2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 * Copyright (C) 2006, 2007 University of Szeged, Hungary
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 as published by
9 * the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along with
17 * this program; if not, write to the Free Software Foundation, Inc., 51
18 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Authors: Artem Bityutskiy (Битюцкий Артём)
26 * This file implements UBIFS I/O subsystem which provides various I/O-related
27 * helper functions (reading/writing/checking/validating nodes) and implements
28 * write-buffering support. Write buffers help to save space which otherwise
29 * would have been wasted for padding to the nearest minimal I/O unit boundary.
30 * Instead, data first goes to the write-buffer and is flushed when the
31 * buffer is full or when it is not used for some time (by timer). This is
32 * similar to the mechanism is used by JFFS2.
34 * UBIFS distinguishes between minimum write size (@c->min_io_size) and maximum
35 * write size (@c->max_write_size). The latter is the maximum amount of bytes
36 * the underlying flash is able to program at a time, and writing in
37 * @c->max_write_size units should presumably be faster. Obviously,
38 * @c->min_io_size <= @c->max_write_size. Write-buffers are of
39 * @c->max_write_size bytes in size for maximum performance. However, when a
40 * write-buffer is flushed, only the portion of it (aligned to @c->min_io_size
41 * boundary) which contains data is written, not the whole write-buffer,
42 * because this is more space-efficient.
44 * This optimization adds few complications to the code. Indeed, on the one
45 * hand, we want to write in optimal @c->max_write_size bytes chunks, which
46 * also means aligning writes at the @c->max_write_size bytes offsets. On the
47 * other hand, we do not want to waste space when synchronizing the write
48 * buffer, so during synchronization we writes in smaller chunks. And this makes
49 * the next write offset to be not aligned to @c->max_write_size bytes. So the
50 * have to make sure that the write-buffer offset (@wbuf->offs) becomes aligned
51 * to @c->max_write_size bytes again. We do this by temporarily shrinking
52 * write-buffer size (@wbuf->size).
54 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
55 * mutexes defined inside these objects. Since sometimes upper-level code
56 * has to lock the write-buffer (e.g. journal space reservation code), many
57 * functions related to write-buffers have "nolock" suffix which means that the
58 * caller has to lock the write-buffer before calling this function.
60 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
61 * aligned, UBIFS starts the next node from the aligned address, and the padded
62 * bytes may contain any rubbish. In other words, UBIFS does not put padding
63 * bytes in those small gaps. Common headers of nodes store real node lengths,
64 * not aligned lengths. Indexing nodes also store real lengths in branches.
66 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
67 * uses padding nodes or padding bytes, if the padding node does not fit.
69 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
70 * they are read from the flash media.
73 #include <linux/crc32.h>
74 #include <linux/slab.h>
78 * ubifs_ro_mode - switch UBIFS to read read-only mode.
79 * @c: UBIFS file-system description object
80 * @err: error code which is the reason of switching to R/O mode
82 void ubifs_ro_mode(struct ubifs_info *c, int err)
86 c->no_chk_data_crc = 0;
87 c->vfs_sb->s_flags |= MS_RDONLY;
88 ubifs_warn(c, "switched to read-only mode, error %d", err);
94 * Below are simple wrappers over UBI I/O functions which include some
95 * additional checks and UBIFS debugging stuff. See corresponding UBI function
96 * for more information.
99 int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
100 int len, int even_ebadmsg)
104 err = ubi_read(c->ubi, lnum, buf, offs, len);
106 * In case of %-EBADMSG print the error message only if the
107 * @even_ebadmsg is true.
109 if (err && (err != -EBADMSG || even_ebadmsg)) {
110 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
111 len, lnum, offs, err);
117 int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
122 ubifs_assert(!c->ro_media && !c->ro_mount);
125 if (!dbg_is_tst_rcvry(c))
126 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
128 err = dbg_leb_write(c, lnum, buf, offs, len);
130 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
131 len, lnum, offs, err);
132 ubifs_ro_mode(c, err);
138 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
142 ubifs_assert(!c->ro_media && !c->ro_mount);
145 if (!dbg_is_tst_rcvry(c))
146 err = ubi_leb_change(c->ubi, lnum, buf, len);
148 err = dbg_leb_change(c, lnum, buf, len);
150 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
152 ubifs_ro_mode(c, err);
158 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
162 ubifs_assert(!c->ro_media && !c->ro_mount);
165 if (!dbg_is_tst_rcvry(c))
166 err = ubi_leb_unmap(c->ubi, lnum);
168 err = dbg_leb_unmap(c, lnum);
170 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
171 ubifs_ro_mode(c, err);
177 int ubifs_leb_map(struct ubifs_info *c, int lnum)
181 ubifs_assert(!c->ro_media && !c->ro_mount);
184 if (!dbg_is_tst_rcvry(c))
185 err = ubi_leb_map(c->ubi, lnum);
187 err = dbg_leb_map(c, lnum);
189 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
190 ubifs_ro_mode(c, err);
196 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
200 err = ubi_is_mapped(c->ubi, lnum);
202 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
210 * ubifs_check_node - check node.
211 * @c: UBIFS file-system description object
212 * @buf: node to check
213 * @lnum: logical eraseblock number
214 * @offs: offset within the logical eraseblock
215 * @quiet: print no messages
216 * @must_chk_crc: indicates whether to always check the CRC
218 * This function checks node magic number and CRC checksum. This function also
219 * validates node length to prevent UBIFS from becoming crazy when an attacker
220 * feeds it a file-system image with incorrect nodes. For example, too large
221 * node length in the common header could cause UBIFS to read memory outside of
222 * allocated buffer when checking the CRC checksum.
224 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
225 * true, which is controlled by corresponding UBIFS mount option. However, if
226 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
227 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
228 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
229 * is checked. This is because during mounting or re-mounting from R/O mode to
230 * R/W mode we may read journal nodes (when replying the journal or doing the
231 * recovery) and the journal nodes may potentially be corrupted, so checking is
234 * This function returns zero in case of success and %-EUCLEAN in case of bad
237 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
238 int offs, int quiet, int must_chk_crc)
240 int err = -EINVAL, type, node_len, dump_node = 1;
241 uint32_t crc, node_crc, magic;
242 const struct ubifs_ch *ch = buf;
244 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
245 ubifs_assert(!(offs & 7) && offs < c->leb_size);
247 magic = le32_to_cpu(ch->magic);
248 if (magic != UBIFS_NODE_MAGIC) {
250 ubifs_err(c, "bad magic %#08x, expected %#08x",
251 magic, UBIFS_NODE_MAGIC);
256 type = ch->node_type;
257 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
259 ubifs_err(c, "bad node type %d", type);
263 node_len = le32_to_cpu(ch->len);
264 if (node_len + offs > c->leb_size)
267 if (c->ranges[type].max_len == 0) {
268 if (node_len != c->ranges[type].len)
270 } else if (node_len < c->ranges[type].min_len ||
271 node_len > c->ranges[type].max_len)
274 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
275 !c->remounting_rw && c->no_chk_data_crc)
278 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
279 node_crc = le32_to_cpu(ch->crc);
280 if (crc != node_crc) {
282 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
292 ubifs_err(c, "bad node length %d", node_len);
293 if (type == UBIFS_DATA_NODE && node_len > UBIFS_DATA_NODE_SZ)
297 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
299 ubifs_dump_node(c, buf);
301 int safe_len = min3(node_len, c->leb_size - offs,
302 (int)UBIFS_MAX_DATA_NODE_SZ);
303 pr_err("\tprevent out-of-bounds memory access\n");
304 pr_err("\ttruncated data node length %d\n", safe_len);
305 pr_err("\tcorrupted data node:\n");
306 print_hex_dump(KERN_ERR, "\t", DUMP_PREFIX_OFFSET, 32, 1,
315 * ubifs_pad - pad flash space.
316 * @c: UBIFS file-system description object
317 * @buf: buffer to put padding to
318 * @pad: how many bytes to pad
320 * The flash media obliges us to write only in chunks of %c->min_io_size and
321 * when we have to write less data we add padding node to the write-buffer and
322 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
323 * media is being scanned. If the amount of wasted space is not enough to fit a
324 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
325 * pattern (%UBIFS_PADDING_BYTE).
327 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
330 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
334 ubifs_assert(pad >= 0);
336 if (pad >= UBIFS_PAD_NODE_SZ) {
337 struct ubifs_ch *ch = buf;
338 struct ubifs_pad_node *pad_node = buf;
340 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
341 ch->node_type = UBIFS_PAD_NODE;
342 ch->group_type = UBIFS_NO_NODE_GROUP;
343 ch->padding[0] = ch->padding[1] = 0;
345 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
346 pad -= UBIFS_PAD_NODE_SZ;
347 pad_node->pad_len = cpu_to_le32(pad);
348 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
349 ch->crc = cpu_to_le32(crc);
350 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
352 /* Too little space, padding node won't fit */
353 memset(buf, UBIFS_PADDING_BYTE, pad);
357 * next_sqnum - get next sequence number.
358 * @c: UBIFS file-system description object
360 static unsigned long long next_sqnum(struct ubifs_info *c)
362 unsigned long long sqnum;
364 spin_lock(&c->cnt_lock);
365 sqnum = ++c->max_sqnum;
366 spin_unlock(&c->cnt_lock);
368 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
369 if (sqnum >= SQNUM_WATERMARK) {
370 ubifs_err(c, "sequence number overflow %llu, end of life",
372 ubifs_ro_mode(c, -EINVAL);
374 ubifs_warn(c, "running out of sequence numbers, end of life soon");
381 * ubifs_prepare_node - prepare node to be written to flash.
382 * @c: UBIFS file-system description object
383 * @node: the node to pad
385 * @pad: if the buffer has to be padded
387 * This function prepares node at @node to be written to the media - it
388 * calculates node CRC, fills the common header, and adds proper padding up to
389 * the next minimum I/O unit if @pad is not zero.
391 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
394 struct ubifs_ch *ch = node;
395 unsigned long long sqnum = next_sqnum(c);
397 ubifs_assert(len >= UBIFS_CH_SZ);
399 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
400 ch->len = cpu_to_le32(len);
401 ch->group_type = UBIFS_NO_NODE_GROUP;
402 ch->sqnum = cpu_to_le64(sqnum);
403 ch->padding[0] = ch->padding[1] = 0;
404 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
405 ch->crc = cpu_to_le32(crc);
409 pad = ALIGN(len, c->min_io_size) - len;
410 ubifs_pad(c, node + len, pad);
415 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
416 * @c: UBIFS file-system description object
417 * @node: the node to pad
419 * @last: indicates the last node of the group
421 * This function prepares node at @node to be written to the media - it
422 * calculates node CRC and fills the common header.
424 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
427 struct ubifs_ch *ch = node;
428 unsigned long long sqnum = next_sqnum(c);
430 ubifs_assert(len >= UBIFS_CH_SZ);
432 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
433 ch->len = cpu_to_le32(len);
435 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
437 ch->group_type = UBIFS_IN_NODE_GROUP;
438 ch->sqnum = cpu_to_le64(sqnum);
439 ch->padding[0] = ch->padding[1] = 0;
440 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
441 ch->crc = cpu_to_le32(crc);
445 * wbuf_timer_callback - write-buffer timer callback function.
446 * @timer: timer data (write-buffer descriptor)
448 * This function is called when the write-buffer timer expires.
450 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
452 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
454 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
456 wbuf->c->need_wbuf_sync = 1;
457 ubifs_wake_up_bgt(wbuf->c);
458 return HRTIMER_NORESTART;
462 * new_wbuf_timer - start new write-buffer timer.
463 * @wbuf: write-buffer descriptor
465 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
467 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
468 unsigned long long delta = dirty_writeback_interval;
470 /* centi to milli, milli to nano, then 10% */
471 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
473 ubifs_assert(!hrtimer_active(&wbuf->timer));
474 ubifs_assert(delta <= ULONG_MAX);
478 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
479 dbg_jhead(wbuf->jhead),
480 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
481 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
482 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
487 * cancel_wbuf_timer - cancel write-buffer timer.
488 * @wbuf: write-buffer descriptor
490 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
495 hrtimer_cancel(&wbuf->timer);
499 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
500 * @wbuf: write-buffer to synchronize
502 * This function synchronizes write-buffer @buf and returns zero in case of
503 * success or a negative error code in case of failure.
505 * Note, although write-buffers are of @c->max_write_size, this function does
506 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
507 * if the write-buffer is only partially filled with data, only the used part
508 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
509 * This way we waste less space.
511 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
513 struct ubifs_info *c = wbuf->c;
514 int err, dirt, sync_len;
516 cancel_wbuf_timer_nolock(wbuf);
517 if (!wbuf->used || wbuf->lnum == -1)
518 /* Write-buffer is empty or not seeked */
521 dbg_io("LEB %d:%d, %d bytes, jhead %s",
522 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
523 ubifs_assert(!(wbuf->avail & 7));
524 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
525 ubifs_assert(wbuf->size >= c->min_io_size);
526 ubifs_assert(wbuf->size <= c->max_write_size);
527 ubifs_assert(wbuf->size % c->min_io_size == 0);
528 ubifs_assert(!c->ro_media && !c->ro_mount);
529 if (c->leb_size - wbuf->offs >= c->max_write_size)
530 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
536 * Do not write whole write buffer but write only the minimum necessary
537 * amount of min. I/O units.
539 sync_len = ALIGN(wbuf->used, c->min_io_size);
540 dirt = sync_len - wbuf->used;
542 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
543 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
547 spin_lock(&wbuf->lock);
548 wbuf->offs += sync_len;
550 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
551 * But our goal is to optimize writes and make sure we write in
552 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
553 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
554 * sure that @wbuf->offs + @wbuf->size is aligned to
555 * @c->max_write_size. This way we make sure that after next
556 * write-buffer flush we are again at the optimal offset (aligned to
557 * @c->max_write_size).
559 if (c->leb_size - wbuf->offs < c->max_write_size)
560 wbuf->size = c->leb_size - wbuf->offs;
561 else if (wbuf->offs & (c->max_write_size - 1))
562 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
564 wbuf->size = c->max_write_size;
565 wbuf->avail = wbuf->size;
568 spin_unlock(&wbuf->lock);
570 if (wbuf->sync_callback)
571 err = wbuf->sync_callback(c, wbuf->lnum,
572 c->leb_size - wbuf->offs, dirt);
577 * ubifs_wbuf_seek_nolock - seek write-buffer.
578 * @wbuf: write-buffer
579 * @lnum: logical eraseblock number to seek to
580 * @offs: logical eraseblock offset to seek to
582 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
583 * The write-buffer has to be empty. Returns zero in case of success and a
584 * negative error code in case of failure.
586 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
588 const struct ubifs_info *c = wbuf->c;
590 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
591 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
592 ubifs_assert(offs >= 0 && offs <= c->leb_size);
593 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
594 ubifs_assert(lnum != wbuf->lnum);
595 ubifs_assert(wbuf->used == 0);
597 spin_lock(&wbuf->lock);
600 if (c->leb_size - wbuf->offs < c->max_write_size)
601 wbuf->size = c->leb_size - wbuf->offs;
602 else if (wbuf->offs & (c->max_write_size - 1))
603 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
605 wbuf->size = c->max_write_size;
606 wbuf->avail = wbuf->size;
608 spin_unlock(&wbuf->lock);
614 * ubifs_bg_wbufs_sync - synchronize write-buffers.
615 * @c: UBIFS file-system description object
617 * This function is called by background thread to synchronize write-buffers.
618 * Returns zero in case of success and a negative error code in case of
621 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
625 ubifs_assert(!c->ro_media && !c->ro_mount);
626 if (!c->need_wbuf_sync)
628 c->need_wbuf_sync = 0;
635 dbg_io("synchronize");
636 for (i = 0; i < c->jhead_cnt; i++) {
637 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
642 * If the mutex is locked then wbuf is being changed, so
643 * synchronization is not necessary.
645 if (mutex_is_locked(&wbuf->io_mutex))
648 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
649 if (!wbuf->need_sync) {
650 mutex_unlock(&wbuf->io_mutex);
654 err = ubifs_wbuf_sync_nolock(wbuf);
655 mutex_unlock(&wbuf->io_mutex);
657 ubifs_err(c, "cannot sync write-buffer, error %d", err);
658 ubifs_ro_mode(c, err);
666 /* Cancel all timers to prevent repeated errors */
667 for (i = 0; i < c->jhead_cnt; i++) {
668 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
670 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
671 cancel_wbuf_timer_nolock(wbuf);
672 mutex_unlock(&wbuf->io_mutex);
678 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
679 * @wbuf: write-buffer
680 * @buf: node to write
683 * This function writes data to flash via write-buffer @wbuf. This means that
684 * the last piece of the node won't reach the flash media immediately if it
685 * does not take whole max. write unit (@c->max_write_size). Instead, the node
686 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
687 * because more data are appended to the write-buffer).
689 * This function returns zero in case of success and a negative error code in
690 * case of failure. If the node cannot be written because there is no more
691 * space in this logical eraseblock, %-ENOSPC is returned.
693 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
695 struct ubifs_info *c = wbuf->c;
696 int err, written, n, aligned_len = ALIGN(len, 8);
698 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
699 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
700 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
701 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
702 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
703 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
704 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
705 ubifs_assert(wbuf->size >= c->min_io_size);
706 ubifs_assert(wbuf->size <= c->max_write_size);
707 ubifs_assert(wbuf->size % c->min_io_size == 0);
708 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
709 ubifs_assert(!c->ro_media && !c->ro_mount);
710 ubifs_assert(!c->space_fixup);
711 if (c->leb_size - wbuf->offs >= c->max_write_size)
712 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
714 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
719 cancel_wbuf_timer_nolock(wbuf);
724 if (aligned_len <= wbuf->avail) {
726 * The node is not very large and fits entirely within
729 memcpy(wbuf->buf + wbuf->used, buf, len);
730 if (aligned_len > len) {
731 ubifs_assert(aligned_len - len < 8);
732 ubifs_pad(c, wbuf->buf + wbuf->used + len, aligned_len - len);
735 if (aligned_len == wbuf->avail) {
736 dbg_io("flush jhead %s wbuf to LEB %d:%d",
737 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
738 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
739 wbuf->offs, wbuf->size);
743 spin_lock(&wbuf->lock);
744 wbuf->offs += wbuf->size;
745 if (c->leb_size - wbuf->offs >= c->max_write_size)
746 wbuf->size = c->max_write_size;
748 wbuf->size = c->leb_size - wbuf->offs;
749 wbuf->avail = wbuf->size;
752 spin_unlock(&wbuf->lock);
754 spin_lock(&wbuf->lock);
755 wbuf->avail -= aligned_len;
756 wbuf->used += aligned_len;
757 spin_unlock(&wbuf->lock);
767 * The node is large enough and does not fit entirely within
768 * current available space. We have to fill and flush
769 * write-buffer and switch to the next max. write unit.
771 dbg_io("flush jhead %s wbuf to LEB %d:%d",
772 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
773 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
774 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
779 wbuf->offs += wbuf->size;
781 aligned_len -= wbuf->avail;
782 written += wbuf->avail;
783 } else if (wbuf->offs & (c->max_write_size - 1)) {
785 * The write-buffer offset is not aligned to
786 * @c->max_write_size and @wbuf->size is less than
787 * @c->max_write_size. Write @wbuf->size bytes to make sure the
788 * following writes are done in optimal @c->max_write_size
791 dbg_io("write %d bytes to LEB %d:%d",
792 wbuf->size, wbuf->lnum, wbuf->offs);
793 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
798 wbuf->offs += wbuf->size;
800 aligned_len -= wbuf->size;
801 written += wbuf->size;
805 * The remaining data may take more whole max. write units, so write the
806 * remains multiple to max. write unit size directly to the flash media.
807 * We align node length to 8-byte boundary because we anyway flash wbuf
808 * if the remaining space is less than 8 bytes.
810 n = aligned_len >> c->max_write_shift;
812 n <<= c->max_write_shift;
813 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
815 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
825 spin_lock(&wbuf->lock);
828 * And now we have what's left and what does not take whole
829 * max. write unit, so write it to the write-buffer and we are
832 memcpy(wbuf->buf, buf + written, len);
833 if (aligned_len > len) {
834 ubifs_assert(aligned_len - len < 8);
835 ubifs_pad(c, wbuf->buf + len, aligned_len - len);
839 if (c->leb_size - wbuf->offs >= c->max_write_size)
840 wbuf->size = c->max_write_size;
842 wbuf->size = c->leb_size - wbuf->offs;
843 wbuf->avail = wbuf->size - aligned_len;
844 wbuf->used = aligned_len;
846 spin_unlock(&wbuf->lock);
849 if (wbuf->sync_callback) {
850 int free = c->leb_size - wbuf->offs - wbuf->used;
852 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
858 new_wbuf_timer_nolock(wbuf);
863 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
864 len, wbuf->lnum, wbuf->offs, err);
865 ubifs_dump_node(c, buf);
867 ubifs_dump_leb(c, wbuf->lnum);
872 * ubifs_write_node - write node to the media.
873 * @c: UBIFS file-system description object
874 * @buf: the node to write
876 * @lnum: logical eraseblock number
877 * @offs: offset within the logical eraseblock
879 * This function automatically fills node magic number, assigns sequence
880 * number, and calculates node CRC checksum. The length of the @buf buffer has
881 * to be aligned to the minimal I/O unit size. This function automatically
882 * appends padding node and padding bytes if needed. Returns zero in case of
883 * success and a negative error code in case of failure.
885 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
888 int err, buf_len = ALIGN(len, c->min_io_size);
890 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
891 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
893 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
894 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
895 ubifs_assert(!c->ro_media && !c->ro_mount);
896 ubifs_assert(!c->space_fixup);
901 ubifs_prepare_node(c, buf, len, 1);
902 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
904 ubifs_dump_node(c, buf);
910 * ubifs_read_node_wbuf - read node from the media or write-buffer.
911 * @wbuf: wbuf to check for un-written data
912 * @buf: buffer to read to
915 * @lnum: logical eraseblock number
916 * @offs: offset within the logical eraseblock
918 * This function reads a node of known type and length, checks it and stores
919 * in @buf. If the node partially or fully sits in the write-buffer, this
920 * function takes data from the buffer, otherwise it reads the flash media.
921 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
922 * error code in case of failure.
924 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
927 const struct ubifs_info *c = wbuf->c;
928 int err, rlen, overlap;
929 struct ubifs_ch *ch = buf;
931 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
932 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
933 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
934 ubifs_assert(!(offs & 7) && offs < c->leb_size);
935 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
937 spin_lock(&wbuf->lock);
938 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
940 /* We may safely unlock the write-buffer and read the data */
941 spin_unlock(&wbuf->lock);
942 return ubifs_read_node(c, buf, type, len, lnum, offs);
945 /* Don't read under wbuf */
946 rlen = wbuf->offs - offs;
950 /* Copy the rest from the write-buffer */
951 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
952 spin_unlock(&wbuf->lock);
955 /* Read everything that goes before write-buffer */
956 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
957 if (err && err != -EBADMSG)
961 if (type != ch->node_type) {
962 ubifs_err(c, "bad node type (%d but expected %d)",
963 ch->node_type, type);
967 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
969 ubifs_err(c, "expected node type %d", type);
973 rlen = le32_to_cpu(ch->len);
975 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
982 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
983 ubifs_dump_node(c, buf);
989 * ubifs_read_node - read node.
990 * @c: UBIFS file-system description object
991 * @buf: buffer to read to
993 * @len: node length (not aligned)
994 * @lnum: logical eraseblock number
995 * @offs: offset within the logical eraseblock
997 * This function reads a node of known type and and length, checks it and
998 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
999 * and a negative error code in case of failure.
1001 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1005 struct ubifs_ch *ch = buf;
1007 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1008 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1009 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1010 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1011 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1013 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1014 if (err && err != -EBADMSG)
1017 if (type != ch->node_type) {
1018 ubifs_errc(c, "bad node type (%d but expected %d)",
1019 ch->node_type, type);
1023 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1025 ubifs_errc(c, "expected node type %d", type);
1029 l = le32_to_cpu(ch->len);
1031 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1038 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1039 offs, ubi_is_mapped(c->ubi, lnum));
1041 ubifs_dump_node(c, buf);
1048 * ubifs_wbuf_init - initialize write-buffer.
1049 * @c: UBIFS file-system description object
1050 * @wbuf: write-buffer to initialize
1052 * This function initializes write-buffer. Returns zero in case of success
1053 * %-ENOMEM in case of failure.
1055 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1059 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1063 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1064 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1065 if (!wbuf->inodes) {
1072 wbuf->lnum = wbuf->offs = -1;
1074 * If the LEB starts at the max. write size aligned address, then
1075 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1076 * set it to something smaller so that it ends at the closest max.
1077 * write size boundary.
1079 size = c->max_write_size - (c->leb_start % c->max_write_size);
1080 wbuf->avail = wbuf->size = size;
1081 wbuf->sync_callback = NULL;
1082 mutex_init(&wbuf->io_mutex);
1083 spin_lock_init(&wbuf->lock);
1087 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1088 wbuf->timer.function = wbuf_timer_callback_nolock;
1093 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1094 * @wbuf: the write-buffer where to add
1095 * @inum: the inode number
1097 * This function adds an inode number to the inode array of the write-buffer.
1099 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1102 /* NOR flash or something similar */
1105 spin_lock(&wbuf->lock);
1107 wbuf->inodes[wbuf->next_ino++] = inum;
1108 spin_unlock(&wbuf->lock);
1112 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1113 * @wbuf: the write-buffer
1114 * @inum: the inode number
1116 * This function returns with %1 if the write-buffer contains some data from the
1117 * given inode otherwise it returns with %0.
1119 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1123 spin_lock(&wbuf->lock);
1124 for (i = 0; i < wbuf->next_ino; i++)
1125 if (inum == wbuf->inodes[i]) {
1129 spin_unlock(&wbuf->lock);
1135 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1136 * @c: UBIFS file-system description object
1137 * @inode: inode to synchronize
1139 * This function synchronizes write-buffers which contain nodes belonging to
1140 * @inode. Returns zero in case of success and a negative error code in case of
1143 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1147 for (i = 0; i < c->jhead_cnt; i++) {
1148 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1152 * GC head is special, do not look at it. Even if the
1153 * head contains something related to this inode, it is
1154 * a _copy_ of corresponding on-flash node which sits
1159 if (!wbuf_has_ino(wbuf, inode->i_ino))
1162 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1163 if (wbuf_has_ino(wbuf, inode->i_ino))
1164 err = ubifs_wbuf_sync_nolock(wbuf);
1165 mutex_unlock(&wbuf->io_mutex);
1168 ubifs_ro_mode(c, err);