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 static struct kmem_cache *ubifs_inode_slab;
50 /* UBIFS TNC shrinker description */
51 static struct shrinker ubifs_shrinker_info = {
52 .scan_objects = ubifs_shrink_scan,
53 .count_objects = ubifs_shrink_count,
54 .seeks = DEFAULT_SEEKS,
58 * validate_inode - validate inode.
59 * @c: UBIFS file-system description object
60 * @inode: the inode to validate
62 * This is a helper function for 'ubifs_iget()' which validates various fields
63 * of a newly built inode to make sure they contain sane values and prevent
64 * possible vulnerabilities. Returns zero if the inode is all right and
65 * a non-zero error code if not.
67 static int validate_inode(struct ubifs_info *c, const struct inode *inode)
70 const struct ubifs_inode *ui = ubifs_inode(inode);
72 if (inode->i_size > c->max_inode_sz) {
73 ubifs_err(c, "inode is too large (%lld)",
74 (long long)inode->i_size);
78 if (ui->compr_type >= UBIFS_COMPR_TYPES_CNT) {
79 ubifs_err(c, "unknown compression type %d", ui->compr_type);
83 if (ui->xattr_names + ui->xattr_cnt > XATTR_LIST_MAX)
86 if (ui->data_len < 0 || ui->data_len > UBIFS_MAX_INO_DATA)
89 if (ui->xattr && !S_ISREG(inode->i_mode))
92 if (!ubifs_compr_present(c, ui->compr_type)) {
93 ubifs_warn(c, "inode %lu uses '%s' compression, but it was not compiled in",
94 inode->i_ino, ubifs_compr_name(c, ui->compr_type));
97 err = dbg_check_dir(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;
132 #ifndef CONFIG_UBIFS_ATIME_SUPPORT
133 inode->i_flags |= S_NOATIME;
135 set_nlink(inode, le32_to_cpu(ino->nlink));
136 i_uid_write(inode, le32_to_cpu(ino->uid));
137 i_gid_write(inode, le32_to_cpu(ino->gid));
138 inode->i_atime.tv_sec = (int64_t)le64_to_cpu(ino->atime_sec);
139 inode->i_atime.tv_nsec = le32_to_cpu(ino->atime_nsec);
140 inode->i_mtime.tv_sec = (int64_t)le64_to_cpu(ino->mtime_sec);
141 inode->i_mtime.tv_nsec = le32_to_cpu(ino->mtime_nsec);
142 inode->i_ctime.tv_sec = (int64_t)le64_to_cpu(ino->ctime_sec);
143 inode->i_ctime.tv_nsec = le32_to_cpu(ino->ctime_nsec);
144 inode->i_mode = le32_to_cpu(ino->mode);
145 inode->i_size = le64_to_cpu(ino->size);
147 ui->data_len = le32_to_cpu(ino->data_len);
148 ui->flags = le32_to_cpu(ino->flags);
149 ui->compr_type = le16_to_cpu(ino->compr_type);
150 ui->creat_sqnum = le64_to_cpu(ino->creat_sqnum);
151 ui->xattr_cnt = le32_to_cpu(ino->xattr_cnt);
152 ui->xattr_size = le32_to_cpu(ino->xattr_size);
153 ui->xattr_names = le32_to_cpu(ino->xattr_names);
154 ui->synced_i_size = ui->ui_size = inode->i_size;
156 ui->xattr = (ui->flags & UBIFS_XATTR_FL) ? 1 : 0;
158 err = validate_inode(c, inode);
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(c, "inode %lu validation failed, error %d", inode->i_ino, err);
249 ubifs_dump_node(c, ino);
250 ubifs_dump_inode(c, inode);
255 ubifs_err(c, "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_i_callback(struct rcu_head *head)
277 struct inode *inode = container_of(head, struct inode, i_rcu);
278 struct ubifs_inode *ui = ubifs_inode(inode);
280 kmem_cache_free(ubifs_inode_slab, ui);
283 static void ubifs_destroy_inode(struct inode *inode)
285 call_rcu(&inode->i_rcu, ubifs_i_callback);
289 * Note, Linux write-back code calls this without 'i_mutex'.
291 static int ubifs_write_inode(struct inode *inode, struct writeback_control *wbc)
294 struct ubifs_info *c = inode->i_sb->s_fs_info;
295 struct ubifs_inode *ui = ubifs_inode(inode);
297 ubifs_assert(c, !ui->xattr);
298 if (is_bad_inode(inode))
301 mutex_lock(&ui->ui_mutex);
303 * Due to races between write-back forced by budgeting
304 * (see 'sync_some_inodes()') and background write-back, the inode may
305 * have already been synchronized, do not do this again. This might
306 * also happen if it was synchronized in an VFS operation, e.g.
310 mutex_unlock(&ui->ui_mutex);
315 * As an optimization, do not write orphan inodes to the media just
316 * because this is not needed.
318 dbg_gen("inode %lu, mode %#x, nlink %u",
319 inode->i_ino, (int)inode->i_mode, inode->i_nlink);
320 if (inode->i_nlink) {
321 err = ubifs_jnl_write_inode(c, inode);
323 ubifs_err(c, "can't write inode %lu, error %d",
326 err = dbg_check_inode_size(c, inode, ui->ui_size);
330 mutex_unlock(&ui->ui_mutex);
331 ubifs_release_dirty_inode_budget(c, ui);
335 static void ubifs_evict_inode(struct inode *inode)
338 struct ubifs_info *c = inode->i_sb->s_fs_info;
339 struct ubifs_inode *ui = ubifs_inode(inode);
343 * Extended attribute inode deletions are fully handled in
344 * 'ubifs_removexattr()'. These inodes are special and have
345 * limited usage, so there is nothing to do here.
349 dbg_gen("inode %lu, mode %#x", inode->i_ino, (int)inode->i_mode);
350 ubifs_assert(c, !atomic_read(&inode->i_count));
352 truncate_inode_pages_final(&inode->i_data);
357 if (is_bad_inode(inode))
360 ui->ui_size = inode->i_size = 0;
361 err = ubifs_jnl_delete_inode(c, inode);
364 * Worst case we have a lost orphan inode wasting space, so a
365 * simple error message is OK here.
367 ubifs_err(c, "can't delete inode %lu, error %d",
372 ubifs_release_dirty_inode_budget(c, ui);
374 /* We've deleted something - clean the "no space" flags */
375 c->bi.nospace = c->bi.nospace_rp = 0;
380 fscrypt_put_encryption_info(inode);
383 static void ubifs_dirty_inode(struct inode *inode, int flags)
385 struct ubifs_info *c = inode->i_sb->s_fs_info;
386 struct ubifs_inode *ui = ubifs_inode(inode);
388 ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
391 dbg_gen("inode %lu", inode->i_ino);
395 static int ubifs_statfs(struct dentry *dentry, struct kstatfs *buf)
397 struct ubifs_info *c = dentry->d_sb->s_fs_info;
398 unsigned long long free;
399 __le32 *uuid = (__le32 *)c->uuid;
401 free = ubifs_get_free_space(c);
402 dbg_gen("free space %lld bytes (%lld blocks)",
403 free, free >> UBIFS_BLOCK_SHIFT);
405 buf->f_type = UBIFS_SUPER_MAGIC;
406 buf->f_bsize = UBIFS_BLOCK_SIZE;
407 buf->f_blocks = c->block_cnt;
408 buf->f_bfree = free >> UBIFS_BLOCK_SHIFT;
409 if (free > c->report_rp_size)
410 buf->f_bavail = (free - c->report_rp_size) >> UBIFS_BLOCK_SHIFT;
415 buf->f_namelen = UBIFS_MAX_NLEN;
416 buf->f_fsid.val[0] = le32_to_cpu(uuid[0]) ^ le32_to_cpu(uuid[2]);
417 buf->f_fsid.val[1] = le32_to_cpu(uuid[1]) ^ le32_to_cpu(uuid[3]);
418 ubifs_assert(c, buf->f_bfree <= c->block_cnt);
422 static int ubifs_show_options(struct seq_file *s, struct dentry *root)
424 struct ubifs_info *c = root->d_sb->s_fs_info;
426 if (c->mount_opts.unmount_mode == 2)
427 seq_puts(s, ",fast_unmount");
428 else if (c->mount_opts.unmount_mode == 1)
429 seq_puts(s, ",norm_unmount");
431 if (c->mount_opts.bulk_read == 2)
432 seq_puts(s, ",bulk_read");
433 else if (c->mount_opts.bulk_read == 1)
434 seq_puts(s, ",no_bulk_read");
436 if (c->mount_opts.chk_data_crc == 2)
437 seq_puts(s, ",chk_data_crc");
438 else if (c->mount_opts.chk_data_crc == 1)
439 seq_puts(s, ",no_chk_data_crc");
441 if (c->mount_opts.override_compr) {
442 seq_printf(s, ",compr=%s",
443 ubifs_compr_name(c, c->mount_opts.compr_type));
446 seq_printf(s, ",assert=%s", ubifs_assert_action_name(c));
447 seq_printf(s, ",ubi=%d,vol=%d", c->vi.ubi_num, c->vi.vol_id);
452 static int ubifs_sync_fs(struct super_block *sb, int wait)
455 struct ubifs_info *c = sb->s_fs_info;
458 * Zero @wait is just an advisory thing to help the file system shove
459 * lots of data into the queues, and there will be the second
460 * '->sync_fs()' call, with non-zero @wait.
466 * Synchronize write buffers, because 'ubifs_run_commit()' does not
467 * do this if it waits for an already running commit.
469 for (i = 0; i < c->jhead_cnt; i++) {
470 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
476 * Strictly speaking, it is not necessary to commit the journal here,
477 * synchronizing write-buffers would be enough. But committing makes
478 * UBIFS free space predictions much more accurate, so we want to let
479 * the user be able to get more accurate results of 'statfs()' after
480 * they synchronize the file system.
482 err = ubifs_run_commit(c);
486 return ubi_sync(c->vi.ubi_num);
490 * init_constants_early - initialize UBIFS constants.
491 * @c: UBIFS file-system description object
493 * This function initialize UBIFS constants which do not need the superblock to
494 * be read. It also checks that the UBI volume satisfies basic UBIFS
495 * requirements. Returns zero in case of success and a negative error code in
498 static int init_constants_early(struct ubifs_info *c)
500 if (c->vi.corrupted) {
501 ubifs_warn(c, "UBI volume is corrupted - read-only mode");
506 ubifs_msg(c, "read-only UBI device");
510 if (c->vi.vol_type == UBI_STATIC_VOLUME) {
511 ubifs_msg(c, "static UBI volume - read-only mode");
515 c->leb_cnt = c->vi.size;
516 c->leb_size = c->vi.usable_leb_size;
517 c->leb_start = c->di.leb_start;
518 c->half_leb_size = c->leb_size / 2;
519 c->min_io_size = c->di.min_io_size;
520 c->min_io_shift = fls(c->min_io_size) - 1;
521 c->max_write_size = c->di.max_write_size;
522 c->max_write_shift = fls(c->max_write_size) - 1;
524 if (c->leb_size < UBIFS_MIN_LEB_SZ) {
525 ubifs_errc(c, "too small LEBs (%d bytes), min. is %d bytes",
526 c->leb_size, UBIFS_MIN_LEB_SZ);
530 if (c->leb_cnt < UBIFS_MIN_LEB_CNT) {
531 ubifs_errc(c, "too few LEBs (%d), min. is %d",
532 c->leb_cnt, UBIFS_MIN_LEB_CNT);
536 if (!is_power_of_2(c->min_io_size)) {
537 ubifs_errc(c, "bad min. I/O size %d", c->min_io_size);
542 * Maximum write size has to be greater or equivalent to min. I/O
543 * size, and be multiple of min. I/O size.
545 if (c->max_write_size < c->min_io_size ||
546 c->max_write_size % c->min_io_size ||
547 !is_power_of_2(c->max_write_size)) {
548 ubifs_errc(c, "bad write buffer size %d for %d min. I/O unit",
549 c->max_write_size, c->min_io_size);
554 * UBIFS aligns all node to 8-byte boundary, so to make function in
555 * io.c simpler, assume minimum I/O unit size to be 8 bytes if it is
558 if (c->min_io_size < 8) {
561 if (c->max_write_size < c->min_io_size) {
562 c->max_write_size = c->min_io_size;
563 c->max_write_shift = c->min_io_shift;
567 c->ref_node_alsz = ALIGN(UBIFS_REF_NODE_SZ, c->min_io_size);
568 c->mst_node_alsz = ALIGN(UBIFS_MST_NODE_SZ, c->min_io_size);
571 * Initialize node length ranges which are mostly needed for node
574 c->ranges[UBIFS_PAD_NODE].len = UBIFS_PAD_NODE_SZ;
575 c->ranges[UBIFS_SB_NODE].len = UBIFS_SB_NODE_SZ;
576 c->ranges[UBIFS_MST_NODE].len = UBIFS_MST_NODE_SZ;
577 c->ranges[UBIFS_REF_NODE].len = UBIFS_REF_NODE_SZ;
578 c->ranges[UBIFS_TRUN_NODE].len = UBIFS_TRUN_NODE_SZ;
579 c->ranges[UBIFS_CS_NODE].len = UBIFS_CS_NODE_SZ;
580 c->ranges[UBIFS_AUTH_NODE].min_len = UBIFS_AUTH_NODE_SZ;
581 c->ranges[UBIFS_AUTH_NODE].max_len = UBIFS_AUTH_NODE_SZ +
584 c->ranges[UBIFS_INO_NODE].min_len = UBIFS_INO_NODE_SZ;
585 c->ranges[UBIFS_INO_NODE].max_len = UBIFS_MAX_INO_NODE_SZ;
586 c->ranges[UBIFS_ORPH_NODE].min_len =
587 UBIFS_ORPH_NODE_SZ + sizeof(__le64);
588 c->ranges[UBIFS_ORPH_NODE].max_len = c->leb_size;
589 c->ranges[UBIFS_DENT_NODE].min_len = UBIFS_DENT_NODE_SZ;
590 c->ranges[UBIFS_DENT_NODE].max_len = UBIFS_MAX_DENT_NODE_SZ;
591 c->ranges[UBIFS_XENT_NODE].min_len = UBIFS_XENT_NODE_SZ;
592 c->ranges[UBIFS_XENT_NODE].max_len = UBIFS_MAX_XENT_NODE_SZ;
593 c->ranges[UBIFS_DATA_NODE].min_len = UBIFS_DATA_NODE_SZ;
594 c->ranges[UBIFS_DATA_NODE].max_len = UBIFS_MAX_DATA_NODE_SZ;
596 * Minimum indexing node size is amended later when superblock is
597 * read and the key length is known.
599 c->ranges[UBIFS_IDX_NODE].min_len = UBIFS_IDX_NODE_SZ + UBIFS_BRANCH_SZ;
601 * Maximum indexing node size is amended later when superblock is
602 * read and the fanout is known.
604 c->ranges[UBIFS_IDX_NODE].max_len = INT_MAX;
607 * Initialize dead and dark LEB space watermarks. See gc.c for comments
608 * about these values.
610 c->dead_wm = ALIGN(MIN_WRITE_SZ, c->min_io_size);
611 c->dark_wm = ALIGN(UBIFS_MAX_NODE_SZ, c->min_io_size);
614 * Calculate how many bytes would be wasted at the end of LEB if it was
615 * fully filled with data nodes of maximum size. This is used in
616 * calculations when reporting free space.
618 c->leb_overhead = c->leb_size % UBIFS_MAX_DATA_NODE_SZ;
620 /* Buffer size for bulk-reads */
621 c->max_bu_buf_len = UBIFS_MAX_BULK_READ * UBIFS_MAX_DATA_NODE_SZ;
622 if (c->max_bu_buf_len > c->leb_size)
623 c->max_bu_buf_len = c->leb_size;
628 * bud_wbuf_callback - bud LEB write-buffer synchronization call-back.
629 * @c: UBIFS file-system description object
630 * @lnum: LEB the write-buffer was synchronized to
631 * @free: how many free bytes left in this LEB
632 * @pad: how many bytes were padded
634 * This is a callback function which is called by the I/O unit when the
635 * write-buffer is synchronized. We need this to correctly maintain space
636 * accounting in bud logical eraseblocks. This function returns zero in case of
637 * success and a negative error code in case of failure.
639 * This function actually belongs to the journal, but we keep it here because
640 * we want to keep it static.
642 static int bud_wbuf_callback(struct ubifs_info *c, int lnum, int free, int pad)
644 return ubifs_update_one_lp(c, lnum, free, pad, 0, 0);
648 * init_constants_sb - initialize UBIFS constants.
649 * @c: UBIFS file-system description object
651 * This is a helper function which initializes various UBIFS constants after
652 * the superblock has been read. It also checks various UBIFS parameters and
653 * makes sure they are all right. Returns zero in case of success and a
654 * negative error code in case of failure.
656 static int init_constants_sb(struct ubifs_info *c)
661 c->main_bytes = (long long)c->main_lebs * c->leb_size;
662 c->max_znode_sz = sizeof(struct ubifs_znode) +
663 c->fanout * sizeof(struct ubifs_zbranch);
665 tmp = ubifs_idx_node_sz(c, 1);
666 c->ranges[UBIFS_IDX_NODE].min_len = tmp;
667 c->min_idx_node_sz = ALIGN(tmp, 8);
669 tmp = ubifs_idx_node_sz(c, c->fanout);
670 c->ranges[UBIFS_IDX_NODE].max_len = tmp;
671 c->max_idx_node_sz = ALIGN(tmp, 8);
673 /* Make sure LEB size is large enough to fit full commit */
674 tmp = UBIFS_CS_NODE_SZ + UBIFS_REF_NODE_SZ * c->jhead_cnt;
675 tmp = ALIGN(tmp, c->min_io_size);
676 if (tmp > c->leb_size) {
677 ubifs_err(c, "too small LEB size %d, at least %d needed",
683 * Make sure that the log is large enough to fit reference nodes for
684 * all buds plus one reserved LEB.
686 tmp64 = c->max_bud_bytes + c->leb_size - 1;
687 c->max_bud_cnt = div_u64(tmp64, c->leb_size);
688 tmp = (c->ref_node_alsz * c->max_bud_cnt + c->leb_size - 1);
691 if (c->log_lebs < tmp) {
692 ubifs_err(c, "too small log %d LEBs, required min. %d LEBs",
698 * When budgeting we assume worst-case scenarios when the pages are not
699 * be compressed and direntries are of the maximum size.
701 * Note, data, which may be stored in inodes is budgeted separately, so
702 * it is not included into 'c->bi.inode_budget'.
704 c->bi.page_budget = UBIFS_MAX_DATA_NODE_SZ * UBIFS_BLOCKS_PER_PAGE;
705 c->bi.inode_budget = UBIFS_INO_NODE_SZ;
706 c->bi.dent_budget = UBIFS_MAX_DENT_NODE_SZ;
709 * When the amount of flash space used by buds becomes
710 * 'c->max_bud_bytes', UBIFS just blocks all writers and starts commit.
711 * The writers are unblocked when the commit is finished. To avoid
712 * writers to be blocked UBIFS initiates background commit in advance,
713 * when number of bud bytes becomes above the limit defined below.
715 c->bg_bud_bytes = (c->max_bud_bytes * 13) >> 4;
718 * Ensure minimum journal size. All the bytes in the journal heads are
719 * considered to be used, when calculating the current journal usage.
720 * Consequently, if the journal is too small, UBIFS will treat it as
723 tmp64 = (long long)(c->jhead_cnt + 1) * c->leb_size + 1;
724 if (c->bg_bud_bytes < tmp64)
725 c->bg_bud_bytes = tmp64;
726 if (c->max_bud_bytes < tmp64 + c->leb_size)
727 c->max_bud_bytes = tmp64 + c->leb_size;
729 err = ubifs_calc_lpt_geom(c);
733 /* Initialize effective LEB size used in budgeting calculations */
734 c->idx_leb_size = c->leb_size - c->max_idx_node_sz;
739 * init_constants_master - initialize UBIFS constants.
740 * @c: UBIFS file-system description object
742 * This is a helper function which initializes various UBIFS constants after
743 * the master node has been read. It also checks various UBIFS parameters and
744 * makes sure they are all right.
746 static void init_constants_master(struct ubifs_info *c)
750 c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c);
751 c->report_rp_size = ubifs_reported_space(c, c->rp_size);
754 * Calculate total amount of FS blocks. This number is not used
755 * internally because it does not make much sense for UBIFS, but it is
756 * necessary to report something for the 'statfs()' call.
758 * Subtract the LEB reserved for GC, the LEB which is reserved for
759 * deletions, minimum LEBs for the index, and assume only one journal
762 tmp64 = c->main_lebs - 1 - 1 - MIN_INDEX_LEBS - c->jhead_cnt + 1;
763 tmp64 *= (long long)c->leb_size - c->leb_overhead;
764 tmp64 = ubifs_reported_space(c, tmp64);
765 c->block_cnt = tmp64 >> UBIFS_BLOCK_SHIFT;
769 * take_gc_lnum - reserve GC LEB.
770 * @c: UBIFS file-system description object
772 * This function ensures that the LEB reserved for garbage collection is marked
773 * as "taken" in lprops. We also have to set free space to LEB size and dirty
774 * space to zero, because lprops may contain out-of-date information if the
775 * file-system was un-mounted before it has been committed. This function
776 * returns zero in case of success and a negative error code in case of
779 static int take_gc_lnum(struct ubifs_info *c)
783 if (c->gc_lnum == -1) {
784 ubifs_err(c, "no LEB for GC");
788 /* And we have to tell lprops that this LEB is taken */
789 err = ubifs_change_one_lp(c, c->gc_lnum, c->leb_size, 0,
795 * alloc_wbufs - allocate write-buffers.
796 * @c: UBIFS file-system description object
798 * This helper function allocates and initializes UBIFS write-buffers. Returns
799 * zero in case of success and %-ENOMEM in case of failure.
801 static int alloc_wbufs(struct ubifs_info *c)
805 c->jheads = kcalloc(c->jhead_cnt, sizeof(struct ubifs_jhead),
810 /* Initialize journal heads */
811 for (i = 0; i < c->jhead_cnt; i++) {
812 INIT_LIST_HEAD(&c->jheads[i].buds_list);
813 err = ubifs_wbuf_init(c, &c->jheads[i].wbuf);
817 c->jheads[i].wbuf.sync_callback = &bud_wbuf_callback;
818 c->jheads[i].wbuf.jhead = i;
819 c->jheads[i].grouped = 1;
820 c->jheads[i].log_hash = ubifs_hash_get_desc(c);
821 if (IS_ERR(c->jheads[i].log_hash))
826 * Garbage Collector head does not need to be synchronized by timer.
827 * Also GC head nodes are not grouped.
829 c->jheads[GCHD].wbuf.no_timer = 1;
830 c->jheads[GCHD].grouped = 0;
836 kfree(c->jheads[i].log_hash);
842 * free_wbufs - free write-buffers.
843 * @c: UBIFS file-system description object
845 static void free_wbufs(struct ubifs_info *c)
850 for (i = 0; i < c->jhead_cnt; i++) {
851 kfree(c->jheads[i].wbuf.buf);
852 kfree(c->jheads[i].wbuf.inodes);
853 kfree(c->jheads[i].log_hash);
861 * free_orphans - free orphans.
862 * @c: UBIFS file-system description object
864 static void free_orphans(struct ubifs_info *c)
866 struct ubifs_orphan *orph;
868 while (c->orph_dnext) {
869 orph = c->orph_dnext;
870 c->orph_dnext = orph->dnext;
871 list_del(&orph->list);
875 while (!list_empty(&c->orph_list)) {
876 orph = list_entry(c->orph_list.next, struct ubifs_orphan, list);
877 list_del(&orph->list);
879 ubifs_err(c, "orphan list not empty at unmount");
887 * free_buds - free per-bud objects.
888 * @c: UBIFS file-system description object
890 static void free_buds(struct ubifs_info *c)
892 struct ubifs_bud *bud, *n;
894 rbtree_postorder_for_each_entry_safe(bud, n, &c->buds, rb)
899 * check_volume_empty - check if the UBI volume is empty.
900 * @c: UBIFS file-system description object
902 * This function checks if the UBIFS volume is empty by looking if its LEBs are
903 * mapped or not. The result of checking is stored in the @c->empty variable.
904 * Returns zero in case of success and a negative error code in case of
907 static int check_volume_empty(struct ubifs_info *c)
912 for (lnum = 0; lnum < c->leb_cnt; lnum++) {
913 err = ubifs_is_mapped(c, lnum);
914 if (unlikely(err < 0))
928 * UBIFS mount options.
930 * Opt_fast_unmount: do not run a journal commit before un-mounting
931 * Opt_norm_unmount: run a journal commit before un-mounting
932 * Opt_bulk_read: enable bulk-reads
933 * Opt_no_bulk_read: disable bulk-reads
934 * Opt_chk_data_crc: check CRCs when reading data nodes
935 * Opt_no_chk_data_crc: do not check CRCs when reading data nodes
936 * Opt_override_compr: override default compressor
937 * Opt_assert: set ubifs_assert() action
938 * Opt_auth_key: The key name used for authentication
939 * Opt_auth_hash_name: The hash type used for authentication
940 * Opt_err: just end of array marker
957 static const match_table_t tokens = {
958 {Opt_fast_unmount, "fast_unmount"},
959 {Opt_norm_unmount, "norm_unmount"},
960 {Opt_bulk_read, "bulk_read"},
961 {Opt_no_bulk_read, "no_bulk_read"},
962 {Opt_chk_data_crc, "chk_data_crc"},
963 {Opt_no_chk_data_crc, "no_chk_data_crc"},
964 {Opt_override_compr, "compr=%s"},
965 {Opt_auth_key, "auth_key=%s"},
966 {Opt_auth_hash_name, "auth_hash_name=%s"},
967 {Opt_ignore, "ubi=%s"},
968 {Opt_ignore, "vol=%s"},
969 {Opt_assert, "assert=%s"},
974 * parse_standard_option - parse a standard mount option.
975 * @option: the option to parse
977 * Normally, standard mount options like "sync" are passed to file-systems as
978 * flags. However, when a "rootflags=" kernel boot parameter is used, they may
979 * be present in the options string. This function tries to deal with this
980 * situation and parse standard options. Returns 0 if the option was not
981 * recognized, and the corresponding integer flag if it was.
983 * UBIFS is only interested in the "sync" option, so do not check for anything
986 static int parse_standard_option(const char *option)
989 pr_notice("UBIFS: parse %s\n", option);
990 if (!strcmp(option, "sync"))
991 return SB_SYNCHRONOUS;
996 * ubifs_parse_options - parse mount parameters.
997 * @c: UBIFS file-system description object
998 * @options: parameters to parse
999 * @is_remount: non-zero if this is FS re-mount
1001 * This function parses UBIFS mount options and returns zero in case success
1002 * and a negative error code in case of failure.
1004 static int ubifs_parse_options(struct ubifs_info *c, char *options,
1008 substring_t args[MAX_OPT_ARGS];
1013 while ((p = strsep(&options, ","))) {
1019 token = match_token(p, tokens, args);
1022 * %Opt_fast_unmount and %Opt_norm_unmount options are ignored.
1023 * We accept them in order to be backward-compatible. But this
1024 * should be removed at some point.
1026 case Opt_fast_unmount:
1027 c->mount_opts.unmount_mode = 2;
1029 case Opt_norm_unmount:
1030 c->mount_opts.unmount_mode = 1;
1033 c->mount_opts.bulk_read = 2;
1036 case Opt_no_bulk_read:
1037 c->mount_opts.bulk_read = 1;
1040 case Opt_chk_data_crc:
1041 c->mount_opts.chk_data_crc = 2;
1042 c->no_chk_data_crc = 0;
1044 case Opt_no_chk_data_crc:
1045 c->mount_opts.chk_data_crc = 1;
1046 c->no_chk_data_crc = 1;
1048 case Opt_override_compr:
1050 char *name = match_strdup(&args[0]);
1054 if (!strcmp(name, "none"))
1055 c->mount_opts.compr_type = UBIFS_COMPR_NONE;
1056 else if (!strcmp(name, "lzo"))
1057 c->mount_opts.compr_type = UBIFS_COMPR_LZO;
1058 else if (!strcmp(name, "zlib"))
1059 c->mount_opts.compr_type = UBIFS_COMPR_ZLIB;
1061 ubifs_err(c, "unknown compressor \"%s\"", name); //FIXME: is c ready?
1066 c->mount_opts.override_compr = 1;
1067 c->default_compr = c->mount_opts.compr_type;
1072 char *act = match_strdup(&args[0]);
1076 if (!strcmp(act, "report"))
1077 c->assert_action = ASSACT_REPORT;
1078 else if (!strcmp(act, "read-only"))
1079 c->assert_action = ASSACT_RO;
1080 else if (!strcmp(act, "panic"))
1081 c->assert_action = ASSACT_PANIC;
1083 ubifs_err(c, "unknown assert action \"%s\"", act);
1091 c->auth_key_name = kstrdup(args[0].from, GFP_KERNEL);
1092 if (!c->auth_key_name)
1095 case Opt_auth_hash_name:
1096 c->auth_hash_name = kstrdup(args[0].from, GFP_KERNEL);
1097 if (!c->auth_hash_name)
1105 struct super_block *sb = c->vfs_sb;
1107 flag = parse_standard_option(p);
1109 ubifs_err(c, "unrecognized mount option \"%s\" or missing value",
1113 sb->s_flags |= flag;
1123 * destroy_journal - destroy journal data structures.
1124 * @c: UBIFS file-system description object
1126 * This function destroys journal data structures including those that may have
1127 * been created by recovery functions.
1129 static void destroy_journal(struct ubifs_info *c)
1131 while (!list_empty(&c->unclean_leb_list)) {
1132 struct ubifs_unclean_leb *ucleb;
1134 ucleb = list_entry(c->unclean_leb_list.next,
1135 struct ubifs_unclean_leb, list);
1136 list_del(&ucleb->list);
1139 while (!list_empty(&c->old_buds)) {
1140 struct ubifs_bud *bud;
1142 bud = list_entry(c->old_buds.next, struct ubifs_bud, list);
1143 list_del(&bud->list);
1146 ubifs_destroy_idx_gc(c);
1147 ubifs_destroy_size_tree(c);
1153 * bu_init - initialize bulk-read information.
1154 * @c: UBIFS file-system description object
1156 static void bu_init(struct ubifs_info *c)
1158 ubifs_assert(c, c->bulk_read == 1);
1161 return; /* Already initialized */
1164 c->bu.buf = kmalloc(c->max_bu_buf_len, GFP_KERNEL | __GFP_NOWARN);
1166 if (c->max_bu_buf_len > UBIFS_KMALLOC_OK) {
1167 c->max_bu_buf_len = UBIFS_KMALLOC_OK;
1171 /* Just disable bulk-read */
1172 ubifs_warn(c, "cannot allocate %d bytes of memory for bulk-read, disabling it",
1174 c->mount_opts.bulk_read = 1;
1181 * check_free_space - check if there is enough free space to mount.
1182 * @c: UBIFS file-system description object
1184 * This function makes sure UBIFS has enough free space to be mounted in
1185 * read/write mode. UBIFS must always have some free space to allow deletions.
1187 static int check_free_space(struct ubifs_info *c)
1189 ubifs_assert(c, c->dark_wm > 0);
1190 if (c->lst.total_free + c->lst.total_dirty < c->dark_wm) {
1191 ubifs_err(c, "insufficient free space to mount in R/W mode");
1192 ubifs_dump_budg(c, &c->bi);
1193 ubifs_dump_lprops(c);
1200 * mount_ubifs - mount UBIFS file-system.
1201 * @c: UBIFS file-system description object
1203 * This function mounts UBIFS file system. Returns zero in case of success and
1204 * a negative error code in case of failure.
1206 static int mount_ubifs(struct ubifs_info *c)
1212 c->ro_mount = !!sb_rdonly(c->vfs_sb);
1213 /* Suppress error messages while probing if SB_SILENT is set */
1214 c->probing = !!(c->vfs_sb->s_flags & SB_SILENT);
1216 err = init_constants_early(c);
1220 err = ubifs_debugging_init(c);
1224 err = check_volume_empty(c);
1228 if (c->empty && (c->ro_mount || c->ro_media)) {
1230 * This UBI volume is empty, and read-only, or the file system
1231 * is mounted read-only - we cannot format it.
1233 ubifs_err(c, "can't format empty UBI volume: read-only %s",
1234 c->ro_media ? "UBI volume" : "mount");
1239 if (c->ro_media && !c->ro_mount) {
1240 ubifs_err(c, "cannot mount read-write - read-only media");
1246 * The requirement for the buffer is that it should fit indexing B-tree
1247 * height amount of integers. We assume the height if the TNC tree will
1251 c->bottom_up_buf = kmalloc_array(BOTTOM_UP_HEIGHT, sizeof(int),
1253 if (!c->bottom_up_buf)
1256 c->sbuf = vmalloc(c->leb_size);
1261 c->ileb_buf = vmalloc(c->leb_size);
1266 if (c->bulk_read == 1)
1270 c->write_reserve_buf = kmalloc(COMPRESSED_DATA_NODE_BUF_SZ + \
1271 UBIFS_CIPHER_BLOCK_SIZE,
1273 if (!c->write_reserve_buf)
1279 if (c->auth_key_name) {
1280 if (IS_ENABLED(CONFIG_UBIFS_FS_AUTHENTICATION)) {
1281 err = ubifs_init_authentication(c);
1285 ubifs_err(c, "auth_key_name, but UBIFS is built without"
1286 " authentication support");
1292 err = ubifs_read_superblock(c);
1299 * Make sure the compressor which is set as default in the superblock
1300 * or overridden by mount options is actually compiled in.
1302 if (!ubifs_compr_present(c, c->default_compr)) {
1303 ubifs_err(c, "'compressor \"%s\" is not compiled in",
1304 ubifs_compr_name(c, c->default_compr));
1309 err = init_constants_sb(c);
1313 sz = ALIGN(c->max_idx_node_sz, c->min_io_size);
1314 sz = ALIGN(sz + c->max_idx_node_sz, c->min_io_size);
1315 c->cbuf = kmalloc(sz, GFP_NOFS);
1321 err = alloc_wbufs(c);
1325 sprintf(c->bgt_name, BGT_NAME_PATTERN, c->vi.ubi_num, c->vi.vol_id);
1327 /* Create background thread */
1328 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1329 if (IS_ERR(c->bgt)) {
1330 err = PTR_ERR(c->bgt);
1332 ubifs_err(c, "cannot spawn \"%s\", error %d",
1336 wake_up_process(c->bgt);
1339 err = ubifs_read_master(c);
1343 init_constants_master(c);
1345 if ((c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY)) != 0) {
1346 ubifs_msg(c, "recovery needed");
1347 c->need_recovery = 1;
1350 if (c->need_recovery && !c->ro_mount) {
1351 err = ubifs_recover_inl_heads(c, c->sbuf);
1356 err = ubifs_lpt_init(c, 1, !c->ro_mount);
1360 if (!c->ro_mount && c->space_fixup) {
1361 err = ubifs_fixup_free_space(c);
1366 if (!c->ro_mount && !c->need_recovery) {
1368 * Set the "dirty" flag so that if we reboot uncleanly we
1369 * will notice this immediately on the next mount.
1371 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1372 err = ubifs_write_master(c);
1377 err = dbg_check_idx_size(c, c->bi.old_idx_sz);
1381 err = ubifs_replay_journal(c);
1385 /* Calculate 'min_idx_lebs' after journal replay */
1386 c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c);
1388 err = ubifs_mount_orphans(c, c->need_recovery, c->ro_mount);
1395 err = check_free_space(c);
1399 /* Check for enough log space */
1400 lnum = c->lhead_lnum + 1;
1401 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1402 lnum = UBIFS_LOG_LNUM;
1403 if (lnum == c->ltail_lnum) {
1404 err = ubifs_consolidate_log(c);
1409 if (c->need_recovery) {
1410 if (!ubifs_authenticated(c)) {
1411 err = ubifs_recover_size(c, true);
1416 err = ubifs_rcvry_gc_commit(c);
1420 if (ubifs_authenticated(c)) {
1421 err = ubifs_recover_size(c, false);
1426 err = take_gc_lnum(c);
1431 * GC LEB may contain garbage if there was an unclean
1432 * reboot, and it should be un-mapped.
1434 err = ubifs_leb_unmap(c, c->gc_lnum);
1439 err = dbg_check_lprops(c);
1442 } else if (c->need_recovery) {
1443 err = ubifs_recover_size(c, false);
1448 * Even if we mount read-only, we have to set space in GC LEB
1449 * to proper value because this affects UBIFS free space
1450 * reporting. We do not want to have a situation when
1451 * re-mounting from R/O to R/W changes amount of free space.
1453 err = take_gc_lnum(c);
1458 spin_lock(&ubifs_infos_lock);
1459 list_add_tail(&c->infos_list, &ubifs_infos);
1460 spin_unlock(&ubifs_infos_lock);
1462 if (c->need_recovery) {
1464 ubifs_msg(c, "recovery deferred");
1466 c->need_recovery = 0;
1467 ubifs_msg(c, "recovery completed");
1469 * GC LEB has to be empty and taken at this point. But
1470 * the journal head LEBs may also be accounted as
1471 * "empty taken" if they are empty.
1473 ubifs_assert(c, c->lst.taken_empty_lebs > 0);
1476 ubifs_assert(c, c->lst.taken_empty_lebs > 0);
1478 err = dbg_check_filesystem(c);
1482 err = dbg_debugfs_init_fs(c);
1488 ubifs_msg(c, "UBIFS: mounted UBI device %d, volume %d, name \"%s\"%s",
1489 c->vi.ubi_num, c->vi.vol_id, c->vi.name,
1490 c->ro_mount ? ", R/O mode" : "");
1491 x = (long long)c->main_lebs * c->leb_size;
1492 y = (long long)c->log_lebs * c->leb_size + c->max_bud_bytes;
1493 ubifs_msg(c, "LEB size: %d bytes (%d KiB), min./max. I/O unit sizes: %d bytes/%d bytes",
1494 c->leb_size, c->leb_size >> 10, c->min_io_size,
1496 ubifs_msg(c, "FS size: %lld bytes (%lld MiB, %d LEBs), journal size %lld bytes (%lld MiB, %d LEBs)",
1497 x, x >> 20, c->main_lebs,
1498 y, y >> 20, c->log_lebs + c->max_bud_cnt);
1499 ubifs_msg(c, "reserved for root: %llu bytes (%llu KiB)",
1500 c->report_rp_size, c->report_rp_size >> 10);
1501 ubifs_msg(c, "media format: w%d/r%d (latest is w%d/r%d), UUID %pUB%s",
1502 c->fmt_version, c->ro_compat_version,
1503 UBIFS_FORMAT_VERSION, UBIFS_RO_COMPAT_VERSION, c->uuid,
1504 c->big_lpt ? ", big LPT model" : ", small LPT model");
1506 dbg_gen("default compressor: %s", ubifs_compr_name(c, c->default_compr));
1507 dbg_gen("data journal heads: %d",
1508 c->jhead_cnt - NONDATA_JHEADS_CNT);
1509 dbg_gen("log LEBs: %d (%d - %d)",
1510 c->log_lebs, UBIFS_LOG_LNUM, c->log_last);
1511 dbg_gen("LPT area LEBs: %d (%d - %d)",
1512 c->lpt_lebs, c->lpt_first, c->lpt_last);
1513 dbg_gen("orphan area LEBs: %d (%d - %d)",
1514 c->orph_lebs, c->orph_first, c->orph_last);
1515 dbg_gen("main area LEBs: %d (%d - %d)",
1516 c->main_lebs, c->main_first, c->leb_cnt - 1);
1517 dbg_gen("index LEBs: %d", c->lst.idx_lebs);
1518 dbg_gen("total index bytes: %lld (%lld KiB, %lld MiB)",
1519 c->bi.old_idx_sz, c->bi.old_idx_sz >> 10,
1520 c->bi.old_idx_sz >> 20);
1521 dbg_gen("key hash type: %d", c->key_hash_type);
1522 dbg_gen("tree fanout: %d", c->fanout);
1523 dbg_gen("reserved GC LEB: %d", c->gc_lnum);
1524 dbg_gen("max. znode size %d", c->max_znode_sz);
1525 dbg_gen("max. index node size %d", c->max_idx_node_sz);
1526 dbg_gen("node sizes: data %zu, inode %zu, dentry %zu",
1527 UBIFS_DATA_NODE_SZ, UBIFS_INO_NODE_SZ, UBIFS_DENT_NODE_SZ);
1528 dbg_gen("node sizes: trun %zu, sb %zu, master %zu",
1529 UBIFS_TRUN_NODE_SZ, UBIFS_SB_NODE_SZ, UBIFS_MST_NODE_SZ);
1530 dbg_gen("node sizes: ref %zu, cmt. start %zu, orph %zu",
1531 UBIFS_REF_NODE_SZ, UBIFS_CS_NODE_SZ, UBIFS_ORPH_NODE_SZ);
1532 dbg_gen("max. node sizes: data %zu, inode %zu dentry %zu, idx %d",
1533 UBIFS_MAX_DATA_NODE_SZ, UBIFS_MAX_INO_NODE_SZ,
1534 UBIFS_MAX_DENT_NODE_SZ, ubifs_idx_node_sz(c, c->fanout));
1535 dbg_gen("dead watermark: %d", c->dead_wm);
1536 dbg_gen("dark watermark: %d", c->dark_wm);
1537 dbg_gen("LEB overhead: %d", c->leb_overhead);
1538 x = (long long)c->main_lebs * c->dark_wm;
1539 dbg_gen("max. dark space: %lld (%lld KiB, %lld MiB)",
1540 x, x >> 10, x >> 20);
1541 dbg_gen("maximum bud bytes: %lld (%lld KiB, %lld MiB)",
1542 c->max_bud_bytes, c->max_bud_bytes >> 10,
1543 c->max_bud_bytes >> 20);
1544 dbg_gen("BG commit bud bytes: %lld (%lld KiB, %lld MiB)",
1545 c->bg_bud_bytes, c->bg_bud_bytes >> 10,
1546 c->bg_bud_bytes >> 20);
1547 dbg_gen("current bud bytes %lld (%lld KiB, %lld MiB)",
1548 c->bud_bytes, c->bud_bytes >> 10, c->bud_bytes >> 20);
1549 dbg_gen("max. seq. number: %llu", c->max_sqnum);
1550 dbg_gen("commit number: %llu", c->cmt_no);
1555 spin_lock(&ubifs_infos_lock);
1556 list_del(&c->infos_list);
1557 spin_unlock(&ubifs_infos_lock);
1563 ubifs_lpt_free(c, 0);
1566 kfree(c->rcvrd_mst_node);
1568 kthread_stop(c->bgt);
1574 kfree(c->write_reserve_buf);
1578 kfree(c->bottom_up_buf);
1579 ubifs_debugging_exit(c);
1584 * ubifs_umount - un-mount UBIFS file-system.
1585 * @c: UBIFS file-system description object
1587 * Note, this function is called to free allocated resourced when un-mounting,
1588 * as well as free resources when an error occurred while we were half way
1589 * through mounting (error path cleanup function). So it has to make sure the
1590 * resource was actually allocated before freeing it.
1592 static void ubifs_umount(struct ubifs_info *c)
1594 dbg_gen("un-mounting UBI device %d, volume %d", c->vi.ubi_num,
1597 dbg_debugfs_exit_fs(c);
1598 spin_lock(&ubifs_infos_lock);
1599 list_del(&c->infos_list);
1600 spin_unlock(&ubifs_infos_lock);
1603 kthread_stop(c->bgt);
1608 ubifs_lpt_free(c, 0);
1609 ubifs_exit_authentication(c);
1611 kfree(c->auth_key_name);
1612 kfree(c->auth_hash_name);
1614 kfree(c->rcvrd_mst_node);
1616 kfree(c->write_reserve_buf);
1620 kfree(c->bottom_up_buf);
1621 ubifs_debugging_exit(c);
1625 * ubifs_remount_rw - re-mount in read-write mode.
1626 * @c: UBIFS file-system description object
1628 * UBIFS avoids allocating many unnecessary resources when mounted in read-only
1629 * mode. This function allocates the needed resources and re-mounts UBIFS in
1632 static int ubifs_remount_rw(struct ubifs_info *c)
1636 if (c->rw_incompat) {
1637 ubifs_err(c, "the file-system is not R/W-compatible");
1638 ubifs_msg(c, "on-flash format version is w%d/r%d, but software only supports up to version w%d/r%d",
1639 c->fmt_version, c->ro_compat_version,
1640 UBIFS_FORMAT_VERSION, UBIFS_RO_COMPAT_VERSION);
1644 mutex_lock(&c->umount_mutex);
1645 dbg_save_space_info(c);
1646 c->remounting_rw = 1;
1649 if (c->space_fixup) {
1650 err = ubifs_fixup_free_space(c);
1655 err = check_free_space(c);
1659 if (c->old_leb_cnt != c->leb_cnt) {
1660 struct ubifs_sb_node *sup = c->sup_node;
1662 sup->leb_cnt = cpu_to_le32(c->leb_cnt);
1663 err = ubifs_write_sb_node(c, sup);
1668 if (c->need_recovery) {
1669 ubifs_msg(c, "completing deferred recovery");
1670 err = ubifs_write_rcvrd_mst_node(c);
1673 if (!ubifs_authenticated(c)) {
1674 err = ubifs_recover_size(c, true);
1678 err = ubifs_clean_lebs(c, c->sbuf);
1681 err = ubifs_recover_inl_heads(c, c->sbuf);
1685 /* A readonly mount is not allowed to have orphans */
1686 ubifs_assert(c, c->tot_orphans == 0);
1687 err = ubifs_clear_orphans(c);
1692 if (!(c->mst_node->flags & cpu_to_le32(UBIFS_MST_DIRTY))) {
1693 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
1694 err = ubifs_write_master(c);
1699 c->ileb_buf = vmalloc(c->leb_size);
1705 c->write_reserve_buf = kmalloc(COMPRESSED_DATA_NODE_BUF_SZ + \
1706 UBIFS_CIPHER_BLOCK_SIZE, GFP_KERNEL);
1707 if (!c->write_reserve_buf) {
1712 err = ubifs_lpt_init(c, 0, 1);
1716 /* Create background thread */
1717 c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
1718 if (IS_ERR(c->bgt)) {
1719 err = PTR_ERR(c->bgt);
1721 ubifs_err(c, "cannot spawn \"%s\", error %d",
1725 wake_up_process(c->bgt);
1727 c->orph_buf = vmalloc(c->leb_size);
1733 /* Check for enough log space */
1734 lnum = c->lhead_lnum + 1;
1735 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs)
1736 lnum = UBIFS_LOG_LNUM;
1737 if (lnum == c->ltail_lnum) {
1738 err = ubifs_consolidate_log(c);
1743 if (c->need_recovery) {
1744 err = ubifs_rcvry_gc_commit(c);
1748 if (ubifs_authenticated(c)) {
1749 err = ubifs_recover_size(c, false);
1754 err = ubifs_leb_unmap(c, c->gc_lnum);
1759 dbg_gen("re-mounted read-write");
1760 c->remounting_rw = 0;
1762 if (c->need_recovery) {
1763 c->need_recovery = 0;
1764 ubifs_msg(c, "deferred recovery completed");
1767 * Do not run the debugging space check if the were doing
1768 * recovery, because when we saved the information we had the
1769 * file-system in a state where the TNC and lprops has been
1770 * modified in memory, but all the I/O operations (including a
1771 * commit) were deferred. So the file-system was in
1772 * "non-committed" state. Now the file-system is in committed
1773 * state, and of course the amount of free space will change
1774 * because, for example, the old index size was imprecise.
1776 err = dbg_check_space_info(c);
1779 mutex_unlock(&c->umount_mutex);
1787 kthread_stop(c->bgt);
1791 kfree(c->write_reserve_buf);
1792 c->write_reserve_buf = NULL;
1795 ubifs_lpt_free(c, 1);
1796 c->remounting_rw = 0;
1797 mutex_unlock(&c->umount_mutex);
1802 * ubifs_remount_ro - re-mount in read-only mode.
1803 * @c: UBIFS file-system description object
1805 * We assume VFS has stopped writing. Possibly the background thread could be
1806 * running a commit, however kthread_stop will wait in that case.
1808 static void ubifs_remount_ro(struct ubifs_info *c)
1812 ubifs_assert(c, !c->need_recovery);
1813 ubifs_assert(c, !c->ro_mount);
1815 mutex_lock(&c->umount_mutex);
1817 kthread_stop(c->bgt);
1821 dbg_save_space_info(c);
1823 for (i = 0; i < c->jhead_cnt; i++) {
1824 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
1826 ubifs_ro_mode(c, err);
1829 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1830 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1831 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1832 err = ubifs_write_master(c);
1834 ubifs_ro_mode(c, err);
1838 kfree(c->write_reserve_buf);
1839 c->write_reserve_buf = NULL;
1842 ubifs_lpt_free(c, 1);
1844 err = dbg_check_space_info(c);
1846 ubifs_ro_mode(c, err);
1847 mutex_unlock(&c->umount_mutex);
1850 static void ubifs_put_super(struct super_block *sb)
1853 struct ubifs_info *c = sb->s_fs_info;
1855 ubifs_msg(c, "un-mount UBI device %d", c->vi.ubi_num);
1858 * The following asserts are only valid if there has not been a failure
1859 * of the media. For example, there will be dirty inodes if we failed
1860 * to write them back because of I/O errors.
1863 ubifs_assert(c, c->bi.idx_growth == 0);
1864 ubifs_assert(c, c->bi.dd_growth == 0);
1865 ubifs_assert(c, c->bi.data_growth == 0);
1869 * The 'c->umount_lock' prevents races between UBIFS memory shrinker
1870 * and file system un-mount. Namely, it prevents the shrinker from
1871 * picking this superblock for shrinking - it will be just skipped if
1872 * the mutex is locked.
1874 mutex_lock(&c->umount_mutex);
1877 * First of all kill the background thread to make sure it does
1878 * not interfere with un-mounting and freeing resources.
1881 kthread_stop(c->bgt);
1886 * On fatal errors c->ro_error is set to 1, in which case we do
1887 * not write the master node.
1892 /* Synchronize write-buffers */
1893 for (i = 0; i < c->jhead_cnt; i++) {
1894 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
1896 ubifs_ro_mode(c, err);
1900 * We are being cleanly unmounted which means the
1901 * orphans were killed - indicate this in the master
1902 * node. Also save the reserved GC LEB number.
1904 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_DIRTY);
1905 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
1906 c->mst_node->gc_lnum = cpu_to_le32(c->gc_lnum);
1907 err = ubifs_write_master(c);
1910 * Recovery will attempt to fix the master area
1911 * next mount, so we just print a message and
1912 * continue to unmount normally.
1914 ubifs_err(c, "failed to write master node, error %d",
1917 for (i = 0; i < c->jhead_cnt; i++)
1918 /* Make sure write-buffer timers are canceled */
1919 hrtimer_cancel(&c->jheads[i].wbuf.timer);
1924 ubi_close_volume(c->ubi);
1925 mutex_unlock(&c->umount_mutex);
1928 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
1931 struct ubifs_info *c = sb->s_fs_info;
1933 sync_filesystem(sb);
1934 dbg_gen("old flags %#lx, new flags %#x", sb->s_flags, *flags);
1936 err = ubifs_parse_options(c, data, 1);
1938 ubifs_err(c, "invalid or unknown remount parameter");
1942 if (c->ro_mount && !(*flags & SB_RDONLY)) {
1944 ubifs_msg(c, "cannot re-mount R/W due to prior errors");
1948 ubifs_msg(c, "cannot re-mount R/W - UBI volume is R/O");
1951 err = ubifs_remount_rw(c);
1954 } else if (!c->ro_mount && (*flags & SB_RDONLY)) {
1956 ubifs_msg(c, "cannot re-mount R/O due to prior errors");
1959 ubifs_remount_ro(c);
1962 if (c->bulk_read == 1)
1965 dbg_gen("disable bulk-read");
1966 mutex_lock(&c->bu_mutex);
1969 mutex_unlock(&c->bu_mutex);
1972 if (!c->need_recovery)
1973 ubifs_assert(c, c->lst.taken_empty_lebs > 0);
1978 const struct super_operations ubifs_super_operations = {
1979 .alloc_inode = ubifs_alloc_inode,
1980 .destroy_inode = ubifs_destroy_inode,
1981 .put_super = ubifs_put_super,
1982 .write_inode = ubifs_write_inode,
1983 .evict_inode = ubifs_evict_inode,
1984 .statfs = ubifs_statfs,
1985 .dirty_inode = ubifs_dirty_inode,
1986 .remount_fs = ubifs_remount_fs,
1987 .show_options = ubifs_show_options,
1988 .sync_fs = ubifs_sync_fs,
1992 * open_ubi - parse UBI device name string and open the UBI device.
1993 * @name: UBI volume name
1994 * @mode: UBI volume open mode
1996 * The primary method of mounting UBIFS is by specifying the UBI volume
1997 * character device node path. However, UBIFS may also be mounted withoug any
1998 * character device node using one of the following methods:
2000 * o ubiX_Y - mount UBI device number X, volume Y;
2001 * o ubiY - mount UBI device number 0, volume Y;
2002 * o ubiX:NAME - mount UBI device X, volume with name NAME;
2003 * o ubi:NAME - mount UBI device 0, volume with name NAME.
2005 * Alternative '!' separator may be used instead of ':' (because some shells
2006 * like busybox may interpret ':' as an NFS host name separator). This function
2007 * returns UBI volume description object in case of success and a negative
2008 * error code in case of failure.
2010 static struct ubi_volume_desc *open_ubi(const char *name, int mode)
2012 struct ubi_volume_desc *ubi;
2016 if (!name || !*name)
2017 return ERR_PTR(-EINVAL);
2019 /* First, try to open using the device node path method */
2020 ubi = ubi_open_volume_path(name, mode);
2024 /* Try the "nodev" method */
2025 if (name[0] != 'u' || name[1] != 'b' || name[2] != 'i')
2026 return ERR_PTR(-EINVAL);
2028 /* ubi:NAME method */
2029 if ((name[3] == ':' || name[3] == '!') && name[4] != '\0')
2030 return ubi_open_volume_nm(0, name + 4, mode);
2032 if (!isdigit(name[3]))
2033 return ERR_PTR(-EINVAL);
2035 dev = simple_strtoul(name + 3, &endptr, 0);
2038 if (*endptr == '\0')
2039 return ubi_open_volume(0, dev, mode);
2042 if (*endptr == '_' && isdigit(endptr[1])) {
2043 vol = simple_strtoul(endptr + 1, &endptr, 0);
2044 if (*endptr != '\0')
2045 return ERR_PTR(-EINVAL);
2046 return ubi_open_volume(dev, vol, mode);
2049 /* ubiX:NAME method */
2050 if ((*endptr == ':' || *endptr == '!') && endptr[1] != '\0')
2051 return ubi_open_volume_nm(dev, ++endptr, mode);
2053 return ERR_PTR(-EINVAL);
2056 static struct ubifs_info *alloc_ubifs_info(struct ubi_volume_desc *ubi)
2058 struct ubifs_info *c;
2060 c = kzalloc(sizeof(struct ubifs_info), GFP_KERNEL);
2062 spin_lock_init(&c->cnt_lock);
2063 spin_lock_init(&c->cs_lock);
2064 spin_lock_init(&c->buds_lock);
2065 spin_lock_init(&c->space_lock);
2066 spin_lock_init(&c->orphan_lock);
2067 init_rwsem(&c->commit_sem);
2068 mutex_init(&c->lp_mutex);
2069 mutex_init(&c->tnc_mutex);
2070 mutex_init(&c->log_mutex);
2071 mutex_init(&c->umount_mutex);
2072 mutex_init(&c->bu_mutex);
2073 mutex_init(&c->write_reserve_mutex);
2074 init_waitqueue_head(&c->cmt_wq);
2076 c->old_idx = RB_ROOT;
2077 c->size_tree = RB_ROOT;
2078 c->orph_tree = RB_ROOT;
2079 INIT_LIST_HEAD(&c->infos_list);
2080 INIT_LIST_HEAD(&c->idx_gc);
2081 INIT_LIST_HEAD(&c->replay_list);
2082 INIT_LIST_HEAD(&c->replay_buds);
2083 INIT_LIST_HEAD(&c->uncat_list);
2084 INIT_LIST_HEAD(&c->empty_list);
2085 INIT_LIST_HEAD(&c->freeable_list);
2086 INIT_LIST_HEAD(&c->frdi_idx_list);
2087 INIT_LIST_HEAD(&c->unclean_leb_list);
2088 INIT_LIST_HEAD(&c->old_buds);
2089 INIT_LIST_HEAD(&c->orph_list);
2090 INIT_LIST_HEAD(&c->orph_new);
2091 c->no_chk_data_crc = 1;
2092 c->assert_action = ASSACT_RO;
2094 c->highest_inum = UBIFS_FIRST_INO;
2095 c->lhead_lnum = c->ltail_lnum = UBIFS_LOG_LNUM;
2097 ubi_get_volume_info(ubi, &c->vi);
2098 ubi_get_device_info(c->vi.ubi_num, &c->di);
2103 static int ubifs_fill_super(struct super_block *sb, void *data, int silent)
2105 struct ubifs_info *c = sb->s_fs_info;
2110 /* Re-open the UBI device in read-write mode */
2111 c->ubi = ubi_open_volume(c->vi.ubi_num, c->vi.vol_id, UBI_READWRITE);
2112 if (IS_ERR(c->ubi)) {
2113 err = PTR_ERR(c->ubi);
2117 err = ubifs_parse_options(c, data, 0);
2122 * UBIFS provides 'backing_dev_info' in order to disable read-ahead. For
2123 * UBIFS, I/O is not deferred, it is done immediately in readpage,
2124 * which means the user would have to wait not just for their own I/O
2125 * but the read-ahead I/O as well i.e. completely pointless.
2127 * Read-ahead will be disabled because @sb->s_bdi->ra_pages is 0. Also
2128 * @sb->s_bdi->capabilities are initialized to 0 so there won't be any
2129 * writeback happening.
2131 err = super_setup_bdi_name(sb, "ubifs_%d_%d", c->vi.ubi_num,
2137 sb->s_magic = UBIFS_SUPER_MAGIC;
2138 sb->s_blocksize = UBIFS_BLOCK_SIZE;
2139 sb->s_blocksize_bits = UBIFS_BLOCK_SHIFT;
2140 sb->s_maxbytes = c->max_inode_sz = key_max_inode_size(c);
2141 if (c->max_inode_sz > MAX_LFS_FILESIZE)
2142 sb->s_maxbytes = c->max_inode_sz = MAX_LFS_FILESIZE;
2143 sb->s_op = &ubifs_super_operations;
2144 #ifdef CONFIG_UBIFS_FS_XATTR
2145 sb->s_xattr = ubifs_xattr_handlers;
2147 #ifdef CONFIG_FS_ENCRYPTION
2148 sb->s_cop = &ubifs_crypt_operations;
2151 mutex_lock(&c->umount_mutex);
2152 err = mount_ubifs(c);
2154 ubifs_assert(c, err < 0);
2158 /* Read the root inode */
2159 root = ubifs_iget(sb, UBIFS_ROOT_INO);
2161 err = PTR_ERR(root);
2165 sb->s_root = d_make_root(root);
2171 mutex_unlock(&c->umount_mutex);
2177 mutex_unlock(&c->umount_mutex);
2179 ubi_close_volume(c->ubi);
2184 static int sb_test(struct super_block *sb, void *data)
2186 struct ubifs_info *c1 = data;
2187 struct ubifs_info *c = sb->s_fs_info;
2189 return c->vi.cdev == c1->vi.cdev;
2192 static int sb_set(struct super_block *sb, void *data)
2194 sb->s_fs_info = data;
2195 return set_anon_super(sb, NULL);
2198 static struct dentry *ubifs_mount(struct file_system_type *fs_type, int flags,
2199 const char *name, void *data)
2201 struct ubi_volume_desc *ubi;
2202 struct ubifs_info *c;
2203 struct super_block *sb;
2206 dbg_gen("name %s, flags %#x", name, flags);
2209 * Get UBI device number and volume ID. Mount it read-only so far
2210 * because this might be a new mount point, and UBI allows only one
2211 * read-write user at a time.
2213 ubi = open_ubi(name, UBI_READONLY);
2215 if (!(flags & SB_SILENT))
2216 pr_err("UBIFS error (pid: %d): cannot open \"%s\", error %d",
2217 current->pid, name, (int)PTR_ERR(ubi));
2218 return ERR_CAST(ubi);
2221 c = alloc_ubifs_info(ubi);
2227 dbg_gen("opened ubi%d_%d", c->vi.ubi_num, c->vi.vol_id);
2229 sb = sget(fs_type, sb_test, sb_set, flags, c);
2237 struct ubifs_info *c1 = sb->s_fs_info;
2239 /* A new mount point for already mounted UBIFS */
2240 dbg_gen("this ubi volume is already mounted");
2241 if (!!(flags & SB_RDONLY) != c1->ro_mount) {
2246 err = ubifs_fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
2249 /* We do not support atime */
2250 sb->s_flags |= SB_ACTIVE;
2251 #ifndef CONFIG_UBIFS_ATIME_SUPPORT
2252 sb->s_flags |= SB_NOATIME;
2254 ubifs_msg(c, "full atime support is enabled.");
2258 /* 'fill_super()' opens ubi again so we must close it here */
2259 ubi_close_volume(ubi);
2261 return dget(sb->s_root);
2264 deactivate_locked_super(sb);
2266 ubi_close_volume(ubi);
2267 return ERR_PTR(err);
2270 static void kill_ubifs_super(struct super_block *s)
2272 struct ubifs_info *c = s->s_fs_info;
2277 static struct file_system_type ubifs_fs_type = {
2279 .owner = THIS_MODULE,
2280 .mount = ubifs_mount,
2281 .kill_sb = kill_ubifs_super,
2283 MODULE_ALIAS_FS("ubifs");
2286 * Inode slab cache constructor.
2288 static void inode_slab_ctor(void *obj)
2290 struct ubifs_inode *ui = obj;
2291 inode_init_once(&ui->vfs_inode);
2294 static int __init ubifs_init(void)
2298 BUILD_BUG_ON(sizeof(struct ubifs_ch) != 24);
2300 /* Make sure node sizes are 8-byte aligned */
2301 BUILD_BUG_ON(UBIFS_CH_SZ & 7);
2302 BUILD_BUG_ON(UBIFS_INO_NODE_SZ & 7);
2303 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ & 7);
2304 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ & 7);
2305 BUILD_BUG_ON(UBIFS_DATA_NODE_SZ & 7);
2306 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ & 7);
2307 BUILD_BUG_ON(UBIFS_SB_NODE_SZ & 7);
2308 BUILD_BUG_ON(UBIFS_MST_NODE_SZ & 7);
2309 BUILD_BUG_ON(UBIFS_REF_NODE_SZ & 7);
2310 BUILD_BUG_ON(UBIFS_CS_NODE_SZ & 7);
2311 BUILD_BUG_ON(UBIFS_ORPH_NODE_SZ & 7);
2313 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ & 7);
2314 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ & 7);
2315 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ & 7);
2316 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ & 7);
2317 BUILD_BUG_ON(UBIFS_MAX_NODE_SZ & 7);
2318 BUILD_BUG_ON(MIN_WRITE_SZ & 7);
2320 /* Check min. node size */
2321 BUILD_BUG_ON(UBIFS_INO_NODE_SZ < MIN_WRITE_SZ);
2322 BUILD_BUG_ON(UBIFS_DENT_NODE_SZ < MIN_WRITE_SZ);
2323 BUILD_BUG_ON(UBIFS_XENT_NODE_SZ < MIN_WRITE_SZ);
2324 BUILD_BUG_ON(UBIFS_TRUN_NODE_SZ < MIN_WRITE_SZ);
2326 BUILD_BUG_ON(UBIFS_MAX_DENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2327 BUILD_BUG_ON(UBIFS_MAX_XENT_NODE_SZ > UBIFS_MAX_NODE_SZ);
2328 BUILD_BUG_ON(UBIFS_MAX_DATA_NODE_SZ > UBIFS_MAX_NODE_SZ);
2329 BUILD_BUG_ON(UBIFS_MAX_INO_NODE_SZ > UBIFS_MAX_NODE_SZ);
2331 /* Defined node sizes */
2332 BUILD_BUG_ON(UBIFS_SB_NODE_SZ != 4096);
2333 BUILD_BUG_ON(UBIFS_MST_NODE_SZ != 512);
2334 BUILD_BUG_ON(UBIFS_INO_NODE_SZ != 160);
2335 BUILD_BUG_ON(UBIFS_REF_NODE_SZ != 64);
2338 * We use 2 bit wide bit-fields to store compression type, which should
2339 * be amended if more compressors are added. The bit-fields are:
2340 * @compr_type in 'struct ubifs_inode', @default_compr in
2341 * 'struct ubifs_info' and @compr_type in 'struct ubifs_mount_opts'.
2343 BUILD_BUG_ON(UBIFS_COMPR_TYPES_CNT > 4);
2346 * We require that PAGE_SIZE is greater-than-or-equal-to
2347 * UBIFS_BLOCK_SIZE. It is assumed that both are powers of 2.
2349 if (PAGE_SIZE < UBIFS_BLOCK_SIZE) {
2350 pr_err("UBIFS error (pid %d): VFS page cache size is %u bytes, but UBIFS requires at least 4096 bytes",
2351 current->pid, (unsigned int)PAGE_SIZE);
2355 ubifs_inode_slab = kmem_cache_create("ubifs_inode_slab",
2356 sizeof(struct ubifs_inode), 0,
2357 SLAB_MEM_SPREAD | SLAB_RECLAIM_ACCOUNT |
2358 SLAB_ACCOUNT, &inode_slab_ctor);
2359 if (!ubifs_inode_slab)
2362 err = register_shrinker(&ubifs_shrinker_info);
2366 err = ubifs_compressors_init();
2370 err = dbg_debugfs_init();
2374 err = register_filesystem(&ubifs_fs_type);
2376 pr_err("UBIFS error (pid %d): cannot register file system, error %d",
2385 ubifs_compressors_exit();
2387 unregister_shrinker(&ubifs_shrinker_info);
2389 kmem_cache_destroy(ubifs_inode_slab);
2392 /* late_initcall to let compressors initialize first */
2393 late_initcall(ubifs_init);
2395 static void __exit ubifs_exit(void)
2397 WARN_ON(!list_empty(&ubifs_infos));
2398 WARN_ON(atomic_long_read(&ubifs_clean_zn_cnt) != 0);
2401 ubifs_compressors_exit();
2402 unregister_shrinker(&ubifs_shrinker_info);
2405 * Make sure all delayed rcu free inodes are flushed before we
2409 kmem_cache_destroy(ubifs_inode_slab);
2410 unregister_filesystem(&ubifs_fs_type);
2412 module_exit(ubifs_exit);
2414 MODULE_LICENSE("GPL");
2415 MODULE_VERSION(__stringify(UBIFS_VERSION));
2416 MODULE_AUTHOR("Artem Bityutskiy, Adrian Hunter");
2417 MODULE_DESCRIPTION("UBIFS - UBI File System");