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 |= SB_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, !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, !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, !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, !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(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
245 ubifs_assert(c, !(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(c, 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(c, 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(c, 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 * @c: UBIFS file-system description object
464 * @wbuf: write-buffer descriptor
466 static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
468 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
469 unsigned long long delta = dirty_writeback_interval;
471 /* centi to milli, milli to nano, then 10% */
472 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
474 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
475 ubifs_assert(c, delta <= ULONG_MAX);
479 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
480 dbg_jhead(wbuf->jhead),
481 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
482 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
483 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
488 * cancel_wbuf_timer - cancel write-buffer timer.
489 * @wbuf: write-buffer descriptor
491 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
496 hrtimer_cancel(&wbuf->timer);
500 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
501 * @wbuf: write-buffer to synchronize
503 * This function synchronizes write-buffer @buf and returns zero in case of
504 * success or a negative error code in case of failure.
506 * Note, although write-buffers are of @c->max_write_size, this function does
507 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
508 * if the write-buffer is only partially filled with data, only the used part
509 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
510 * This way we waste less space.
512 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
514 struct ubifs_info *c = wbuf->c;
515 int err, dirt, sync_len;
517 cancel_wbuf_timer_nolock(wbuf);
518 if (!wbuf->used || wbuf->lnum == -1)
519 /* Write-buffer is empty or not seeked */
522 dbg_io("LEB %d:%d, %d bytes, jhead %s",
523 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
524 ubifs_assert(c, !(wbuf->avail & 7));
525 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
526 ubifs_assert(c, wbuf->size >= c->min_io_size);
527 ubifs_assert(c, wbuf->size <= c->max_write_size);
528 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
529 ubifs_assert(c, !c->ro_media && !c->ro_mount);
530 if (c->leb_size - wbuf->offs >= c->max_write_size)
531 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
537 * Do not write whole write buffer but write only the minimum necessary
538 * amount of min. I/O units.
540 sync_len = ALIGN(wbuf->used, c->min_io_size);
541 dirt = sync_len - wbuf->used;
543 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
544 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
548 spin_lock(&wbuf->lock);
549 wbuf->offs += sync_len;
551 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
552 * But our goal is to optimize writes and make sure we write in
553 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
554 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
555 * sure that @wbuf->offs + @wbuf->size is aligned to
556 * @c->max_write_size. This way we make sure that after next
557 * write-buffer flush we are again at the optimal offset (aligned to
558 * @c->max_write_size).
560 if (c->leb_size - wbuf->offs < c->max_write_size)
561 wbuf->size = c->leb_size - wbuf->offs;
562 else if (wbuf->offs & (c->max_write_size - 1))
563 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
565 wbuf->size = c->max_write_size;
566 wbuf->avail = wbuf->size;
569 spin_unlock(&wbuf->lock);
571 if (wbuf->sync_callback)
572 err = wbuf->sync_callback(c, wbuf->lnum,
573 c->leb_size - wbuf->offs, dirt);
578 * ubifs_wbuf_seek_nolock - seek write-buffer.
579 * @wbuf: write-buffer
580 * @lnum: logical eraseblock number to seek to
581 * @offs: logical eraseblock offset to seek to
583 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
584 * The write-buffer has to be empty. Returns zero in case of success and a
585 * negative error code in case of failure.
587 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
589 const struct ubifs_info *c = wbuf->c;
591 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
592 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
593 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
594 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
595 ubifs_assert(c, lnum != wbuf->lnum);
596 ubifs_assert(c, wbuf->used == 0);
598 spin_lock(&wbuf->lock);
601 if (c->leb_size - wbuf->offs < c->max_write_size)
602 wbuf->size = c->leb_size - wbuf->offs;
603 else if (wbuf->offs & (c->max_write_size - 1))
604 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
606 wbuf->size = c->max_write_size;
607 wbuf->avail = wbuf->size;
609 spin_unlock(&wbuf->lock);
615 * ubifs_bg_wbufs_sync - synchronize write-buffers.
616 * @c: UBIFS file-system description object
618 * This function is called by background thread to synchronize write-buffers.
619 * Returns zero in case of success and a negative error code in case of
622 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
626 ubifs_assert(c, !c->ro_media && !c->ro_mount);
627 if (!c->need_wbuf_sync)
629 c->need_wbuf_sync = 0;
636 dbg_io("synchronize");
637 for (i = 0; i < c->jhead_cnt; i++) {
638 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
643 * If the mutex is locked then wbuf is being changed, so
644 * synchronization is not necessary.
646 if (mutex_is_locked(&wbuf->io_mutex))
649 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
650 if (!wbuf->need_sync) {
651 mutex_unlock(&wbuf->io_mutex);
655 err = ubifs_wbuf_sync_nolock(wbuf);
656 mutex_unlock(&wbuf->io_mutex);
658 ubifs_err(c, "cannot sync write-buffer, error %d", err);
659 ubifs_ro_mode(c, err);
667 /* Cancel all timers to prevent repeated errors */
668 for (i = 0; i < c->jhead_cnt; i++) {
669 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
671 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
672 cancel_wbuf_timer_nolock(wbuf);
673 mutex_unlock(&wbuf->io_mutex);
679 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
680 * @wbuf: write-buffer
681 * @buf: node to write
684 * This function writes data to flash via write-buffer @wbuf. This means that
685 * the last piece of the node won't reach the flash media immediately if it
686 * does not take whole max. write unit (@c->max_write_size). Instead, the node
687 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
688 * because more data are appended to the write-buffer).
690 * This function returns zero in case of success and a negative error code in
691 * case of failure. If the node cannot be written because there is no more
692 * space in this logical eraseblock, %-ENOSPC is returned.
694 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
696 struct ubifs_info *c = wbuf->c;
697 int err, written, n, aligned_len = ALIGN(len, 8);
699 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
700 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
701 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
702 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
703 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
704 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
705 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
706 ubifs_assert(c, wbuf->size >= c->min_io_size);
707 ubifs_assert(c, wbuf->size <= c->max_write_size);
708 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
709 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
710 ubifs_assert(c, !c->ro_media && !c->ro_mount);
711 ubifs_assert(c, !c->space_fixup);
712 if (c->leb_size - wbuf->offs >= c->max_write_size)
713 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
715 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
720 cancel_wbuf_timer_nolock(wbuf);
725 if (aligned_len <= wbuf->avail) {
727 * The node is not very large and fits entirely within
730 memcpy(wbuf->buf + wbuf->used, buf, len);
731 if (aligned_len > len) {
732 ubifs_assert(c, aligned_len - len < 8);
733 ubifs_pad(c, wbuf->buf + wbuf->used + len, aligned_len - len);
736 if (aligned_len == wbuf->avail) {
737 dbg_io("flush jhead %s wbuf to LEB %d:%d",
738 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
739 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
740 wbuf->offs, wbuf->size);
744 spin_lock(&wbuf->lock);
745 wbuf->offs += wbuf->size;
746 if (c->leb_size - wbuf->offs >= c->max_write_size)
747 wbuf->size = c->max_write_size;
749 wbuf->size = c->leb_size - wbuf->offs;
750 wbuf->avail = wbuf->size;
753 spin_unlock(&wbuf->lock);
755 spin_lock(&wbuf->lock);
756 wbuf->avail -= aligned_len;
757 wbuf->used += aligned_len;
758 spin_unlock(&wbuf->lock);
768 * The node is large enough and does not fit entirely within
769 * current available space. We have to fill and flush
770 * write-buffer and switch to the next max. write unit.
772 dbg_io("flush jhead %s wbuf to LEB %d:%d",
773 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
774 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
775 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
780 wbuf->offs += wbuf->size;
782 aligned_len -= wbuf->avail;
783 written += wbuf->avail;
784 } else if (wbuf->offs & (c->max_write_size - 1)) {
786 * The write-buffer offset is not aligned to
787 * @c->max_write_size and @wbuf->size is less than
788 * @c->max_write_size. Write @wbuf->size bytes to make sure the
789 * following writes are done in optimal @c->max_write_size
792 dbg_io("write %d bytes to LEB %d:%d",
793 wbuf->size, wbuf->lnum, wbuf->offs);
794 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
799 wbuf->offs += wbuf->size;
801 aligned_len -= wbuf->size;
802 written += wbuf->size;
806 * The remaining data may take more whole max. write units, so write the
807 * remains multiple to max. write unit size directly to the flash media.
808 * We align node length to 8-byte boundary because we anyway flash wbuf
809 * if the remaining space is less than 8 bytes.
811 n = aligned_len >> c->max_write_shift;
813 n <<= c->max_write_shift;
814 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
816 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
826 spin_lock(&wbuf->lock);
829 * And now we have what's left and what does not take whole
830 * max. write unit, so write it to the write-buffer and we are
833 memcpy(wbuf->buf, buf + written, len);
834 if (aligned_len > len) {
835 ubifs_assert(c, aligned_len - len < 8);
836 ubifs_pad(c, wbuf->buf + len, aligned_len - len);
840 if (c->leb_size - wbuf->offs >= c->max_write_size)
841 wbuf->size = c->max_write_size;
843 wbuf->size = c->leb_size - wbuf->offs;
844 wbuf->avail = wbuf->size - aligned_len;
845 wbuf->used = aligned_len;
847 spin_unlock(&wbuf->lock);
850 if (wbuf->sync_callback) {
851 int free = c->leb_size - wbuf->offs - wbuf->used;
853 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
859 new_wbuf_timer_nolock(c, wbuf);
864 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
865 len, wbuf->lnum, wbuf->offs, err);
866 ubifs_dump_node(c, buf);
868 ubifs_dump_leb(c, wbuf->lnum);
873 * ubifs_write_node - write node to the media.
874 * @c: UBIFS file-system description object
875 * @buf: the node to write
877 * @lnum: logical eraseblock number
878 * @offs: offset within the logical eraseblock
880 * This function automatically fills node magic number, assigns sequence
881 * number, and calculates node CRC checksum. The length of the @buf buffer has
882 * to be aligned to the minimal I/O unit size. This function automatically
883 * appends padding node and padding bytes if needed. Returns zero in case of
884 * success and a negative error code in case of failure.
886 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
889 int err, buf_len = ALIGN(len, c->min_io_size);
891 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
892 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
894 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
895 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
896 ubifs_assert(c, !c->ro_media && !c->ro_mount);
897 ubifs_assert(c, !c->space_fixup);
902 ubifs_prepare_node(c, buf, len, 1);
903 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
905 ubifs_dump_node(c, buf);
911 * ubifs_read_node_wbuf - read node from the media or write-buffer.
912 * @wbuf: wbuf to check for un-written data
913 * @buf: buffer to read to
916 * @lnum: logical eraseblock number
917 * @offs: offset within the logical eraseblock
919 * This function reads a node of known type and length, checks it and stores
920 * in @buf. If the node partially or fully sits in the write-buffer, this
921 * function takes data from the buffer, otherwise it reads the flash media.
922 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
923 * error code in case of failure.
925 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
928 const struct ubifs_info *c = wbuf->c;
929 int err, rlen, overlap;
930 struct ubifs_ch *ch = buf;
932 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
933 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
934 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
935 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
936 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
938 spin_lock(&wbuf->lock);
939 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
941 /* We may safely unlock the write-buffer and read the data */
942 spin_unlock(&wbuf->lock);
943 return ubifs_read_node(c, buf, type, len, lnum, offs);
946 /* Don't read under wbuf */
947 rlen = wbuf->offs - offs;
951 /* Copy the rest from the write-buffer */
952 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
953 spin_unlock(&wbuf->lock);
956 /* Read everything that goes before write-buffer */
957 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
958 if (err && err != -EBADMSG)
962 if (type != ch->node_type) {
963 ubifs_err(c, "bad node type (%d but expected %d)",
964 ch->node_type, type);
968 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
970 ubifs_err(c, "expected node type %d", type);
974 rlen = le32_to_cpu(ch->len);
976 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
983 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
984 ubifs_dump_node(c, buf);
990 * ubifs_read_node - read node.
991 * @c: UBIFS file-system description object
992 * @buf: buffer to read to
994 * @len: node length (not aligned)
995 * @lnum: logical eraseblock number
996 * @offs: offset within the logical eraseblock
998 * This function reads a node of known type and and length, checks it and
999 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1000 * and a negative error code in case of failure.
1002 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1006 struct ubifs_ch *ch = buf;
1008 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1009 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1010 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1011 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1012 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1014 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1015 if (err && err != -EBADMSG)
1018 if (type != ch->node_type) {
1019 ubifs_errc(c, "bad node type (%d but expected %d)",
1020 ch->node_type, type);
1024 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1026 ubifs_errc(c, "expected node type %d", type);
1030 l = le32_to_cpu(ch->len);
1032 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1039 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1040 offs, ubi_is_mapped(c->ubi, lnum));
1042 ubifs_dump_node(c, buf);
1049 * ubifs_wbuf_init - initialize write-buffer.
1050 * @c: UBIFS file-system description object
1051 * @wbuf: write-buffer to initialize
1053 * This function initializes write-buffer. Returns zero in case of success
1054 * %-ENOMEM in case of failure.
1056 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1060 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1064 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1065 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1066 if (!wbuf->inodes) {
1073 wbuf->lnum = wbuf->offs = -1;
1075 * If the LEB starts at the max. write size aligned address, then
1076 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1077 * set it to something smaller so that it ends at the closest max.
1078 * write size boundary.
1080 size = c->max_write_size - (c->leb_start % c->max_write_size);
1081 wbuf->avail = wbuf->size = size;
1082 wbuf->sync_callback = NULL;
1083 mutex_init(&wbuf->io_mutex);
1084 spin_lock_init(&wbuf->lock);
1088 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1089 wbuf->timer.function = wbuf_timer_callback_nolock;
1094 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1095 * @wbuf: the write-buffer where to add
1096 * @inum: the inode number
1098 * This function adds an inode number to the inode array of the write-buffer.
1100 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1103 /* NOR flash or something similar */
1106 spin_lock(&wbuf->lock);
1108 wbuf->inodes[wbuf->next_ino++] = inum;
1109 spin_unlock(&wbuf->lock);
1113 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1114 * @wbuf: the write-buffer
1115 * @inum: the inode number
1117 * This function returns with %1 if the write-buffer contains some data from the
1118 * given inode otherwise it returns with %0.
1120 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1124 spin_lock(&wbuf->lock);
1125 for (i = 0; i < wbuf->next_ino; i++)
1126 if (inum == wbuf->inodes[i]) {
1130 spin_unlock(&wbuf->lock);
1136 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1137 * @c: UBIFS file-system description object
1138 * @inode: inode to synchronize
1140 * This function synchronizes write-buffers which contain nodes belonging to
1141 * @inode. Returns zero in case of success and a negative error code in case of
1144 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1148 for (i = 0; i < c->jhead_cnt; i++) {
1149 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1153 * GC head is special, do not look at it. Even if the
1154 * head contains something related to this inode, it is
1155 * a _copy_ of corresponding on-flash node which sits
1160 if (!wbuf_has_ino(wbuf, inode->i_ino))
1163 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1164 if (wbuf_has_ino(wbuf, inode->i_ino))
1165 err = ubifs_wbuf_sync_nolock(wbuf);
1166 mutex_unlock(&wbuf->io_mutex);
1169 ubifs_ro_mode(c, err);