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.
451 * VFS calls '->sync_fs()' before synchronizing all dirty inodes and
452 * pages, so synchronize them first, then commit the journal. Strictly
453 * speaking, it is not necessary to commit the journal here,
454 * synchronizing write-buffers would be enough. But committing makes
455 * UBIFS free space predictions much more accurate, so we want to let
456 * the user be able to get more accurate results of 'statfs()' after
457 * they synchronize the file system.
459 generic_sync_sb_inodes(sb, &wbc);
462 * Synchronize write buffers, because 'ubifs_run_commit()' does not
463 * do this if it waits for an already running commit.
465 for (i = 0; i < c->jhead_cnt; i++) {
466 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
471 err = ubifs_run_commit(c);
475 return ubi_sync(c->vi.ubi_num);
479 * init_constants_early - initialize UBIFS constants.
480 * @c: UBIFS file-system description object
482 * This function initialize UBIFS constants which do not need the superblock to
483 * be read. It also checks that the UBI volume satisfies basic UBIFS
484 * requirements. Returns zero in case of success and a negative error code in
487 static int init_constants_early(struct ubifs_info *c)
489 if (c->vi.corrupted) {
490 ubifs_warn("UBI volume is corrupted - read-only mode");
495 ubifs_msg("read-only UBI device");
499 if (c->vi.vol_type == UBI_STATIC_VOLUME) {
500 ubifs_msg("static UBI volume - read-only mode");
504 c->leb_cnt = c->vi.size;
505 c->leb_size = c->vi.usable_leb_size;
506 c->half_leb_size = c->leb_size / 2;
507 c->min_io_size = c->di.min_io_size;
508 c->min_io_shift = fls(c->min_io_size) - 1;
510 if (c->leb_size < UBIFS_MIN_LEB_SZ) {
511 ubifs_err("too small LEBs (%d bytes), min. is %d bytes",
512 c->leb_size, UBIFS_MIN_LEB_SZ);
516 if (c->leb_cnt < UBIFS_MIN_LEB_CNT) {
517 ubifs_err("too few LEBs (%d), min. is %d",
518 c->leb_cnt, UBIFS_MIN_LEB_CNT);
522 if (!is_power_of_2(c->min_io_size)) {
523 ubifs_err("bad min. I/O size %d", c->min_io_size);
528 * UBIFS aligns all node to 8-byte boundary, so to make function in
529 * io.c simpler, assume minimum I/O unit size to be 8 bytes if it is
532 if (c->min_io_size < 8) {
537 c->ref_node_alsz = ALIGN(UBIFS_REF_NODE_SZ, c->min_io_size);
538 c->mst_node_alsz = ALIGN(UBIFS_MST_NODE_SZ, c->min_io_size);
541 * Initialize node length ranges which are mostly needed for node
544 c->ranges[UBIFS_PAD_NODE].len = UBIFS_PAD_NODE_SZ;
545 c->ranges[UBIFS_SB_NODE].len = UBIFS_SB_NODE_SZ;
546 c->ranges[UBIFS_MST_NODE].len = UBIFS_MST_NODE_SZ;
547 c->ranges[UBIFS_REF_NODE].len = UBIFS_REF_NODE_SZ;
548 c->ranges[UBIFS_TRUN_NODE].len = UBIFS_TRUN_NODE_SZ;
549 c->ranges[UBIFS_CS_NODE].len = UBIFS_CS_NODE_SZ;
551 c->ranges[UBIFS_INO_NODE].min_len = UBIFS_INO_NODE_SZ;
552 c->ranges[UBIFS_INO_NODE].max_len = UBIFS_MAX_INO_NODE_SZ;
553 c->ranges[UBIFS_ORPH_NODE].min_len =
554 UBIFS_ORPH_NODE_SZ + sizeof(__le64);
555 c->ranges[UBIFS_ORPH_NODE].max_len = c->leb_size;
556 c->ranges[UBIFS_DENT_NODE].min_len = UBIFS_DENT_NODE_SZ;
557 c->ranges[UBIFS_DENT_NODE].max_len = UBIFS_MAX_DENT_NODE_SZ;
558 c->ranges[UBIFS_XENT_NODE].min_len = UBIFS_XENT_NODE_SZ;
559 c->ranges[UBIFS_XENT_NODE].max_len = UBIFS_MAX_XENT_NODE_SZ;
560 c->ranges[UBIFS_DATA_NODE].min_len = UBIFS_DATA_NODE_SZ;
561 c->ranges[UBIFS_DATA_NODE].max_len = UBIFS_MAX_DATA_NODE_SZ;
563 * Minimum indexing node size is amended later when superblock is
564 * read and the key length is known.
566 c->ranges[UBIFS_IDX_NODE].min_len = UBIFS_IDX_NODE_SZ + UBIFS_BRANCH_SZ;
568 * Maximum indexing node size is amended later when superblock is
569 * read and the fanout is known.
571 c->ranges[UBIFS_IDX_NODE].max_len = INT_MAX;
574 * Initialize dead and dark LEB space watermarks. See gc.c for comments
575 * about these values.
577 c->dead_wm = ALIGN(MIN_WRITE_SZ, c->min_io_size);
578 c->dark_wm = ALIGN(UBIFS_MAX_NODE_SZ, c->min_io_size);
581 * Calculate how many bytes would be wasted at the end of LEB if it was
582 * fully filled with data nodes of maximum size. This is used in
583 * calculations when reporting free space.
585 c->leb_overhead = c->leb_size % UBIFS_MAX_DATA_NODE_SZ;
587 /* Buffer size for bulk-reads */
588 c->max_bu_buf_len = UBIFS_MAX_BULK_READ * UBIFS_MAX_DATA_NODE_SZ;
589 if (c->max_bu_buf_len > c->leb_size)
590 c->max_bu_buf_len = c->leb_size;
595 * bud_wbuf_callback - bud LEB write-buffer synchronization call-back.
596 * @c: UBIFS file-system description object
597 * @lnum: LEB the write-buffer was synchronized to
598 * @free: how many free bytes left in this LEB
599 * @pad: how many bytes were padded
601 * This is a callback function which is called by the I/O unit when the
602 * write-buffer is synchronized. We need this to correctly maintain space
603 * accounting in bud logical eraseblocks. This function returns zero in case of
604 * success and a negative error code in case of failure.
606 * This function actually belongs to the journal, but we keep it here because
607 * we want to keep it static.
609 static int bud_wbuf_callback(struct ubifs_info *c, int lnum, int free, int pad)
611 return ubifs_update_one_lp(c, lnum, free, pad, 0, 0);
615 * init_constants_sb - initialize UBIFS constants.
616 * @c: UBIFS file-system description object
618 * This is a helper function which initializes various UBIFS constants after
619 * the superblock has been read. It also checks various UBIFS parameters and
620 * makes sure they are all right. Returns zero in case of success and a
621 * negative error code in case of failure.
623 static int init_constants_sb(struct ubifs_info *c)
628 c->main_bytes = (long long)c->main_lebs * c->leb_size;
629 c->max_znode_sz = sizeof(struct ubifs_znode) +
630 c->fanout * sizeof(struct ubifs_zbranch);
632 tmp = ubifs_idx_node_sz(c, 1);
633 c->ranges[UBIFS_IDX_NODE].min_len = tmp;
634 c->min_idx_node_sz = ALIGN(tmp, 8);
636 tmp = ubifs_idx_node_sz(c, c->fanout);
637 c->ranges[UBIFS_IDX_NODE].max_len = tmp;
638 c->max_idx_node_sz = ALIGN(tmp, 8);
640 /* Make sure LEB size is large enough to fit full commit */
641 tmp = UBIFS_CS_NODE_SZ + UBIFS_REF_NODE_SZ * c->jhead_cnt;
642 tmp = ALIGN(tmp, c->min_io_size);
643 if (tmp > c->leb_size) {
644 dbg_err("too small LEB size %d, at least %d needed",
650 * Make sure that the log is large enough to fit reference nodes for
651 * all buds plus one reserved LEB.
653 tmp64 = c->max_bud_bytes + c->leb_size - 1;
654 c->max_bud_cnt = div_u64(tmp64, c->leb_size);
655 tmp = (c->ref_node_alsz * c->max_bud_cnt + c->leb_size - 1);
658 if (c->log_lebs < tmp) {
659 dbg_err("too small log %d LEBs, required min. %d LEBs",
665 * When budgeting we assume worst-case scenarios when the pages are not
666 * be compressed and direntries are of the maximum size.
668 * Note, data, which may be stored in inodes is budgeted separately, so
669 * it is not included into 'c->inode_budget'.
671 c->page_budget = UBIFS_MAX_DATA_NODE_SZ * UBIFS_BLOCKS_PER_PAGE;
672 c->inode_budget = UBIFS_INO_NODE_SZ;
673 c->dent_budget = UBIFS_MAX_DENT_NODE_SZ;
676 * When the amount of flash space used by buds becomes
677 * 'c->max_bud_bytes', UBIFS just blocks all writers and starts commit.
678 * The writers are unblocked when the commit is finished. To avoid
679 * writers to be blocked UBIFS initiates background commit in advance,
680 * when number of bud bytes becomes above the limit defined below.
682 c->bg_bud_bytes = (c->max_bud_bytes * 13) >> 4;
685 * Ensure minimum journal size. All the bytes in the journal heads are
686 * considered to be used, when calculating the current journal usage.
687 * Consequently, if the journal is too small, UBIFS will treat it as
690 tmp64 = (long long)(c->jhead_cnt + 1) * c->leb_size + 1;
691 if (c->bg_bud_bytes < tmp64)
692 c->bg_bud_bytes = tmp64;
693 if (c->max_bud_bytes < tmp64 + c->leb_size)
694 c->max_bud_bytes = tmp64 + c->leb_size;
696 err = ubifs_calc_lpt_geom(c);
700 /* Initialize effective LEB size used in budgeting calculations */
701 c->idx_leb_size = c->leb_size - c->max_idx_node_sz;
706 * init_constants_master - initialize UBIFS constants.
707 * @c: UBIFS file-system description object
709 * This is a helper function which initializes various UBIFS constants after
710 * the master node has been read. It also checks various UBIFS parameters and
711 * makes sure they are all right.
713 static void init_constants_master(struct ubifs_info *c)
717 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
718 c->report_rp_size = ubifs_reported_space(c, c->rp_size);
721 * Calculate total amount of FS blocks. This number is not used
722 * internally because it does not make much sense for UBIFS, but it is
723 * necessary to report something for the 'statfs()' call.
725 * Subtract the LEB reserved for GC, the LEB which is reserved for
726 * deletions, minimum LEBs for the index, and assume only one journal
729 tmp64 = c->main_lebs - 1 - 1 - MIN_INDEX_LEBS - c->jhead_cnt + 1;
730 tmp64 *= (long long)c->leb_size - c->leb_overhead;
731 tmp64 = ubifs_reported_space(c, tmp64);
732 c->block_cnt = tmp64 >> UBIFS_BLOCK_SHIFT;
736 * take_gc_lnum - reserve GC LEB.
737 * @c: UBIFS file-system description object
739 * This function ensures that the LEB reserved for garbage collection is marked
740 * as "taken" in lprops. We also have to set free space to LEB size and dirty
741 * space to zero, because lprops may contain out-of-date information if the
742 * file-system was un-mounted before it has been committed. This function
743 * returns zero in case of success and a negative error code in case of
746 static int take_gc_lnum(struct ubifs_info *c)
750 if (c->gc_lnum == -1) {
751 ubifs_err("no LEB for GC");
755 /* And we have to tell lprops that this LEB is taken */
756 err = ubifs_change_one_lp(c, c->gc_lnum, c->leb_size, 0,
762 * alloc_wbufs - allocate write-buffers.
763 * @c: UBIFS file-system description object
765 * This helper function allocates and initializes UBIFS write-buffers. Returns
766 * zero in case of success and %-ENOMEM in case of failure.
768 static int alloc_wbufs(struct ubifs_info *c)
772 c->jheads = kzalloc(c->jhead_cnt * sizeof(struct ubifs_jhead),
777 /* Initialize journal heads */
778 for (i = 0; i < c->jhead_cnt; i++) {
779 INIT_LIST_HEAD(&c->jheads[i].buds_list);
780 err = ubifs_wbuf_init(c, &c->jheads[i].wbuf);
784 c->jheads[i].wbuf.sync_callback = &bud_wbuf_callback;
785 c->jheads[i].wbuf.jhead = i;
788 c->jheads[BASEHD].wbuf.dtype = UBI_SHORTTERM;
790 * Garbage Collector head likely contains long-term data and
791 * does not need to be synchronized by timer.
793 c->jheads[GCHD].wbuf.dtype = UBI_LONGTERM;
794 c->jheads[GCHD].wbuf.timeout = 0;
800 * free_wbufs - free write-buffers.
801 * @c: UBIFS file-system description object
803 static void free_wbufs(struct ubifs_info *c)
808 for (i = 0; i < c->jhead_cnt; i++) {
809 kfree(c->jheads[i].wbuf.buf);
810 kfree(c->jheads[i].wbuf.inodes);
818 * free_orphans - free orphans.
819 * @c: UBIFS file-system description object
821 static void free_orphans(struct ubifs_info *c)
823 struct ubifs_orphan *orph;
825 while (c->orph_dnext) {
826 orph = c->orph_dnext;
827 c->orph_dnext = orph->dnext;
828 list_del(&orph->list);
832 while (!list_empty(&c->orph_list)) {
833 orph = list_entry(c->orph_list.next, struct ubifs_orphan, list);
834 list_del(&orph->list);
836 dbg_err("orphan list not empty at unmount");
844 * free_buds - free per-bud objects.
845 * @c: UBIFS file-system description object
847 static void free_buds(struct ubifs_info *c)
849 struct rb_node *this = c->buds.rb_node;
850 struct ubifs_bud *bud;
854 this = this->rb_left;
855 else if (this->rb_right)
856 this = this->rb_right;
858 bud = rb_entry(this, struct ubifs_bud, rb);
859 this = rb_parent(this);
861 if (this->rb_left == &bud->rb)
862 this->rb_left = NULL;
864 this->rb_right = NULL;
872 * check_volume_empty - check if the UBI volume is empty.
873 * @c: UBIFS file-system description object
875 * This function checks if the UBIFS volume is empty by looking if its LEBs are
876 * mapped or not. The result of checking is stored in the @c->empty variable.
877 * Returns zero in case of success and a negative error code in case of
880 static int check_volume_empty(struct ubifs_info *c)
885 for (lnum = 0; lnum < c->leb_cnt; lnum++) {
886 err = ubi_is_mapped(c->ubi, lnum);
887 if (unlikely(err < 0))
901 * UBIFS mount options.
903 * Opt_fast_unmount: do not run a journal commit before un-mounting
904 * Opt_norm_unmount: run a journal commit before un-mounting
905 * Opt_bulk_read: enable bulk-reads
906 * Opt_no_bulk_read: disable bulk-reads
907 * Opt_chk_data_crc: check CRCs when reading data nodes
908 * Opt_no_chk_data_crc: do not check CRCs when reading data nodes
909 * Opt_override_compr: override default compressor
910 * Opt_err: just end of array marker
923 static const match_table_t tokens = {
924 {Opt_fast_unmount, "fast_unmount"},
925 {Opt_norm_unmount, "norm_unmount"},
926 {Opt_bulk_read, "bulk_read"},
927 {Opt_no_bulk_read, "no_bulk_read"},
928 {Opt_chk_data_crc, "chk_data_crc"},
929 {Opt_no_chk_data_crc, "no_chk_data_crc"},
930 {Opt_override_compr, "compr=%s"},
935 * ubifs_parse_options - parse mount parameters.
936 * @c: UBIFS file-system description object
937 * @options: parameters to parse
938 * @is_remount: non-zero if this is FS re-mount
940 * This function parses UBIFS mount options and returns zero in case success
941 * and a negative error code in case of failure.
943 static int ubifs_parse_options(struct ubifs_info *c, char *options,
947 substring_t args[MAX_OPT_ARGS];
952 while ((p = strsep(&options, ","))) {
958 token = match_token(p, tokens, args);
961 * %Opt_fast_unmount and %Opt_norm_unmount options are ignored.
962 * We accepte them in order to be backware-compatible. But this
963 * should be removed at some point.
965 case Opt_fast_unmount:
966 c->mount_opts.unmount_mode = 2;
968 case Opt_norm_unmount:
969 c->mount_opts.unmount_mode = 1;
972 c->mount_opts.bulk_read = 2;
975 case Opt_no_bulk_read:
976 c->mount_opts.bulk_read = 1;
979 case Opt_chk_data_crc:
980 c->mount_opts.chk_data_crc = 2;
981 c->no_chk_data_crc = 0;
983 case Opt_no_chk_data_crc:
984 c->mount_opts.chk_data_crc = 1;
985 c->no_chk_data_crc = 1;
987 case Opt_override_compr:
989 char *name = match_strdup(&args[0]);
993 if (!strcmp(name, "none"))
994 c->mount_opts.compr_type = UBIFS_COMPR_NONE;
995 else if (!strcmp(name, "lzo"))
996 c->mount_opts.compr_type = UBIFS_COMPR_LZO;
997 else if (!strcmp(name, "zlib"))
998 c->mount_opts.compr_type = UBIFS_COMPR_ZLIB;
1000 ubifs_err("unknown compressor \"%s\"", name);
1005 c->mount_opts.override_compr = 1;
1006 c->default_compr = c->mount_opts.compr_type;
1010 ubifs_err("unrecognized mount option \"%s\" "
1011 "or missing value", p);
1020 * destroy_journal - destroy journal data structures.
1021 * @c: UBIFS file-system description object
1023 * This function destroys journal data structures including those that may have
1024 * been created by recovery functions.
1026 static void destroy_journal(struct ubifs_info *c)
1028 while (!list_empty(&c->unclean_leb_list)) {
1029 struct ubifs_unclean_leb *ucleb;
1031 ucleb = list_entry(c->unclean_leb_list.next,
1032 struct ubifs_unclean_leb, list);
1033 list_del(&ucleb->list);
1036 while (!list_empty(&c->old_buds)) {
1037 struct ubifs_bud *bud;
1039 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
1040 list_del(&bud->list);
1043 ubifs_destroy_idx_gc(c);
1044 ubifs_destroy_size_tree(c);
1050 * bu_init - initialize bulk-read information.
1051 * @c: UBIFS file-system description object
1053 static void bu_init(struct ubifs_info *c)
1055 ubifs_assert(c->bulk_read == 1);
1058 return; /* Already initialized */
1061 c->bu.buf = kmalloc(c->max_bu_buf_len, GFP_KERNEL | __GFP_NOWARN);
1063 if (c->max_bu_buf_len > UBIFS_KMALLOC_OK) {
1064 c->max_bu_buf_len = UBIFS_KMALLOC_OK;
1068 /* Just disable bulk-read */
1069 ubifs_warn("Cannot allocate %d bytes of memory for bulk-read, "
1070 "disabling it", c->max_bu_buf_len);
1071 c->mount_opts.bulk_read = 1;
1078 * check_free_space - check if there is enough free space to mount.
1079 * @c: UBIFS file-system description object
1081 * This function makes sure UBIFS has enough free space to be mounted in
1082 * read/write mode. UBIFS must always have some free space to allow deletions.
1084 static int check_free_space(struct ubifs_info *c)
1086 ubifs_assert(c->dark_wm > 0);
1087 if (c->lst.total_free + c->lst.total_dirty < c->dark_wm) {
1088 ubifs_err("insufficient free space to mount in read/write mode");
1097 * mount_ubifs - mount UBIFS file-system.
1098 * @c: UBIFS file-system description object
1100 * This function mounts UBIFS file system. Returns zero in case of success and
1101 * a negative error code in case of failure.
1103 * Note, the function does not de-allocate resources it it fails half way
1104 * through, and the caller has to do this instead.
1106 static int mount_ubifs(struct ubifs_info *c)
1108 struct super_block *sb = c->vfs_sb;
1109 int err, mounted_read_only = (sb->s_flags & MS_RDONLY);
1113 err = init_constants_early(c);
1117 err = ubifs_debugging_init(c);
1121 err = check_volume_empty(c);
1125 if (c->empty && (mounted_read_only || c->ro_media)) {
1127 * This UBI volume is empty, and read-only, or the file system
1128 * is mounted read-only - we cannot format it.
1130 ubifs_err("can't format empty UBI volume: read-only %s",
1131 c->ro_media ? "UBI volume" : "mount");
1136 if (c->ro_media && !mounted_read_only) {
1137 ubifs_err("cannot mount read-write - read-only media");
1143 * The requirement for the buffer is that it should fit indexing B-tree
1144 * height amount of integers. We assume the height if the TNC tree will
1148 c->bottom_up_buf = kmalloc(BOTTOM_UP_HEIGHT * sizeof(int), GFP_KERNEL);
1149 if (!c->bottom_up_buf)
1152 c->sbuf = vmalloc(c->leb_size);
1156 if (!mounted_read_only) {
1157 c->ileb_buf = vmalloc(c->leb_size);
1162 if (c->bulk_read == 1)
1166 * We have to check all CRCs, even for data nodes, when we mount the FS
1167 * (specifically, when we are replaying).
1169 c->always_chk_crc = 1;
1171 err = ubifs_read_superblock(c);
1176 * Make sure the compressor which is set as default in the superblock
1177 * or overridden by mount options is actually compiled in.
1179 if (!ubifs_compr_present(c->default_compr)) {
1180 ubifs_err("'compressor \"%s\" is not compiled in",
1181 ubifs_compr_name(c->default_compr));
1185 err = init_constants_sb(c);
1189 sz = ALIGN(c->max_idx_node_sz, c->min_io_size);
1190 sz = ALIGN(sz + c->max_idx_node_sz, c->min_io_size);
1191 c->cbuf = kmalloc(sz, GFP_NOFS);
1197 sprintf(c->bgt_name, BGT_NAME_PATTERN, c->vi.ubi_num, c->vi.vol_id);
1198 if (!mounted_read_only) {
1199 err = alloc_wbufs(c);
1203 /* Create background thread */
1204 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1205 if (IS_ERR(c->bgt)) {
1206 err = PTR_ERR(c->bgt);
1208 ubifs_err("cannot spawn \"%s\", error %d",
1212 wake_up_process(c->bgt);
1215 err = ubifs_read_master(c);
1219 init_constants_master(c);
1221 if ((c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY)) != 0) {
1222 ubifs_msg("recovery needed");
1223 c->need_recovery = 1;
1224 if (!mounted_read_only) {
1225 err = ubifs_recover_inl_heads(c, c->sbuf);
1229 } else if (!mounted_read_only) {
1231 * Set the "dirty" flag so that if we reboot uncleanly we
1232 * will notice this immediately on the next mount.
1234 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1235 err = ubifs_write_master(c);
1240 err = ubifs_lpt_init(c, 1, !mounted_read_only);
1244 err = dbg_check_idx_size(c, c->old_idx_sz);
1248 err = ubifs_replay_journal(c);
1252 err = ubifs_mount_orphans(c, c->need_recovery, mounted_read_only);
1256 if (!mounted_read_only) {
1259 err = check_free_space(c);
1263 /* Check for enough log space */
1264 lnum = c->lhead_lnum + 1;
1265 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1266 lnum = UBIFS_LOG_LNUM;
1267 if (lnum == c->ltail_lnum) {
1268 err = ubifs_consolidate_log(c);
1273 if (c->need_recovery) {
1274 err = ubifs_recover_size(c);
1277 err = ubifs_rcvry_gc_commit(c);
1279 err = take_gc_lnum(c);
1284 * GC LEB may contain garbage if there was an unclean
1285 * reboot, and it should be un-mapped.
1287 err = ubifs_leb_unmap(c, c->gc_lnum);
1292 err = dbg_check_lprops(c);
1295 } else if (c->need_recovery) {
1296 err = ubifs_recover_size(c);
1301 * Even if we mount read-only, we have to set space in GC LEB
1302 * to proper value because this affects UBIFS free space
1303 * reporting. We do not want to have a situation when
1304 * re-mounting from R/O to R/W changes amount of free space.
1306 err = take_gc_lnum(c);
1311 spin_lock(&ubifs_infos_lock);
1312 list_add_tail(&c->infos_list, &ubifs_infos);
1313 spin_unlock(&ubifs_infos_lock);
1315 if (c->need_recovery) {
1316 if (mounted_read_only)
1317 ubifs_msg("recovery deferred");
1319 c->need_recovery = 0;
1320 ubifs_msg("recovery completed");
1322 * GC LEB has to be empty and taken at this point. But
1323 * the journal head LEBs may also be accounted as
1324 * "empty taken" if they are empty.
1326 ubifs_assert(c->lst.taken_empty_lebs > 0);
1329 ubifs_assert(c->lst.taken_empty_lebs > 0);
1331 err = dbg_check_filesystem(c);
1335 err = dbg_debugfs_init_fs(c);
1339 c->always_chk_crc = 0;
1341 ubifs_msg("mounted UBI device %d, volume %d, name \"%s\"",
1342 c->vi.ubi_num, c->vi.vol_id, c->vi.name);
1343 if (mounted_read_only)
1344 ubifs_msg("mounted read-only");
1345 x = (long long)c->main_lebs * c->leb_size;
1346 ubifs_msg("file system size: %lld bytes (%lld KiB, %lld MiB, %d "
1347 "LEBs)", x, x >> 10, x >> 20, c->main_lebs);
1348 x = (long long)c->log_lebs * c->leb_size + c->max_bud_bytes;
1349 ubifs_msg("journal size: %lld bytes (%lld KiB, %lld MiB, %d "
1350 "LEBs)", x, x >> 10, x >> 20, c->log_lebs + c->max_bud_cnt);
1351 ubifs_msg("media format: w%d/r%d (latest is w%d/r%d)",
1352 c->fmt_version, c->ro_compat_version,
1353 UBIFS_FORMAT_VERSION, UBIFS_RO_COMPAT_VERSION);
1354 ubifs_msg("default compressor: %s", ubifs_compr_name(c->default_compr));
1355 ubifs_msg("reserved for root: %llu bytes (%llu KiB)",
1356 c->report_rp_size, c->report_rp_size >> 10);
1358 dbg_msg("compiled on: " __DATE__ " at " __TIME__);
1359 dbg_msg("min. I/O unit size: %d bytes", c->min_io_size);
1360 dbg_msg("LEB size: %d bytes (%d KiB)",
1361 c->leb_size, c->leb_size >> 10);
1362 dbg_msg("data journal heads: %d",
1363 c->jhead_cnt - NONDATA_JHEADS_CNT);
1364 dbg_msg("UUID: %02X%02X%02X%02X-%02X%02X"
1365 "-%02X%02X-%02X%02X-%02X%02X%02X%02X%02X%02X",
1366 c->uuid[0], c->uuid[1], c->uuid[2], c->uuid[3],
1367 c->uuid[4], c->uuid[5], c->uuid[6], c->uuid[7],
1368 c->uuid[8], c->uuid[9], c->uuid[10], c->uuid[11],
1369 c->uuid[12], c->uuid[13], c->uuid[14], c->uuid[15]);
1370 dbg_msg("big_lpt %d", c->big_lpt);
1371 dbg_msg("log LEBs: %d (%d - %d)",
1372 c->log_lebs, UBIFS_LOG_LNUM, c->log_last);
1373 dbg_msg("LPT area LEBs: %d (%d - %d)",
1374 c->lpt_lebs, c->lpt_first, c->lpt_last);
1375 dbg_msg("orphan area LEBs: %d (%d - %d)",
1376 c->orph_lebs, c->orph_first, c->orph_last);
1377 dbg_msg("main area LEBs: %d (%d - %d)",
1378 c->main_lebs, c->main_first, c->leb_cnt - 1);
1379 dbg_msg("index LEBs: %d", c->lst.idx_lebs);
1380 dbg_msg("total index bytes: %lld (%lld KiB, %lld MiB)",
1381 c->old_idx_sz, c->old_idx_sz >> 10, c->old_idx_sz >> 20);
1382 dbg_msg("key hash type: %d", c->key_hash_type);
1383 dbg_msg("tree fanout: %d", c->fanout);
1384 dbg_msg("reserved GC LEB: %d", c->gc_lnum);
1385 dbg_msg("first main LEB: %d", c->main_first);
1386 dbg_msg("max. znode size %d", c->max_znode_sz);
1387 dbg_msg("max. index node size %d", c->max_idx_node_sz);
1388 dbg_msg("node sizes: data %zu, inode %zu, dentry %zu",
1389 UBIFS_DATA_NODE_SZ, UBIFS_INO_NODE_SZ, UBIFS_DENT_NODE_SZ);
1390 dbg_msg("node sizes: trun %zu, sb %zu, master %zu",
1391 UBIFS_TRUN_NODE_SZ, UBIFS_SB_NODE_SZ, UBIFS_MST_NODE_SZ);
1392 dbg_msg("node sizes: ref %zu, cmt. start %zu, orph %zu",
1393 UBIFS_REF_NODE_SZ, UBIFS_CS_NODE_SZ, UBIFS_ORPH_NODE_SZ);
1394 dbg_msg("max. node sizes: data %zu, inode %zu dentry %zu",
1395 UBIFS_MAX_DATA_NODE_SZ, UBIFS_MAX_INO_NODE_SZ,
1396 UBIFS_MAX_DENT_NODE_SZ);
1397 dbg_msg("dead watermark: %d", c->dead_wm);
1398 dbg_msg("dark watermark: %d", c->dark_wm);
1399 dbg_msg("LEB overhead: %d", c->leb_overhead);
1400 x = (long long)c->main_lebs * c->dark_wm;
1401 dbg_msg("max. dark space: %lld (%lld KiB, %lld MiB)",
1402 x, x >> 10, x >> 20);
1403 dbg_msg("maximum bud bytes: %lld (%lld KiB, %lld MiB)",
1404 c->max_bud_bytes, c->max_bud_bytes >> 10,
1405 c->max_bud_bytes >> 20);
1406 dbg_msg("BG commit bud bytes: %lld (%lld KiB, %lld MiB)",
1407 c->bg_bud_bytes, c->bg_bud_bytes >> 10,
1408 c->bg_bud_bytes >> 20);
1409 dbg_msg("current bud bytes %lld (%lld KiB, %lld MiB)",
1410 c->bud_bytes, c->bud_bytes >> 10, c->bud_bytes >> 20);
1411 dbg_msg("max. seq. number: %llu", c->max_sqnum);
1412 dbg_msg("commit number: %llu", c->cmt_no);
1417 spin_lock(&ubifs_infos_lock);
1418 list_del(&c->infos_list);
1419 spin_unlock(&ubifs_infos_lock);
1425 ubifs_lpt_free(c, 0);
1428 kfree(c->rcvrd_mst_node);
1430 kthread_stop(c->bgt);
1439 kfree(c->bottom_up_buf);
1440 ubifs_debugging_exit(c);
1445 * ubifs_umount - un-mount UBIFS file-system.
1446 * @c: UBIFS file-system description object
1448 * Note, this function is called to free allocated resourced when un-mounting,
1449 * as well as free resources when an error occurred while we were half way
1450 * through mounting (error path cleanup function). So it has to make sure the
1451 * resource was actually allocated before freeing it.
1453 static void ubifs_umount(struct ubifs_info *c)
1455 dbg_gen("un-mounting UBI device %d, volume %d", c->vi.ubi_num,
1458 dbg_debugfs_exit_fs(c);
1459 spin_lock(&ubifs_infos_lock);
1460 list_del(&c->infos_list);
1461 spin_unlock(&ubifs_infos_lock);
1464 kthread_stop(c->bgt);
1469 ubifs_lpt_free(c, 0);
1472 kfree(c->rcvrd_mst_node);
1477 kfree(c->bottom_up_buf);
1478 ubifs_debugging_exit(c);
1482 * ubifs_remount_rw - re-mount in read-write mode.
1483 * @c: UBIFS file-system description object
1485 * UBIFS avoids allocating many unnecessary resources when mounted in read-only
1486 * mode. This function allocates the needed resources and re-mounts UBIFS in
1489 static int ubifs_remount_rw(struct ubifs_info *c)
1493 if (c->rw_incompat) {
1494 ubifs_err("the file-system is not R/W-compatible");
1495 ubifs_msg("on-flash format version is w%d/r%d, but software "
1496 "only supports up to version w%d/r%d", c->fmt_version,
1497 c->ro_compat_version, UBIFS_FORMAT_VERSION,
1498 UBIFS_RO_COMPAT_VERSION);
1502 mutex_lock(&c->umount_mutex);
1503 dbg_save_space_info(c);
1504 c->remounting_rw = 1;
1505 c->always_chk_crc = 1;
1507 err = check_free_space(c);
1511 if (c->old_leb_cnt != c->leb_cnt) {
1512 struct ubifs_sb_node *sup;
1514 sup = ubifs_read_sb_node(c);
1519 sup->leb_cnt = cpu_to_le32(c->leb_cnt);
1520 err = ubifs_write_sb_node(c, sup);
1525 if (c->need_recovery) {
1526 ubifs_msg("completing deferred recovery");
1527 err = ubifs_write_rcvrd_mst_node(c);
1530 err = ubifs_recover_size(c);
1533 err = ubifs_clean_lebs(c, c->sbuf);
1536 err = ubifs_recover_inl_heads(c, c->sbuf);
1540 /* A readonly mount is not allowed to have orphans */
1541 ubifs_assert(c->tot_orphans == 0);
1542 err = ubifs_clear_orphans(c);
1547 if (!(c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY))) {
1548 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1549 err = ubifs_write_master(c);
1554 c->ileb_buf = vmalloc(c->leb_size);
1560 err = ubifs_lpt_init(c, 0, 1);
1564 err = alloc_wbufs(c);
1568 ubifs_create_buds_lists(c);
1570 /* Create background thread */
1571 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1572 if (IS_ERR(c->bgt)) {
1573 err = PTR_ERR(c->bgt);
1575 ubifs_err("cannot spawn \"%s\", error %d",
1579 wake_up_process(c->bgt);
1581 c->orph_buf = vmalloc(c->leb_size);
1587 /* Check for enough log space */
1588 lnum = c->lhead_lnum + 1;
1589 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1590 lnum = UBIFS_LOG_LNUM;
1591 if (lnum == c->ltail_lnum) {
1592 err = ubifs_consolidate_log(c);
1597 if (c->need_recovery)
1598 err = ubifs_rcvry_gc_commit(c);
1600 err = ubifs_leb_unmap(c, c->gc_lnum);
1604 if (c->need_recovery) {
1605 c->need_recovery = 0;
1606 ubifs_msg("deferred recovery completed");
1609 dbg_gen("re-mounted read-write");
1610 c->vfs_sb->s_flags &= ~MS_RDONLY;
1611 c->remounting_rw = 0;
1612 c->always_chk_crc = 0;
1613 err = dbg_check_space_info(c);
1614 mutex_unlock(&c->umount_mutex);
1621 kthread_stop(c->bgt);
1627 ubifs_lpt_free(c, 1);
1628 c->remounting_rw = 0;
1629 c->always_chk_crc = 0;
1630 mutex_unlock(&c->umount_mutex);
1635 * ubifs_remount_ro - re-mount in read-only mode.
1636 * @c: UBIFS file-system description object
1638 * We assume VFS has stopped writing. Possibly the background thread could be
1639 * running a commit, however kthread_stop will wait in that case.
1641 static void ubifs_remount_ro(struct ubifs_info *c)
1645 ubifs_assert(!c->need_recovery);
1646 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY));
1648 mutex_lock(&c->umount_mutex);
1650 kthread_stop(c->bgt);
1654 dbg_save_space_info(c);
1656 for (i = 0; i < c->jhead_cnt; i++) {
1657 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1658 del_timer_sync(&c->jheads[i].wbuf.timer);
1661 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1662 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1663 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1664 err = ubifs_write_master(c);
1666 ubifs_ro_mode(c, err);
1673 ubifs_lpt_free(c, 1);
1674 err = dbg_check_space_info(c);
1676 ubifs_ro_mode(c, err);
1677 mutex_unlock(&c->umount_mutex);
1680 static void ubifs_put_super(struct super_block *sb)
1683 struct ubifs_info *c = sb->s_fs_info;
1685 ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
1688 * The following asserts are only valid if there has not been a failure
1689 * of the media. For example, there will be dirty inodes if we failed
1690 * to write them back because of I/O errors.
1692 ubifs_assert(atomic_long_read(&c->dirty_pg_cnt) == 0);
1693 ubifs_assert(c->budg_idx_growth == 0);
1694 ubifs_assert(c->budg_dd_growth == 0);
1695 ubifs_assert(c->budg_data_growth == 0);
1698 * The 'c->umount_lock' prevents races between UBIFS memory shrinker
1699 * and file system un-mount. Namely, it prevents the shrinker from
1700 * picking this superblock for shrinking - it will be just skipped if
1701 * the mutex is locked.
1703 mutex_lock(&c->umount_mutex);
1704 if (!(c->vfs_sb->s_flags & MS_RDONLY)) {
1706 * First of all kill the background thread to make sure it does
1707 * not interfere with un-mounting and freeing resources.
1710 kthread_stop(c->bgt);
1714 /* Synchronize write-buffers */
1716 for (i = 0; i < c->jhead_cnt; i++) {
1717 ubifs_wbuf_sync(&c->jheads[i].wbuf);
1718 del_timer_sync(&c->jheads[i].wbuf.timer);
1722 * On fatal errors c->ro_media is set to 1, in which case we do
1723 * not write the master node.
1727 * We are being cleanly unmounted which means the
1728 * orphans were killed - indicate this in the master
1729 * node. Also save the reserved GC LEB number.
1733 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1734 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1735 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1736 err = ubifs_write_master(c);
1739 * Recovery will attempt to fix the master area
1740 * next mount, so we just print a message and
1741 * continue to unmount normally.
1743 ubifs_err("failed to write master node, "
1749 bdi_destroy(&c->bdi);
1750 ubi_close_volume(c->ubi);
1751 mutex_unlock(&c->umount_mutex);
1755 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
1758 struct ubifs_info *c = sb->s_fs_info;
1760 dbg_gen("old flags %#lx, new flags %#x", sb->s_flags, *flags);
1762 err = ubifs_parse_options(c, data, 1);
1764 ubifs_err("invalid or unknown remount parameter");
1768 if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
1770 ubifs_msg("cannot re-mount due to prior errors");
1773 err = ubifs_remount_rw(c);
1776 } else if (!(sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY)) {
1778 ubifs_msg("cannot re-mount due to prior errors");
1781 ubifs_remount_ro(c);
1784 if (c->bulk_read == 1)
1787 dbg_gen("disable bulk-read");
1792 ubifs_assert(c->lst.taken_empty_lebs > 0);
1796 const struct super_operations ubifs_super_operations = {
1797 .alloc_inode = ubifs_alloc_inode,
1798 .destroy_inode = ubifs_destroy_inode,
1799 .put_super = ubifs_put_super,
1800 .write_inode = ubifs_write_inode,
1801 .delete_inode = ubifs_delete_inode,
1802 .statfs = ubifs_statfs,
1803 .dirty_inode = ubifs_dirty_inode,
1804 .remount_fs = ubifs_remount_fs,
1805 .show_options = ubifs_show_options,
1806 .sync_fs = ubifs_sync_fs,
1810 * open_ubi - parse UBI device name string and open the UBI device.
1811 * @name: UBI volume name
1812 * @mode: UBI volume open mode
1814 * There are several ways to specify UBI volumes when mounting UBIFS:
1815 * o ubiX_Y - UBI device number X, volume Y;
1816 * o ubiY - UBI device number 0, volume Y;
1817 * o ubiX:NAME - mount UBI device X, volume with name NAME;
1818 * o ubi:NAME - mount UBI device 0, volume with name NAME.
1820 * Alternative '!' separator may be used instead of ':' (because some shells
1821 * like busybox may interpret ':' as an NFS host name separator). This function
1822 * returns ubi volume object in case of success and a negative error code in
1825 static struct ubi_volume_desc *open_ubi(const char *name, int mode)
1830 if (name[0] != 'u' || name[1] != 'b' || name[2] != 'i')
1831 return ERR_PTR(-EINVAL);
1833 /* ubi:NAME method */
1834 if ((name[3] == ':' || name[3] == '!') && name[4] != '\0')
1835 return ubi_open_volume_nm(0, name + 4, mode);
1837 if (!isdigit(name[3]))
1838 return ERR_PTR(-EINVAL);
1840 dev = simple_strtoul(name + 3, &endptr, 0);
1843 if (*endptr == '\0')
1844 return ubi_open_volume(0, dev, mode);
1847 if (*endptr == '_' && isdigit(endptr[1])) {
1848 vol = simple_strtoul(endptr + 1, &endptr, 0);
1849 if (*endptr != '\0')
1850 return ERR_PTR(-EINVAL);
1851 return ubi_open_volume(dev, vol, mode);
1854 /* ubiX:NAME method */
1855 if ((*endptr == ':' || *endptr == '!') && endptr[1] != '\0')
1856 return ubi_open_volume_nm(dev, ++endptr, mode);
1858 return ERR_PTR(-EINVAL);
1861 static int ubifs_fill_super(struct super_block *sb, void *data, int silent)
1863 struct ubi_volume_desc *ubi = sb->s_fs_info;
1864 struct ubifs_info *c;
1868 c = kzalloc(sizeof(struct ubifs_info), GFP_KERNEL);
1872 spin_lock_init(&c->cnt_lock);
1873 spin_lock_init(&c->cs_lock);
1874 spin_lock_init(&c->buds_lock);
1875 spin_lock_init(&c->space_lock);
1876 spin_lock_init(&c->orphan_lock);
1877 init_rwsem(&c->commit_sem);
1878 mutex_init(&c->lp_mutex);
1879 mutex_init(&c->tnc_mutex);
1880 mutex_init(&c->log_mutex);
1881 mutex_init(&c->mst_mutex);
1882 mutex_init(&c->umount_mutex);
1883 mutex_init(&c->bu_mutex);
1884 init_waitqueue_head(&c->cmt_wq);
1886 c->old_idx = RB_ROOT;
1887 c->size_tree = RB_ROOT;
1888 c->orph_tree = RB_ROOT;
1889 INIT_LIST_HEAD(&c->infos_list);
1890 INIT_LIST_HEAD(&c->idx_gc);
1891 INIT_LIST_HEAD(&c->replay_list);
1892 INIT_LIST_HEAD(&c->replay_buds);
1893 INIT_LIST_HEAD(&c->uncat_list);
1894 INIT_LIST_HEAD(&c->empty_list);
1895 INIT_LIST_HEAD(&c->freeable_list);
1896 INIT_LIST_HEAD(&c->frdi_idx_list);
1897 INIT_LIST_HEAD(&c->unclean_leb_list);
1898 INIT_LIST_HEAD(&c->old_buds);
1899 INIT_LIST_HEAD(&c->orph_list);
1900 INIT_LIST_HEAD(&c->orph_new);
1902 c->highest_inum = UBIFS_FIRST_INO;
1903 c->lhead_lnum = c->ltail_lnum = UBIFS_LOG_LNUM;
1905 ubi_get_volume_info(ubi, &c->vi);
1906 ubi_get_device_info(c->vi.ubi_num, &c->di);
1908 /* Re-open the UBI device in read-write mode */
1909 c->ubi = ubi_open_volume(c->vi.ubi_num, c->vi.vol_id, UBI_READWRITE);
1910 if (IS_ERR(c->ubi)) {
1911 err = PTR_ERR(c->ubi);
1916 * UBIFS provides 'backing_dev_info' in order to disable read-ahead. For
1917 * UBIFS, I/O is not deferred, it is done immediately in readpage,
1918 * which means the user would have to wait not just for their own I/O
1919 * but the read-ahead I/O as well i.e. completely pointless.
1921 * Read-ahead will be disabled because @c->bdi.ra_pages is 0.
1923 c->bdi.capabilities = BDI_CAP_MAP_COPY;
1924 c->bdi.unplug_io_fn = default_unplug_io_fn;
1925 err = bdi_init(&c->bdi);
1929 err = ubifs_parse_options(c, data, 0);
1936 sb->s_magic = UBIFS_SUPER_MAGIC;
1937 sb->s_blocksize = UBIFS_BLOCK_SIZE;
1938 sb->s_blocksize_bits = UBIFS_BLOCK_SHIFT;
1939 sb->s_dev = c->vi.cdev;
1940 sb->s_maxbytes = c->max_inode_sz = key_max_inode_size(c);
1941 if (c->max_inode_sz > MAX_LFS_FILESIZE)
1942 sb->s_maxbytes = c->max_inode_sz = MAX_LFS_FILESIZE;
1943 sb->s_op = &ubifs_super_operations;
1945 mutex_lock(&c->umount_mutex);
1946 err = mount_ubifs(c);
1948 ubifs_assert(err < 0);
1952 /* Read the root inode */
1953 root = ubifs_iget(sb, UBIFS_ROOT_INO);
1955 err = PTR_ERR(root);
1959 sb->s_root = d_alloc_root(root);
1963 mutex_unlock(&c->umount_mutex);
1971 mutex_unlock(&c->umount_mutex);
1973 bdi_destroy(&c->bdi);
1975 ubi_close_volume(c->ubi);
1981 static int sb_test(struct super_block *sb, void *data)
1985 return sb->s_dev == *dev;
1988 static int sb_set(struct super_block *sb, void *data)
1996 static int ubifs_get_sb(struct file_system_type *fs_type, int flags,
1997 const char *name, void *data, struct vfsmount *mnt)
1999 struct ubi_volume_desc *ubi;
2000 struct ubi_volume_info vi;
2001 struct super_block *sb;
2004 dbg_gen("name %s, flags %#x", name, flags);
2007 * Get UBI device number and volume ID. Mount it read-only so far
2008 * because this might be a new mount point, and UBI allows only one
2009 * read-write user at a time.
2011 ubi = open_ubi(name, UBI_READONLY);
2013 ubifs_err("cannot open \"%s\", error %d",
2014 name, (int)PTR_ERR(ubi));
2015 return PTR_ERR(ubi);
2017 ubi_get_volume_info(ubi, &vi);
2019 dbg_gen("opened ubi%d_%d", vi.ubi_num, vi.vol_id);
2021 sb = sget(fs_type, &sb_test, &sb_set, &vi.cdev);
2028 /* A new mount point for already mounted UBIFS */
2029 dbg_gen("this ubi volume is already mounted");
2030 if ((flags ^ sb->s_flags) & MS_RDONLY) {
2035 sb->s_flags = flags;
2037 * Pass 'ubi' to 'fill_super()' in sb->s_fs_info where it is
2040 sb->s_fs_info = ubi;
2041 err = ubifs_fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
2044 /* We do not support atime */
2045 sb->s_flags |= MS_ACTIVE | MS_NOATIME;
2048 /* 'fill_super()' opens ubi again so we must close it here */
2049 ubi_close_volume(ubi);
2051 simple_set_mnt(mnt, sb);
2055 deactivate_locked_super(sb);
2057 ubi_close_volume(ubi);
2061 static void ubifs_kill_sb(struct super_block *sb)
2063 generic_shutdown_super(sb);
2066 static struct file_system_type ubifs_fs_type = {
2068 .owner = THIS_MODULE,
2069 .get_sb = ubifs_get_sb,
2070 .kill_sb = ubifs_kill_sb
2074 * Inode slab cache constructor.
2076 static void inode_slab_ctor(void *obj)
2078 struct ubifs_inode *ui = obj;
2079 inode_init_once(&ui->vfs_inode);
2082 static int __init ubifs_init(void)
2086 BUILD_BUG_ON(sizeof(struct ubifs_ch) != 24);
2088 /* Make sure node sizes are 8-byte aligned */
2089 BUILD_BUG_ON(UBIFS_CH_SZ & 7);
2090 BUILD_BUG_ON(UBIFS_INO_NODE_SZ & 7);
2091 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ & 7);
2092 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ & 7);
2093 BUILD_BUG_ON(UBIFS_DATA_NODE_SZ & 7);
2094 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ & 7);
2095 BUILD_BUG_ON(UBIFS_SB_NODE_SZ & 7);
2096 BUILD_BUG_ON(UBIFS_MST_NODE_SZ & 7);
2097 BUILD_BUG_ON(UBIFS_REF_NODE_SZ & 7);
2098 BUILD_BUG_ON(UBIFS_CS_NODE_SZ & 7);
2099 BUILD_BUG_ON(UBIFS_ORPH_NODE_SZ & 7);
2101 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ & 7);
2102 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ & 7);
2103 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ & 7);
2104 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ & 7);
2105 BUILD_BUG_ON(UBIFS_MAX_NODE_SZ & 7);
2106 BUILD_BUG_ON(MIN_WRITE_SZ & 7);
2108 /* Check min. node size */
2109 BUILD_BUG_ON(UBIFS_INO_NODE_SZ < MIN_WRITE_SZ);
2110 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ < MIN_WRITE_SZ);
2111 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ < MIN_WRITE_SZ);
2112 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ < MIN_WRITE_SZ);
2114 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2115 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2116 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ > UBIFS_MAX_NODE_SZ);
2117 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ > UBIFS_MAX_NODE_SZ);
2119 /* Defined node sizes */
2120 BUILD_BUG_ON(UBIFS_SB_NODE_SZ != 4096);
2121 BUILD_BUG_ON(UBIFS_MST_NODE_SZ != 512);
2122 BUILD_BUG_ON(UBIFS_INO_NODE_SZ != 160);
2123 BUILD_BUG_ON(UBIFS_REF_NODE_SZ != 64);
2126 * We use 2 bit wide bit-fields to store compression type, which should
2127 * be amended if more compressors are added. The bit-fields are:
2128 * @compr_type in 'struct ubifs_inode', @default_compr in
2129 * 'struct ubifs_info' and @compr_type in 'struct ubifs_mount_opts'.
2131 BUILD_BUG_ON(UBIFS_COMPR_TYPES_CNT > 4);
2134 * We require that PAGE_CACHE_SIZE is greater-than-or-equal-to
2135 * UBIFS_BLOCK_SIZE. It is assumed that both are powers of 2.
2137 if (PAGE_CACHE_SIZE < UBIFS_BLOCK_SIZE) {
2138 ubifs_err("VFS page cache size is %u bytes, but UBIFS requires"
2139 " at least 4096 bytes",
2140 (unsigned int)PAGE_CACHE_SIZE);
2144 err = register_filesystem(&ubifs_fs_type);
2146 ubifs_err("cannot register file system, error %d", err);
2151 ubifs_inode_slab = kmem_cache_create("ubifs_inode_slab",
2152 sizeof(struct ubifs_inode), 0,
2153 SLAB_MEM_SPREAD | SLAB_RECLAIM_ACCOUNT,
2155 if (!ubifs_inode_slab)
2158 register_shrinker(&ubifs_shrinker_info);
2160 err = ubifs_compressors_init();
2164 err = dbg_debugfs_init();
2171 ubifs_compressors_exit();
2173 unregister_shrinker(&ubifs_shrinker_info);
2174 kmem_cache_destroy(ubifs_inode_slab);
2176 unregister_filesystem(&ubifs_fs_type);
2179 /* late_initcall to let compressors initialize first */
2180 late_initcall(ubifs_init);
2182 static void __exit ubifs_exit(void)
2184 ubifs_assert(list_empty(&ubifs_infos));
2185 ubifs_assert(atomic_long_read(&ubifs_clean_zn_cnt) == 0);
2188 ubifs_compressors_exit();
2189 unregister_shrinker(&ubifs_shrinker_info);
2190 kmem_cache_destroy(ubifs_inode_slab);
2191 unregister_filesystem(&ubifs_fs_type);
2193 module_exit(ubifs_exit);
2195 MODULE_LICENSE("GPL");
2196 MODULE_VERSION(__stringify(UBIFS_VERSION));
2197 MODULE_AUTHOR("Artem Bityutskiy, Adrian Hunter");
2198 MODULE_DESCRIPTION("UBIFS - UBI File System");