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("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("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, dtype);
128 err = dbg_leb_write(c->ubi, lnum, buf, offs, len, dtype);
130 ubifs_err("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,
143 ubifs_assert(!c->ro_media && !c->ro_mount);
146 if (!dbg_is_tst_rcvry(c))
147 err = ubi_leb_change(c->ubi, lnum, buf, len, dtype);
149 err = dbg_leb_change(c->ubi, lnum, buf, len, dtype);
151 ubifs_err("changing %d bytes in LEB %d failed, error %d",
153 ubifs_ro_mode(c, err);
159 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
163 ubifs_assert(!c->ro_media && !c->ro_mount);
166 if (!dbg_is_tst_rcvry(c))
167 err = ubi_leb_unmap(c->ubi, lnum);
169 err = dbg_leb_unmap(c->ubi, lnum);
171 ubifs_err("unmap LEB %d failed, error %d", lnum, err);
172 ubifs_ro_mode(c, err);
178 int ubifs_leb_map(struct ubifs_info *c, int lnum, int dtype)
182 ubifs_assert(!c->ro_media && !c->ro_mount);
185 if (!dbg_is_tst_rcvry(c))
186 err = ubi_leb_map(c->ubi, lnum, dtype);
188 err = dbg_leb_map(c->ubi, lnum, dtype);
190 ubifs_err("mapping LEB %d failed, error %d", lnum, err);
191 ubifs_ro_mode(c, err);
197 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
201 err = ubi_is_mapped(c->ubi, lnum);
203 ubifs_err("ubi_is_mapped failed for LEB %d, error %d",
211 * ubifs_check_node - check node.
212 * @c: UBIFS file-system description object
213 * @buf: node to check
214 * @lnum: logical eraseblock number
215 * @offs: offset within the logical eraseblock
216 * @quiet: print no messages
217 * @must_chk_crc: indicates whether to always check the CRC
219 * This function checks node magic number and CRC checksum. This function also
220 * validates node length to prevent UBIFS from becoming crazy when an attacker
221 * feeds it a file-system image with incorrect nodes. For example, too large
222 * node length in the common header could cause UBIFS to read memory outside of
223 * allocated buffer when checking the CRC checksum.
225 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
226 * true, which is controlled by corresponding UBIFS mount option. However, if
227 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
228 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
229 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
230 * is checked. This is because during mounting or re-mounting from R/O mode to
231 * R/W mode we may read journal nodes (when replying the journal or doing the
232 * recovery) and the journal nodes may potentially be corrupted, so checking is
235 * This function returns zero in case of success and %-EUCLEAN in case of bad
238 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
239 int offs, int quiet, int must_chk_crc)
241 int err = -EINVAL, type, node_len;
242 uint32_t crc, node_crc, magic;
243 const struct ubifs_ch *ch = buf;
245 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
246 ubifs_assert(!(offs & 7) && offs < c->leb_size);
248 magic = le32_to_cpu(ch->magic);
249 if (magic != UBIFS_NODE_MAGIC) {
251 ubifs_err("bad magic %#08x, expected %#08x",
252 magic, UBIFS_NODE_MAGIC);
257 type = ch->node_type;
258 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
260 ubifs_err("bad node type %d", type);
264 node_len = le32_to_cpu(ch->len);
265 if (node_len + offs > c->leb_size)
268 if (c->ranges[type].max_len == 0) {
269 if (node_len != c->ranges[type].len)
271 } else if (node_len < c->ranges[type].min_len ||
272 node_len > c->ranges[type].max_len)
275 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
276 !c->remounting_rw && c->no_chk_data_crc)
279 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
280 node_crc = le32_to_cpu(ch->crc);
281 if (crc != node_crc) {
283 ubifs_err("bad CRC: calculated %#08x, read %#08x",
293 ubifs_err("bad node length %d", node_len);
296 ubifs_err("bad node at LEB %d:%d", lnum, offs);
297 dbg_dump_node(c, buf);
304 * ubifs_pad - pad flash space.
305 * @c: UBIFS file-system description object
306 * @buf: buffer to put padding to
307 * @pad: how many bytes to pad
309 * The flash media obliges us to write only in chunks of %c->min_io_size and
310 * when we have to write less data we add padding node to the write-buffer and
311 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
312 * media is being scanned. If the amount of wasted space is not enough to fit a
313 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
314 * pattern (%UBIFS_PADDING_BYTE).
316 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
319 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
323 ubifs_assert(pad >= 0 && !(pad & 7));
325 if (pad >= UBIFS_PAD_NODE_SZ) {
326 struct ubifs_ch *ch = buf;
327 struct ubifs_pad_node *pad_node = buf;
329 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
330 ch->node_type = UBIFS_PAD_NODE;
331 ch->group_type = UBIFS_NO_NODE_GROUP;
332 ch->padding[0] = ch->padding[1] = 0;
334 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
335 pad -= UBIFS_PAD_NODE_SZ;
336 pad_node->pad_len = cpu_to_le32(pad);
337 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
338 ch->crc = cpu_to_le32(crc);
339 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
341 /* Too little space, padding node won't fit */
342 memset(buf, UBIFS_PADDING_BYTE, pad);
346 * next_sqnum - get next sequence number.
347 * @c: UBIFS file-system description object
349 static unsigned long long next_sqnum(struct ubifs_info *c)
351 unsigned long long sqnum;
353 spin_lock(&c->cnt_lock);
354 sqnum = ++c->max_sqnum;
355 spin_unlock(&c->cnt_lock);
357 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
358 if (sqnum >= SQNUM_WATERMARK) {
359 ubifs_err("sequence number overflow %llu, end of life",
361 ubifs_ro_mode(c, -EINVAL);
363 ubifs_warn("running out of sequence numbers, end of life soon");
370 * ubifs_prepare_node - prepare node to be written to flash.
371 * @c: UBIFS file-system description object
372 * @node: the node to pad
374 * @pad: if the buffer has to be padded
376 * This function prepares node at @node to be written to the media - it
377 * calculates node CRC, fills the common header, and adds proper padding up to
378 * the next minimum I/O unit if @pad is not zero.
380 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
383 struct ubifs_ch *ch = node;
384 unsigned long long sqnum = next_sqnum(c);
386 ubifs_assert(len >= UBIFS_CH_SZ);
388 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
389 ch->len = cpu_to_le32(len);
390 ch->group_type = UBIFS_NO_NODE_GROUP;
391 ch->sqnum = cpu_to_le64(sqnum);
392 ch->padding[0] = ch->padding[1] = 0;
393 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
394 ch->crc = cpu_to_le32(crc);
398 pad = ALIGN(len, c->min_io_size) - len;
399 ubifs_pad(c, node + len, pad);
404 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
405 * @c: UBIFS file-system description object
406 * @node: the node to pad
408 * @last: indicates the last node of the group
410 * This function prepares node at @node to be written to the media - it
411 * calculates node CRC and fills the common header.
413 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
416 struct ubifs_ch *ch = node;
417 unsigned long long sqnum = next_sqnum(c);
419 ubifs_assert(len >= UBIFS_CH_SZ);
421 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
422 ch->len = cpu_to_le32(len);
424 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
426 ch->group_type = UBIFS_IN_NODE_GROUP;
427 ch->sqnum = cpu_to_le64(sqnum);
428 ch->padding[0] = ch->padding[1] = 0;
429 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
430 ch->crc = cpu_to_le32(crc);
434 * wbuf_timer_callback - write-buffer timer callback function.
435 * @data: timer data (write-buffer descriptor)
437 * This function is called when the write-buffer timer expires.
439 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
441 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
443 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
445 wbuf->c->need_wbuf_sync = 1;
446 ubifs_wake_up_bgt(wbuf->c);
447 return HRTIMER_NORESTART;
451 * new_wbuf_timer - start new write-buffer timer.
452 * @wbuf: write-buffer descriptor
454 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
456 ubifs_assert(!hrtimer_active(&wbuf->timer));
460 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
461 dbg_jhead(wbuf->jhead),
462 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
463 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
465 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
470 * cancel_wbuf_timer - cancel write-buffer timer.
471 * @wbuf: write-buffer descriptor
473 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
478 hrtimer_cancel(&wbuf->timer);
482 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
483 * @wbuf: write-buffer to synchronize
485 * This function synchronizes write-buffer @buf and returns zero in case of
486 * success or a negative error code in case of failure.
488 * Note, although write-buffers are of @c->max_write_size, this function does
489 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
490 * if the write-buffer is only partially filled with data, only the used part
491 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
492 * This way we waste less space.
494 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
496 struct ubifs_info *c = wbuf->c;
497 int err, dirt, sync_len;
499 cancel_wbuf_timer_nolock(wbuf);
500 if (!wbuf->used || wbuf->lnum == -1)
501 /* Write-buffer is empty or not seeked */
504 dbg_io("LEB %d:%d, %d bytes, jhead %s",
505 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
506 ubifs_assert(!(wbuf->avail & 7));
507 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
508 ubifs_assert(wbuf->size >= c->min_io_size);
509 ubifs_assert(wbuf->size <= c->max_write_size);
510 ubifs_assert(wbuf->size % c->min_io_size == 0);
511 ubifs_assert(!c->ro_media && !c->ro_mount);
512 if (c->leb_size - wbuf->offs >= c->max_write_size)
513 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
519 * Do not write whole write buffer but write only the minimum necessary
520 * amount of min. I/O units.
522 sync_len = ALIGN(wbuf->used, c->min_io_size);
523 dirt = sync_len - wbuf->used;
525 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
526 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len,
531 spin_lock(&wbuf->lock);
532 wbuf->offs += sync_len;
534 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
535 * But our goal is to optimize writes and make sure we write in
536 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
537 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
538 * sure that @wbuf->offs + @wbuf->size is aligned to
539 * @c->max_write_size. This way we make sure that after next
540 * write-buffer flush we are again at the optimal offset (aligned to
541 * @c->max_write_size).
543 if (c->leb_size - wbuf->offs < c->max_write_size)
544 wbuf->size = c->leb_size - wbuf->offs;
545 else if (wbuf->offs & (c->max_write_size - 1))
546 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
548 wbuf->size = c->max_write_size;
549 wbuf->avail = wbuf->size;
552 spin_unlock(&wbuf->lock);
554 if (wbuf->sync_callback)
555 err = wbuf->sync_callback(c, wbuf->lnum,
556 c->leb_size - wbuf->offs, dirt);
561 * ubifs_wbuf_seek_nolock - seek write-buffer.
562 * @wbuf: write-buffer
563 * @lnum: logical eraseblock number to seek to
564 * @offs: logical eraseblock offset to seek to
567 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
568 * The write-buffer has to be empty. Returns zero in case of success and a
569 * negative error code in case of failure.
571 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs,
574 const struct ubifs_info *c = wbuf->c;
576 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
577 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
578 ubifs_assert(offs >= 0 && offs <= c->leb_size);
579 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
580 ubifs_assert(lnum != wbuf->lnum);
581 ubifs_assert(wbuf->used == 0);
583 spin_lock(&wbuf->lock);
586 if (c->leb_size - wbuf->offs < c->max_write_size)
587 wbuf->size = c->leb_size - wbuf->offs;
588 else if (wbuf->offs & (c->max_write_size - 1))
589 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
591 wbuf->size = c->max_write_size;
592 wbuf->avail = wbuf->size;
594 spin_unlock(&wbuf->lock);
601 * ubifs_bg_wbufs_sync - synchronize write-buffers.
602 * @c: UBIFS file-system description object
604 * This function is called by background thread to synchronize write-buffers.
605 * Returns zero in case of success and a negative error code in case of
608 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
612 ubifs_assert(!c->ro_media && !c->ro_mount);
613 if (!c->need_wbuf_sync)
615 c->need_wbuf_sync = 0;
622 dbg_io("synchronize");
623 for (i = 0; i < c->jhead_cnt; i++) {
624 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
629 * If the mutex is locked then wbuf is being changed, so
630 * synchronization is not necessary.
632 if (mutex_is_locked(&wbuf->io_mutex))
635 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
636 if (!wbuf->need_sync) {
637 mutex_unlock(&wbuf->io_mutex);
641 err = ubifs_wbuf_sync_nolock(wbuf);
642 mutex_unlock(&wbuf->io_mutex);
644 ubifs_err("cannot sync write-buffer, error %d", err);
645 ubifs_ro_mode(c, err);
653 /* Cancel all timers to prevent repeated errors */
654 for (i = 0; i < c->jhead_cnt; i++) {
655 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
657 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
658 cancel_wbuf_timer_nolock(wbuf);
659 mutex_unlock(&wbuf->io_mutex);
665 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
666 * @wbuf: write-buffer
667 * @buf: node to write
670 * This function writes data to flash via write-buffer @wbuf. This means that
671 * the last piece of the node won't reach the flash media immediately if it
672 * does not take whole max. write unit (@c->max_write_size). Instead, the node
673 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
674 * because more data are appended to the write-buffer).
676 * This function returns zero in case of success and a negative error code in
677 * case of failure. If the node cannot be written because there is no more
678 * space in this logical eraseblock, %-ENOSPC is returned.
680 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
682 struct ubifs_info *c = wbuf->c;
683 int err, written, n, aligned_len = ALIGN(len, 8);
685 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
686 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
687 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
688 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
689 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
690 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
691 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
692 ubifs_assert(wbuf->size >= c->min_io_size);
693 ubifs_assert(wbuf->size <= c->max_write_size);
694 ubifs_assert(wbuf->size % c->min_io_size == 0);
695 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
696 ubifs_assert(!c->ro_media && !c->ro_mount);
697 ubifs_assert(!c->space_fixup);
698 if (c->leb_size - wbuf->offs >= c->max_write_size)
699 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
701 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
706 cancel_wbuf_timer_nolock(wbuf);
711 if (aligned_len <= wbuf->avail) {
713 * The node is not very large and fits entirely within
716 memcpy(wbuf->buf + wbuf->used, buf, len);
718 if (aligned_len == wbuf->avail) {
719 dbg_io("flush jhead %s wbuf to LEB %d:%d",
720 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
721 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
722 wbuf->offs, wbuf->size,
727 spin_lock(&wbuf->lock);
728 wbuf->offs += wbuf->size;
729 if (c->leb_size - wbuf->offs >= c->max_write_size)
730 wbuf->size = c->max_write_size;
732 wbuf->size = c->leb_size - wbuf->offs;
733 wbuf->avail = wbuf->size;
736 spin_unlock(&wbuf->lock);
738 spin_lock(&wbuf->lock);
739 wbuf->avail -= aligned_len;
740 wbuf->used += aligned_len;
741 spin_unlock(&wbuf->lock);
751 * The node is large enough and does not fit entirely within
752 * current available space. We have to fill and flush
753 * write-buffer and switch to the next max. write unit.
755 dbg_io("flush jhead %s wbuf to LEB %d:%d",
756 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
757 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
758 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
759 wbuf->size, wbuf->dtype);
763 wbuf->offs += wbuf->size;
765 aligned_len -= wbuf->avail;
766 written += wbuf->avail;
767 } else if (wbuf->offs & (c->max_write_size - 1)) {
769 * The write-buffer offset is not aligned to
770 * @c->max_write_size and @wbuf->size is less than
771 * @c->max_write_size. Write @wbuf->size bytes to make sure the
772 * following writes are done in optimal @c->max_write_size
775 dbg_io("write %d bytes to LEB %d:%d",
776 wbuf->size, wbuf->lnum, wbuf->offs);
777 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
778 wbuf->size, wbuf->dtype);
782 wbuf->offs += wbuf->size;
784 aligned_len -= wbuf->size;
785 written += wbuf->size;
789 * The remaining data may take more whole max. write units, so write the
790 * remains multiple to max. write unit size directly to the flash media.
791 * We align node length to 8-byte boundary because we anyway flash wbuf
792 * if the remaining space is less than 8 bytes.
794 n = aligned_len >> c->max_write_shift;
796 n <<= c->max_write_shift;
797 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
799 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
800 wbuf->offs, n, wbuf->dtype);
809 spin_lock(&wbuf->lock);
812 * And now we have what's left and what does not take whole
813 * max. write unit, so write it to the write-buffer and we are
816 memcpy(wbuf->buf, buf + written, len);
818 if (c->leb_size - wbuf->offs >= c->max_write_size)
819 wbuf->size = c->max_write_size;
821 wbuf->size = c->leb_size - wbuf->offs;
822 wbuf->avail = wbuf->size - aligned_len;
823 wbuf->used = aligned_len;
825 spin_unlock(&wbuf->lock);
828 if (wbuf->sync_callback) {
829 int free = c->leb_size - wbuf->offs - wbuf->used;
831 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
837 new_wbuf_timer_nolock(wbuf);
842 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
843 len, wbuf->lnum, wbuf->offs, err);
844 dbg_dump_node(c, buf);
846 dbg_dump_leb(c, wbuf->lnum);
851 * ubifs_write_node - write node to the media.
852 * @c: UBIFS file-system description object
853 * @buf: the node to write
855 * @lnum: logical eraseblock number
856 * @offs: offset within the logical eraseblock
857 * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
859 * This function automatically fills node magic number, assigns sequence
860 * number, and calculates node CRC checksum. The length of the @buf buffer has
861 * to be aligned to the minimal I/O unit size. This function automatically
862 * appends padding node and padding bytes if needed. Returns zero in case of
863 * success and a negative error code in case of failure.
865 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
868 int err, buf_len = ALIGN(len, c->min_io_size);
870 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
871 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
873 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
874 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
875 ubifs_assert(!c->ro_media && !c->ro_mount);
876 ubifs_assert(!c->space_fixup);
881 ubifs_prepare_node(c, buf, len, 1);
882 err = ubifs_leb_write(c, lnum, buf, offs, buf_len, dtype);
884 dbg_dump_node(c, buf);
890 * ubifs_read_node_wbuf - read node from the media or write-buffer.
891 * @wbuf: wbuf to check for un-written data
892 * @buf: buffer to read to
895 * @lnum: logical eraseblock number
896 * @offs: offset within the logical eraseblock
898 * This function reads a node of known type and length, checks it and stores
899 * in @buf. If the node partially or fully sits in the write-buffer, this
900 * function takes data from the buffer, otherwise it reads the flash media.
901 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
902 * error code in case of failure.
904 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
907 const struct ubifs_info *c = wbuf->c;
908 int err, rlen, overlap;
909 struct ubifs_ch *ch = buf;
911 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
912 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
913 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
914 ubifs_assert(!(offs & 7) && offs < c->leb_size);
915 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
917 spin_lock(&wbuf->lock);
918 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
920 /* We may safely unlock the write-buffer and read the data */
921 spin_unlock(&wbuf->lock);
922 return ubifs_read_node(c, buf, type, len, lnum, offs);
925 /* Don't read under wbuf */
926 rlen = wbuf->offs - offs;
930 /* Copy the rest from the write-buffer */
931 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
932 spin_unlock(&wbuf->lock);
935 /* Read everything that goes before write-buffer */
936 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
937 if (err && err != -EBADMSG)
941 if (type != ch->node_type) {
942 ubifs_err("bad node type (%d but expected %d)",
943 ch->node_type, type);
947 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
949 ubifs_err("expected node type %d", type);
953 rlen = le32_to_cpu(ch->len);
955 ubifs_err("bad node length %d, expected %d", rlen, len);
962 ubifs_err("bad node at LEB %d:%d", lnum, offs);
963 dbg_dump_node(c, buf);
969 * ubifs_read_node - read node.
970 * @c: UBIFS file-system description object
971 * @buf: buffer to read to
973 * @len: node length (not aligned)
974 * @lnum: logical eraseblock number
975 * @offs: offset within the logical eraseblock
977 * This function reads a node of known type and and length, checks it and
978 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
979 * and a negative error code in case of failure.
981 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
985 struct ubifs_ch *ch = buf;
987 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
988 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
989 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
990 ubifs_assert(!(offs & 7) && offs < c->leb_size);
991 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
993 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
994 if (err && err != -EBADMSG)
997 if (type != ch->node_type) {
998 ubifs_err("bad node type (%d but expected %d)",
999 ch->node_type, type);
1003 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1005 ubifs_err("expected node type %d", type);
1009 l = le32_to_cpu(ch->len);
1011 ubifs_err("bad node length %d, expected %d", l, len);
1018 ubifs_err("bad node at LEB %d:%d, LEB mapping status %d", lnum, offs,
1019 ubi_is_mapped(c->ubi, lnum));
1020 dbg_dump_node(c, buf);
1026 * ubifs_wbuf_init - initialize write-buffer.
1027 * @c: UBIFS file-system description object
1028 * @wbuf: write-buffer to initialize
1030 * This function initializes write-buffer. Returns zero in case of success
1031 * %-ENOMEM in case of failure.
1033 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1037 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1041 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1042 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1043 if (!wbuf->inodes) {
1050 wbuf->lnum = wbuf->offs = -1;
1052 * If the LEB starts at the max. write size aligned address, then
1053 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1054 * set it to something smaller so that it ends at the closest max.
1055 * write size boundary.
1057 size = c->max_write_size - (c->leb_start % c->max_write_size);
1058 wbuf->avail = wbuf->size = size;
1059 wbuf->dtype = UBI_UNKNOWN;
1060 wbuf->sync_callback = NULL;
1061 mutex_init(&wbuf->io_mutex);
1062 spin_lock_init(&wbuf->lock);
1066 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1067 wbuf->timer.function = wbuf_timer_callback_nolock;
1068 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1069 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1070 wbuf->delta *= 1000000000ULL;
1071 ubifs_assert(wbuf->delta <= ULONG_MAX);
1076 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1077 * @wbuf: the write-buffer where to add
1078 * @inum: the inode number
1080 * This function adds an inode number to the inode array of the write-buffer.
1082 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1085 /* NOR flash or something similar */
1088 spin_lock(&wbuf->lock);
1090 wbuf->inodes[wbuf->next_ino++] = inum;
1091 spin_unlock(&wbuf->lock);
1095 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1096 * @wbuf: the write-buffer
1097 * @inum: the inode number
1099 * This function returns with %1 if the write-buffer contains some data from the
1100 * given inode otherwise it returns with %0.
1102 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1106 spin_lock(&wbuf->lock);
1107 for (i = 0; i < wbuf->next_ino; i++)
1108 if (inum == wbuf->inodes[i]) {
1112 spin_unlock(&wbuf->lock);
1118 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1119 * @c: UBIFS file-system description object
1120 * @inode: inode to synchronize
1122 * This function synchronizes write-buffers which contain nodes belonging to
1123 * @inode. Returns zero in case of success and a negative error code in case of
1126 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1130 for (i = 0; i < c->jhead_cnt; i++) {
1131 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1135 * GC head is special, do not look at it. Even if the
1136 * head contains something related to this inode, it is
1137 * a _copy_ of corresponding on-flash node which sits
1142 if (!wbuf_has_ino(wbuf, inode->i_ino))
1145 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1146 if (wbuf_has_ino(wbuf, inode->i_ino))
1147 err = ubifs_wbuf_sync_nolock(wbuf);
1148 mutex_unlock(&wbuf->io_mutex);
1151 ubifs_ro_mode(c, err);