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]);
400 ubifs_assert(buf->f_bfree <= c->block_cnt);
404 static int ubifs_show_options(struct seq_file *s, struct vfsmount *mnt)
406 struct ubifs_info *c = mnt->mnt_sb->s_fs_info;
408 if (c->mount_opts.unmount_mode == 2)
409 seq_printf(s, ",fast_unmount");
410 else if (c->mount_opts.unmount_mode == 1)
411 seq_printf(s, ",norm_unmount");
413 if (c->mount_opts.bulk_read == 2)
414 seq_printf(s, ",bulk_read");
415 else if (c->mount_opts.bulk_read == 1)
416 seq_printf(s, ",no_bulk_read");
418 if (c->mount_opts.chk_data_crc == 2)
419 seq_printf(s, ",chk_data_crc");
420 else if (c->mount_opts.chk_data_crc == 1)
421 seq_printf(s, ",no_chk_data_crc");
423 if (c->mount_opts.override_compr) {
424 seq_printf(s, ",compr=%s",
425 ubifs_compr_name(c->mount_opts.compr_type));
431 static int ubifs_sync_fs(struct super_block *sb, int wait)
434 struct ubifs_info *c = sb->s_fs_info;
435 struct writeback_control wbc = {
436 .sync_mode = WB_SYNC_ALL,
438 .range_end = LLONG_MAX,
439 .nr_to_write = LONG_MAX,
443 * Zero @wait is just an advisory thing to help the file system shove
444 * lots of data into the queues, and there will be the second
445 * '->sync_fs()' call, with non-zero @wait.
450 if (sb->s_flags & MS_RDONLY)
454 * VFS calls '->sync_fs()' before synchronizing all dirty inodes and
455 * pages, so synchronize them first, then commit the journal. Strictly
456 * speaking, it is not necessary to commit the journal here,
457 * synchronizing write-buffers would be enough. But committing makes
458 * UBIFS free space predictions much more accurate, so we want to let
459 * the user be able to get more accurate results of 'statfs()' after
460 * they synchronize the file system.
462 generic_sync_sb_inodes(sb, &wbc);
465 * Synchronize write buffers, because 'ubifs_run_commit()' does not
466 * do this if it waits for an already running commit.
468 for (i = 0; i < c->jhead_cnt; i++) {
469 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
474 err = ubifs_run_commit(c);
478 return ubi_sync(c->vi.ubi_num);
482 * init_constants_early - initialize UBIFS constants.
483 * @c: UBIFS file-system description object
485 * This function initialize UBIFS constants which do not need the superblock to
486 * be read. It also checks that the UBI volume satisfies basic UBIFS
487 * requirements. Returns zero in case of success and a negative error code in
490 static int init_constants_early(struct ubifs_info *c)
492 if (c->vi.corrupted) {
493 ubifs_warn("UBI volume is corrupted - read-only mode");
498 ubifs_msg("read-only UBI device");
502 if (c->vi.vol_type == UBI_STATIC_VOLUME) {
503 ubifs_msg("static UBI volume - read-only mode");
507 c->leb_cnt = c->vi.size;
508 c->leb_size = c->vi.usable_leb_size;
509 c->half_leb_size = c->leb_size / 2;
510 c->min_io_size = c->di.min_io_size;
511 c->min_io_shift = fls(c->min_io_size) - 1;
513 if (c->leb_size < UBIFS_MIN_LEB_SZ) {
514 ubifs_err("too small LEBs (%d bytes), min. is %d bytes",
515 c->leb_size, UBIFS_MIN_LEB_SZ);
519 if (c->leb_cnt < UBIFS_MIN_LEB_CNT) {
520 ubifs_err("too few LEBs (%d), min. is %d",
521 c->leb_cnt, UBIFS_MIN_LEB_CNT);
525 if (!is_power_of_2(c->min_io_size)) {
526 ubifs_err("bad min. I/O size %d", c->min_io_size);
531 * UBIFS aligns all node to 8-byte boundary, so to make function in
532 * io.c simpler, assume minimum I/O unit size to be 8 bytes if it is
535 if (c->min_io_size < 8) {
540 c->ref_node_alsz = ALIGN(UBIFS_REF_NODE_SZ, c->min_io_size);
541 c->mst_node_alsz = ALIGN(UBIFS_MST_NODE_SZ, c->min_io_size);
544 * Initialize node length ranges which are mostly needed for node
547 c->ranges[UBIFS_PAD_NODE].len = UBIFS_PAD_NODE_SZ;
548 c->ranges[UBIFS_SB_NODE].len = UBIFS_SB_NODE_SZ;
549 c->ranges[UBIFS_MST_NODE].len = UBIFS_MST_NODE_SZ;
550 c->ranges[UBIFS_REF_NODE].len = UBIFS_REF_NODE_SZ;
551 c->ranges[UBIFS_TRUN_NODE].len = UBIFS_TRUN_NODE_SZ;
552 c->ranges[UBIFS_CS_NODE].len = UBIFS_CS_NODE_SZ;
554 c->ranges[UBIFS_INO_NODE].min_len = UBIFS_INO_NODE_SZ;
555 c->ranges[UBIFS_INO_NODE].max_len = UBIFS_MAX_INO_NODE_SZ;
556 c->ranges[UBIFS_ORPH_NODE].min_len =
557 UBIFS_ORPH_NODE_SZ + sizeof(__le64);
558 c->ranges[UBIFS_ORPH_NODE].max_len = c->leb_size;
559 c->ranges[UBIFS_DENT_NODE].min_len = UBIFS_DENT_NODE_SZ;
560 c->ranges[UBIFS_DENT_NODE].max_len = UBIFS_MAX_DENT_NODE_SZ;
561 c->ranges[UBIFS_XENT_NODE].min_len = UBIFS_XENT_NODE_SZ;
562 c->ranges[UBIFS_XENT_NODE].max_len = UBIFS_MAX_XENT_NODE_SZ;
563 c->ranges[UBIFS_DATA_NODE].min_len = UBIFS_DATA_NODE_SZ;
564 c->ranges[UBIFS_DATA_NODE].max_len = UBIFS_MAX_DATA_NODE_SZ;
566 * Minimum indexing node size is amended later when superblock is
567 * read and the key length is known.
569 c->ranges[UBIFS_IDX_NODE].min_len = UBIFS_IDX_NODE_SZ + UBIFS_BRANCH_SZ;
571 * Maximum indexing node size is amended later when superblock is
572 * read and the fanout is known.
574 c->ranges[UBIFS_IDX_NODE].max_len = INT_MAX;
577 * Initialize dead and dark LEB space watermarks. See gc.c for comments
578 * about these values.
580 c->dead_wm = ALIGN(MIN_WRITE_SZ, c->min_io_size);
581 c->dark_wm = ALIGN(UBIFS_MAX_NODE_SZ, c->min_io_size);
584 * Calculate how many bytes would be wasted at the end of LEB if it was
585 * fully filled with data nodes of maximum size. This is used in
586 * calculations when reporting free space.
588 c->leb_overhead = c->leb_size % UBIFS_MAX_DATA_NODE_SZ;
590 /* Buffer size for bulk-reads */
591 c->max_bu_buf_len = UBIFS_MAX_BULK_READ * UBIFS_MAX_DATA_NODE_SZ;
592 if (c->max_bu_buf_len > c->leb_size)
593 c->max_bu_buf_len = c->leb_size;
598 * bud_wbuf_callback - bud LEB write-buffer synchronization call-back.
599 * @c: UBIFS file-system description object
600 * @lnum: LEB the write-buffer was synchronized to
601 * @free: how many free bytes left in this LEB
602 * @pad: how many bytes were padded
604 * This is a callback function which is called by the I/O unit when the
605 * write-buffer is synchronized. We need this to correctly maintain space
606 * accounting in bud logical eraseblocks. This function returns zero in case of
607 * success and a negative error code in case of failure.
609 * This function actually belongs to the journal, but we keep it here because
610 * we want to keep it static.
612 static int bud_wbuf_callback(struct ubifs_info *c, int lnum, int free, int pad)
614 return ubifs_update_one_lp(c, lnum, free, pad, 0, 0);
618 * init_constants_sb - initialize UBIFS constants.
619 * @c: UBIFS file-system description object
621 * This is a helper function which initializes various UBIFS constants after
622 * the superblock has been read. It also checks various UBIFS parameters and
623 * makes sure they are all right. Returns zero in case of success and a
624 * negative error code in case of failure.
626 static int init_constants_sb(struct ubifs_info *c)
631 c->main_bytes = (long long)c->main_lebs * c->leb_size;
632 c->max_znode_sz = sizeof(struct ubifs_znode) +
633 c->fanout * sizeof(struct ubifs_zbranch);
635 tmp = ubifs_idx_node_sz(c, 1);
636 c->ranges[UBIFS_IDX_NODE].min_len = tmp;
637 c->min_idx_node_sz = ALIGN(tmp, 8);
639 tmp = ubifs_idx_node_sz(c, c->fanout);
640 c->ranges[UBIFS_IDX_NODE].max_len = tmp;
641 c->max_idx_node_sz = ALIGN(tmp, 8);
643 /* Make sure LEB size is large enough to fit full commit */
644 tmp = UBIFS_CS_NODE_SZ + UBIFS_REF_NODE_SZ * c->jhead_cnt;
645 tmp = ALIGN(tmp, c->min_io_size);
646 if (tmp > c->leb_size) {
647 dbg_err("too small LEB size %d, at least %d needed",
653 * Make sure that the log is large enough to fit reference nodes for
654 * all buds plus one reserved LEB.
656 tmp64 = c->max_bud_bytes + c->leb_size - 1;
657 c->max_bud_cnt = div_u64(tmp64, c->leb_size);
658 tmp = (c->ref_node_alsz * c->max_bud_cnt + c->leb_size - 1);
661 if (c->log_lebs < tmp) {
662 dbg_err("too small log %d LEBs, required min. %d LEBs",
668 * When budgeting we assume worst-case scenarios when the pages are not
669 * be compressed and direntries are of the maximum size.
671 * Note, data, which may be stored in inodes is budgeted separately, so
672 * it is not included into 'c->inode_budget'.
674 c->page_budget = UBIFS_MAX_DATA_NODE_SZ * UBIFS_BLOCKS_PER_PAGE;
675 c->inode_budget = UBIFS_INO_NODE_SZ;
676 c->dent_budget = UBIFS_MAX_DENT_NODE_SZ;
679 * When the amount of flash space used by buds becomes
680 * 'c->max_bud_bytes', UBIFS just blocks all writers and starts commit.
681 * The writers are unblocked when the commit is finished. To avoid
682 * writers to be blocked UBIFS initiates background commit in advance,
683 * when number of bud bytes becomes above the limit defined below.
685 c->bg_bud_bytes = (c->max_bud_bytes * 13) >> 4;
688 * Ensure minimum journal size. All the bytes in the journal heads are
689 * considered to be used, when calculating the current journal usage.
690 * Consequently, if the journal is too small, UBIFS will treat it as
693 tmp64 = (long long)(c->jhead_cnt + 1) * c->leb_size + 1;
694 if (c->bg_bud_bytes < tmp64)
695 c->bg_bud_bytes = tmp64;
696 if (c->max_bud_bytes < tmp64 + c->leb_size)
697 c->max_bud_bytes = tmp64 + c->leb_size;
699 err = ubifs_calc_lpt_geom(c);
703 /* Initialize effective LEB size used in budgeting calculations */
704 c->idx_leb_size = c->leb_size - c->max_idx_node_sz;
709 * init_constants_master - initialize UBIFS constants.
710 * @c: UBIFS file-system description object
712 * This is a helper function which initializes various UBIFS constants after
713 * the master node has been read. It also checks various UBIFS parameters and
714 * makes sure they are all right.
716 static void init_constants_master(struct ubifs_info *c)
720 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
721 c->report_rp_size = ubifs_reported_space(c, c->rp_size);
724 * Calculate total amount of FS blocks. This number is not used
725 * internally because it does not make much sense for UBIFS, but it is
726 * necessary to report something for the 'statfs()' call.
728 * Subtract the LEB reserved for GC, the LEB which is reserved for
729 * deletions, minimum LEBs for the index, and assume only one journal
732 tmp64 = c->main_lebs - 1 - 1 - MIN_INDEX_LEBS - c->jhead_cnt + 1;
733 tmp64 *= (long long)c->leb_size - c->leb_overhead;
734 tmp64 = ubifs_reported_space(c, tmp64);
735 c->block_cnt = tmp64 >> UBIFS_BLOCK_SHIFT;
739 * take_gc_lnum - reserve GC LEB.
740 * @c: UBIFS file-system description object
742 * This function ensures that the LEB reserved for garbage collection is marked
743 * as "taken" in lprops. We also have to set free space to LEB size and dirty
744 * space to zero, because lprops may contain out-of-date information if the
745 * file-system was un-mounted before it has been committed. This function
746 * returns zero in case of success and a negative error code in case of
749 static int take_gc_lnum(struct ubifs_info *c)
753 if (c->gc_lnum == -1) {
754 ubifs_err("no LEB for GC");
758 /* And we have to tell lprops that this LEB is taken */
759 err = ubifs_change_one_lp(c, c->gc_lnum, c->leb_size, 0,
765 * alloc_wbufs - allocate write-buffers.
766 * @c: UBIFS file-system description object
768 * This helper function allocates and initializes UBIFS write-buffers. Returns
769 * zero in case of success and %-ENOMEM in case of failure.
771 static int alloc_wbufs(struct ubifs_info *c)
775 c->jheads = kzalloc(c->jhead_cnt * sizeof(struct ubifs_jhead),
780 /* Initialize journal heads */
781 for (i = 0; i < c->jhead_cnt; i++) {
782 INIT_LIST_HEAD(&c->jheads[i].buds_list);
783 err = ubifs_wbuf_init(c, &c->jheads[i].wbuf);
787 c->jheads[i].wbuf.sync_callback = &bud_wbuf_callback;
788 c->jheads[i].wbuf.jhead = i;
791 c->jheads[BASEHD].wbuf.dtype = UBI_SHORTTERM;
793 * Garbage Collector head likely contains long-term data and
794 * does not need to be synchronized by timer.
796 c->jheads[GCHD].wbuf.dtype = UBI_LONGTERM;
797 c->jheads[GCHD].wbuf.timeout = 0;
803 * free_wbufs - free write-buffers.
804 * @c: UBIFS file-system description object
806 static void free_wbufs(struct ubifs_info *c)
811 for (i = 0; i < c->jhead_cnt; i++) {
812 kfree(c->jheads[i].wbuf.buf);
813 kfree(c->jheads[i].wbuf.inodes);
821 * free_orphans - free orphans.
822 * @c: UBIFS file-system description object
824 static void free_orphans(struct ubifs_info *c)
826 struct ubifs_orphan *orph;
828 while (c->orph_dnext) {
829 orph = c->orph_dnext;
830 c->orph_dnext = orph->dnext;
831 list_del(&orph->list);
835 while (!list_empty(&c->orph_list)) {
836 orph = list_entry(c->orph_list.next, struct ubifs_orphan, list);
837 list_del(&orph->list);
839 dbg_err("orphan list not empty at unmount");
847 * free_buds - free per-bud objects.
848 * @c: UBIFS file-system description object
850 static void free_buds(struct ubifs_info *c)
852 struct rb_node *this = c->buds.rb_node;
853 struct ubifs_bud *bud;
857 this = this->rb_left;
858 else if (this->rb_right)
859 this = this->rb_right;
861 bud = rb_entry(this, struct ubifs_bud, rb);
862 this = rb_parent(this);
864 if (this->rb_left == &bud->rb)
865 this->rb_left = NULL;
867 this->rb_right = NULL;
875 * check_volume_empty - check if the UBI volume is empty.
876 * @c: UBIFS file-system description object
878 * This function checks if the UBIFS volume is empty by looking if its LEBs are
879 * mapped or not. The result of checking is stored in the @c->empty variable.
880 * Returns zero in case of success and a negative error code in case of
883 static int check_volume_empty(struct ubifs_info *c)
888 for (lnum = 0; lnum < c->leb_cnt; lnum++) {
889 err = ubi_is_mapped(c->ubi, lnum);
890 if (unlikely(err < 0))
904 * UBIFS mount options.
906 * Opt_fast_unmount: do not run a journal commit before un-mounting
907 * Opt_norm_unmount: run a journal commit before un-mounting
908 * Opt_bulk_read: enable bulk-reads
909 * Opt_no_bulk_read: disable bulk-reads
910 * Opt_chk_data_crc: check CRCs when reading data nodes
911 * Opt_no_chk_data_crc: do not check CRCs when reading data nodes
912 * Opt_override_compr: override default compressor
913 * Opt_err: just end of array marker
926 static const match_table_t tokens = {
927 {Opt_fast_unmount, "fast_unmount"},
928 {Opt_norm_unmount, "norm_unmount"},
929 {Opt_bulk_read, "bulk_read"},
930 {Opt_no_bulk_read, "no_bulk_read"},
931 {Opt_chk_data_crc, "chk_data_crc"},
932 {Opt_no_chk_data_crc, "no_chk_data_crc"},
933 {Opt_override_compr, "compr=%s"},
938 * ubifs_parse_options - parse mount parameters.
939 * @c: UBIFS file-system description object
940 * @options: parameters to parse
941 * @is_remount: non-zero if this is FS re-mount
943 * This function parses UBIFS mount options and returns zero in case success
944 * and a negative error code in case of failure.
946 static int ubifs_parse_options(struct ubifs_info *c, char *options,
950 substring_t args[MAX_OPT_ARGS];
955 while ((p = strsep(&options, ","))) {
961 token = match_token(p, tokens, args);
964 * %Opt_fast_unmount and %Opt_norm_unmount options are ignored.
965 * We accepte them in order to be backware-compatible. But this
966 * should be removed at some point.
968 case Opt_fast_unmount:
969 c->mount_opts.unmount_mode = 2;
971 case Opt_norm_unmount:
972 c->mount_opts.unmount_mode = 1;
975 c->mount_opts.bulk_read = 2;
978 case Opt_no_bulk_read:
979 c->mount_opts.bulk_read = 1;
982 case Opt_chk_data_crc:
983 c->mount_opts.chk_data_crc = 2;
984 c->no_chk_data_crc = 0;
986 case Opt_no_chk_data_crc:
987 c->mount_opts.chk_data_crc = 1;
988 c->no_chk_data_crc = 1;
990 case Opt_override_compr:
992 char *name = match_strdup(&args[0]);
996 if (!strcmp(name, "none"))
997 c->mount_opts.compr_type = UBIFS_COMPR_NONE;
998 else if (!strcmp(name, "lzo"))
999 c->mount_opts.compr_type = UBIFS_COMPR_LZO;
1000 else if (!strcmp(name, "zlib"))
1001 c->mount_opts.compr_type = UBIFS_COMPR_ZLIB;
1003 ubifs_err("unknown compressor \"%s\"", name);
1008 c->mount_opts.override_compr = 1;
1009 c->default_compr = c->mount_opts.compr_type;
1013 ubifs_err("unrecognized mount option \"%s\" "
1014 "or missing value", p);
1023 * destroy_journal - destroy journal data structures.
1024 * @c: UBIFS file-system description object
1026 * This function destroys journal data structures including those that may have
1027 * been created by recovery functions.
1029 static void destroy_journal(struct ubifs_info *c)
1031 while (!list_empty(&c->unclean_leb_list)) {
1032 struct ubifs_unclean_leb *ucleb;
1034 ucleb = list_entry(c->unclean_leb_list.next,
1035 struct ubifs_unclean_leb, list);
1036 list_del(&ucleb->list);
1039 while (!list_empty(&c->old_buds)) {
1040 struct ubifs_bud *bud;
1042 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
1043 list_del(&bud->list);
1046 ubifs_destroy_idx_gc(c);
1047 ubifs_destroy_size_tree(c);
1053 * bu_init - initialize bulk-read information.
1054 * @c: UBIFS file-system description object
1056 static void bu_init(struct ubifs_info *c)
1058 ubifs_assert(c->bulk_read == 1);
1061 return; /* Already initialized */
1064 c->bu.buf = kmalloc(c->max_bu_buf_len, GFP_KERNEL | __GFP_NOWARN);
1066 if (c->max_bu_buf_len > UBIFS_KMALLOC_OK) {
1067 c->max_bu_buf_len = UBIFS_KMALLOC_OK;
1071 /* Just disable bulk-read */
1072 ubifs_warn("Cannot allocate %d bytes of memory for bulk-read, "
1073 "disabling it", c->max_bu_buf_len);
1074 c->mount_opts.bulk_read = 1;
1081 * check_free_space - check if there is enough free space to mount.
1082 * @c: UBIFS file-system description object
1084 * This function makes sure UBIFS has enough free space to be mounted in
1085 * read/write mode. UBIFS must always have some free space to allow deletions.
1087 static int check_free_space(struct ubifs_info *c)
1089 ubifs_assert(c->dark_wm > 0);
1090 if (c->lst.total_free + c->lst.total_dirty < c->dark_wm) {
1091 ubifs_err("insufficient free space to mount in read/write mode");
1100 * mount_ubifs - mount UBIFS file-system.
1101 * @c: UBIFS file-system description object
1103 * This function mounts UBIFS file system. Returns zero in case of success and
1104 * a negative error code in case of failure.
1106 * Note, the function does not de-allocate resources it it fails half way
1107 * through, and the caller has to do this instead.
1109 static int mount_ubifs(struct ubifs_info *c)
1111 struct super_block *sb = c->vfs_sb;
1112 int err, mounted_read_only = (sb->s_flags & MS_RDONLY);
1116 err = init_constants_early(c);
1120 err = ubifs_debugging_init(c);
1124 err = check_volume_empty(c);
1128 if (c->empty && (mounted_read_only || c->ro_media)) {
1130 * This UBI volume is empty, and read-only, or the file system
1131 * is mounted read-only - we cannot format it.
1133 ubifs_err("can't format empty UBI volume: read-only %s",
1134 c->ro_media ? "UBI volume" : "mount");
1139 if (c->ro_media && !mounted_read_only) {
1140 ubifs_err("cannot mount read-write - read-only media");
1146 * The requirement for the buffer is that it should fit indexing B-tree
1147 * height amount of integers. We assume the height if the TNC tree will
1151 c->bottom_up_buf = kmalloc(BOTTOM_UP_HEIGHT * sizeof(int), GFP_KERNEL);
1152 if (!c->bottom_up_buf)
1155 c->sbuf = vmalloc(c->leb_size);
1159 if (!mounted_read_only) {
1160 c->ileb_buf = vmalloc(c->leb_size);
1165 if (c->bulk_read == 1)
1169 * We have to check all CRCs, even for data nodes, when we mount the FS
1170 * (specifically, when we are replaying).
1172 c->always_chk_crc = 1;
1174 err = ubifs_read_superblock(c);
1179 * Make sure the compressor which is set as default in the superblock
1180 * or overridden by mount options is actually compiled in.
1182 if (!ubifs_compr_present(c->default_compr)) {
1183 ubifs_err("'compressor \"%s\" is not compiled in",
1184 ubifs_compr_name(c->default_compr));
1188 err = init_constants_sb(c);
1192 sz = ALIGN(c->max_idx_node_sz, c->min_io_size);
1193 sz = ALIGN(sz + c->max_idx_node_sz, c->min_io_size);
1194 c->cbuf = kmalloc(sz, GFP_NOFS);
1200 sprintf(c->bgt_name, BGT_NAME_PATTERN, c->vi.ubi_num, c->vi.vol_id);
1201 if (!mounted_read_only) {
1202 err = alloc_wbufs(c);
1206 /* Create background thread */
1207 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1208 if (IS_ERR(c->bgt)) {
1209 err = PTR_ERR(c->bgt);
1211 ubifs_err("cannot spawn \"%s\", error %d",
1215 wake_up_process(c->bgt);
1218 err = ubifs_read_master(c);
1222 init_constants_master(c);
1224 if ((c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY)) != 0) {
1225 ubifs_msg("recovery needed");
1226 c->need_recovery = 1;
1227 if (!mounted_read_only) {
1228 err = ubifs_recover_inl_heads(c, c->sbuf);
1232 } else if (!mounted_read_only) {
1234 * Set the "dirty" flag so that if we reboot uncleanly we
1235 * will notice this immediately on the next mount.
1237 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1238 err = ubifs_write_master(c);
1243 err = ubifs_lpt_init(c, 1, !mounted_read_only);
1247 err = dbg_check_idx_size(c, c->old_idx_sz);
1251 err = ubifs_replay_journal(c);
1255 err = ubifs_mount_orphans(c, c->need_recovery, mounted_read_only);
1259 if (!mounted_read_only) {
1262 err = check_free_space(c);
1266 /* Check for enough log space */
1267 lnum = c->lhead_lnum + 1;
1268 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1269 lnum = UBIFS_LOG_LNUM;
1270 if (lnum == c->ltail_lnum) {
1271 err = ubifs_consolidate_log(c);
1276 if (c->need_recovery) {
1277 err = ubifs_recover_size(c);
1280 err = ubifs_rcvry_gc_commit(c);
1282 err = take_gc_lnum(c);
1287 * GC LEB may contain garbage if there was an unclean
1288 * reboot, and it should be un-mapped.
1290 err = ubifs_leb_unmap(c, c->gc_lnum);
1295 err = dbg_check_lprops(c);
1298 } else if (c->need_recovery) {
1299 err = ubifs_recover_size(c);
1304 * Even if we mount read-only, we have to set space in GC LEB
1305 * to proper value because this affects UBIFS free space
1306 * reporting. We do not want to have a situation when
1307 * re-mounting from R/O to R/W changes amount of free space.
1309 err = take_gc_lnum(c);
1314 spin_lock(&ubifs_infos_lock);
1315 list_add_tail(&c->infos_list, &ubifs_infos);
1316 spin_unlock(&ubifs_infos_lock);
1318 if (c->need_recovery) {
1319 if (mounted_read_only)
1320 ubifs_msg("recovery deferred");
1322 c->need_recovery = 0;
1323 ubifs_msg("recovery completed");
1325 * GC LEB has to be empty and taken at this point. But
1326 * the journal head LEBs may also be accounted as
1327 * "empty taken" if they are empty.
1329 ubifs_assert(c->lst.taken_empty_lebs > 0);
1332 ubifs_assert(c->lst.taken_empty_lebs > 0);
1334 err = dbg_check_filesystem(c);
1338 err = dbg_debugfs_init_fs(c);
1342 c->always_chk_crc = 0;
1344 ubifs_msg("mounted UBI device %d, volume %d, name \"%s\"",
1345 c->vi.ubi_num, c->vi.vol_id, c->vi.name);
1346 if (mounted_read_only)
1347 ubifs_msg("mounted read-only");
1348 x = (long long)c->main_lebs * c->leb_size;
1349 ubifs_msg("file system size: %lld bytes (%lld KiB, %lld MiB, %d "
1350 "LEBs)", x, x >> 10, x >> 20, c->main_lebs);
1351 x = (long long)c->log_lebs * c->leb_size + c->max_bud_bytes;
1352 ubifs_msg("journal size: %lld bytes (%lld KiB, %lld MiB, %d "
1353 "LEBs)", x, x >> 10, x >> 20, c->log_lebs + c->max_bud_cnt);
1354 ubifs_msg("media format: %d (latest is %d)",
1355 c->fmt_version, UBIFS_FORMAT_VERSION);
1356 ubifs_msg("default compressor: %s", ubifs_compr_name(c->default_compr));
1357 ubifs_msg("reserved for root: %llu bytes (%llu KiB)",
1358 c->report_rp_size, c->report_rp_size >> 10);
1360 dbg_msg("compiled on: " __DATE__ " at " __TIME__);
1361 dbg_msg("min. I/O unit size: %d bytes", c->min_io_size);
1362 dbg_msg("LEB size: %d bytes (%d KiB)",
1363 c->leb_size, c->leb_size >> 10);
1364 dbg_msg("data journal heads: %d",
1365 c->jhead_cnt - NONDATA_JHEADS_CNT);
1366 dbg_msg("UUID: %02X%02X%02X%02X-%02X%02X"
1367 "-%02X%02X-%02X%02X-%02X%02X%02X%02X%02X%02X",
1368 c->uuid[0], c->uuid[1], c->uuid[2], c->uuid[3],
1369 c->uuid[4], c->uuid[5], c->uuid[6], c->uuid[7],
1370 c->uuid[8], c->uuid[9], c->uuid[10], c->uuid[11],
1371 c->uuid[12], c->uuid[13], c->uuid[14], c->uuid[15]);
1372 dbg_msg("big_lpt %d", c->big_lpt);
1373 dbg_msg("log LEBs: %d (%d - %d)",
1374 c->log_lebs, UBIFS_LOG_LNUM, c->log_last);
1375 dbg_msg("LPT area LEBs: %d (%d - %d)",
1376 c->lpt_lebs, c->lpt_first, c->lpt_last);
1377 dbg_msg("orphan area LEBs: %d (%d - %d)",
1378 c->orph_lebs, c->orph_first, c->orph_last);
1379 dbg_msg("main area LEBs: %d (%d - %d)",
1380 c->main_lebs, c->main_first, c->leb_cnt - 1);
1381 dbg_msg("index LEBs: %d", c->lst.idx_lebs);
1382 dbg_msg("total index bytes: %lld (%lld KiB, %lld MiB)",
1383 c->old_idx_sz, c->old_idx_sz >> 10, c->old_idx_sz >> 20);
1384 dbg_msg("key hash type: %d", c->key_hash_type);
1385 dbg_msg("tree fanout: %d", c->fanout);
1386 dbg_msg("reserved GC LEB: %d", c->gc_lnum);
1387 dbg_msg("first main LEB: %d", c->main_first);
1388 dbg_msg("max. znode size %d", c->max_znode_sz);
1389 dbg_msg("max. index node size %d", c->max_idx_node_sz);
1390 dbg_msg("node sizes: data %zu, inode %zu, dentry %zu",
1391 UBIFS_DATA_NODE_SZ, UBIFS_INO_NODE_SZ, UBIFS_DENT_NODE_SZ);
1392 dbg_msg("node sizes: trun %zu, sb %zu, master %zu",
1393 UBIFS_TRUN_NODE_SZ, UBIFS_SB_NODE_SZ, UBIFS_MST_NODE_SZ);
1394 dbg_msg("node sizes: ref %zu, cmt. start %zu, orph %zu",
1395 UBIFS_REF_NODE_SZ, UBIFS_CS_NODE_SZ, UBIFS_ORPH_NODE_SZ);
1396 dbg_msg("max. node sizes: data %zu, inode %zu dentry %zu",
1397 UBIFS_MAX_DATA_NODE_SZ, UBIFS_MAX_INO_NODE_SZ,
1398 UBIFS_MAX_DENT_NODE_SZ);
1399 dbg_msg("dead watermark: %d", c->dead_wm);
1400 dbg_msg("dark watermark: %d", c->dark_wm);
1401 dbg_msg("LEB overhead: %d", c->leb_overhead);
1402 x = (long long)c->main_lebs * c->dark_wm;
1403 dbg_msg("max. dark space: %lld (%lld KiB, %lld MiB)",
1404 x, x >> 10, x >> 20);
1405 dbg_msg("maximum bud bytes: %lld (%lld KiB, %lld MiB)",
1406 c->max_bud_bytes, c->max_bud_bytes >> 10,
1407 c->max_bud_bytes >> 20);
1408 dbg_msg("BG commit bud bytes: %lld (%lld KiB, %lld MiB)",
1409 c->bg_bud_bytes, c->bg_bud_bytes >> 10,
1410 c->bg_bud_bytes >> 20);
1411 dbg_msg("current bud bytes %lld (%lld KiB, %lld MiB)",
1412 c->bud_bytes, c->bud_bytes >> 10, c->bud_bytes >> 20);
1413 dbg_msg("max. seq. number: %llu", c->max_sqnum);
1414 dbg_msg("commit number: %llu", c->cmt_no);
1419 spin_lock(&ubifs_infos_lock);
1420 list_del(&c->infos_list);
1421 spin_unlock(&ubifs_infos_lock);
1427 ubifs_lpt_free(c, 0);
1430 kfree(c->rcvrd_mst_node);
1432 kthread_stop(c->bgt);
1441 kfree(c->bottom_up_buf);
1442 ubifs_debugging_exit(c);
1447 * ubifs_umount - un-mount UBIFS file-system.
1448 * @c: UBIFS file-system description object
1450 * Note, this function is called to free allocated resourced when un-mounting,
1451 * as well as free resources when an error occurred while we were half way
1452 * through mounting (error path cleanup function). So it has to make sure the
1453 * resource was actually allocated before freeing it.
1455 static void ubifs_umount(struct ubifs_info *c)
1457 dbg_gen("un-mounting UBI device %d, volume %d", c->vi.ubi_num,
1460 dbg_debugfs_exit_fs(c);
1461 spin_lock(&ubifs_infos_lock);
1462 list_del(&c->infos_list);
1463 spin_unlock(&ubifs_infos_lock);
1466 kthread_stop(c->bgt);
1471 ubifs_lpt_free(c, 0);
1474 kfree(c->rcvrd_mst_node);
1479 kfree(c->bottom_up_buf);
1480 ubifs_debugging_exit(c);
1484 * ubifs_remount_rw - re-mount in read-write mode.
1485 * @c: UBIFS file-system description object
1487 * UBIFS avoids allocating many unnecessary resources when mounted in read-only
1488 * mode. This function allocates the needed resources and re-mounts UBIFS in
1491 static int ubifs_remount_rw(struct ubifs_info *c)
1495 mutex_lock(&c->umount_mutex);
1496 dbg_save_space_info(c);
1497 c->remounting_rw = 1;
1498 c->always_chk_crc = 1;
1500 err = check_free_space(c);
1504 if (c->old_leb_cnt != c->leb_cnt) {
1505 struct ubifs_sb_node *sup;
1507 sup = ubifs_read_sb_node(c);
1512 sup->leb_cnt = cpu_to_le32(c->leb_cnt);
1513 err = ubifs_write_sb_node(c, sup);
1518 if (c->need_recovery) {
1519 ubifs_msg("completing deferred recovery");
1520 err = ubifs_write_rcvrd_mst_node(c);
1523 err = ubifs_recover_size(c);
1526 err = ubifs_clean_lebs(c, c->sbuf);
1529 err = ubifs_recover_inl_heads(c, c->sbuf);
1533 /* A readonly mount is not allowed to have orphans */
1534 ubifs_assert(c->tot_orphans == 0);
1535 err = ubifs_clear_orphans(c);
1540 if (!(c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY))) {
1541 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1542 err = ubifs_write_master(c);
1547 c->ileb_buf = vmalloc(c->leb_size);
1553 err = ubifs_lpt_init(c, 0, 1);
1557 err = alloc_wbufs(c);
1561 ubifs_create_buds_lists(c);
1563 /* Create background thread */
1564 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1565 if (IS_ERR(c->bgt)) {
1566 err = PTR_ERR(c->bgt);
1568 ubifs_err("cannot spawn \"%s\", error %d",
1572 wake_up_process(c->bgt);
1574 c->orph_buf = vmalloc(c->leb_size);
1580 /* Check for enough log space */
1581 lnum = c->lhead_lnum + 1;
1582 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1583 lnum = UBIFS_LOG_LNUM;
1584 if (lnum == c->ltail_lnum) {
1585 err = ubifs_consolidate_log(c);
1590 if (c->need_recovery)
1591 err = ubifs_rcvry_gc_commit(c);
1593 err = ubifs_leb_unmap(c, c->gc_lnum);
1597 if (c->need_recovery) {
1598 c->need_recovery = 0;
1599 ubifs_msg("deferred recovery completed");
1602 dbg_gen("re-mounted read-write");
1603 c->vfs_sb->s_flags &= ~MS_RDONLY;
1604 c->remounting_rw = 0;
1605 c->always_chk_crc = 0;
1606 err = dbg_check_space_info(c);
1607 mutex_unlock(&c->umount_mutex);
1614 kthread_stop(c->bgt);
1620 ubifs_lpt_free(c, 1);
1621 c->remounting_rw = 0;
1622 c->always_chk_crc = 0;
1623 mutex_unlock(&c->umount_mutex);
1628 * ubifs_remount_ro - re-mount in read-only mode.
1629 * @c: UBIFS file-system description object
1631 * We assume VFS has stopped writing. Possibly the background thread could be
1632 * running a commit, however kthread_stop will wait in that case.
1634 static void ubifs_remount_ro(struct ubifs_info *c)
1638 ubifs_assert(!c->need_recovery);
1639 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY));
1641 mutex_lock(&c->umount_mutex);
1643 kthread_stop(c->bgt);
1647 dbg_save_space_info(c);
1649 for (i = 0; i < c->jhead_cnt; i++) {
1650 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1651 del_timer_sync(&c->jheads[i].wbuf.timer);
1654 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1655 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1656 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1657 err = ubifs_write_master(c);
1659 ubifs_ro_mode(c, err);
1666 ubifs_lpt_free(c, 1);
1667 err = dbg_check_space_info(c);
1669 ubifs_ro_mode(c, err);
1670 mutex_unlock(&c->umount_mutex);
1673 static void ubifs_put_super(struct super_block *sb)
1676 struct ubifs_info *c = sb->s_fs_info;
1678 ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
1681 * The following asserts are only valid if there has not been a failure
1682 * of the media. For example, there will be dirty inodes if we failed
1683 * to write them back because of I/O errors.
1685 ubifs_assert(atomic_long_read(&c->dirty_pg_cnt) == 0);
1686 ubifs_assert(c->budg_idx_growth == 0);
1687 ubifs_assert(c->budg_dd_growth == 0);
1688 ubifs_assert(c->budg_data_growth == 0);
1691 * The 'c->umount_lock' prevents races between UBIFS memory shrinker
1692 * and file system un-mount. Namely, it prevents the shrinker from
1693 * picking this superblock for shrinking - it will be just skipped if
1694 * the mutex is locked.
1696 mutex_lock(&c->umount_mutex);
1697 if (!(c->vfs_sb->s_flags & MS_RDONLY)) {
1699 * First of all kill the background thread to make sure it does
1700 * not interfere with un-mounting and freeing resources.
1703 kthread_stop(c->bgt);
1707 /* Synchronize write-buffers */
1709 for (i = 0; i < c->jhead_cnt; i++) {
1710 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1711 del_timer_sync(&c->jheads[i].wbuf.timer);
1715 * On fatal errors c->ro_media is set to 1, in which case we do
1716 * not write the master node.
1720 * We are being cleanly unmounted which means the
1721 * orphans were killed - indicate this in the master
1722 * node. Also save the reserved GC LEB number.
1726 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1727 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1728 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1729 err = ubifs_write_master(c);
1732 * Recovery will attempt to fix the master area
1733 * next mount, so we just print a message and
1734 * continue to unmount normally.
1736 ubifs_err("failed to write master node, "
1742 bdi_destroy(&c->bdi);
1743 ubi_close_volume(c->ubi);
1744 mutex_unlock(&c->umount_mutex);
1748 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
1751 struct ubifs_info *c = sb->s_fs_info;
1753 dbg_gen("old flags %#lx, new flags %#x", sb->s_flags, *flags);
1755 err = ubifs_parse_options(c, data, 1);
1757 ubifs_err("invalid or unknown remount parameter");
1761 if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
1763 ubifs_msg("cannot re-mount due to prior errors");
1766 err = ubifs_remount_rw(c);
1769 } else if (!(sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY)) {
1771 ubifs_msg("cannot re-mount due to prior errors");
1774 ubifs_remount_ro(c);
1777 if (c->bulk_read == 1)
1780 dbg_gen("disable bulk-read");
1785 ubifs_assert(c->lst.taken_empty_lebs > 0);
1789 const struct super_operations ubifs_super_operations = {
1790 .alloc_inode = ubifs_alloc_inode,
1791 .destroy_inode = ubifs_destroy_inode,
1792 .put_super = ubifs_put_super,
1793 .write_inode = ubifs_write_inode,
1794 .delete_inode = ubifs_delete_inode,
1795 .statfs = ubifs_statfs,
1796 .dirty_inode = ubifs_dirty_inode,
1797 .remount_fs = ubifs_remount_fs,
1798 .show_options = ubifs_show_options,
1799 .sync_fs = ubifs_sync_fs,
1803 * open_ubi - parse UBI device name string and open the UBI device.
1804 * @name: UBI volume name
1805 * @mode: UBI volume open mode
1807 * There are several ways to specify UBI volumes when mounting UBIFS:
1808 * o ubiX_Y - UBI device number X, volume Y;
1809 * o ubiY - UBI device number 0, volume Y;
1810 * o ubiX:NAME - mount UBI device X, volume with name NAME;
1811 * o ubi:NAME - mount UBI device 0, volume with name NAME.
1813 * Alternative '!' separator may be used instead of ':' (because some shells
1814 * like busybox may interpret ':' as an NFS host name separator). This function
1815 * returns ubi volume object in case of success and a negative error code in
1818 static struct ubi_volume_desc *open_ubi(const char *name, int mode)
1823 if (name[0] != 'u' || name[1] != 'b' || name[2] != 'i')
1824 return ERR_PTR(-EINVAL);
1826 /* ubi:NAME method */
1827 if ((name[3] == ':' || name[3] == '!') && name[4] != '\0')
1828 return ubi_open_volume_nm(0, name + 4, mode);
1830 if (!isdigit(name[3]))
1831 return ERR_PTR(-EINVAL);
1833 dev = simple_strtoul(name + 3, &endptr, 0);
1836 if (*endptr == '\0')
1837 return ubi_open_volume(0, dev, mode);
1840 if (*endptr == '_' && isdigit(endptr[1])) {
1841 vol = simple_strtoul(endptr + 1, &endptr, 0);
1842 if (*endptr != '\0')
1843 return ERR_PTR(-EINVAL);
1844 return ubi_open_volume(dev, vol, mode);
1847 /* ubiX:NAME method */
1848 if ((*endptr == ':' || *endptr == '!') && endptr[1] != '\0')
1849 return ubi_open_volume_nm(dev, ++endptr, mode);
1851 return ERR_PTR(-EINVAL);
1854 static int ubifs_fill_super(struct super_block *sb, void *data, int silent)
1856 struct ubi_volume_desc *ubi = sb->s_fs_info;
1857 struct ubifs_info *c;
1861 c = kzalloc(sizeof(struct ubifs_info), GFP_KERNEL);
1865 spin_lock_init(&c->cnt_lock);
1866 spin_lock_init(&c->cs_lock);
1867 spin_lock_init(&c->buds_lock);
1868 spin_lock_init(&c->space_lock);
1869 spin_lock_init(&c->orphan_lock);
1870 init_rwsem(&c->commit_sem);
1871 mutex_init(&c->lp_mutex);
1872 mutex_init(&c->tnc_mutex);
1873 mutex_init(&c->log_mutex);
1874 mutex_init(&c->mst_mutex);
1875 mutex_init(&c->umount_mutex);
1876 mutex_init(&c->bu_mutex);
1877 init_waitqueue_head(&c->cmt_wq);
1879 c->old_idx = RB_ROOT;
1880 c->size_tree = RB_ROOT;
1881 c->orph_tree = RB_ROOT;
1882 INIT_LIST_HEAD(&c->infos_list);
1883 INIT_LIST_HEAD(&c->idx_gc);
1884 INIT_LIST_HEAD(&c->replay_list);
1885 INIT_LIST_HEAD(&c->replay_buds);
1886 INIT_LIST_HEAD(&c->uncat_list);
1887 INIT_LIST_HEAD(&c->empty_list);
1888 INIT_LIST_HEAD(&c->freeable_list);
1889 INIT_LIST_HEAD(&c->frdi_idx_list);
1890 INIT_LIST_HEAD(&c->unclean_leb_list);
1891 INIT_LIST_HEAD(&c->old_buds);
1892 INIT_LIST_HEAD(&c->orph_list);
1893 INIT_LIST_HEAD(&c->orph_new);
1895 c->highest_inum = UBIFS_FIRST_INO;
1896 c->lhead_lnum = c->ltail_lnum = UBIFS_LOG_LNUM;
1898 ubi_get_volume_info(ubi, &c->vi);
1899 ubi_get_device_info(c->vi.ubi_num, &c->di);
1901 /* Re-open the UBI device in read-write mode */
1902 c->ubi = ubi_open_volume(c->vi.ubi_num, c->vi.vol_id, UBI_READWRITE);
1903 if (IS_ERR(c->ubi)) {
1904 err = PTR_ERR(c->ubi);
1909 * UBIFS provides 'backing_dev_info' in order to disable read-ahead. For
1910 * UBIFS, I/O is not deferred, it is done immediately in readpage,
1911 * which means the user would have to wait not just for their own I/O
1912 * but the read-ahead I/O as well i.e. completely pointless.
1914 * Read-ahead will be disabled because @c->bdi.ra_pages is 0.
1916 c->bdi.capabilities = BDI_CAP_MAP_COPY;
1917 c->bdi.unplug_io_fn = default_unplug_io_fn;
1918 err = bdi_init(&c->bdi);
1922 err = ubifs_parse_options(c, data, 0);
1929 sb->s_magic = UBIFS_SUPER_MAGIC;
1930 sb->s_blocksize = UBIFS_BLOCK_SIZE;
1931 sb->s_blocksize_bits = UBIFS_BLOCK_SHIFT;
1932 sb->s_dev = c->vi.cdev;
1933 sb->s_maxbytes = c->max_inode_sz = key_max_inode_size(c);
1934 if (c->max_inode_sz > MAX_LFS_FILESIZE)
1935 sb->s_maxbytes = c->max_inode_sz = MAX_LFS_FILESIZE;
1936 sb->s_op = &ubifs_super_operations;
1938 mutex_lock(&c->umount_mutex);
1939 err = mount_ubifs(c);
1941 ubifs_assert(err < 0);
1945 /* Read the root inode */
1946 root = ubifs_iget(sb, UBIFS_ROOT_INO);
1948 err = PTR_ERR(root);
1952 sb->s_root = d_alloc_root(root);
1956 mutex_unlock(&c->umount_mutex);
1964 mutex_unlock(&c->umount_mutex);
1966 bdi_destroy(&c->bdi);
1968 ubi_close_volume(c->ubi);
1974 static int sb_test(struct super_block *sb, void *data)
1978 return sb->s_dev == *dev;
1981 static int sb_set(struct super_block *sb, void *data)
1989 static int ubifs_get_sb(struct file_system_type *fs_type, int flags,
1990 const char *name, void *data, struct vfsmount *mnt)
1992 struct ubi_volume_desc *ubi;
1993 struct ubi_volume_info vi;
1994 struct super_block *sb;
1997 dbg_gen("name %s, flags %#x", name, flags);
2000 * Get UBI device number and volume ID. Mount it read-only so far
2001 * because this might be a new mount point, and UBI allows only one
2002 * read-write user at a time.
2004 ubi = open_ubi(name, UBI_READONLY);
2006 ubifs_err("cannot open \"%s\", error %d",
2007 name, (int)PTR_ERR(ubi));
2008 return PTR_ERR(ubi);
2010 ubi_get_volume_info(ubi, &vi);
2012 dbg_gen("opened ubi%d_%d", vi.ubi_num, vi.vol_id);
2014 sb = sget(fs_type, &sb_test, &sb_set, &vi.cdev);
2021 /* A new mount point for already mounted UBIFS */
2022 dbg_gen("this ubi volume is already mounted");
2023 if ((flags ^ sb->s_flags) & MS_RDONLY) {
2028 sb->s_flags = flags;
2030 * Pass 'ubi' to 'fill_super()' in sb->s_fs_info where it is
2033 sb->s_fs_info = ubi;
2034 err = ubifs_fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
2037 /* We do not support atime */
2038 sb->s_flags |= MS_ACTIVE | MS_NOATIME;
2041 /* 'fill_super()' opens ubi again so we must close it here */
2042 ubi_close_volume(ubi);
2044 return simple_set_mnt(mnt, sb);
2047 up_write(&sb->s_umount);
2048 deactivate_super(sb);
2050 ubi_close_volume(ubi);
2054 static void ubifs_kill_sb(struct super_block *sb)
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");