c = dentry->d_sb->s_fs_info;
391 unsigned long long free;
392 __le32 *uuid = (__le32 *)c->uuid;
394 free = ubifs_get_free_space(c);
395 dbg_gen("free space %lld bytes (%lld blocks)",
396 free, free >> UBIFS_BLOCK_SHIFT);
398 buf->f_type = UBIFS_SUPER_MAGIC;
399 buf->f_bsize = UBIFS_BLOCK_SIZE;
400 buf->f_blocks = c->block_cnt;
401 buf->f_bfree = free >> UBIFS_BLOCK_SHIFT;
402 if (free > c->report_rp_size)
403 buf->f_bavail = (free - c->report_rp_size) >> UBIFS_BLOCK_SHIFT;
408 buf->f_namelen = UBIFS_MAX_NLEN;
409 buf->f_fsid.val[0] = le32_to_cpu(uuid[0]) ^ le32_to_cpu(uuid[2]);
410 buf->f_fsid.val[1] = le32_to_cpu(uuid[1]) ^ le32_to_cpu(uuid[3]);
411 ubifs_assert(buf->f_bfree <= c->block_cnt);
415 static int ubifs_show_options(struct seq_file *s, struct vfsmount *mnt)
417 struct ubifs_info *c = mnt->mnt_sb->s_fs_info;
419 if (c->mount_opts.unmount_mode == 2)
420 seq_printf(s, ",fast_unmount");
421 else if (c->mount_opts.unmount_mode == 1)
422 seq_printf(s, ",norm_unmount");
424 if (c->mount_opts.bulk_read == 2)
425 seq_printf(s, ",bulk_read");
426 else if (c->mount_opts.bulk_read == 1)
427 seq_printf(s, ",no_bulk_read");
429 if (c->mount_opts.chk_data_crc == 2)
430 seq_printf(s, ",chk_data_crc");
431 else if (c->mount_opts.chk_data_crc == 1)
432 seq_printf(s, ",no_chk_data_crc");
434 if (c->mount_opts.override_compr) {
435 seq_printf(s, ",compr=%s",
436 ubifs_compr_name(c->mount_opts.compr_type));
442 static int ubifs_sync_fs(struct super_block *sb, int wait)
445 struct ubifs_info *c = sb->s_fs_info;
448 * Zero @wait is just an advisory thing to help the file system shove
449 * lots of data into the queues, and there will be the second
450 * '->sync_fs()' call, with non-zero @wait.
456 * Synchronize write buffers, because 'ubifs_run_commit()' does not
457 * do this if it waits for an already running commit.
459 for (i = 0; i < c->jhead_cnt; i++) {
460 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
466 * Strictly speaking, it is not necessary to commit the journal here,
467 * synchronizing write-buffers would be enough. But committing makes
468 * UBIFS free space predictions much more accurate, so we want to let
469 * the user be able to get more accurate results of 'statfs()' after
470 * they synchronize the file system.
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->leb_start = c->di.leb_start;
508 c->half_leb_size = c->leb_size / 2;
509 c->min_io_size = c->di.min_io_size;
510 c->min_io_shift = fls(c->min_io_size) - 1;
511 c->max_write_size = c->di.max_write_size;
512 c->max_write_shift = fls(c->max_write_size) - 1;
514 if (c->leb_size < UBIFS_MIN_LEB_SZ) {
515 ubifs_err("too small LEBs (%d bytes), min. is %d bytes",
516 c->leb_size, UBIFS_MIN_LEB_SZ);
520 if (c->leb_cnt < UBIFS_MIN_LEB_CNT) {
521 ubifs_err("too few LEBs (%d), min. is %d",
522 c->leb_cnt, UBIFS_MIN_LEB_CNT);
526 if (!is_power_of_2(c->min_io_size)) {
527 ubifs_err("bad min. I/O size %d", c->min_io_size);
532 * Maximum write size has to be greater or equivalent to min. I/O
533 * size, and be multiple of min. I/O size.
535 if (c->max_write_size < c->min_io_size ||
536 c->max_write_size % c->min_io_size ||
537 !is_power_of_2(c->max_write_size)) {
538 ubifs_err("bad write buffer size %d for %d min. I/O unit",
539 c->max_write_size, c->min_io_size);
544 * UBIFS aligns all node to 8-byte boundary, so to make function in
545 * io.c simpler, assume minimum I/O unit size to be 8 bytes if it is
548 if (c->min_io_size < 8) {
551 if (c->max_write_size < c->min_io_size) {
552 c->max_write_size = c->min_io_size;
553 c->max_write_shift = c->min_io_shift;
557 c->ref_node_alsz = ALIGN(UBIFS_REF_NODE_SZ, c->min_io_size);
558 c->mst_node_alsz = ALIGN(UBIFS_MST_NODE_SZ, c->min_io_size);
561 * Initialize node length ranges which are mostly needed for node
564 c->ranges[UBIFS_PAD_NODE].len = UBIFS_PAD_NODE_SZ;
565 c->ranges[UBIFS_SB_NODE].len = UBIFS_SB_NODE_SZ;
566 c->ranges[UBIFS_MST_NODE].len = UBIFS_MST_NODE_SZ;
567 c->ranges[UBIFS_REF_NODE].len = UBIFS_REF_NODE_SZ;
568 c->ranges[UBIFS_TRUN_NODE].len = UBIFS_TRUN_NODE_SZ;
569 c->ranges[UBIFS_CS_NODE].len = UBIFS_CS_NODE_SZ;
571 c->ranges[UBIFS_INO_NODE].min_len = UBIFS_INO_NODE_SZ;
572 c->ranges[UBIFS_INO_NODE].max_len = UBIFS_MAX_INO_NODE_SZ;
573 c->ranges[UBIFS_ORPH_NODE].min_len =
574 UBIFS_ORPH_NODE_SZ + sizeof(__le64);
575 c->ranges[UBIFS_ORPH_NODE].max_len = c->leb_size;
576 c->ranges[UBIFS_DENT_NODE].min_len = UBIFS_DENT_NODE_SZ;
577 c->ranges[UBIFS_DENT_NODE].max_len = UBIFS_MAX_DENT_NODE_SZ;
578 c->ranges[UBIFS_XENT_NODE].min_len = UBIFS_XENT_NODE_SZ;
579 c->ranges[UBIFS_XENT_NODE].max_len = UBIFS_MAX_XENT_NODE_SZ;
580 c->ranges[UBIFS_DATA_NODE].min_len = UBIFS_DATA_NODE_SZ;
581 c->ranges[UBIFS_DATA_NODE].max_len = UBIFS_MAX_DATA_NODE_SZ;
583 * Minimum indexing node size is amended later when superblock is
584 * read and the key length is known.
586 c->ranges[UBIFS_IDX_NODE].min_len = UBIFS_IDX_NODE_SZ + UBIFS_BRANCH_SZ;
588 * Maximum indexing node size is amended later when superblock is
589 * read and the fanout is known.
591 c->ranges[UBIFS_IDX_NODE].max_len = INT_MAX;
594 * Initialize dead and dark LEB space watermarks. See gc.c for comments
595 * about these values.
597 c->dead_wm = ALIGN(MIN_WRITE_SZ, c->min_io_size);
598 c->dark_wm = ALIGN(UBIFS_MAX_NODE_SZ, c->min_io_size);
601 * Calculate how many bytes would be wasted at the end of LEB if it was
602 * fully filled with data nodes of maximum size. This is used in
603 * calculations when reporting free space.
605 c->leb_overhead = c->leb_size % UBIFS_MAX_DATA_NODE_SZ;
607 /* Buffer size for bulk-reads */
608 c->max_bu_buf_len = UBIFS_MAX_BULK_READ * UBIFS_MAX_DATA_NODE_SZ;
609 if (c->max_bu_buf_len > c->leb_size)
610 c->max_bu_buf_len = c->leb_size;
615 * bud_wbuf_callback - bud LEB write-buffer synchronization call-back.
616 * @c: UBIFS file-system description object
617 * @lnum: LEB the write-buffer was synchronized to
618 * @free: how many free bytes left in this LEB
619 * @pad: how many bytes were padded
621 * This is a callback function which is called by the I/O unit when the
622 * write-buffer is synchronized. We need this to correctly maintain space
623 * accounting in bud logical eraseblocks. This function returns zero in case of
624 * success and a negative error code in case of failure.
626 * This function actually belongs to the journal, but we keep it here because
627 * we want to keep it static.
629 static int bud_wbuf_callback(struct ubifs_info *c, int lnum, int free, int pad)
631 return ubifs_update_one_lp(c, lnum, free, pad, 0, 0);
635 * init_constants_sb - initialize UBIFS constants.
636 * @c: UBIFS file-system description object
638 * This is a helper function which initializes various UBIFS constants after
639 * the superblock has been read. It also checks various UBIFS parameters and
640 * makes sure they are all right. Returns zero in case of success and a
641 * negative error code in case of failure.
643 static int init_constants_sb(struct ubifs_info *c)
648 c->main_bytes = (long long)c->main_lebs * c->leb_size;
649 c->max_znode_sz = sizeof(struct ubifs_znode) +
650 c->fanout * sizeof(struct ubifs_zbranch);
652 tmp = ubifs_idx_node_sz(c, 1);
653 c->ranges[UBIFS_IDX_NODE].min_len = tmp;
654 c->min_idx_node_sz = ALIGN(tmp, 8);
656 tmp = ubifs_idx_node_sz(c, c->fanout);
657 c->ranges[UBIFS_IDX_NODE].max_len = tmp;
658 c->max_idx_node_sz = ALIGN(tmp, 8);
660 /* Make sure LEB size is large enough to fit full commit */
661 tmp = UBIFS_CS_NODE_SZ + UBIFS_REF_NODE_SZ * c->jhead_cnt;
662 tmp = ALIGN(tmp, c->min_io_size);
663 if (tmp > c->leb_size) {
664 dbg_err("too small LEB size %d, at least %d needed",
670 * Make sure that the log is large enough to fit reference nodes for
671 * all buds plus one reserved LEB.
673 tmp64 = c->max_bud_bytes + c->leb_size - 1;
674 c->max_bud_cnt = div_u64(tmp64, c->leb_size);
675 tmp = (c->ref_node_alsz * c->max_bud_cnt + c->leb_size - 1);
678 if (c->log_lebs < tmp) {
679 dbg_err("too small log %d LEBs, required min. %d LEBs",
685 * When budgeting we assume worst-case scenarios when the pages are not
686 * be compressed and direntries are of the maximum size.
688 * Note, data, which may be stored in inodes is budgeted separately, so
689 * it is not included into 'c->inode_budget'.
691 c->page_budget = UBIFS_MAX_DATA_NODE_SZ * UBIFS_BLOCKS_PER_PAGE;
692 c->inode_budget = UBIFS_INO_NODE_SZ;
693 c->dent_budget = UBIFS_MAX_DENT_NODE_SZ;
696 * When the amount of flash space used by buds becomes
697 * 'c->max_bud_bytes', UBIFS just blocks all writers and starts commit.
698 * The writers are unblocked when the commit is finished. To avoid
699 * writers to be blocked UBIFS initiates background commit in advance,
700 * when number of bud bytes becomes above the limit defined below.
702 c->bg_bud_bytes = (c->max_bud_bytes * 13) >> 4;
705 * Ensure minimum journal size. All the bytes in the journal heads are
706 * considered to be used, when calculating the current journal usage.
707 * Consequently, if the journal is too small, UBIFS will treat it as
710 tmp64 = (long long)(c->jhead_cnt + 1) * c->leb_size + 1;
711 if (c->bg_bud_bytes < tmp64)
712 c->bg_bud_bytes = tmp64;
713 if (c->max_bud_bytes < tmp64 + c->leb_size)
714 c->max_bud_bytes = tmp64 + c->leb_size;
716 err = ubifs_calc_lpt_geom(c);
720 /* Initialize effective LEB size used in budgeting calculations */
721 c->idx_leb_size = c->leb_size - c->max_idx_node_sz;
726 * init_constants_master - initialize UBIFS constants.
727 * @c: UBIFS file-system description object
729 * This is a helper function which initializes various UBIFS constants after
730 * the master node has been read. It also checks various UBIFS parameters and
731 * makes sure they are all right.
733 static void init_constants_master(struct ubifs_info *c)
737 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
738 c->report_rp_size = ubifs_reported_space(c, c->rp_size);
741 * Calculate total amount of FS blocks. This number is not used
742 * internally because it does not make much sense for UBIFS, but it is
743 * necessary to report something for the 'statfs()' call.
745 * Subtract the LEB reserved for GC, the LEB which is reserved for
746 * deletions, minimum LEBs for the index, and assume only one journal
749 tmp64 = c->main_lebs - 1 - 1 - MIN_INDEX_LEBS - c->jhead_cnt + 1;
750 tmp64 *= (long long)c->leb_size - c->leb_overhead;
751 tmp64 = ubifs_reported_space(c, tmp64);
752 c->block_cnt = tmp64 >> UBIFS_BLOCK_SHIFT;
756 * take_gc_lnum - reserve GC LEB.
757 * @c: UBIFS file-system description object
759 * This function ensures that the LEB reserved for garbage collection is marked
760 * as "taken" in lprops. We also have to set free space to LEB size and dirty
761 * space to zero, because lprops may contain out-of-date information if the
762 * file-system was un-mounted before it has been committed. This function
763 * returns zero in case of success and a negative error code in case of
766 static int take_gc_lnum(struct ubifs_info *c)
770 if (c->gc_lnum == -1) {
771 ubifs_err("no LEB for GC");
775 /* And we have to tell lprops that this LEB is taken */
776 err = ubifs_change_one_lp(c, c->gc_lnum, c->leb_size, 0,
782 * alloc_wbufs - allocate write-buffers.
783 * @c: UBIFS file-system description object
785 * This helper function allocates and initializes UBIFS write-buffers. Returns
786 * zero in case of success and %-ENOMEM in case of failure.
788 static int alloc_wbufs(struct ubifs_info *c)
792 c->jheads = kzalloc(c->jhead_cnt * sizeof(struct ubifs_jhead),
797 /* Initialize journal heads */
798 for (i = 0; i < c->jhead_cnt; i++) {
799 INIT_LIST_HEAD(&c->jheads[i].buds_list);
800 err = ubifs_wbuf_init(c, &c->jheads[i].wbuf);
804 c->jheads[i].wbuf.sync_callback = &bud_wbuf_callback;
805 c->jheads[i].wbuf.jhead = i;
808 c->jheads[BASEHD].wbuf.dtype = UBI_SHORTTERM;
810 * Garbage Collector head likely contains long-term data and
811 * does not need to be synchronized by timer.
813 c->jheads[GCHD].wbuf.dtype = UBI_LONGTERM;
814 c->jheads[GCHD].wbuf.no_timer = 1;
820 * free_wbufs - free write-buffers.
821 * @c: UBIFS file-system description object
823 static void free_wbufs(struct ubifs_info *c)
828 for (i = 0; i < c->jhead_cnt; i++) {
829 kfree(c->jheads[i].wbuf.buf);
830 kfree(c->jheads[i].wbuf.inodes);
838 * free_orphans - free orphans.
839 * @c: UBIFS file-system description object
841 static void free_orphans(struct ubifs_info *c)
843 struct ubifs_orphan *orph;
845 while (c->orph_dnext) {
846 orph = c->orph_dnext;
847 c->orph_dnext = orph->dnext;
848 list_del(&orph->list);
852 while (!list_empty(&c->orph_list)) {
853 orph = list_entry(c->orph_list.next, struct ubifs_orphan, list);
854 list_del(&orph->list);
856 dbg_err("orphan list not empty at unmount");
864 * free_buds - free per-bud objects.
865 * @c: UBIFS file-system description object
867 static void free_buds(struct ubifs_info *c)
869 struct rb_node *this = c->buds.rb_node;
870 struct ubifs_bud *bud;
874 this = this->rb_left;
875 else if (this->rb_right)
876 this = this->rb_right;
878 bud = rb_entry(this, struct ubifs_bud, rb);
879 this = rb_parent(this);
881 if (this->rb_left == &bud->rb)
882 this->rb_left = NULL;
884 this->rb_right = NULL;
892 * check_volume_empty - check if the UBI volume is empty.
893 * @c: UBIFS file-system description object
895 * This function checks if the UBIFS volume is empty by looking if its LEBs are
896 * mapped or not. The result of checking is stored in the @c->empty variable.
897 * Returns zero in case of success and a negative error code in case of
900 static int check_volume_empty(struct ubifs_info *c)
905 for (lnum = 0; lnum < c->leb_cnt; lnum++) {
906 err = ubi_is_mapped(c->ubi, lnum);
907 if (unlikely(err < 0))
921 * UBIFS mount options.
923 * Opt_fast_unmount: do not run a journal commit before un-mounting
924 * Opt_norm_unmount: run a journal commit before un-mounting
925 * Opt_bulk_read: enable bulk-reads
926 * Opt_no_bulk_read: disable bulk-reads
927 * Opt_chk_data_crc: check CRCs when reading data nodes
928 * Opt_no_chk_data_crc: do not check CRCs when reading data nodes
929 * Opt_override_compr: override default compressor
930 * Opt_err: just end of array marker
943 static const match_table_t tokens = {
944 {Opt_fast_unmount, "fast_unmount"},
945 {Opt_norm_unmount, "norm_unmount"},
946 {Opt_bulk_read, "bulk_read"},
947 {Opt_no_bulk_read, "no_bulk_read"},
948 {Opt_chk_data_crc, "chk_data_crc"},
949 {Opt_no_chk_data_crc, "no_chk_data_crc"},
950 {Opt_override_compr, "compr=%s"},
955 * parse_standard_option - parse a standard mount option.
956 * @option: the option to parse
958 * Normally, standard mount options like "sync" are passed to file-systems as
959 * flags. However, when a "rootflags=" kernel boot parameter is used, they may
960 * be present in the options string. This function tries to deal with this
961 * situation and parse standard options. Returns 0 if the option was not
962 * recognized, and the corresponding integer flag if it was.
964 * UBIFS is only interested in the "sync" option, so do not check for anything
967 static int parse_standard_option(const char *option)
969 ubifs_msg("parse %s", option);
970 if (!strcmp(option, "sync"))
971 return MS_SYNCHRONOUS;
976 * ubifs_parse_options - parse mount parameters.
977 * @c: UBIFS file-system description object
978 * @options: parameters to parse
979 * @is_remount: non-zero if this is FS re-mount
981 * This function parses UBIFS mount options and returns zero in case success
982 * and a negative error code in case of failure.
984 static int ubifs_parse_options(struct ubifs_info *c, char *options,
988 substring_t args[MAX_OPT_ARGS];
993 while ((p = strsep(&options, ","))) {
999 token = match_token(p, tokens, args);
1002 * %Opt_fast_unmount and %Opt_norm_unmount options are ignored.
1003 * We accept them in order to be backward-compatible. But this
1004 * should be removed at some point.
1006 case Opt_fast_unmount:
1007 c->mount_opts.unmount_mode = 2;
1009 case Opt_norm_unmount:
1010 c->mount_opts.unmount_mode = 1;
1013 c->mount_opts.bulk_read = 2;
1016 case Opt_no_bulk_read:
1017 c->mount_opts.bulk_read = 1;
1020 case Opt_chk_data_crc:
1021 c->mount_opts.chk_data_crc = 2;
1022 c->no_chk_data_crc = 0;
1024 case Opt_no_chk_data_crc:
1025 c->mount_opts.chk_data_crc = 1;
1026 c->no_chk_data_crc = 1;
1028 case Opt_override_compr:
1030 char *name = match_strdup(&args[0]);
1034 if (!strcmp(name, "none"))
1035 c->mount_opts.compr_type = UBIFS_COMPR_NONE;
1036 else if (!strcmp(name, "lzo"))
1037 c->mount_opts.compr_type = UBIFS_COMPR_LZO;
1038 else if (!strcmp(name, "zlib"))
1039 c->mount_opts.compr_type = UBIFS_COMPR_ZLIB;
1041 ubifs_err("unknown compressor \"%s\"", name);
1046 c->mount_opts.override_compr = 1;
1047 c->default_compr = c->mount_opts.compr_type;
1053 struct super_block *sb = c->vfs_sb;
1055 flag = parse_standard_option(p);
1057 ubifs_err("unrecognized mount option \"%s\" "
1058 "or missing value", p);
1061 sb->s_flags |= flag;
1071 * destroy_journal - destroy journal data structures.
1072 * @c: UBIFS file-system description object
1074 * This function destroys journal data structures including those that may have
1075 * been created by recovery functions.
1077 static void destroy_journal(struct ubifs_info *c)
1079 while (!list_empty(&c->unclean_leb_list)) {
1080 struct ubifs_unclean_leb *ucleb;
1082 ucleb = list_entry(c->unclean_leb_list.next,
1083 struct ubifs_unclean_leb, list);
1084 list_del(&ucleb->list);
1087 while (!list_empty(&c->old_buds)) {
1088 struct ubifs_bud *bud;
1090 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
1091 list_del(&bud->list);
1094 ubifs_destroy_idx_gc(c);
1095 ubifs_destroy_size_tree(c);
1101 * bu_init - initialize bulk-read information.
1102 * @c: UBIFS file-system description object
1104 static void bu_init(struct ubifs_info *c)
1106 ubifs_assert(c->bulk_read == 1);
1109 return; /* Already initialized */
1112 c->bu.buf = kmalloc(c->max_bu_buf_len, GFP_KERNEL | __GFP_NOWARN);
1114 if (c->max_bu_buf_len > UBIFS_KMALLOC_OK) {
1115 c->max_bu_buf_len = UBIFS_KMALLOC_OK;
1119 /* Just disable bulk-read */
1120 ubifs_warn("Cannot allocate %d bytes of memory for bulk-read, "
1121 "disabling it", c->max_bu_buf_len);
1122 c->mount_opts.bulk_read = 1;
1129 * check_free_space - check if there is enough free space to mount.
1130 * @c: UBIFS file-system description object
1132 * This function makes sure UBIFS has enough free space to be mounted in
1133 * read/write mode. UBIFS must always have some free space to allow deletions.
1135 static int check_free_space(struct ubifs_info *c)
1137 ubifs_assert(c->dark_wm > 0);
1138 if (c->lst.total_free + c->lst.total_dirty < c->dark_wm) {
1139 ubifs_err("insufficient free space to mount in read/write mode");
1148 * mount_ubifs - mount UBIFS file-system.
1149 * @c: UBIFS file-system description object
1151 * This function mounts UBIFS file system. Returns zero in case of success and
1152 * a negative error code in case of failure.
1154 * Note, the function does not de-allocate resources it it fails half way
1155 * through, and the caller has to do this instead.
1157 static int mount_ubifs(struct ubifs_info *c)
1163 c->ro_mount = !!(c->vfs_sb->s_flags & MS_RDONLY);
1164 err = init_constants_early(c);
1168 err = ubifs_debugging_init(c);
1172 err = check_volume_empty(c);
1176 if (c->empty && (c->ro_mount || c->ro_media)) {
1178 * This UBI volume is empty, and read-only, or the file system
1179 * is mounted read-only - we cannot format it.
1181 ubifs_err("can't format empty UBI volume: read-only %s",
1182 c->ro_media ? "UBI volume" : "mount");
1187 if (c->ro_media && !c->ro_mount) {
1188 ubifs_err("cannot mount read-write - read-only media");
1194 * The requirement for the buffer is that it should fit indexing B-tree
1195 * height amount of integers. We assume the height if the TNC tree will
1199 c->bottom_up_buf = kmalloc(BOTTOM_UP_HEIGHT * sizeof(int), GFP_KERNEL);
1200 if (!c->bottom_up_buf)
1203 c->sbuf = vmalloc(c->leb_size);
1208 c->ileb_buf = vmalloc(c->leb_size);
1213 if (c->bulk_read == 1)
1217 c->write_reserve_buf = kmalloc(COMPRESSED_DATA_NODE_BUF_SZ,
1219 if (!c->write_reserve_buf)
1225 err = ubifs_read_superblock(c);
1230 * Make sure the compressor which is set as default in the superblock
1231 * or overridden by mount options is actually compiled in.
1233 if (!ubifs_compr_present(c->default_compr)) {
1234 ubifs_err("'compressor \"%s\" is not compiled in",
1235 ubifs_compr_name(c->default_compr));
1240 err = init_constants_sb(c);
1244 sz = ALIGN(c->max_idx_node_sz, c->min_io_size);
1245 sz = ALIGN(sz + c->max_idx_node_sz, c->min_io_size);
1246 c->cbuf = kmalloc(sz, GFP_NOFS);
1252 sprintf(c->bgt_name, BGT_NAME_PATTERN, c->vi.ubi_num, c->vi.vol_id);
1254 err = alloc_wbufs(c);
1258 /* Create background thread */
1259 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1260 if (IS_ERR(c->bgt)) {
1261 err = PTR_ERR(c->bgt);
1263 ubifs_err("cannot spawn \"%s\", error %d",
1267 wake_up_process(c->bgt);
1270 err = ubifs_read_master(c);
1274 init_constants_master(c);
1276 if ((c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY)) != 0) {
1277 ubifs_msg("recovery needed");
1278 c->need_recovery = 1;
1280 err = ubifs_recover_inl_heads(c, c->sbuf);
1284 } else if (!c->ro_mount) {
1286 * Set the "dirty" flag so that if we reboot uncleanly we
1287 * will notice this immediately on the next mount.
1289 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1290 err = ubifs_write_master(c);
1295 err = ubifs_lpt_init(c, 1, !c->ro_mount);
1299 err = dbg_check_idx_size(c, c->old_idx_sz);
1303 err = ubifs_replay_journal(c);
1307 /* Calculate 'min_idx_lebs' after journal replay */
1308 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
1310 err = ubifs_mount_orphans(c, c->need_recovery, c->ro_mount);
1317 err = check_free_space(c);
1321 /* Check for enough log space */
1322 lnum = c->lhead_lnum + 1;
1323 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1324 lnum = UBIFS_LOG_LNUM;
1325 if (lnum == c->ltail_lnum) {
1326 err = ubifs_consolidate_log(c);
1331 if (c->need_recovery) {
1332 err = ubifs_recover_size(c);
1335 err = ubifs_rcvry_gc_commit(c);
1339 err = take_gc_lnum(c);
1344 * GC LEB may contain garbage if there was an unclean
1345 * reboot, and it should be un-mapped.
1347 err = ubifs_leb_unmap(c, c->gc_lnum);
1352 err = dbg_check_lprops(c);
1355 } else if (c->need_recovery) {
1356 err = ubifs_recover_size(c);
1361 * Even if we mount read-only, we have to set space in GC LEB
1362 * to proper value because this affects UBIFS free space
1363 * reporting. We do not want to have a situation when
1364 * re-mounting from R/O to R/W changes amount of free space.
1366 err = take_gc_lnum(c);
1371 spin_lock(&ubifs_infos_lock);
1372 list_add_tail(&c->infos_list, &ubifs_infos);
1373 spin_unlock(&ubifs_infos_lock);
1375 if (c->need_recovery) {
1377 ubifs_msg("recovery deferred");
1379 c->need_recovery = 0;
1380 ubifs_msg("recovery completed");
1382 * GC LEB has to be empty and taken at this point. But
1383 * the journal head LEBs may also be accounted as
1384 * "empty taken" if they are empty.
1386 ubifs_assert(c->lst.taken_empty_lebs > 0);
1389 ubifs_assert(c->lst.taken_empty_lebs > 0);
1391 err = dbg_check_filesystem(c);
1395 err = dbg_debugfs_init_fs(c);
1401 ubifs_msg("mounted UBI device %d, volume %d, name \"%s\"",
1402 c->vi.ubi_num, c->vi.vol_id, c->vi.name);
1404 ubifs_msg("mounted read-only");
1405 x = (long long)c->main_lebs * c->leb_size;
1406 ubifs_msg("file system size: %lld bytes (%lld KiB, %lld MiB, %d "
1407 "LEBs)", x, x >> 10, x >> 20, c->main_lebs);
1408 x = (long long)c->log_lebs * c->leb_size + c->max_bud_bytes;
1409 ubifs_msg("journal size: %lld bytes (%lld KiB, %lld MiB, %d "
1410 "LEBs)", x, x >> 10, x >> 20, c->log_lebs + c->max_bud_cnt);
1411 ubifs_msg("media format: w%d/r%d (latest is w%d/r%d)",
1412 c->fmt_version, c->ro_compat_version,
1413 UBIFS_FORMAT_VERSION, UBIFS_RO_COMPAT_VERSION);
1414 ubifs_msg("default compressor: %s", ubifs_compr_name(c->default_compr));
1415 ubifs_msg("reserved for root: %llu bytes (%llu KiB)",
1416 c->report_rp_size, c->report_rp_size >> 10);
1418 dbg_msg("compiled on: " __DATE__ " at " __TIME__);
1419 dbg_msg("min. I/O unit size: %d bytes", c->min_io_size);
1420 dbg_msg("max. write size: %d bytes", c->max_write_size);
1421 dbg_msg("LEB size: %d bytes (%d KiB)",
1422 c->leb_size, c->leb_size >> 10);
1423 dbg_msg("data journal heads: %d",
1424 c->jhead_cnt - NONDATA_JHEADS_CNT);
1425 dbg_msg("UUID: %pUB", c->uuid);
1426 dbg_msg("big_lpt %d", c->big_lpt);
1427 dbg_msg("log LEBs: %d (%d - %d)",
1428 c->log_lebs, UBIFS_LOG_LNUM, c->log_last);
1429 dbg_msg("LPT area LEBs: %d (%d - %d)",
1430 c->lpt_lebs, c->lpt_first, c->lpt_last);
1431 dbg_msg("orphan area LEBs: %d (%d - %d)",
1432 c->orph_lebs, c->orph_first, c->orph_last);
1433 dbg_msg("main area LEBs: %d (%d - %d)",
1434 c->main_lebs, c->main_first, c->leb_cnt - 1);
1435 dbg_msg("index LEBs: %d", c->lst.idx_lebs);
1436 dbg_msg("total index bytes: %lld (%lld KiB, %lld MiB)",
1437 c->old_idx_sz, c->old_idx_sz >> 10, c->old_idx_sz >> 20);
1438 dbg_msg("key hash type: %d", c->key_hash_type);
1439 dbg_msg("tree fanout: %d", c->fanout);
1440 dbg_msg("reserved GC LEB: %d", c->gc_lnum);
1441 dbg_msg("first main LEB: %d", c->main_first);
1442 dbg_msg("max. znode size %d", c->max_znode_sz);
1443 dbg_msg("max. index node size %d", c->max_idx_node_sz);
1444 dbg_msg("node sizes: data %zu, inode %zu, dentry %zu",
1445 UBIFS_DATA_NODE_SZ, UBIFS_INO_NODE_SZ, UBIFS_DENT_NODE_SZ);
1446 dbg_msg("node sizes: trun %zu, sb %zu, master %zu",
1447 UBIFS_TRUN_NODE_SZ, UBIFS_SB_NODE_SZ, UBIFS_MST_NODE_SZ);
1448 dbg_msg("node sizes: ref %zu, cmt. start %zu, orph %zu",
1449 UBIFS_REF_NODE_SZ, UBIFS_CS_NODE_SZ, UBIFS_ORPH_NODE_SZ);
1450 dbg_msg("max. node sizes: data %zu, inode %zu dentry %zu, idx %d",
1451 UBIFS_MAX_DATA_NODE_SZ, UBIFS_MAX_INO_NODE_SZ,
1452 UBIFS_MAX_DENT_NODE_SZ, ubifs_idx_node_sz(c, c->fanout));
1453 dbg_msg("dead watermark: %d", c->dead_wm);
1454 dbg_msg("dark watermark: %d", c->dark_wm);
1455 dbg_msg("LEB overhead: %d", c->leb_overhead);
1456 x = (long long)c->main_lebs * c->dark_wm;
1457 dbg_msg("max. dark space: %lld (%lld KiB, %lld MiB)",
1458 x, x >> 10, x >> 20);
1459 dbg_msg("maximum bud bytes: %lld (%lld KiB, %lld MiB)",
1460 c->max_bud_bytes, c->max_bud_bytes >> 10,
1461 c->max_bud_bytes >> 20);
1462 dbg_msg("BG commit bud bytes: %lld (%lld KiB, %lld MiB)",
1463 c->bg_bud_bytes, c->bg_bud_bytes >> 10,
1464 c->bg_bud_bytes >> 20);
1465 dbg_msg("current bud bytes %lld (%lld KiB, %lld MiB)",
1466 c->bud_bytes, c->bud_bytes >> 10, c->bud_bytes >> 20);
1467 dbg_msg("max. seq. number: %llu", c->max_sqnum);
1468 dbg_msg("commit number: %llu", c->cmt_no);
1473 spin_lock(&ubifs_infos_lock);
1474 list_del(&c->infos_list);
1475 spin_unlock(&ubifs_infos_lock);
1481 ubifs_lpt_free(c, 0);
1484 kfree(c->rcvrd_mst_node);
1486 kthread_stop(c->bgt);
1492 kfree(c->write_reserve_buf);
1496 kfree(c->bottom_up_buf);
1497 ubifs_debugging_exit(c);
1502 * ubifs_umount - un-mount UBIFS file-system.
1503 * @c: UBIFS file-system description object
1505 * Note, this function is called to free allocated resourced when un-mounting,
1506 * as well as free resources when an error occurred while we were half way
1507 * through mounting (error path cleanup function). So it has to make sure the
1508 * resource was actually allocated before freeing it.
1510 static void ubifs_umount(struct ubifs_info *c)
1512 dbg_gen("un-mounting UBI device %d, volume %d", c->vi.ubi_num,
1515 dbg_debugfs_exit_fs(c);
1516 spin_lock(&ubifs_infos_lock);
1517 list_del(&c->infos_list);
1518 spin_unlock(&ubifs_infos_lock);
1521 kthread_stop(c->bgt);
1526 ubifs_lpt_free(c, 0);
1529 kfree(c->rcvrd_mst_node);
1531 kfree(c->write_reserve_buf);
1535 kfree(c->bottom_up_buf);
1536 ubifs_debugging_exit(c);
1540 * ubifs_remount_rw - re-mount in read-write mode.
1541 * @c: UBIFS file-system description object
1543 * UBIFS avoids allocating many unnecessary resources when mounted in read-only
1544 * mode. This function allocates the needed resources and re-mounts UBIFS in
1547 static int ubifs_remount_rw(struct ubifs_info *c)
1551 if (c->rw_incompat) {
1552 ubifs_err("the file-system is not R/W-compatible");
1553 ubifs_msg("on-flash format version is w%d/r%d, but software "
1554 "only supports up to version w%d/r%d", c->fmt_version,
1555 c->ro_compat_version, UBIFS_FORMAT_VERSION,
1556 UBIFS_RO_COMPAT_VERSION);
1560 mutex_lock(&c->umount_mutex);
1561 dbg_save_space_info(c);
1562 c->remounting_rw = 1;
1564 err = check_free_space(c);
1568 if (c->old_leb_cnt != c->leb_cnt) {
1569 struct ubifs_sb_node *sup;
1571 sup = ubifs_read_sb_node(c);
1576 sup->leb_cnt = cpu_to_le32(c->leb_cnt);
1577 err = ubifs_write_sb_node(c, sup);
1582 if (c->need_recovery) {
1583 ubifs_msg("completing deferred recovery");
1584 err = ubifs_write_rcvrd_mst_node(c);
1587 err = ubifs_recover_size(c);
1590 err = ubifs_clean_lebs(c, c->sbuf);
1593 err = ubifs_recover_inl_heads(c, c->sbuf);
1597 /* A readonly mount is not allowed to have orphans */
1598 ubifs_assert(c->tot_orphans == 0);
1599 err = ubifs_clear_orphans(c);
1604 if (!(c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY))) {
1605 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1606 err = ubifs_write_master(c);
1611 c->ileb_buf = vmalloc(c->leb_size);
1617 c->write_reserve_buf = kmalloc(COMPRESSED_DATA_NODE_BUF_SZ, GFP_KERNEL);
1618 if (!c->write_reserve_buf)
1621 err = ubifs_lpt_init(c, 0, 1);
1625 err = alloc_wbufs(c);
1629 ubifs_create_buds_lists(c);
1631 /* Create background thread */
1632 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1633 if (IS_ERR(c->bgt)) {
1634 err = PTR_ERR(c->bgt);
1636 ubifs_err("cannot spawn \"%s\", error %d",
1640 wake_up_process(c->bgt);
1642 c->orph_buf = vmalloc(c->leb_size);
1648 /* Check for enough log space */
1649 lnum = c->lhead_lnum + 1;
1650 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1651 lnum = UBIFS_LOG_LNUM;
1652 if (lnum == c->ltail_lnum) {
1653 err = ubifs_consolidate_log(c);
1658 if (c->need_recovery)
1659 err = ubifs_rcvry_gc_commit(c);
1661 err = ubifs_leb_unmap(c, c->gc_lnum);
1665 if (c->need_recovery) {
1666 c->need_recovery = 0;
1667 ubifs_msg("deferred recovery completed");
1670 dbg_gen("re-mounted read-write");
1672 c->remounting_rw = 0;
1673 err = dbg_check_space_info(c);
1674 mutex_unlock(&c->umount_mutex);
1681 kthread_stop(c->bgt);
1685 kfree(c->write_reserve_buf);
1686 c->write_reserve_buf = NULL;
1689 ubifs_lpt_free(c, 1);
1690 c->remounting_rw = 0;
1691 mutex_unlock(&c->umount_mutex);
1696 * ubifs_remount_ro - re-mount in read-only mode.
1697 * @c: UBIFS file-system description object
1699 * We assume VFS has stopped writing. Possibly the background thread could be
1700 * running a commit, however kthread_stop will wait in that case.
1702 static void ubifs_remount_ro(struct ubifs_info *c)
1706 ubifs_assert(!c->need_recovery);
1707 ubifs_assert(!c->ro_mount);
1709 mutex_lock(&c->umount_mutex);
1711 kthread_stop(c->bgt);
1715 dbg_save_space_info(c);
1717 for (i = 0; i < c->jhead_cnt; i++)
1718 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1720 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1721 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1722 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1723 err = ubifs_write_master(c);
1725 ubifs_ro_mode(c, err);
1730 kfree(c->write_reserve_buf);
1731 c->write_reserve_buf = NULL;
1734 ubifs_lpt_free(c, 1);
1736 err = dbg_check_space_info(c);
1738 ubifs_ro_mode(c, err);
1739 mutex_unlock(&c->umount_mutex);
1742 static void ubifs_put_super(struct super_block *sb)
1745 struct ubifs_info *c = sb->s_fs_info;
1747 ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
1751 * The following asserts are only valid if there has not been a failure
1752 * of the media. For example, there will be dirty inodes if we failed
1753 * to write them back because of I/O errors.
1755 ubifs_assert(atomic_long_read(&c->dirty_pg_cnt) == 0);
1756 ubifs_assert(c->budg_idx_growth == 0);
1757 ubifs_assert(c->budg_dd_growth == 0);
1758 ubifs_assert(c->budg_data_growth == 0);
1761 * The 'c->umount_lock' prevents races between UBIFS memory shrinker
1762 * and file system un-mount. Namely, it prevents the shrinker from
1763 * picking this superblock for shrinking - it will be just skipped if
1764 * the mutex is locked.
1766 mutex_lock(&c->umount_mutex);
1769 * First of all kill the background thread to make sure it does
1770 * not interfere with un-mounting and freeing resources.
1773 kthread_stop(c->bgt);
1778 * On fatal errors c->ro_error is set to 1, in which case we do
1779 * not write the master node.
1784 /* Synchronize write-buffers */
1785 for (i = 0; i < c->jhead_cnt; i++)
1786 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1789 * We are being cleanly unmounted which means the
1790 * orphans were killed - indicate this in the master
1791 * node. Also save the reserved GC LEB number.
1793 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1794 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1795 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1796 err = ubifs_write_master(c);
1799 * Recovery will attempt to fix the master area
1800 * next mount, so we just print a message and
1801 * continue to unmount normally.
1803 ubifs_err("failed to write master node, "
1806 for (i = 0; i < c->jhead_cnt; i++)
1807 /* Make sure write-buffer timers are canceled */
1808 hrtimer_cancel(&c->jheads[i].wbuf.timer);
1813 bdi_destroy(&c->bdi);
1814 ubi_close_volume(c->ubi);
1815 mutex_unlock(&c->umount_mutex);
1819 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
1822 struct ubifs_info *c = sb->s_fs_info;
1824 dbg_gen("old flags %#lx, new flags %#x", sb->s_flags, *flags);
1826 err = ubifs_parse_options(c, data, 1);
1828 ubifs_err("invalid or unknown remount parameter");
1832 if (c->ro_mount && !(*flags & MS_RDONLY)) {
1834 ubifs_msg("cannot re-mount R/W due to prior errors");
1838 ubifs_msg("cannot re-mount R/W - UBI volume is R/O");
1841 err = ubifs_remount_rw(c);
1844 } else if (!c->ro_mount && (*flags & MS_RDONLY)) {
1846 ubifs_msg("cannot re-mount R/O due to prior errors");
1849 ubifs_remount_ro(c);
1852 if (c->bulk_read == 1)
1855 dbg_gen("disable bulk-read");
1860 ubifs_assert(c->lst.taken_empty_lebs > 0);
1864 const struct super_operations ubifs_super_operations = {
1865 .alloc_inode = ubifs_alloc_inode,
1866 .destroy_inode = ubifs_destroy_inode,
1867 .put_super = ubifs_put_super,
1868 .write_inode = ubifs_write_inode,
1869 .evict_inode = ubifs_evict_inode,
1870 .statfs = ubifs_statfs,
1871 .dirty_inode = ubifs_dirty_inode,
1872 .remount_fs = ubifs_remount_fs,
1873 .show_options = ubifs_show_options,
1874 .sync_fs = ubifs_sync_fs,
1878 * open_ubi - parse UBI device name string and open the UBI device.
1879 * @name: UBI volume name
1880 * @mode: UBI volume open mode
1882 * The primary method of mounting UBIFS is by specifying the UBI volume
1883 * character device node path. However, UBIFS may also be mounted withoug any
1884 * character device node using one of the following methods:
1886 * o ubiX_Y - mount UBI device number X, volume Y;
1887 * o ubiY - mount UBI device number 0, volume Y;
1888 * o ubiX:NAME - mount UBI device X, volume with name NAME;
1889 * o ubi:NAME - mount UBI device 0, volume with name NAME.
1891 * Alternative '!' separator may be used instead of ':' (because some shells
1892 * like busybox may interpret ':' as an NFS host name separator). This function
1893 * returns UBI volume description object in case of success and a negative
1894 * error code in case of failure.
1896 static struct ubi_volume_desc *open_ubi(const char *name, int mode)
1898 struct ubi_volume_desc *ubi;
1902 /* First, try to open using the device node path method */
1903 ubi = ubi_open_volume_path(name, mode);
1907 /* Try the "nodev" method */
1908 if (name[0] != 'u' || name[1] != 'b' || name[2] != 'i')
1909 return ERR_PTR(-EINVAL);
1911 /* ubi:NAME method */
1912 if ((name[3] == ':' || name[3] == '!') && name[4] != '\0')
1913 return ubi_open_volume_nm(0, name + 4, mode);
1915 if (!isdigit(name[3]))
1916 return ERR_PTR(-EINVAL);
1918 dev = simple_strtoul(name + 3, &endptr, 0);
1921 if (*endptr == '\0')
1922 return ubi_open_volume(0, dev, mode);
1925 if (*endptr == '_' && isdigit(endptr[1])) {
1926 vol = simple_strtoul(endptr + 1, &endptr, 0);
1927 if (*endptr != '\0')
1928 return ERR_PTR(-EINVAL);
1929 return ubi_open_volume(dev, vol, mode);
1932 /* ubiX:NAME method */
1933 if ((*endptr == ':' || *endptr == '!') && endptr[1] != '\0')
1934 return ubi_open_volume_nm(dev, ++endptr, mode);
1936 return ERR_PTR(-EINVAL);
1939 static int ubifs_fill_super(struct super_block *sb, void *data, int silent)
1941 struct ubi_volume_desc *ubi = sb->s_fs_info;
1942 struct ubifs_info *c;
1946 c = kzalloc(sizeof(struct ubifs_info), GFP_KERNEL);
1950 spin_lock_init(&c->cnt_lock);
1951 spin_lock_init(&c->cs_lock);
1952 spin_lock_init(&c->buds_lock);
1953 spin_lock_init(&c->space_lock);
1954 spin_lock_init(&c->orphan_lock);
1955 init_rwsem(&c->commit_sem);
1956 mutex_init(&c->lp_mutex);
1957 mutex_init(&c->tnc_mutex);
1958 mutex_init(&c->log_mutex);
1959 mutex_init(&c->mst_mutex);
1960 mutex_init(&c->umount_mutex);
1961 mutex_init(&c->bu_mutex);
1962 mutex_init(&c->write_reserve_mutex);
1963 init_waitqueue_head(&c->cmt_wq);
1965 c->old_idx = RB_ROOT;
1966 c->size_tree = RB_ROOT;
1967 c->orph_tree = RB_ROOT;
1968 INIT_LIST_HEAD(&c->infos_list);
1969 INIT_LIST_HEAD(&c->idx_gc);
1970 INIT_LIST_HEAD(&c->replay_list);
1971 INIT_LIST_HEAD(&c->replay_buds);
1972 INIT_LIST_HEAD(&c->uncat_list);
1973 INIT_LIST_HEAD(&c->empty_list);
1974 INIT_LIST_HEAD(&c->freeable_list);
1975 INIT_LIST_HEAD(&c->frdi_idx_list);
1976 INIT_LIST_HEAD(&c->unclean_leb_list);
1977 INIT_LIST_HEAD(&c->old_buds);
1978 INIT_LIST_HEAD(&c->orph_list);
1979 INIT_LIST_HEAD(&c->orph_new);
1980 c->no_chk_data_crc = 1;
1983 c->highest_inum = UBIFS_FIRST_INO;
1984 c->lhead_lnum = c->ltail_lnum = UBIFS_LOG_LNUM;
1986 ubi_get_volume_info(ubi, &c->vi);
1987 ubi_get_device_info(c->vi.ubi_num, &c->di);
1989 /* Re-open the UBI device in read-write mode */
1990 c->ubi = ubi_open_volume(c->vi.ubi_num, c->vi.vol_id, UBI_READWRITE);
1991 if (IS_ERR(c->ubi)) {
1992 err = PTR_ERR(c->ubi);
1997 * UBIFS provides 'backing_dev_info' in order to disable read-ahead. For
1998 * UBIFS, I/O is not deferred, it is done immediately in readpage,
1999 * which means the user would have to wait not just for their own I/O
2000 * but the read-ahead I/O as well i.e. completely pointless.
2002 * Read-ahead will be disabled because @c->bdi.ra_pages is 0.
2004 c->bdi.name = "ubifs",
2005 c->bdi.capabilities = BDI_CAP_MAP_COPY;
2006 c->bdi.unplug_io_fn = default_unplug_io_fn;
2007 err = bdi_init(&c->bdi);
2010 err = bdi_register(&c->bdi, NULL, "ubifs_%d_%d",
2011 c->vi.ubi_num, c->vi.vol_id);
2015 err = ubifs_parse_options(c, data, 0);
2019 sb->s_bdi = &c->bdi;
2021 sb->s_magic = UBIFS_SUPER_MAGIC;
2022 sb->s_blocksize = UBIFS_BLOCK_SIZE;
2023 sb->s_blocksize_bits = UBIFS_BLOCK_SHIFT;
2024 sb->s_maxbytes = c->max_inode_sz = key_max_inode_size(c);
2025 if (c->max_inode_sz > MAX_LFS_FILESIZE)
2026 sb->s_maxbytes = c->max_inode_sz = MAX_LFS_FILESIZE;
2027 sb->s_op = &ubifs_super_operations;
2029 mutex_lock(&c->umount_mutex);
2030 err = mount_ubifs(c);
2032 ubifs_assert(err < 0);
2036 /* Read the root inode */
2037 root = ubifs_iget(sb, UBIFS_ROOT_INO);
2039 err = PTR_ERR(root);
2043 sb->s_root = d_alloc_root(root);
2047 mutex_unlock(&c->umount_mutex);
2055 mutex_unlock(&c->umount_mutex);
2057 bdi_destroy(&c->bdi);
2059 ubi_close_volume(c->ubi);
2065 static int sb_test(struct super_block *sb, void *data)
2068 struct ubifs_info *c = sb->s_fs_info;
2070 return c->vi.cdev == *dev;
2073 static struct dentry *ubifs_mount(struct file_system_type *fs_type, int flags,
2074 const char *name, void *data)
2076 struct ubi_volume_desc *ubi;
2077 struct ubi_volume_info vi;
2078 struct super_block *sb;
2081 dbg_gen("name %s, flags %#x", name, flags);
2084 * Get UBI device number and volume ID. Mount it read-only so far
2085 * because this might be a new mount point, and UBI allows only one
2086 * read-write user at a time.
2088 ubi = open_ubi(name, UBI_READONLY);
2090 dbg_err("cannot open \"%s\", error %d",
2091 name, (int)PTR_ERR(ubi));
2092 return ERR_CAST(ubi);
2094 ubi_get_volume_info(ubi, &vi);
2096 dbg_gen("opened ubi%d_%d", vi.ubi_num, vi.vol_id);
2098 sb = sget(fs_type, &sb_test, &set_anon_super, &vi.cdev);
2105 struct ubifs_info *c1 = sb->s_fs_info;
2107 /* A new mount point for already mounted UBIFS */
2108 dbg_gen("this ubi volume is already mounted");
2109 if (!!(flags & MS_RDONLY) != c1->ro_mount) {
2114 sb->s_flags = flags;
2116 * Pass 'ubi' to 'fill_super()' in sb->s_fs_info where it is
2119 sb->s_fs_info = ubi;
2120 err = ubifs_fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
2123 /* We do not support atime */
2124 sb->s_flags |= MS_ACTIVE | MS_NOATIME;
2127 /* 'fill_super()' opens ubi again so we must close it here */
2128 ubi_close_volume(ubi);
2130 return dget(sb->s_root);
2133 deactivate_locked_super(sb);
2135 ubi_close_volume(ubi);
2136 return ERR_PTR(err);
2139 static struct file_system_type ubifs_fs_type = {
2141 .owner = THIS_MODULE,
2142 .mount = ubifs_mount,
2143 .kill_sb = kill_anon_super,
2147 * Inode slab cache constructor.
2149 static void inode_slab_ctor(void *obj)
2151 struct ubifs_inode *ui = obj;
2152 inode_init_once(&ui->vfs_inode);
2155 static int __init ubifs_init(void)
2159 BUILD_BUG_ON(sizeof(struct ubifs_ch) != 24);
2161 /* Make sure node sizes are 8-byte aligned */
2162 BUILD_BUG_ON(UBIFS_CH_SZ & 7);
2163 BUILD_BUG_ON(UBIFS_INO_NODE_SZ & 7);
2164 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ & 7);
2165 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ & 7);
2166 BUILD_BUG_ON(UBIFS_DATA_NODE_SZ & 7);
2167 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ & 7);
2168 BUILD_BUG_ON(UBIFS_SB_NODE_SZ & 7);
2169 BUILD_BUG_ON(UBIFS_MST_NODE_SZ & 7);
2170 BUILD_BUG_ON(UBIFS_REF_NODE_SZ & 7);
2171 BUILD_BUG_ON(UBIFS_CS_NODE_SZ & 7);
2172 BUILD_BUG_ON(UBIFS_ORPH_NODE_SZ & 7);
2174 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ & 7);
2175 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ & 7);
2176 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ & 7);
2177 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ & 7);
2178 BUILD_BUG_ON(UBIFS_MAX_NODE_SZ & 7);
2179 BUILD_BUG_ON(MIN_WRITE_SZ & 7);
2181 /* Check min. node size */
2182 BUILD_BUG_ON(UBIFS_INO_NODE_SZ < MIN_WRITE_SZ);
2183 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ < MIN_WRITE_SZ);
2184 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ < MIN_WRITE_SZ);
2185 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ < MIN_WRITE_SZ);
2187 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2188 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2189 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ > UBIFS_MAX_NODE_SZ);
2190 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ > UBIFS_MAX_NODE_SZ);
2192 /* Defined node sizes */
2193 BUILD_BUG_ON(UBIFS_SB_NODE_SZ != 4096);
2194 BUILD_BUG_ON(UBIFS_MST_NODE_SZ != 512);
2195 BUILD_BUG_ON(UBIFS_INO_NODE_SZ != 160);
2196 BUILD_BUG_ON(UBIFS_REF_NODE_SZ != 64);
2199 * We use 2 bit wide bit-fields to store compression type, which should
2200 * be amended if more compressors are added. The bit-fields are:
2201 * @compr_type in 'struct ubifs_inode', @default_compr in
2202 * 'struct ubifs_info' and @compr_type in 'struct ubifs_mount_opts'.
2204 BUILD_BUG_ON(UBIFS_COMPR_TYPES_CNT > 4);
2207 * We require that PAGE_CACHE_SIZE is greater-than-or-equal-to
2208 * UBIFS_BLOCK_SIZE. It is assumed that both are powers of 2.
2210 if (PAGE_CACHE_SIZE < UBIFS_BLOCK_SIZE) {
2211 ubifs_err("VFS page cache size is %u bytes, but UBIFS requires"
2212 " at least 4096 bytes",
2213 (unsigned int)PAGE_CACHE_SIZE);
2217 err = register_filesystem(&ubifs_fs_type);
2219 ubifs_err("cannot register file system, error %d", err);
2224 ubifs_inode_slab = kmem_cache_create("ubifs_inode_slab",
2225 sizeof(struct ubifs_inode), 0,
2226 SLAB_MEM_SPREAD | SLAB_RECLAIM_ACCOUNT,
2228 if (!ubifs_inode_slab)
2231 register_shrinker(&ubifs_shrinker_info);
2233 err = ubifs_compressors_init();
2237 err = dbg_debugfs_init();
2244 ubifs_compressors_exit();
2246 unregister_shrinker(&ubifs_shrinker_info);
2247 kmem_cache_destroy(ubifs_inode_slab);
2249 unregister_filesystem(&ubifs_fs_type);
2252 /* late_initcall to let compressors initialize first */
2253 late_initcall(ubifs_init);
2255 static void __exit ubifs_exit(void)
2257 ubifs_assert(list_empty(&ubifs_infos));
2258 ubifs_assert(atomic_long_read(&ubifs_clean_zn_cnt) == 0);
2261 ubifs_compressors_exit();
2262 unregister_shrinker(&ubifs_shrinker_info);
2263 kmem_cache_destroy(ubifs_inode_slab);
2264 unregister_filesystem(&ubifs_fs_type);
2266 module_exit(ubifs_exit);
2268 MODULE_LICENSE("GPL");
2269 MODULE_VERSION(__stringify(UBIFS_VERSION));
2270 MODULE_AUTHOR("Artem Bityutskiy, Adrian Hunter");
2271 MODULE_DESCRIPTION("UBIFS - UBI File System");