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.
64 #include <linux/crc32.h>
65 #include <linux/slab.h>
67 #include <linux/compat.h>
68 #include <linux/err.h>
73 * ubifs_ro_mode - switch UBIFS to read read-only mode.
74 * @c: UBIFS file-system description object
75 * @err: error code which is the reason of switching to R/O mode
77 void ubifs_ro_mode(struct ubifs_info *c, int err)
81 c->no_chk_data_crc = 0;
82 c->vfs_sb->s_flags |= MS_RDONLY;
83 ubifs_warn("switched to read-only mode, error %d", err);
89 * Below are simple wrappers over UBI I/O functions which include some
90 * additional checks and UBIFS debugging stuff. See corresponding UBI function
91 * for more information.
94 int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
95 int len, int even_ebadmsg)
99 err = ubi_read(c->ubi, lnum, buf, offs, len);
101 * In case of %-EBADMSG print the error message only if the
102 * @even_ebadmsg is true.
104 if (err && (err != -EBADMSG || even_ebadmsg)) {
105 ubifs_err("reading %d bytes from LEB %d:%d failed, error %d",
106 len, lnum, offs, err);
112 int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
117 ubifs_assert(!c->ro_media && !c->ro_mount);
120 if (!dbg_is_tst_rcvry(c))
121 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
123 err = dbg_leb_write(c, lnum, buf, offs, len);
125 ubifs_err("writing %d bytes to LEB %d:%d failed, error %d",
126 len, lnum, offs, err);
127 ubifs_ro_mode(c, err);
133 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
137 ubifs_assert(!c->ro_media && !c->ro_mount);
140 if (!dbg_is_tst_rcvry(c))
141 err = ubi_leb_change(c->ubi, lnum, buf, len);
143 err = dbg_leb_change(c, lnum, buf, len);
145 ubifs_err("changing %d bytes in LEB %d failed, error %d",
147 ubifs_ro_mode(c, err);
153 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
157 ubifs_assert(!c->ro_media && !c->ro_mount);
160 if (!dbg_is_tst_rcvry(c))
161 err = ubi_leb_unmap(c->ubi, lnum);
163 err = dbg_leb_unmap(c, lnum);
165 ubifs_err("unmap LEB %d failed, error %d", lnum, err);
166 ubifs_ro_mode(c, err);
172 int ubifs_leb_map(struct ubifs_info *c, int lnum)
176 ubifs_assert(!c->ro_media && !c->ro_mount);
179 if (!dbg_is_tst_rcvry(c))
180 err = ubi_leb_map(c->ubi, lnum);
182 err = dbg_leb_map(c, lnum);
184 ubifs_err("mapping LEB %d failed, error %d", lnum, err);
185 ubifs_ro_mode(c, err);
191 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
195 err = ubi_is_mapped(c->ubi, lnum);
197 ubifs_err("ubi_is_mapped failed for LEB %d, error %d",
205 * ubifs_check_node - check node.
206 * @c: UBIFS file-system description object
207 * @buf: node to check
208 * @lnum: logical eraseblock number
209 * @offs: offset within the logical eraseblock
210 * @quiet: print no messages
211 * @must_chk_crc: indicates whether to always check the CRC
213 * This function checks node magic number and CRC checksum. This function also
214 * validates node length to prevent UBIFS from becoming crazy when an attacker
215 * feeds it a file-system image with incorrect nodes. For example, too large
216 * node length in the common header could cause UBIFS to read memory outside of
217 * allocated buffer when checking the CRC checksum.
219 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
220 * true, which is controlled by corresponding UBIFS mount option. However, if
221 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
222 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
223 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
224 * is checked. This is because during mounting or re-mounting from R/O mode to
225 * R/W mode we may read journal nodes (when replying the journal or doing the
226 * recovery) and the journal nodes may potentially be corrupted, so checking is
229 * This function returns zero in case of success and %-EUCLEAN in case of bad
232 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
233 int offs, int quiet, int must_chk_crc)
235 int err = -EINVAL, type, node_len;
236 uint32_t crc, node_crc, magic;
237 const struct ubifs_ch *ch = buf;
239 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
240 ubifs_assert(!(offs & 7) && offs < c->leb_size);
242 magic = le32_to_cpu(ch->magic);
243 if (magic != UBIFS_NODE_MAGIC) {
245 ubifs_err("bad magic %#08x, expected %#08x",
246 magic, UBIFS_NODE_MAGIC);
251 type = ch->node_type;
252 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
254 ubifs_err("bad node type %d", type);
258 node_len = le32_to_cpu(ch->len);
259 if (node_len + offs > c->leb_size)
262 if (c->ranges[type].max_len == 0) {
263 if (node_len != c->ranges[type].len)
265 } else if (node_len < c->ranges[type].min_len ||
266 node_len > c->ranges[type].max_len)
269 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
270 !c->remounting_rw && c->no_chk_data_crc)
273 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
274 node_crc = le32_to_cpu(ch->crc);
275 if (crc != node_crc) {
277 ubifs_err("bad CRC: calculated %#08x, read %#08x",
287 ubifs_err("bad node length %d", node_len);
290 ubifs_err("bad node at LEB %d:%d", lnum, offs);
291 ubifs_dump_node(c, buf);
298 * ubifs_pad - pad flash space.
299 * @c: UBIFS file-system description object
300 * @buf: buffer to put padding to
301 * @pad: how many bytes to pad
303 * The flash media obliges us to write only in chunks of %c->min_io_size and
304 * when we have to write less data we add padding node to the write-buffer and
305 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
306 * media is being scanned. If the amount of wasted space is not enough to fit a
307 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
308 * pattern (%UBIFS_PADDING_BYTE).
310 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
313 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
317 ubifs_assert(pad >= 0 && !(pad & 7));
319 if (pad >= UBIFS_PAD_NODE_SZ) {
320 struct ubifs_ch *ch = buf;
321 struct ubifs_pad_node *pad_node = buf;
323 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
324 ch->node_type = UBIFS_PAD_NODE;
325 ch->group_type = UBIFS_NO_NODE_GROUP;
326 ch->padding[0] = ch->padding[1] = 0;
328 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
329 pad -= UBIFS_PAD_NODE_SZ;
330 pad_node->pad_len = cpu_to_le32(pad);
331 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
332 ch->crc = cpu_to_le32(crc);
333 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
335 /* Too little space, padding node won't fit */
336 memset(buf, UBIFS_PADDING_BYTE, pad);
340 * next_sqnum - get next sequence number.
341 * @c: UBIFS file-system description object
343 static unsigned long long next_sqnum(struct ubifs_info *c)
345 unsigned long long sqnum;
347 spin_lock(&c->cnt_lock);
348 sqnum = ++c->max_sqnum;
349 spin_unlock(&c->cnt_lock);
351 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
352 if (sqnum >= SQNUM_WATERMARK) {
353 ubifs_err("sequence number overflow %llu, end of life",
355 ubifs_ro_mode(c, -EINVAL);
357 ubifs_warn("running out of sequence numbers, end of life soon");
364 * ubifs_prepare_node - prepare node to be written to flash.
365 * @c: UBIFS file-system description object
366 * @node: the node to pad
368 * @pad: if the buffer has to be padded
370 * This function prepares node at @node to be written to the media - it
371 * calculates node CRC, fills the common header, and adds proper padding up to
372 * the next minimum I/O unit if @pad is not zero.
374 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
377 struct ubifs_ch *ch = node;
378 unsigned long long sqnum = next_sqnum(c);
380 ubifs_assert(len >= UBIFS_CH_SZ);
382 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
383 ch->len = cpu_to_le32(len);
384 ch->group_type = UBIFS_NO_NODE_GROUP;
385 ch->sqnum = cpu_to_le64(sqnum);
386 ch->padding[0] = ch->padding[1] = 0;
387 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
388 ch->crc = cpu_to_le32(crc);
392 pad = ALIGN(len, c->min_io_size) - len;
393 ubifs_pad(c, node + len, pad);
398 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
399 * @c: UBIFS file-system description object
400 * @node: the node to pad
402 * @last: indicates the last node of the group
404 * This function prepares node at @node to be written to the media - it
405 * calculates node CRC and fills the common header.
407 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
410 struct ubifs_ch *ch = node;
411 unsigned long long sqnum = next_sqnum(c);
413 ubifs_assert(len >= UBIFS_CH_SZ);
415 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
416 ch->len = cpu_to_le32(len);
418 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
420 ch->group_type = UBIFS_IN_NODE_GROUP;
421 ch->sqnum = cpu_to_le64(sqnum);
422 ch->padding[0] = ch->padding[1] = 0;
423 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
424 ch->crc = cpu_to_le32(crc);
429 * wbuf_timer_callback - write-buffer timer callback function.
430 * @data: timer data (write-buffer descriptor)
432 * This function is called when the write-buffer timer expires.
434 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
436 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
438 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
440 wbuf->c->need_wbuf_sync = 1;
441 ubifs_wake_up_bgt(wbuf->c);
442 return HRTIMER_NORESTART;
446 * new_wbuf_timer - start new write-buffer timer.
447 * @wbuf: write-buffer descriptor
449 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
451 ubifs_assert(!hrtimer_active(&wbuf->timer));
455 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
456 dbg_jhead(wbuf->jhead),
457 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
458 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
460 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
466 * cancel_wbuf_timer - cancel write-buffer timer.
467 * @wbuf: write-buffer descriptor
469 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
475 hrtimer_cancel(&wbuf->timer);
480 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
481 * @wbuf: write-buffer to synchronize
483 * This function synchronizes write-buffer @buf and returns zero in case of
484 * success or a negative error code in case of failure.
486 * Note, although write-buffers are of @c->max_write_size, this function does
487 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
488 * if the write-buffer is only partially filled with data, only the used part
489 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
490 * This way we waste less space.
492 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
494 struct ubifs_info *c = wbuf->c;
495 int err, dirt, sync_len;
497 cancel_wbuf_timer_nolock(wbuf);
498 if (!wbuf->used || wbuf->lnum == -1)
499 /* Write-buffer is empty or not seeked */
502 dbg_io("LEB %d:%d, %d bytes, jhead %s",
503 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
504 ubifs_assert(!(wbuf->avail & 7));
505 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
506 ubifs_assert(wbuf->size >= c->min_io_size);
507 ubifs_assert(wbuf->size <= c->max_write_size);
508 ubifs_assert(wbuf->size % c->min_io_size == 0);
509 ubifs_assert(!c->ro_media && !c->ro_mount);
510 if (c->leb_size - wbuf->offs >= c->max_write_size)
511 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
517 * Do not write whole write buffer but write only the minimum necessary
518 * amount of min. I/O units.
520 sync_len = ALIGN(wbuf->used, c->min_io_size);
521 dirt = sync_len - wbuf->used;
523 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
524 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
528 spin_lock(&wbuf->lock);
529 wbuf->offs += sync_len;
531 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
532 * But our goal is to optimize writes and make sure we write in
533 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
534 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
535 * sure that @wbuf->offs + @wbuf->size is aligned to
536 * @c->max_write_size. This way we make sure that after next
537 * write-buffer flush we are again at the optimal offset (aligned to
538 * @c->max_write_size).
540 if (c->leb_size - wbuf->offs < c->max_write_size)
541 wbuf->size = c->leb_size - wbuf->offs;
542 else if (wbuf->offs & (c->max_write_size - 1))
543 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
545 wbuf->size = c->max_write_size;
546 wbuf->avail = wbuf->size;
549 spin_unlock(&wbuf->lock);
551 if (wbuf->sync_callback)
552 err = wbuf->sync_callback(c, wbuf->lnum,
553 c->leb_size - wbuf->offs, dirt);
558 * ubifs_wbuf_seek_nolock - seek write-buffer.
559 * @wbuf: write-buffer
560 * @lnum: logical eraseblock number to seek to
561 * @offs: logical eraseblock offset to seek to
563 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
564 * The write-buffer has to be empty. Returns zero in case of success and a
565 * negative error code in case of failure.
567 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
569 const struct ubifs_info *c = wbuf->c;
571 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
572 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
573 ubifs_assert(offs >= 0 && offs <= c->leb_size);
574 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
575 ubifs_assert(lnum != wbuf->lnum);
576 ubifs_assert(wbuf->used == 0);
578 spin_lock(&wbuf->lock);
581 if (c->leb_size - wbuf->offs < c->max_write_size)
582 wbuf->size = c->leb_size - wbuf->offs;
583 else if (wbuf->offs & (c->max_write_size - 1))
584 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
586 wbuf->size = c->max_write_size;
587 wbuf->avail = wbuf->size;
589 spin_unlock(&wbuf->lock);
596 * ubifs_bg_wbufs_sync - synchronize write-buffers.
597 * @c: UBIFS file-system description object
599 * This function is called by background thread to synchronize write-buffers.
600 * Returns zero in case of success and a negative error code in case of
603 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
607 ubifs_assert(!c->ro_media && !c->ro_mount);
608 if (!c->need_wbuf_sync)
610 c->need_wbuf_sync = 0;
617 dbg_io("synchronize");
618 for (i = 0; i < c->jhead_cnt; i++) {
619 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
624 * If the mutex is locked then wbuf is being changed, so
625 * synchronization is not necessary.
627 if (mutex_is_locked(&wbuf->io_mutex))
630 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
631 if (!wbuf->need_sync) {
632 mutex_unlock(&wbuf->io_mutex);
636 err = ubifs_wbuf_sync_nolock(wbuf);
637 mutex_unlock(&wbuf->io_mutex);
639 ubifs_err("cannot sync write-buffer, error %d", err);
640 ubifs_ro_mode(c, err);
648 /* Cancel all timers to prevent repeated errors */
649 for (i = 0; i < c->jhead_cnt; i++) {
650 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
652 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
653 cancel_wbuf_timer_nolock(wbuf);
654 mutex_unlock(&wbuf->io_mutex);
660 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
661 * @wbuf: write-buffer
662 * @buf: node to write
665 * This function writes data to flash via write-buffer @wbuf. This means that
666 * the last piece of the node won't reach the flash media immediately if it
667 * does not take whole max. write unit (@c->max_write_size). Instead, the node
668 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
669 * because more data are appended to the write-buffer).
671 * This function returns zero in case of success and a negative error code in
672 * case of failure. If the node cannot be written because there is no more
673 * space in this logical eraseblock, %-ENOSPC is returned.
675 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
677 struct ubifs_info *c = wbuf->c;
678 int err, written, n, aligned_len = ALIGN(len, 8);
680 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
681 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
682 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
683 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
684 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
685 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
686 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
687 ubifs_assert(wbuf->size >= c->min_io_size);
688 ubifs_assert(wbuf->size <= c->max_write_size);
689 ubifs_assert(wbuf->size % c->min_io_size == 0);
690 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
691 ubifs_assert(!c->ro_media && !c->ro_mount);
692 ubifs_assert(!c->space_fixup);
693 if (c->leb_size - wbuf->offs >= c->max_write_size)
694 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
696 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
701 cancel_wbuf_timer_nolock(wbuf);
706 if (aligned_len <= wbuf->avail) {
708 * The node is not very large and fits entirely within
711 memcpy(wbuf->buf + wbuf->used, buf, len);
713 if (aligned_len == wbuf->avail) {
714 dbg_io("flush jhead %s wbuf to LEB %d:%d",
715 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
716 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
717 wbuf->offs, wbuf->size);
721 spin_lock(&wbuf->lock);
722 wbuf->offs += wbuf->size;
723 if (c->leb_size - wbuf->offs >= c->max_write_size)
724 wbuf->size = c->max_write_size;
726 wbuf->size = c->leb_size - wbuf->offs;
727 wbuf->avail = wbuf->size;
730 spin_unlock(&wbuf->lock);
732 spin_lock(&wbuf->lock);
733 wbuf->avail -= aligned_len;
734 wbuf->used += aligned_len;
735 spin_unlock(&wbuf->lock);
745 * The node is large enough and does not fit entirely within
746 * current available space. We have to fill and flush
747 * write-buffer and switch to the next max. write unit.
749 dbg_io("flush jhead %s wbuf to LEB %d:%d",
750 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
751 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
752 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
757 wbuf->offs += wbuf->size;
759 aligned_len -= wbuf->avail;
760 written += wbuf->avail;
761 } else if (wbuf->offs & (c->max_write_size - 1)) {
763 * The write-buffer offset is not aligned to
764 * @c->max_write_size and @wbuf->size is less than
765 * @c->max_write_size. Write @wbuf->size bytes to make sure the
766 * following writes are done in optimal @c->max_write_size
769 dbg_io("write %d bytes to LEB %d:%d",
770 wbuf->size, wbuf->lnum, wbuf->offs);
771 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
776 wbuf->offs += wbuf->size;
778 aligned_len -= wbuf->size;
779 written += wbuf->size;
783 * The remaining data may take more whole max. write units, so write the
784 * remains multiple to max. write unit size directly to the flash media.
785 * We align node length to 8-byte boundary because we anyway flash wbuf
786 * if the remaining space is less than 8 bytes.
788 n = aligned_len >> c->max_write_shift;
790 n <<= c->max_write_shift;
791 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
793 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
803 spin_lock(&wbuf->lock);
806 * And now we have what's left and what does not take whole
807 * max. write unit, so write it to the write-buffer and we are
810 memcpy(wbuf->buf, buf + written, len);
812 if (c->leb_size - wbuf->offs >= c->max_write_size)
813 wbuf->size = c->max_write_size;
815 wbuf->size = c->leb_size - wbuf->offs;
816 wbuf->avail = wbuf->size - aligned_len;
817 wbuf->used = aligned_len;
819 spin_unlock(&wbuf->lock);
822 if (wbuf->sync_callback) {
823 int free = c->leb_size - wbuf->offs - wbuf->used;
825 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
831 new_wbuf_timer_nolock(wbuf);
836 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
837 len, wbuf->lnum, wbuf->offs, err);
838 ubifs_dump_node(c, buf);
840 ubifs_dump_leb(c, wbuf->lnum);
845 * ubifs_write_node - write node to the media.
846 * @c: UBIFS file-system description object
847 * @buf: the node to write
849 * @lnum: logical eraseblock number
850 * @offs: offset within the logical eraseblock
852 * This function automatically fills node magic number, assigns sequence
853 * number, and calculates node CRC checksum. The length of the @buf buffer has
854 * to be aligned to the minimal I/O unit size. This function automatically
855 * appends padding node and padding bytes if needed. Returns zero in case of
856 * success and a negative error code in case of failure.
858 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
861 int err, buf_len = ALIGN(len, c->min_io_size);
863 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
864 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
866 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
867 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
868 ubifs_assert(!c->ro_media && !c->ro_mount);
869 ubifs_assert(!c->space_fixup);
874 ubifs_prepare_node(c, buf, len, 1);
875 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
877 ubifs_dump_node(c, buf);
884 * ubifs_read_node_wbuf - read node from the media or write-buffer.
885 * @wbuf: wbuf to check for un-written data
886 * @buf: buffer to read to
889 * @lnum: logical eraseblock number
890 * @offs: offset within the logical eraseblock
892 * This function reads a node of known type and length, checks it and stores
893 * in @buf. If the node partially or fully sits in the write-buffer, this
894 * function takes data from the buffer, otherwise it reads the flash media.
895 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
896 * error code in case of failure.
898 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
901 const struct ubifs_info *c = wbuf->c;
902 int err, rlen, overlap;
903 struct ubifs_ch *ch = buf;
905 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
906 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
907 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
908 ubifs_assert(!(offs & 7) && offs < c->leb_size);
909 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
911 spin_lock(&wbuf->lock);
912 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
914 /* We may safely unlock the write-buffer and read the data */
915 spin_unlock(&wbuf->lock);
916 return ubifs_read_node(c, buf, type, len, lnum, offs);
919 /* Don't read under wbuf */
920 rlen = wbuf->offs - offs;
924 /* Copy the rest from the write-buffer */
925 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
926 spin_unlock(&wbuf->lock);
929 /* Read everything that goes before write-buffer */
930 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
931 if (err && err != -EBADMSG)
935 if (type != ch->node_type) {
936 ubifs_err("bad node type (%d but expected %d)",
937 ch->node_type, type);
941 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
943 ubifs_err("expected node type %d", type);
947 rlen = le32_to_cpu(ch->len);
949 ubifs_err("bad node length %d, expected %d", rlen, len);
956 ubifs_err("bad node at LEB %d:%d", lnum, offs);
957 ubifs_dump_node(c, buf);
963 * ubifs_read_node - read node.
964 * @c: UBIFS file-system description object
965 * @buf: buffer to read to
967 * @len: node length (not aligned)
968 * @lnum: logical eraseblock number
969 * @offs: offset within the logical eraseblock
971 * This function reads a node of known type and and length, checks it and
972 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
973 * and a negative error code in case of failure.
975 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
979 struct ubifs_ch *ch = buf;
981 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
982 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
983 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
984 ubifs_assert(!(offs & 7) && offs < c->leb_size);
985 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
987 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
988 if (err && err != -EBADMSG)
991 if (type != ch->node_type) {
992 ubifs_err("bad node type (%d but expected %d)",
993 ch->node_type, type);
997 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
999 ubifs_err("expected node type %d", type);
1003 l = le32_to_cpu(ch->len);
1005 ubifs_err("bad node length %d, expected %d", l, len);
1012 ubifs_err("bad node at LEB %d:%d, LEB mapping status %d", lnum, offs,
1013 ubi_is_mapped(c->ubi, lnum));
1014 ubifs_dump_node(c, buf);
1020 * ubifs_wbuf_init - initialize write-buffer.
1021 * @c: UBIFS file-system description object
1022 * @wbuf: write-buffer to initialize
1024 * This function initializes write-buffer. Returns zero in case of success
1025 * %-ENOMEM in case of failure.
1027 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1031 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1035 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1036 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1037 if (!wbuf->inodes) {
1044 wbuf->lnum = wbuf->offs = -1;
1046 * If the LEB starts at the max. write size aligned address, then
1047 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1048 * set it to something smaller so that it ends at the closest max.
1049 * write size boundary.
1051 size = c->max_write_size - (c->leb_start % c->max_write_size);
1052 wbuf->avail = wbuf->size = size;
1053 wbuf->sync_callback = NULL;
1054 mutex_init(&wbuf->io_mutex);
1055 spin_lock_init(&wbuf->lock);
1060 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1061 wbuf->timer.function = wbuf_timer_callback_nolock;
1062 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1063 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1064 wbuf->delta *= 1000000000ULL;
1065 ubifs_assert(wbuf->delta <= ULONG_MAX);
1071 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1072 * @wbuf: the write-buffer where to add
1073 * @inum: the inode number
1075 * This function adds an inode number to the inode array of the write-buffer.
1077 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1080 /* NOR flash or something similar */
1083 spin_lock(&wbuf->lock);
1085 wbuf->inodes[wbuf->next_ino++] = inum;
1086 spin_unlock(&wbuf->lock);
1090 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1091 * @wbuf: the write-buffer
1092 * @inum: the inode number
1094 * This function returns with %1 if the write-buffer contains some data from the
1095 * given inode otherwise it returns with %0.
1097 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1101 spin_lock(&wbuf->lock);
1102 for (i = 0; i < wbuf->next_ino; i++)
1103 if (inum == wbuf->inodes[i]) {
1107 spin_unlock(&wbuf->lock);
1113 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1114 * @c: UBIFS file-system description object
1115 * @inode: inode to synchronize
1117 * This function synchronizes write-buffers which contain nodes belonging to
1118 * @inode. Returns zero in case of success and a negative error code in case of
1121 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1125 for (i = 0; i < c->jhead_cnt; i++) {
1126 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1130 * GC head is special, do not look at it. Even if the
1131 * head contains something related to this inode, it is
1132 * a _copy_ of corresponding on-flash node which sits
1137 if (!wbuf_has_ino(wbuf, inode->i_ino))
1140 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1141 if (wbuf_has_ino(wbuf, inode->i_ino))
1142 err = ubifs_wbuf_sync_nolock(wbuf);
1143 mutex_unlock(&wbuf->io_mutex);
1146 ubifs_ro_mode(c, err);