1 // SPDX-License-Identifier: GPL-2.0+
3 * This file is part of UBIFS.
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 * Copyright (C) 2006, 2007 University of Szeged, Hungary
8 * Authors: Artem Bityutskiy (Битюцкий Артём)
14 * This file implements UBIFS I/O subsystem which provides various I/O-related
15 * helper functions (reading/writing/checking/validating nodes) and implements
16 * write-buffering support. Write buffers help to save space which otherwise
17 * would have been wasted for padding to the nearest minimal I/O unit boundary.
18 * Instead, data first goes to the write-buffer and is flushed when the
19 * buffer is full or when it is not used for some time (by timer). This is
20 * similar to the mechanism is used by JFFS2.
22 * UBIFS distinguishes between minimum write size (@c->min_io_size) and maximum
23 * write size (@c->max_write_size). The latter is the maximum amount of bytes
24 * the underlying flash is able to program at a time, and writing in
25 * @c->max_write_size units should presumably be faster. Obviously,
26 * @c->min_io_size <= @c->max_write_size. Write-buffers are of
27 * @c->max_write_size bytes in size for maximum performance. However, when a
28 * write-buffer is flushed, only the portion of it (aligned to @c->min_io_size
29 * boundary) which contains data is written, not the whole write-buffer,
30 * because this is more space-efficient.
32 * This optimization adds few complications to the code. Indeed, on the one
33 * hand, we want to write in optimal @c->max_write_size bytes chunks, which
34 * also means aligning writes at the @c->max_write_size bytes offsets. On the
35 * other hand, we do not want to waste space when synchronizing the write
36 * buffer, so during synchronization we writes in smaller chunks. And this makes
37 * the next write offset to be not aligned to @c->max_write_size bytes. So the
38 * have to make sure that the write-buffer offset (@wbuf->offs) becomes aligned
39 * to @c->max_write_size bytes again. We do this by temporarily shrinking
40 * write-buffer size (@wbuf->size).
42 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
43 * mutexes defined inside these objects. Since sometimes upper-level code
44 * has to lock the write-buffer (e.g. journal space reservation code), many
45 * functions related to write-buffers have "nolock" suffix which means that the
46 * caller has to lock the write-buffer before calling this function.
48 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
49 * aligned, UBIFS starts the next node from the aligned address, and the padded
50 * bytes may contain any rubbish. In other words, UBIFS does not put padding
51 * bytes in those small gaps. Common headers of nodes store real node lengths,
52 * not aligned lengths. Indexing nodes also store real lengths in branches.
54 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
55 * uses padding nodes or padding bytes, if the padding node does not fit.
57 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
58 * they are read from the flash media.
63 #include <dm/devres.h>
64 #include <linux/crc32.h>
65 #include <linux/slab.h>
66 #include <u-boot/crc.h>
68 #include <linux/compat.h>
69 #include <linux/err.h>
74 * ubifs_ro_mode - switch UBIFS to read read-only mode.
75 * @c: UBIFS file-system description object
76 * @err: error code which is the reason of switching to R/O mode
78 void ubifs_ro_mode(struct ubifs_info *c, int err)
82 c->no_chk_data_crc = 0;
83 c->vfs_sb->s_flags |= MS_RDONLY;
84 ubifs_warn(c, "switched to read-only mode, error %d", err);
90 * Below are simple wrappers over UBI I/O functions which include some
91 * additional checks and UBIFS debugging stuff. See corresponding UBI function
92 * for more information.
95 int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
96 int len, int even_ebadmsg)
100 err = ubi_read(c->ubi, lnum, buf, offs, len);
102 * In case of %-EBADMSG print the error message only if the
103 * @even_ebadmsg is true.
105 if (err && (err != -EBADMSG || even_ebadmsg)) {
106 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
107 len, lnum, offs, err);
113 int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
118 ubifs_assert(!c->ro_media && !c->ro_mount);
121 if (!dbg_is_tst_rcvry(c))
122 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
125 err = dbg_leb_write(c, lnum, buf, offs, len);
128 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
129 len, lnum, offs, err);
130 ubifs_ro_mode(c, err);
136 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
140 ubifs_assert(!c->ro_media && !c->ro_mount);
143 if (!dbg_is_tst_rcvry(c))
144 err = ubi_leb_change(c->ubi, lnum, buf, len);
147 err = dbg_leb_change(c, lnum, buf, len);
150 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
152 ubifs_ro_mode(c, err);
158 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
162 ubifs_assert(!c->ro_media && !c->ro_mount);
165 if (!dbg_is_tst_rcvry(c))
166 err = ubi_leb_unmap(c->ubi, lnum);
169 err = dbg_leb_unmap(c, lnum);
172 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
173 ubifs_ro_mode(c, err);
179 int ubifs_leb_map(struct ubifs_info *c, int lnum)
183 ubifs_assert(!c->ro_media && !c->ro_mount);
186 if (!dbg_is_tst_rcvry(c))
187 err = ubi_leb_map(c->ubi, lnum);
190 err = dbg_leb_map(c, lnum);
193 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
194 ubifs_ro_mode(c, err);
200 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
204 err = ubi_is_mapped(c->ubi, lnum);
206 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
214 * ubifs_check_node - check node.
215 * @c: UBIFS file-system description object
216 * @buf: node to check
217 * @lnum: logical eraseblock number
218 * @offs: offset within the logical eraseblock
219 * @quiet: print no messages
220 * @must_chk_crc: indicates whether to always check the CRC
222 * This function checks node magic number and CRC checksum. This function also
223 * validates node length to prevent UBIFS from becoming crazy when an attacker
224 * feeds it a file-system image with incorrect nodes. For example, too large
225 * node length in the common header could cause UBIFS to read memory outside of
226 * allocated buffer when checking the CRC checksum.
228 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
229 * true, which is controlled by corresponding UBIFS mount option. However, if
230 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
231 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
232 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
233 * is checked. This is because during mounting or re-mounting from R/O mode to
234 * R/W mode we may read journal nodes (when replying the journal or doing the
235 * recovery) and the journal nodes may potentially be corrupted, so checking is
238 * This function returns zero in case of success and %-EUCLEAN in case of bad
241 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
242 int offs, int quiet, int must_chk_crc)
244 int err = -EINVAL, type, node_len;
245 uint32_t crc, node_crc, magic;
246 const struct ubifs_ch *ch = buf;
248 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
249 ubifs_assert(!(offs & 7) && offs < c->leb_size);
251 magic = le32_to_cpu(ch->magic);
252 if (magic != UBIFS_NODE_MAGIC) {
254 ubifs_err(c, "bad magic %#08x, expected %#08x",
255 magic, UBIFS_NODE_MAGIC);
260 type = ch->node_type;
261 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
263 ubifs_err(c, "bad node type %d", type);
267 node_len = le32_to_cpu(ch->len);
268 if (node_len + offs > c->leb_size)
271 if (c->ranges[type].max_len == 0) {
272 if (node_len != c->ranges[type].len)
274 } else if (node_len < c->ranges[type].min_len ||
275 node_len > c->ranges[type].max_len)
278 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
279 !c->remounting_rw && c->no_chk_data_crc)
282 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
283 node_crc = le32_to_cpu(ch->crc);
284 if (crc != node_crc) {
286 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
296 ubifs_err(c, "bad node length %d", node_len);
299 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
300 ubifs_dump_node(c, buf);
307 * ubifs_pad - pad flash space.
308 * @c: UBIFS file-system description object
309 * @buf: buffer to put padding to
310 * @pad: how many bytes to pad
312 * The flash media obliges us to write only in chunks of %c->min_io_size and
313 * when we have to write less data we add padding node to the write-buffer and
314 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
315 * media is being scanned. If the amount of wasted space is not enough to fit a
316 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
317 * pattern (%UBIFS_PADDING_BYTE).
319 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
322 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
326 ubifs_assert(pad >= 0 && !(pad & 7));
328 if (pad >= UBIFS_PAD_NODE_SZ) {
329 struct ubifs_ch *ch = buf;
330 struct ubifs_pad_node *pad_node = buf;
332 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
333 ch->node_type = UBIFS_PAD_NODE;
334 ch->group_type = UBIFS_NO_NODE_GROUP;
335 ch->padding[0] = ch->padding[1] = 0;
337 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
338 pad -= UBIFS_PAD_NODE_SZ;
339 pad_node->pad_len = cpu_to_le32(pad);
340 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
341 ch->crc = cpu_to_le32(crc);
342 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
344 /* Too little space, padding node won't fit */
345 memset(buf, UBIFS_PADDING_BYTE, pad);
349 * next_sqnum - get next sequence number.
350 * @c: UBIFS file-system description object
352 static unsigned long long next_sqnum(struct ubifs_info *c)
354 unsigned long long sqnum;
356 spin_lock(&c->cnt_lock);
357 sqnum = ++c->max_sqnum;
358 spin_unlock(&c->cnt_lock);
360 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
361 if (sqnum >= SQNUM_WATERMARK) {
362 ubifs_err(c, "sequence number overflow %llu, end of life",
364 ubifs_ro_mode(c, -EINVAL);
366 ubifs_warn(c, "running out of sequence numbers, end of life soon");
373 * ubifs_prepare_node - prepare node to be written to flash.
374 * @c: UBIFS file-system description object
375 * @node: the node to pad
377 * @pad: if the buffer has to be padded
379 * This function prepares node at @node to be written to the media - it
380 * calculates node CRC, fills the common header, and adds proper padding up to
381 * the next minimum I/O unit if @pad is not zero.
383 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
386 struct ubifs_ch *ch = node;
387 unsigned long long sqnum = next_sqnum(c);
389 ubifs_assert(len >= UBIFS_CH_SZ);
391 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
392 ch->len = cpu_to_le32(len);
393 ch->group_type = UBIFS_NO_NODE_GROUP;
394 ch->sqnum = cpu_to_le64(sqnum);
395 ch->padding[0] = ch->padding[1] = 0;
396 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
397 ch->crc = cpu_to_le32(crc);
401 pad = ALIGN(len, c->min_io_size) - len;
402 ubifs_pad(c, node + len, pad);
407 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
408 * @c: UBIFS file-system description object
409 * @node: the node to pad
411 * @last: indicates the last node of the group
413 * This function prepares node at @node to be written to the media - it
414 * calculates node CRC and fills the common header.
416 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
419 struct ubifs_ch *ch = node;
420 unsigned long long sqnum = next_sqnum(c);
422 ubifs_assert(len >= UBIFS_CH_SZ);
424 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
425 ch->len = cpu_to_le32(len);
427 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
429 ch->group_type = UBIFS_IN_NODE_GROUP;
430 ch->sqnum = cpu_to_le64(sqnum);
431 ch->padding[0] = ch->padding[1] = 0;
432 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
433 ch->crc = cpu_to_le32(crc);
438 * wbuf_timer_callback - write-buffer timer callback function.
439 * @timer: timer data (write-buffer descriptor)
441 * This function is called when the write-buffer timer expires.
443 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
445 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
447 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
449 wbuf->c->need_wbuf_sync = 1;
450 ubifs_wake_up_bgt(wbuf->c);
451 return HRTIMER_NORESTART;
455 * new_wbuf_timer - start new write-buffer timer.
456 * @wbuf: write-buffer descriptor
458 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
460 ubifs_assert(!hrtimer_active(&wbuf->timer));
464 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
465 dbg_jhead(wbuf->jhead),
466 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
467 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
469 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
475 * cancel_wbuf_timer - cancel write-buffer timer.
476 * @wbuf: write-buffer descriptor
478 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
484 hrtimer_cancel(&wbuf->timer);
489 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
490 * @wbuf: write-buffer to synchronize
492 * This function synchronizes write-buffer @buf and returns zero in case of
493 * success or a negative error code in case of failure.
495 * Note, although write-buffers are of @c->max_write_size, this function does
496 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
497 * if the write-buffer is only partially filled with data, only the used part
498 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
499 * This way we waste less space.
501 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
503 struct ubifs_info *c = wbuf->c;
504 int err, dirt, sync_len;
506 cancel_wbuf_timer_nolock(wbuf);
507 if (!wbuf->used || wbuf->lnum == -1)
508 /* Write-buffer is empty or not seeked */
511 dbg_io("LEB %d:%d, %d bytes, jhead %s",
512 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
513 ubifs_assert(!(wbuf->avail & 7));
514 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
515 ubifs_assert(wbuf->size >= c->min_io_size);
516 ubifs_assert(wbuf->size <= c->max_write_size);
517 ubifs_assert(wbuf->size % c->min_io_size == 0);
518 ubifs_assert(!c->ro_media && !c->ro_mount);
519 if (c->leb_size - wbuf->offs >= c->max_write_size)
520 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
526 * Do not write whole write buffer but write only the minimum necessary
527 * amount of min. I/O units.
529 sync_len = ALIGN(wbuf->used, c->min_io_size);
530 dirt = sync_len - wbuf->used;
532 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
533 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
537 spin_lock(&wbuf->lock);
538 wbuf->offs += sync_len;
540 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
541 * But our goal is to optimize writes and make sure we write in
542 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
543 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
544 * sure that @wbuf->offs + @wbuf->size is aligned to
545 * @c->max_write_size. This way we make sure that after next
546 * write-buffer flush we are again at the optimal offset (aligned to
547 * @c->max_write_size).
549 if (c->leb_size - wbuf->offs < c->max_write_size)
550 wbuf->size = c->leb_size - wbuf->offs;
551 else if (wbuf->offs & (c->max_write_size - 1))
552 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
554 wbuf->size = c->max_write_size;
555 wbuf->avail = wbuf->size;
558 spin_unlock(&wbuf->lock);
560 if (wbuf->sync_callback)
561 err = wbuf->sync_callback(c, wbuf->lnum,
562 c->leb_size - wbuf->offs, dirt);
567 * ubifs_wbuf_seek_nolock - seek write-buffer.
568 * @wbuf: write-buffer
569 * @lnum: logical eraseblock number to seek to
570 * @offs: logical eraseblock offset to seek to
572 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
573 * The write-buffer has to be empty. Returns zero in case of success and a
574 * negative error code in case of failure.
576 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
578 const struct ubifs_info *c = wbuf->c;
580 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
581 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
582 ubifs_assert(offs >= 0 && offs <= c->leb_size);
583 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
584 ubifs_assert(lnum != wbuf->lnum);
585 ubifs_assert(wbuf->used == 0);
587 spin_lock(&wbuf->lock);
590 if (c->leb_size - wbuf->offs < c->max_write_size)
591 wbuf->size = c->leb_size - wbuf->offs;
592 else if (wbuf->offs & (c->max_write_size - 1))
593 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
595 wbuf->size = c->max_write_size;
596 wbuf->avail = wbuf->size;
598 spin_unlock(&wbuf->lock);
605 * ubifs_bg_wbufs_sync - synchronize write-buffers.
606 * @c: UBIFS file-system description object
608 * This function is called by background thread to synchronize write-buffers.
609 * Returns zero in case of success and a negative error code in case of
612 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
616 ubifs_assert(!c->ro_media && !c->ro_mount);
617 if (!c->need_wbuf_sync)
619 c->need_wbuf_sync = 0;
626 dbg_io("synchronize");
627 for (i = 0; i < c->jhead_cnt; i++) {
628 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
633 * If the mutex is locked then wbuf is being changed, so
634 * synchronization is not necessary.
636 if (mutex_is_locked(&wbuf->io_mutex))
639 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
640 if (!wbuf->need_sync) {
641 mutex_unlock(&wbuf->io_mutex);
645 err = ubifs_wbuf_sync_nolock(wbuf);
646 mutex_unlock(&wbuf->io_mutex);
648 ubifs_err(c, "cannot sync write-buffer, error %d", err);
649 ubifs_ro_mode(c, err);
657 /* Cancel all timers to prevent repeated errors */
658 for (i = 0; i < c->jhead_cnt; i++) {
659 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
661 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
662 cancel_wbuf_timer_nolock(wbuf);
663 mutex_unlock(&wbuf->io_mutex);
669 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
670 * @wbuf: write-buffer
671 * @buf: node to write
674 * This function writes data to flash via write-buffer @wbuf. This means that
675 * the last piece of the node won't reach the flash media immediately if it
676 * does not take whole max. write unit (@c->max_write_size). Instead, the node
677 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
678 * because more data are appended to the write-buffer).
680 * This function returns zero in case of success and a negative error code in
681 * case of failure. If the node cannot be written because there is no more
682 * space in this logical eraseblock, %-ENOSPC is returned.
684 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
686 struct ubifs_info *c = wbuf->c;
687 int err, written, n, aligned_len = ALIGN(len, 8);
689 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
690 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
691 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
692 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
693 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
694 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
695 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
696 ubifs_assert(wbuf->size >= c->min_io_size);
697 ubifs_assert(wbuf->size <= c->max_write_size);
698 ubifs_assert(wbuf->size % c->min_io_size == 0);
699 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
700 ubifs_assert(!c->ro_media && !c->ro_mount);
701 ubifs_assert(!c->space_fixup);
702 if (c->leb_size - wbuf->offs >= c->max_write_size)
703 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
705 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
710 cancel_wbuf_timer_nolock(wbuf);
715 if (aligned_len <= wbuf->avail) {
717 * The node is not very large and fits entirely within
720 memcpy(wbuf->buf + wbuf->used, buf, len);
722 if (aligned_len == wbuf->avail) {
723 dbg_io("flush jhead %s wbuf to LEB %d:%d",
724 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
725 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
726 wbuf->offs, wbuf->size);
730 spin_lock(&wbuf->lock);
731 wbuf->offs += wbuf->size;
732 if (c->leb_size - wbuf->offs >= c->max_write_size)
733 wbuf->size = c->max_write_size;
735 wbuf->size = c->leb_size - wbuf->offs;
736 wbuf->avail = wbuf->size;
739 spin_unlock(&wbuf->lock);
741 spin_lock(&wbuf->lock);
742 wbuf->avail -= aligned_len;
743 wbuf->used += aligned_len;
744 spin_unlock(&wbuf->lock);
754 * The node is large enough and does not fit entirely within
755 * current available space. We have to fill and flush
756 * write-buffer and switch to the next max. write unit.
758 dbg_io("flush jhead %s wbuf to LEB %d:%d",
759 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
760 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
761 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
766 wbuf->offs += wbuf->size;
768 aligned_len -= wbuf->avail;
769 written += wbuf->avail;
770 } else if (wbuf->offs & (c->max_write_size - 1)) {
772 * The write-buffer offset is not aligned to
773 * @c->max_write_size and @wbuf->size is less than
774 * @c->max_write_size. Write @wbuf->size bytes to make sure the
775 * following writes are done in optimal @c->max_write_size
778 dbg_io("write %d bytes to LEB %d:%d",
779 wbuf->size, wbuf->lnum, wbuf->offs);
780 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
785 wbuf->offs += wbuf->size;
787 aligned_len -= wbuf->size;
788 written += wbuf->size;
792 * The remaining data may take more whole max. write units, so write the
793 * remains multiple to max. write unit size directly to the flash media.
794 * We align node length to 8-byte boundary because we anyway flash wbuf
795 * if the remaining space is less than 8 bytes.
797 n = aligned_len >> c->max_write_shift;
799 n <<= c->max_write_shift;
800 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
802 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
812 spin_lock(&wbuf->lock);
815 * And now we have what's left and what does not take whole
816 * max. write unit, so write it to the write-buffer and we are
819 memcpy(wbuf->buf, buf + written, len);
821 if (c->leb_size - wbuf->offs >= c->max_write_size)
822 wbuf->size = c->max_write_size;
824 wbuf->size = c->leb_size - wbuf->offs;
825 wbuf->avail = wbuf->size - aligned_len;
826 wbuf->used = aligned_len;
828 spin_unlock(&wbuf->lock);
831 if (wbuf->sync_callback) {
832 int free = c->leb_size - wbuf->offs - wbuf->used;
834 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
840 new_wbuf_timer_nolock(wbuf);
845 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
846 len, wbuf->lnum, wbuf->offs, err);
847 ubifs_dump_node(c, buf);
849 ubifs_dump_leb(c, wbuf->lnum);
854 * ubifs_write_node - write node to the media.
855 * @c: UBIFS file-system description object
856 * @buf: the node to write
858 * @lnum: logical eraseblock number
859 * @offs: offset within the logical eraseblock
861 * This function automatically fills node magic number, assigns sequence
862 * number, and calculates node CRC checksum. The length of the @buf buffer has
863 * to be aligned to the minimal I/O unit size. This function automatically
864 * appends padding node and padding bytes if needed. Returns zero in case of
865 * success and a negative error code in case of failure.
867 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
870 int err, buf_len = ALIGN(len, c->min_io_size);
872 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
873 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
875 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
876 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
877 ubifs_assert(!c->ro_media && !c->ro_mount);
878 ubifs_assert(!c->space_fixup);
883 ubifs_prepare_node(c, buf, len, 1);
884 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
886 ubifs_dump_node(c, buf);
893 * ubifs_read_node_wbuf - read node from the media or write-buffer.
894 * @wbuf: wbuf to check for un-written data
895 * @buf: buffer to read to
898 * @lnum: logical eraseblock number
899 * @offs: offset within the logical eraseblock
901 * This function reads a node of known type and length, checks it and stores
902 * in @buf. If the node partially or fully sits in the write-buffer, this
903 * function takes data from the buffer, otherwise it reads the flash media.
904 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
905 * error code in case of failure.
907 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
910 const struct ubifs_info *c = wbuf->c;
911 int err, rlen, overlap;
912 struct ubifs_ch *ch = buf;
914 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
915 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
916 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
917 ubifs_assert(!(offs & 7) && offs < c->leb_size);
918 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
920 spin_lock(&wbuf->lock);
921 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
923 /* We may safely unlock the write-buffer and read the data */
924 spin_unlock(&wbuf->lock);
925 return ubifs_read_node(c, buf, type, len, lnum, offs);
928 /* Don't read under wbuf */
929 rlen = wbuf->offs - offs;
933 /* Copy the rest from the write-buffer */
934 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
935 spin_unlock(&wbuf->lock);
938 /* Read everything that goes before write-buffer */
939 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
940 if (err && err != -EBADMSG)
944 if (type != ch->node_type) {
945 ubifs_err(c, "bad node type (%d but expected %d)",
946 ch->node_type, type);
950 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
952 ubifs_err(c, "expected node type %d", type);
956 rlen = le32_to_cpu(ch->len);
958 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
965 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
966 ubifs_dump_node(c, buf);
972 * ubifs_read_node - read node.
973 * @c: UBIFS file-system description object
974 * @buf: buffer to read to
976 * @len: node length (not aligned)
977 * @lnum: logical eraseblock number
978 * @offs: offset within the logical eraseblock
980 * This function reads a node of known type and and length, checks it and
981 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
982 * and a negative error code in case of failure.
984 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
988 struct ubifs_ch *ch = buf;
990 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
991 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
992 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
993 ubifs_assert(!(offs & 7) && offs < c->leb_size);
994 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
996 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
997 if (err && err != -EBADMSG)
1000 if (type != ch->node_type) {
1001 ubifs_errc(c, "bad node type (%d but expected %d)",
1002 ch->node_type, type);
1006 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1008 ubifs_errc(c, "expected node type %d", type);
1012 l = le32_to_cpu(ch->len);
1014 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1021 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1022 offs, ubi_is_mapped(c->ubi, lnum));
1024 ubifs_dump_node(c, buf);
1031 * ubifs_wbuf_init - initialize write-buffer.
1032 * @c: UBIFS file-system description object
1033 * @wbuf: write-buffer to initialize
1035 * This function initializes write-buffer. Returns zero in case of success
1036 * %-ENOMEM in case of failure.
1038 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1042 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1046 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1047 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1048 if (!wbuf->inodes) {
1055 wbuf->lnum = wbuf->offs = -1;
1057 * If the LEB starts at the max. write size aligned address, then
1058 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1059 * set it to something smaller so that it ends at the closest max.
1060 * write size boundary.
1062 size = c->max_write_size - (c->leb_start % c->max_write_size);
1063 wbuf->avail = wbuf->size = size;
1064 wbuf->sync_callback = NULL;
1065 mutex_init(&wbuf->io_mutex);
1066 spin_lock_init(&wbuf->lock);
1071 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1072 wbuf->timer.function = wbuf_timer_callback_nolock;
1073 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1074 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1075 wbuf->delta *= 1000000000ULL;
1076 ubifs_assert(wbuf->delta <= ULONG_MAX);
1082 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1083 * @wbuf: the write-buffer where to add
1084 * @inum: the inode number
1086 * This function adds an inode number to the inode array of the write-buffer.
1088 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1091 /* NOR flash or something similar */
1094 spin_lock(&wbuf->lock);
1096 wbuf->inodes[wbuf->next_ino++] = inum;
1097 spin_unlock(&wbuf->lock);
1101 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1102 * @wbuf: the write-buffer
1103 * @inum: the inode number
1105 * This function returns with %1 if the write-buffer contains some data from the
1106 * given inode otherwise it returns with %0.
1108 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1112 spin_lock(&wbuf->lock);
1113 for (i = 0; i < wbuf->next_ino; i++)
1114 if (inum == wbuf->inodes[i]) {
1118 spin_unlock(&wbuf->lock);
1124 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1125 * @c: UBIFS file-system description object
1126 * @inode: inode to synchronize
1128 * This function synchronizes write-buffers which contain nodes belonging to
1129 * @inode. Returns zero in case of success and a negative error code in case of
1132 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1136 for (i = 0; i < c->jhead_cnt; i++) {
1137 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1141 * GC head is special, do not look at it. Even if the
1142 * head contains something related to this inode, it is
1143 * a _copy_ of corresponding on-flash node which sits
1148 if (!wbuf_has_ino(wbuf, inode->i_ino))
1151 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1152 if (wbuf_has_ino(wbuf, inode->i_ino))
1153 err = ubifs_wbuf_sync_nolock(wbuf);
1154 mutex_unlock(&wbuf->io_mutex);
1157 ubifs_ro_mode(c, err);