2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 * Copyright (C) 2006, 2007 University of Szeged, Hungary
7 * SPDX-License-Identifier: GPL-2.0+
9 * Authors: Artem Bityutskiy (Битюцкий Артём)
15 * This file implements UBIFS I/O subsystem which provides various I/O-related
16 * helper functions (reading/writing/checking/validating nodes) and implements
17 * write-buffering support. Write buffers help to save space which otherwise
18 * would have been wasted for padding to the nearest minimal I/O unit boundary.
19 * Instead, data first goes to the write-buffer and is flushed when the
20 * buffer is full or when it is not used for some time (by timer). This is
21 * similar to the mechanism is used by JFFS2.
23 * UBIFS distinguishes between minimum write size (@c->min_io_size) and maximum
24 * write size (@c->max_write_size). The latter is the maximum amount of bytes
25 * the underlying flash is able to program at a time, and writing in
26 * @c->max_write_size units should presumably be faster. Obviously,
27 * @c->min_io_size <= @c->max_write_size. Write-buffers are of
28 * @c->max_write_size bytes in size for maximum performance. However, when a
29 * write-buffer is flushed, only the portion of it (aligned to @c->min_io_size
30 * boundary) which contains data is written, not the whole write-buffer,
31 * because this is more space-efficient.
33 * This optimization adds few complications to the code. Indeed, on the one
34 * hand, we want to write in optimal @c->max_write_size bytes chunks, which
35 * also means aligning writes at the @c->max_write_size bytes offsets. On the
36 * other hand, we do not want to waste space when synchronizing the write
37 * buffer, so during synchronization we writes in smaller chunks. And this makes
38 * the next write offset to be not aligned to @c->max_write_size bytes. So the
39 * have to make sure that the write-buffer offset (@wbuf->offs) becomes aligned
40 * to @c->max_write_size bytes again. We do this by temporarily shrinking
41 * write-buffer size (@wbuf->size).
43 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
44 * mutexes defined inside these objects. Since sometimes upper-level code
45 * has to lock the write-buffer (e.g. journal space reservation code), many
46 * functions related to write-buffers have "nolock" suffix which means that the
47 * caller has to lock the write-buffer before calling this function.
49 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
50 * aligned, UBIFS starts the next node from the aligned address, and the padded
51 * bytes may contain any rubbish. In other words, UBIFS does not put padding
52 * bytes in those small gaps. Common headers of nodes store real node lengths,
53 * not aligned lengths. Indexing nodes also store real lengths in branches.
55 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
56 * uses padding nodes or padding bytes, if the padding node does not fit.
58 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
59 * they are read from the flash media.
63 #include <linux/crc32.h>
64 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/err.h>
72 * ubifs_ro_mode - switch UBIFS to read read-only mode.
73 * @c: UBIFS file-system description object
74 * @err: error code which is the reason of switching to R/O mode
76 void ubifs_ro_mode(struct ubifs_info *c, int err)
80 c->no_chk_data_crc = 0;
81 c->vfs_sb->s_flags |= MS_RDONLY;
82 ubifs_warn("switched to read-only mode, error %d", err);
88 * Below are simple wrappers over UBI I/O functions which include some
89 * additional checks and UBIFS debugging stuff. See corresponding UBI function
90 * for more information.
93 int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
94 int len, int even_ebadmsg)
98 err = ubi_read(c->ubi, lnum, buf, offs, len);
100 * In case of %-EBADMSG print the error message only if the
101 * @even_ebadmsg is true.
103 if (err && (err != -EBADMSG || even_ebadmsg)) {
104 ubifs_err("reading %d bytes from LEB %d:%d failed, error %d",
105 len, lnum, offs, err);
111 int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
116 ubifs_assert(!c->ro_media && !c->ro_mount);
119 if (!dbg_is_tst_rcvry(c))
120 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
122 err = dbg_leb_write(c, lnum, buf, offs, len);
124 ubifs_err("writing %d bytes to LEB %d:%d failed, error %d",
125 len, lnum, offs, err);
126 ubifs_ro_mode(c, err);
132 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
136 ubifs_assert(!c->ro_media && !c->ro_mount);
139 if (!dbg_is_tst_rcvry(c))
140 err = ubi_leb_change(c->ubi, lnum, buf, len);
142 err = dbg_leb_change(c, lnum, buf, len);
144 ubifs_err("changing %d bytes in LEB %d failed, error %d",
146 ubifs_ro_mode(c, err);
152 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
156 ubifs_assert(!c->ro_media && !c->ro_mount);
159 if (!dbg_is_tst_rcvry(c))
160 err = ubi_leb_unmap(c->ubi, lnum);
162 err = dbg_leb_unmap(c, lnum);
164 ubifs_err("unmap LEB %d failed, error %d", lnum, err);
165 ubifs_ro_mode(c, err);
171 int ubifs_leb_map(struct ubifs_info *c, int lnum)
175 ubifs_assert(!c->ro_media && !c->ro_mount);
178 if (!dbg_is_tst_rcvry(c))
179 err = ubi_leb_map(c->ubi, lnum);
181 err = dbg_leb_map(c, lnum);
183 ubifs_err("mapping LEB %d failed, error %d", lnum, err);
184 ubifs_ro_mode(c, err);
190 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
194 err = ubi_is_mapped(c->ubi, lnum);
196 ubifs_err("ubi_is_mapped failed for LEB %d, error %d",
204 * ubifs_check_node - check node.
205 * @c: UBIFS file-system description object
206 * @buf: node to check
207 * @lnum: logical eraseblock number
208 * @offs: offset within the logical eraseblock
209 * @quiet: print no messages
210 * @must_chk_crc: indicates whether to always check the CRC
212 * This function checks node magic number and CRC checksum. This function also
213 * validates node length to prevent UBIFS from becoming crazy when an attacker
214 * feeds it a file-system image with incorrect nodes. For example, too large
215 * node length in the common header could cause UBIFS to read memory outside of
216 * allocated buffer when checking the CRC checksum.
218 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
219 * true, which is controlled by corresponding UBIFS mount option. However, if
220 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
221 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
222 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
223 * is checked. This is because during mounting or re-mounting from R/O mode to
224 * R/W mode we may read journal nodes (when replying the journal or doing the
225 * recovery) and the journal nodes may potentially be corrupted, so checking is
228 * This function returns zero in case of success and %-EUCLEAN in case of bad
231 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
232 int offs, int quiet, int must_chk_crc)
234 int err = -EINVAL, type, node_len;
235 uint32_t crc, node_crc, magic;
236 const struct ubifs_ch *ch = buf;
238 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
239 ubifs_assert(!(offs & 7) && offs < c->leb_size);
241 magic = le32_to_cpu(ch->magic);
242 if (magic != UBIFS_NODE_MAGIC) {
244 ubifs_err("bad magic %#08x, expected %#08x",
245 magic, UBIFS_NODE_MAGIC);
250 type = ch->node_type;
251 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
253 ubifs_err("bad node type %d", type);
257 node_len = le32_to_cpu(ch->len);
258 if (node_len + offs > c->leb_size)
261 if (c->ranges[type].max_len == 0) {
262 if (node_len != c->ranges[type].len)
264 } else if (node_len < c->ranges[type].min_len ||
265 node_len > c->ranges[type].max_len)
268 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
269 !c->remounting_rw && c->no_chk_data_crc)
272 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
273 node_crc = le32_to_cpu(ch->crc);
274 if (crc != node_crc) {
276 ubifs_err("bad CRC: calculated %#08x, read %#08x",
286 ubifs_err("bad node length %d", node_len);
289 ubifs_err("bad node at LEB %d:%d", lnum, offs);
290 ubifs_dump_node(c, buf);
297 * ubifs_pad - pad flash space.
298 * @c: UBIFS file-system description object
299 * @buf: buffer to put padding to
300 * @pad: how many bytes to pad
302 * The flash media obliges us to write only in chunks of %c->min_io_size and
303 * when we have to write less data we add padding node to the write-buffer and
304 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
305 * media is being scanned. If the amount of wasted space is not enough to fit a
306 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
307 * pattern (%UBIFS_PADDING_BYTE).
309 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
312 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
316 ubifs_assert(pad >= 0 && !(pad & 7));
318 if (pad >= UBIFS_PAD_NODE_SZ) {
319 struct ubifs_ch *ch = buf;
320 struct ubifs_pad_node *pad_node = buf;
322 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
323 ch->node_type = UBIFS_PAD_NODE;
324 ch->group_type = UBIFS_NO_NODE_GROUP;
325 ch->padding[0] = ch->padding[1] = 0;
327 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
328 pad -= UBIFS_PAD_NODE_SZ;
329 pad_node->pad_len = cpu_to_le32(pad);
330 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
331 ch->crc = cpu_to_le32(crc);
332 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
334 /* Too little space, padding node won't fit */
335 memset(buf, UBIFS_PADDING_BYTE, pad);
339 * next_sqnum - get next sequence number.
340 * @c: UBIFS file-system description object
342 static unsigned long long next_sqnum(struct ubifs_info *c)
344 unsigned long long sqnum;
346 spin_lock(&c->cnt_lock);
347 sqnum = ++c->max_sqnum;
348 spin_unlock(&c->cnt_lock);
350 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
351 if (sqnum >= SQNUM_WATERMARK) {
352 ubifs_err("sequence number overflow %llu, end of life",
354 ubifs_ro_mode(c, -EINVAL);
356 ubifs_warn("running out of sequence numbers, end of life soon");
363 * ubifs_prepare_node - prepare node to be written to flash.
364 * @c: UBIFS file-system description object
365 * @node: the node to pad
367 * @pad: if the buffer has to be padded
369 * This function prepares node at @node to be written to the media - it
370 * calculates node CRC, fills the common header, and adds proper padding up to
371 * the next minimum I/O unit if @pad is not zero.
373 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
376 struct ubifs_ch *ch = node;
377 unsigned long long sqnum = next_sqnum(c);
379 ubifs_assert(len >= UBIFS_CH_SZ);
381 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
382 ch->len = cpu_to_le32(len);
383 ch->group_type = UBIFS_NO_NODE_GROUP;
384 ch->sqnum = cpu_to_le64(sqnum);
385 ch->padding[0] = ch->padding[1] = 0;
386 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
387 ch->crc = cpu_to_le32(crc);
391 pad = ALIGN(len, c->min_io_size) - len;
392 ubifs_pad(c, node + len, pad);
397 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
398 * @c: UBIFS file-system description object
399 * @node: the node to pad
401 * @last: indicates the last node of the group
403 * This function prepares node at @node to be written to the media - it
404 * calculates node CRC and fills the common header.
406 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
409 struct ubifs_ch *ch = node;
410 unsigned long long sqnum = next_sqnum(c);
412 ubifs_assert(len >= UBIFS_CH_SZ);
414 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
415 ch->len = cpu_to_le32(len);
417 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
419 ch->group_type = UBIFS_IN_NODE_GROUP;
420 ch->sqnum = cpu_to_le64(sqnum);
421 ch->padding[0] = ch->padding[1] = 0;
422 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
423 ch->crc = cpu_to_le32(crc);
428 * wbuf_timer_callback - write-buffer timer callback function.
429 * @data: timer data (write-buffer descriptor)
431 * This function is called when the write-buffer timer expires.
433 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
435 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
437 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
439 wbuf->c->need_wbuf_sync = 1;
440 ubifs_wake_up_bgt(wbuf->c);
441 return HRTIMER_NORESTART;
445 * new_wbuf_timer - start new write-buffer timer.
446 * @wbuf: write-buffer descriptor
448 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
450 ubifs_assert(!hrtimer_active(&wbuf->timer));
454 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
455 dbg_jhead(wbuf->jhead),
456 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
457 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
459 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
465 * cancel_wbuf_timer - cancel write-buffer timer.
466 * @wbuf: write-buffer descriptor
468 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
474 hrtimer_cancel(&wbuf->timer);
479 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
480 * @wbuf: write-buffer to synchronize
482 * This function synchronizes write-buffer @buf and returns zero in case of
483 * success or a negative error code in case of failure.
485 * Note, although write-buffers are of @c->max_write_size, this function does
486 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
487 * if the write-buffer is only partially filled with data, only the used part
488 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
489 * This way we waste less space.
491 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
493 struct ubifs_info *c = wbuf->c;
494 int err, dirt, sync_len;
496 cancel_wbuf_timer_nolock(wbuf);
497 if (!wbuf->used || wbuf->lnum == -1)
498 /* Write-buffer is empty or not seeked */
501 dbg_io("LEB %d:%d, %d bytes, jhead %s",
502 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
503 ubifs_assert(!(wbuf->avail & 7));
504 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
505 ubifs_assert(wbuf->size >= c->min_io_size);
506 ubifs_assert(wbuf->size <= c->max_write_size);
507 ubifs_assert(wbuf->size % c->min_io_size == 0);
508 ubifs_assert(!c->ro_media && !c->ro_mount);
509 if (c->leb_size - wbuf->offs >= c->max_write_size)
510 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
516 * Do not write whole write buffer but write only the minimum necessary
517 * amount of min. I/O units.
519 sync_len = ALIGN(wbuf->used, c->min_io_size);
520 dirt = sync_len - wbuf->used;
522 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
523 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
527 spin_lock(&wbuf->lock);
528 wbuf->offs += sync_len;
530 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
531 * But our goal is to optimize writes and make sure we write in
532 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
533 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
534 * sure that @wbuf->offs + @wbuf->size is aligned to
535 * @c->max_write_size. This way we make sure that after next
536 * write-buffer flush we are again at the optimal offset (aligned to
537 * @c->max_write_size).
539 if (c->leb_size - wbuf->offs < c->max_write_size)
540 wbuf->size = c->leb_size - wbuf->offs;
541 else if (wbuf->offs & (c->max_write_size - 1))
542 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
544 wbuf->size = c->max_write_size;
545 wbuf->avail = wbuf->size;
548 spin_unlock(&wbuf->lock);
550 if (wbuf->sync_callback)
551 err = wbuf->sync_callback(c, wbuf->lnum,
552 c->leb_size - wbuf->offs, dirt);
557 * ubifs_wbuf_seek_nolock - seek write-buffer.
558 * @wbuf: write-buffer
559 * @lnum: logical eraseblock number to seek to
560 * @offs: logical eraseblock offset to seek to
562 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
563 * The write-buffer has to be empty. Returns zero in case of success and a
564 * negative error code in case of failure.
566 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
568 const struct ubifs_info *c = wbuf->c;
570 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
571 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
572 ubifs_assert(offs >= 0 && offs <= c->leb_size);
573 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
574 ubifs_assert(lnum != wbuf->lnum);
575 ubifs_assert(wbuf->used == 0);
577 spin_lock(&wbuf->lock);
580 if (c->leb_size - wbuf->offs < c->max_write_size)
581 wbuf->size = c->leb_size - wbuf->offs;
582 else if (wbuf->offs & (c->max_write_size - 1))
583 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
585 wbuf->size = c->max_write_size;
586 wbuf->avail = wbuf->size;
588 spin_unlock(&wbuf->lock);
595 * ubifs_bg_wbufs_sync - synchronize write-buffers.
596 * @c: UBIFS file-system description object
598 * This function is called by background thread to synchronize write-buffers.
599 * Returns zero in case of success and a negative error code in case of
602 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
606 ubifs_assert(!c->ro_media && !c->ro_mount);
607 if (!c->need_wbuf_sync)
609 c->need_wbuf_sync = 0;
616 dbg_io("synchronize");
617 for (i = 0; i < c->jhead_cnt; i++) {
618 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
623 * If the mutex is locked then wbuf is being changed, so
624 * synchronization is not necessary.
626 if (mutex_is_locked(&wbuf->io_mutex))
629 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
630 if (!wbuf->need_sync) {
631 mutex_unlock(&wbuf->io_mutex);
635 err = ubifs_wbuf_sync_nolock(wbuf);
636 mutex_unlock(&wbuf->io_mutex);
638 ubifs_err("cannot sync write-buffer, error %d", err);
639 ubifs_ro_mode(c, err);
647 /* Cancel all timers to prevent repeated errors */
648 for (i = 0; i < c->jhead_cnt; i++) {
649 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
651 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
652 cancel_wbuf_timer_nolock(wbuf);
653 mutex_unlock(&wbuf->io_mutex);
659 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
660 * @wbuf: write-buffer
661 * @buf: node to write
664 * This function writes data to flash via write-buffer @wbuf. This means that
665 * the last piece of the node won't reach the flash media immediately if it
666 * does not take whole max. write unit (@c->max_write_size). Instead, the node
667 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
668 * because more data are appended to the write-buffer).
670 * This function returns zero in case of success and a negative error code in
671 * case of failure. If the node cannot be written because there is no more
672 * space in this logical eraseblock, %-ENOSPC is returned.
674 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
676 struct ubifs_info *c = wbuf->c;
677 int err, written, n, aligned_len = ALIGN(len, 8);
679 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
680 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
681 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
682 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
683 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
684 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
685 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
686 ubifs_assert(wbuf->size >= c->min_io_size);
687 ubifs_assert(wbuf->size <= c->max_write_size);
688 ubifs_assert(wbuf->size % c->min_io_size == 0);
689 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
690 ubifs_assert(!c->ro_media && !c->ro_mount);
691 ubifs_assert(!c->space_fixup);
692 if (c->leb_size - wbuf->offs >= c->max_write_size)
693 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
695 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
700 cancel_wbuf_timer_nolock(wbuf);
705 if (aligned_len <= wbuf->avail) {
707 * The node is not very large and fits entirely within
710 memcpy(wbuf->buf + wbuf->used, buf, len);
712 if (aligned_len == wbuf->avail) {
713 dbg_io("flush jhead %s wbuf to LEB %d:%d",
714 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
715 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
716 wbuf->offs, wbuf->size);
720 spin_lock(&wbuf->lock);
721 wbuf->offs += wbuf->size;
722 if (c->leb_size - wbuf->offs >= c->max_write_size)
723 wbuf->size = c->max_write_size;
725 wbuf->size = c->leb_size - wbuf->offs;
726 wbuf->avail = wbuf->size;
729 spin_unlock(&wbuf->lock);
731 spin_lock(&wbuf->lock);
732 wbuf->avail -= aligned_len;
733 wbuf->used += aligned_len;
734 spin_unlock(&wbuf->lock);
744 * The node is large enough and does not fit entirely within
745 * current available space. We have to fill and flush
746 * write-buffer and switch to the next max. write unit.
748 dbg_io("flush jhead %s wbuf to LEB %d:%d",
749 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
750 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
751 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
756 wbuf->offs += wbuf->size;
758 aligned_len -= wbuf->avail;
759 written += wbuf->avail;
760 } else if (wbuf->offs & (c->max_write_size - 1)) {
762 * The write-buffer offset is not aligned to
763 * @c->max_write_size and @wbuf->size is less than
764 * @c->max_write_size. Write @wbuf->size bytes to make sure the
765 * following writes are done in optimal @c->max_write_size
768 dbg_io("write %d bytes to LEB %d:%d",
769 wbuf->size, wbuf->lnum, wbuf->offs);
770 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
775 wbuf->offs += wbuf->size;
777 aligned_len -= wbuf->size;
778 written += wbuf->size;
782 * The remaining data may take more whole max. write units, so write the
783 * remains multiple to max. write unit size directly to the flash media.
784 * We align node length to 8-byte boundary because we anyway flash wbuf
785 * if the remaining space is less than 8 bytes.
787 n = aligned_len >> c->max_write_shift;
789 n <<= c->max_write_shift;
790 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
792 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
802 spin_lock(&wbuf->lock);
805 * And now we have what's left and what does not take whole
806 * max. write unit, so write it to the write-buffer and we are
809 memcpy(wbuf->buf, buf + written, len);
811 if (c->leb_size - wbuf->offs >= c->max_write_size)
812 wbuf->size = c->max_write_size;
814 wbuf->size = c->leb_size - wbuf->offs;
815 wbuf->avail = wbuf->size - aligned_len;
816 wbuf->used = aligned_len;
818 spin_unlock(&wbuf->lock);
821 if (wbuf->sync_callback) {
822 int free = c->leb_size - wbuf->offs - wbuf->used;
824 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
830 new_wbuf_timer_nolock(wbuf);
835 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
836 len, wbuf->lnum, wbuf->offs, err);
837 ubifs_dump_node(c, buf);
839 ubifs_dump_leb(c, wbuf->lnum);
844 * ubifs_write_node - write node to the media.
845 * @c: UBIFS file-system description object
846 * @buf: the node to write
848 * @lnum: logical eraseblock number
849 * @offs: offset within the logical eraseblock
851 * This function automatically fills node magic number, assigns sequence
852 * number, and calculates node CRC checksum. The length of the @buf buffer has
853 * to be aligned to the minimal I/O unit size. This function automatically
854 * appends padding node and padding bytes if needed. Returns zero in case of
855 * success and a negative error code in case of failure.
857 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
860 int err, buf_len = ALIGN(len, c->min_io_size);
862 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
863 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
865 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
866 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
867 ubifs_assert(!c->ro_media && !c->ro_mount);
868 ubifs_assert(!c->space_fixup);
873 ubifs_prepare_node(c, buf, len, 1);
874 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
876 ubifs_dump_node(c, buf);
883 * ubifs_read_node_wbuf - read node from the media or write-buffer.
884 * @wbuf: wbuf to check for un-written data
885 * @buf: buffer to read to
888 * @lnum: logical eraseblock number
889 * @offs: offset within the logical eraseblock
891 * This function reads a node of known type and length, checks it and stores
892 * in @buf. If the node partially or fully sits in the write-buffer, this
893 * function takes data from the buffer, otherwise it reads the flash media.
894 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
895 * error code in case of failure.
897 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
900 const struct ubifs_info *c = wbuf->c;
901 int err, rlen, overlap;
902 struct ubifs_ch *ch = buf;
904 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
905 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
906 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
907 ubifs_assert(!(offs & 7) && offs < c->leb_size);
908 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
910 spin_lock(&wbuf->lock);
911 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
913 /* We may safely unlock the write-buffer and read the data */
914 spin_unlock(&wbuf->lock);
915 return ubifs_read_node(c, buf, type, len, lnum, offs);
918 /* Don't read under wbuf */
919 rlen = wbuf->offs - offs;
923 /* Copy the rest from the write-buffer */
924 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
925 spin_unlock(&wbuf->lock);
928 /* Read everything that goes before write-buffer */
929 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
930 if (err && err != -EBADMSG)
934 if (type != ch->node_type) {
935 ubifs_err("bad node type (%d but expected %d)",
936 ch->node_type, type);
940 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
942 ubifs_err("expected node type %d", type);
946 rlen = le32_to_cpu(ch->len);
948 ubifs_err("bad node length %d, expected %d", rlen, len);
955 ubifs_err("bad node at LEB %d:%d", lnum, offs);
956 ubifs_dump_node(c, buf);
962 * ubifs_read_node - read node.
963 * @c: UBIFS file-system description object
964 * @buf: buffer to read to
966 * @len: node length (not aligned)
967 * @lnum: logical eraseblock number
968 * @offs: offset within the logical eraseblock
970 * This function reads a node of known type and and length, checks it and
971 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
972 * and a negative error code in case of failure.
974 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
978 struct ubifs_ch *ch = buf;
980 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
981 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
982 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
983 ubifs_assert(!(offs & 7) && offs < c->leb_size);
984 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
986 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
987 if (err && err != -EBADMSG)
990 if (type != ch->node_type) {
991 ubifs_err("bad node type (%d but expected %d)",
992 ch->node_type, type);
996 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
998 ubifs_err("expected node type %d", type);
1002 l = le32_to_cpu(ch->len);
1004 ubifs_err("bad node length %d, expected %d", l, len);
1011 ubifs_err("bad node at LEB %d:%d, LEB mapping status %d", lnum, offs,
1012 ubi_is_mapped(c->ubi, lnum));
1013 ubifs_dump_node(c, buf);
1019 * ubifs_wbuf_init - initialize write-buffer.
1020 * @c: UBIFS file-system description object
1021 * @wbuf: write-buffer to initialize
1023 * This function initializes write-buffer. Returns zero in case of success
1024 * %-ENOMEM in case of failure.
1026 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1030 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1034 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1035 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1036 if (!wbuf->inodes) {
1043 wbuf->lnum = wbuf->offs = -1;
1045 * If the LEB starts at the max. write size aligned address, then
1046 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1047 * set it to something smaller so that it ends at the closest max.
1048 * write size boundary.
1050 size = c->max_write_size - (c->leb_start % c->max_write_size);
1051 wbuf->avail = wbuf->size = size;
1052 wbuf->sync_callback = NULL;
1053 mutex_init(&wbuf->io_mutex);
1054 spin_lock_init(&wbuf->lock);
1059 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1060 wbuf->timer.function = wbuf_timer_callback_nolock;
1061 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1062 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1063 wbuf->delta *= 1000000000ULL;
1064 ubifs_assert(wbuf->delta <= ULONG_MAX);
1070 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1071 * @wbuf: the write-buffer where to add
1072 * @inum: the inode number
1074 * This function adds an inode number to the inode array of the write-buffer.
1076 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1079 /* NOR flash or something similar */
1082 spin_lock(&wbuf->lock);
1084 wbuf->inodes[wbuf->next_ino++] = inum;
1085 spin_unlock(&wbuf->lock);
1089 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1090 * @wbuf: the write-buffer
1091 * @inum: the inode number
1093 * This function returns with %1 if the write-buffer contains some data from the
1094 * given inode otherwise it returns with %0.
1096 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1100 spin_lock(&wbuf->lock);
1101 for (i = 0; i < wbuf->next_ino; i++)
1102 if (inum == wbuf->inodes[i]) {
1106 spin_unlock(&wbuf->lock);
1112 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1113 * @c: UBIFS file-system description object
1114 * @inode: inode to synchronize
1116 * This function synchronizes write-buffers which contain nodes belonging to
1117 * @inode. Returns zero in case of success and a negative error code in case of
1120 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1124 for (i = 0; i < c->jhead_cnt; i++) {
1125 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1129 * GC head is special, do not look at it. Even if the
1130 * head contains something related to this inode, it is
1131 * a _copy_ of corresponding on-flash node which sits
1136 if (!wbuf_has_ino(wbuf, inode->i_ino))
1139 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1140 if (wbuf_has_ino(wbuf, inode->i_ino))
1141 err = ubifs_wbuf_sync_nolock(wbuf);
1142 mutex_unlock(&wbuf->io_mutex);
1145 ubifs_ro_mode(c, err);