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.
62 #include <dm/devres.h>
63 #include <linux/crc32.h>
64 #include <linux/slab.h>
65 #include <u-boot/crc.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(c, "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(c, "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);
124 err = dbg_leb_write(c, lnum, buf, offs, len);
127 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
128 len, lnum, offs, err);
129 ubifs_ro_mode(c, err);
135 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
139 ubifs_assert(!c->ro_media && !c->ro_mount);
142 if (!dbg_is_tst_rcvry(c))
143 err = ubi_leb_change(c->ubi, lnum, buf, len);
146 err = dbg_leb_change(c, lnum, buf, len);
149 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
151 ubifs_ro_mode(c, err);
157 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
161 ubifs_assert(!c->ro_media && !c->ro_mount);
164 if (!dbg_is_tst_rcvry(c))
165 err = ubi_leb_unmap(c->ubi, lnum);
168 err = dbg_leb_unmap(c, lnum);
171 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
172 ubifs_ro_mode(c, err);
178 int ubifs_leb_map(struct ubifs_info *c, int lnum)
182 ubifs_assert(!c->ro_media && !c->ro_mount);
185 if (!dbg_is_tst_rcvry(c))
186 err = ubi_leb_map(c->ubi, lnum);
189 err = dbg_leb_map(c, lnum);
192 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
193 ubifs_ro_mode(c, err);
199 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
203 err = ubi_is_mapped(c->ubi, lnum);
205 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
213 * ubifs_check_node - check node.
214 * @c: UBIFS file-system description object
215 * @buf: node to check
216 * @lnum: logical eraseblock number
217 * @offs: offset within the logical eraseblock
218 * @quiet: print no messages
219 * @must_chk_crc: indicates whether to always check the CRC
221 * This function checks node magic number and CRC checksum. This function also
222 * validates node length to prevent UBIFS from becoming crazy when an attacker
223 * feeds it a file-system image with incorrect nodes. For example, too large
224 * node length in the common header could cause UBIFS to read memory outside of
225 * allocated buffer when checking the CRC checksum.
227 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
228 * true, which is controlled by corresponding UBIFS mount option. However, if
229 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
230 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
231 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
232 * is checked. This is because during mounting or re-mounting from R/O mode to
233 * R/W mode we may read journal nodes (when replying the journal or doing the
234 * recovery) and the journal nodes may potentially be corrupted, so checking is
237 * This function returns zero in case of success and %-EUCLEAN in case of bad
240 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
241 int offs, int quiet, int must_chk_crc)
243 int err = -EINVAL, type, node_len;
244 uint32_t crc, node_crc, magic;
245 const struct ubifs_ch *ch = buf;
247 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
248 ubifs_assert(!(offs & 7) && offs < c->leb_size);
250 magic = le32_to_cpu(ch->magic);
251 if (magic != UBIFS_NODE_MAGIC) {
253 ubifs_err(c, "bad magic %#08x, expected %#08x",
254 magic, UBIFS_NODE_MAGIC);
259 type = ch->node_type;
260 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
262 ubifs_err(c, "bad node type %d", type);
266 node_len = le32_to_cpu(ch->len);
267 if (node_len + offs > c->leb_size)
270 if (c->ranges[type].max_len == 0) {
271 if (node_len != c->ranges[type].len)
273 } else if (node_len < c->ranges[type].min_len ||
274 node_len > c->ranges[type].max_len)
277 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
278 !c->remounting_rw && c->no_chk_data_crc)
281 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
282 node_crc = le32_to_cpu(ch->crc);
283 if (crc != node_crc) {
285 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
295 ubifs_err(c, "bad node length %d", node_len);
298 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
299 ubifs_dump_node(c, buf);
306 * ubifs_pad - pad flash space.
307 * @c: UBIFS file-system description object
308 * @buf: buffer to put padding to
309 * @pad: how many bytes to pad
311 * The flash media obliges us to write only in chunks of %c->min_io_size and
312 * when we have to write less data we add padding node to the write-buffer and
313 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
314 * media is being scanned. If the amount of wasted space is not enough to fit a
315 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
316 * pattern (%UBIFS_PADDING_BYTE).
318 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
321 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
325 ubifs_assert(pad >= 0 && !(pad & 7));
327 if (pad >= UBIFS_PAD_NODE_SZ) {
328 struct ubifs_ch *ch = buf;
329 struct ubifs_pad_node *pad_node = buf;
331 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
332 ch->node_type = UBIFS_PAD_NODE;
333 ch->group_type = UBIFS_NO_NODE_GROUP;
334 ch->padding[0] = ch->padding[1] = 0;
336 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
337 pad -= UBIFS_PAD_NODE_SZ;
338 pad_node->pad_len = cpu_to_le32(pad);
339 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
340 ch->crc = cpu_to_le32(crc);
341 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
343 /* Too little space, padding node won't fit */
344 memset(buf, UBIFS_PADDING_BYTE, pad);
348 * next_sqnum - get next sequence number.
349 * @c: UBIFS file-system description object
351 static unsigned long long next_sqnum(struct ubifs_info *c)
353 unsigned long long sqnum;
355 spin_lock(&c->cnt_lock);
356 sqnum = ++c->max_sqnum;
357 spin_unlock(&c->cnt_lock);
359 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
360 if (sqnum >= SQNUM_WATERMARK) {
361 ubifs_err(c, "sequence number overflow %llu, end of life",
363 ubifs_ro_mode(c, -EINVAL);
365 ubifs_warn(c, "running out of sequence numbers, end of life soon");
372 * ubifs_prepare_node - prepare node to be written to flash.
373 * @c: UBIFS file-system description object
374 * @node: the node to pad
376 * @pad: if the buffer has to be padded
378 * This function prepares node at @node to be written to the media - it
379 * calculates node CRC, fills the common header, and adds proper padding up to
380 * the next minimum I/O unit if @pad is not zero.
382 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
385 struct ubifs_ch *ch = node;
386 unsigned long long sqnum = next_sqnum(c);
388 ubifs_assert(len >= UBIFS_CH_SZ);
390 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
391 ch->len = cpu_to_le32(len);
392 ch->group_type = UBIFS_NO_NODE_GROUP;
393 ch->sqnum = cpu_to_le64(sqnum);
394 ch->padding[0] = ch->padding[1] = 0;
395 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
396 ch->crc = cpu_to_le32(crc);
400 pad = ALIGN(len, c->min_io_size) - len;
401 ubifs_pad(c, node + len, pad);
406 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
407 * @c: UBIFS file-system description object
408 * @node: the node to pad
410 * @last: indicates the last node of the group
412 * This function prepares node at @node to be written to the media - it
413 * calculates node CRC and fills the common header.
415 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
418 struct ubifs_ch *ch = node;
419 unsigned long long sqnum = next_sqnum(c);
421 ubifs_assert(len >= UBIFS_CH_SZ);
423 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
424 ch->len = cpu_to_le32(len);
426 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
428 ch->group_type = UBIFS_IN_NODE_GROUP;
429 ch->sqnum = cpu_to_le64(sqnum);
430 ch->padding[0] = ch->padding[1] = 0;
431 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
432 ch->crc = cpu_to_le32(crc);
437 * wbuf_timer_callback - write-buffer timer callback function.
438 * @timer: timer data (write-buffer descriptor)
440 * This function is called when the write-buffer timer expires.
442 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
444 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
446 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
448 wbuf->c->need_wbuf_sync = 1;
449 ubifs_wake_up_bgt(wbuf->c);
450 return HRTIMER_NORESTART;
454 * new_wbuf_timer - start new write-buffer timer.
455 * @wbuf: write-buffer descriptor
457 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
459 ubifs_assert(!hrtimer_active(&wbuf->timer));
463 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
464 dbg_jhead(wbuf->jhead),
465 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
466 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
468 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
474 * cancel_wbuf_timer - cancel write-buffer timer.
475 * @wbuf: write-buffer descriptor
477 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
483 hrtimer_cancel(&wbuf->timer);
488 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
489 * @wbuf: write-buffer to synchronize
491 * This function synchronizes write-buffer @buf and returns zero in case of
492 * success or a negative error code in case of failure.
494 * Note, although write-buffers are of @c->max_write_size, this function does
495 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
496 * if the write-buffer is only partially filled with data, only the used part
497 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
498 * This way we waste less space.
500 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
502 struct ubifs_info *c = wbuf->c;
503 int err, dirt, sync_len;
505 cancel_wbuf_timer_nolock(wbuf);
506 if (!wbuf->used || wbuf->lnum == -1)
507 /* Write-buffer is empty or not seeked */
510 dbg_io("LEB %d:%d, %d bytes, jhead %s",
511 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
512 ubifs_assert(!(wbuf->avail & 7));
513 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
514 ubifs_assert(wbuf->size >= c->min_io_size);
515 ubifs_assert(wbuf->size <= c->max_write_size);
516 ubifs_assert(wbuf->size % c->min_io_size == 0);
517 ubifs_assert(!c->ro_media && !c->ro_mount);
518 if (c->leb_size - wbuf->offs >= c->max_write_size)
519 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
525 * Do not write whole write buffer but write only the minimum necessary
526 * amount of min. I/O units.
528 sync_len = ALIGN(wbuf->used, c->min_io_size);
529 dirt = sync_len - wbuf->used;
531 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
532 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
536 spin_lock(&wbuf->lock);
537 wbuf->offs += sync_len;
539 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
540 * But our goal is to optimize writes and make sure we write in
541 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
542 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
543 * sure that @wbuf->offs + @wbuf->size is aligned to
544 * @c->max_write_size. This way we make sure that after next
545 * write-buffer flush we are again at the optimal offset (aligned to
546 * @c->max_write_size).
548 if (c->leb_size - wbuf->offs < c->max_write_size)
549 wbuf->size = c->leb_size - wbuf->offs;
550 else if (wbuf->offs & (c->max_write_size - 1))
551 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
553 wbuf->size = c->max_write_size;
554 wbuf->avail = wbuf->size;
557 spin_unlock(&wbuf->lock);
559 if (wbuf->sync_callback)
560 err = wbuf->sync_callback(c, wbuf->lnum,
561 c->leb_size - wbuf->offs, dirt);
566 * ubifs_wbuf_seek_nolock - seek write-buffer.
567 * @wbuf: write-buffer
568 * @lnum: logical eraseblock number to seek to
569 * @offs: logical eraseblock offset to seek to
571 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
572 * The write-buffer has to be empty. Returns zero in case of success and a
573 * negative error code in case of failure.
575 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
577 const struct ubifs_info *c = wbuf->c;
579 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
580 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
581 ubifs_assert(offs >= 0 && offs <= c->leb_size);
582 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
583 ubifs_assert(lnum != wbuf->lnum);
584 ubifs_assert(wbuf->used == 0);
586 spin_lock(&wbuf->lock);
589 if (c->leb_size - wbuf->offs < c->max_write_size)
590 wbuf->size = c->leb_size - wbuf->offs;
591 else if (wbuf->offs & (c->max_write_size - 1))
592 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
594 wbuf->size = c->max_write_size;
595 wbuf->avail = wbuf->size;
597 spin_unlock(&wbuf->lock);
604 * ubifs_bg_wbufs_sync - synchronize write-buffers.
605 * @c: UBIFS file-system description object
607 * This function is called by background thread to synchronize write-buffers.
608 * Returns zero in case of success and a negative error code in case of
611 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
615 ubifs_assert(!c->ro_media && !c->ro_mount);
616 if (!c->need_wbuf_sync)
618 c->need_wbuf_sync = 0;
625 dbg_io("synchronize");
626 for (i = 0; i < c->jhead_cnt; i++) {
627 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
632 * If the mutex is locked then wbuf is being changed, so
633 * synchronization is not necessary.
635 if (mutex_is_locked(&wbuf->io_mutex))
638 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
639 if (!wbuf->need_sync) {
640 mutex_unlock(&wbuf->io_mutex);
644 err = ubifs_wbuf_sync_nolock(wbuf);
645 mutex_unlock(&wbuf->io_mutex);
647 ubifs_err(c, "cannot sync write-buffer, error %d", err);
648 ubifs_ro_mode(c, err);
656 /* Cancel all timers to prevent repeated errors */
657 for (i = 0; i < c->jhead_cnt; i++) {
658 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
660 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
661 cancel_wbuf_timer_nolock(wbuf);
662 mutex_unlock(&wbuf->io_mutex);
668 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
669 * @wbuf: write-buffer
670 * @buf: node to write
673 * This function writes data to flash via write-buffer @wbuf. This means that
674 * the last piece of the node won't reach the flash media immediately if it
675 * does not take whole max. write unit (@c->max_write_size). Instead, the node
676 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
677 * because more data are appended to the write-buffer).
679 * This function returns zero in case of success and a negative error code in
680 * case of failure. If the node cannot be written because there is no more
681 * space in this logical eraseblock, %-ENOSPC is returned.
683 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
685 struct ubifs_info *c = wbuf->c;
686 int err, written, n, aligned_len = ALIGN(len, 8);
688 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
689 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
690 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
691 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
692 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
693 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
694 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
695 ubifs_assert(wbuf->size >= c->min_io_size);
696 ubifs_assert(wbuf->size <= c->max_write_size);
697 ubifs_assert(wbuf->size % c->min_io_size == 0);
698 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
699 ubifs_assert(!c->ro_media && !c->ro_mount);
700 ubifs_assert(!c->space_fixup);
701 if (c->leb_size - wbuf->offs >= c->max_write_size)
702 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
704 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
709 cancel_wbuf_timer_nolock(wbuf);
714 if (aligned_len <= wbuf->avail) {
716 * The node is not very large and fits entirely within
719 memcpy(wbuf->buf + wbuf->used, buf, len);
721 if (aligned_len == wbuf->avail) {
722 dbg_io("flush jhead %s wbuf to LEB %d:%d",
723 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
724 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
725 wbuf->offs, wbuf->size);
729 spin_lock(&wbuf->lock);
730 wbuf->offs += wbuf->size;
731 if (c->leb_size - wbuf->offs >= c->max_write_size)
732 wbuf->size = c->max_write_size;
734 wbuf->size = c->leb_size - wbuf->offs;
735 wbuf->avail = wbuf->size;
738 spin_unlock(&wbuf->lock);
740 spin_lock(&wbuf->lock);
741 wbuf->avail -= aligned_len;
742 wbuf->used += aligned_len;
743 spin_unlock(&wbuf->lock);
753 * The node is large enough and does not fit entirely within
754 * current available space. We have to fill and flush
755 * write-buffer and switch to the next max. write unit.
757 dbg_io("flush jhead %s wbuf to LEB %d:%d",
758 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
759 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
760 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
765 wbuf->offs += wbuf->size;
767 aligned_len -= wbuf->avail;
768 written += wbuf->avail;
769 } else if (wbuf->offs & (c->max_write_size - 1)) {
771 * The write-buffer offset is not aligned to
772 * @c->max_write_size and @wbuf->size is less than
773 * @c->max_write_size. Write @wbuf->size bytes to make sure the
774 * following writes are done in optimal @c->max_write_size
777 dbg_io("write %d bytes to LEB %d:%d",
778 wbuf->size, wbuf->lnum, wbuf->offs);
779 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
784 wbuf->offs += wbuf->size;
786 aligned_len -= wbuf->size;
787 written += wbuf->size;
791 * The remaining data may take more whole max. write units, so write the
792 * remains multiple to max. write unit size directly to the flash media.
793 * We align node length to 8-byte boundary because we anyway flash wbuf
794 * if the remaining space is less than 8 bytes.
796 n = aligned_len >> c->max_write_shift;
798 n <<= c->max_write_shift;
799 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
801 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
811 spin_lock(&wbuf->lock);
814 * And now we have what's left and what does not take whole
815 * max. write unit, so write it to the write-buffer and we are
818 memcpy(wbuf->buf, buf + written, len);
820 if (c->leb_size - wbuf->offs >= c->max_write_size)
821 wbuf->size = c->max_write_size;
823 wbuf->size = c->leb_size - wbuf->offs;
824 wbuf->avail = wbuf->size - aligned_len;
825 wbuf->used = aligned_len;
827 spin_unlock(&wbuf->lock);
830 if (wbuf->sync_callback) {
831 int free = c->leb_size - wbuf->offs - wbuf->used;
833 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
839 new_wbuf_timer_nolock(wbuf);
844 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
845 len, wbuf->lnum, wbuf->offs, err);
846 ubifs_dump_node(c, buf);
848 ubifs_dump_leb(c, wbuf->lnum);
853 * ubifs_write_node - write node to the media.
854 * @c: UBIFS file-system description object
855 * @buf: the node to write
857 * @lnum: logical eraseblock number
858 * @offs: offset within the logical eraseblock
860 * This function automatically fills node magic number, assigns sequence
861 * number, and calculates node CRC checksum. The length of the @buf buffer has
862 * to be aligned to the minimal I/O unit size. This function automatically
863 * appends padding node and padding bytes if needed. Returns zero in case of
864 * success and a negative error code in case of failure.
866 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
869 int err, buf_len = ALIGN(len, c->min_io_size);
871 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
872 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
874 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
875 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
876 ubifs_assert(!c->ro_media && !c->ro_mount);
877 ubifs_assert(!c->space_fixup);
882 ubifs_prepare_node(c, buf, len, 1);
883 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
885 ubifs_dump_node(c, buf);
892 * ubifs_read_node_wbuf - read node from the media or write-buffer.
893 * @wbuf: wbuf to check for un-written data
894 * @buf: buffer to read to
897 * @lnum: logical eraseblock number
898 * @offs: offset within the logical eraseblock
900 * This function reads a node of known type and length, checks it and stores
901 * in @buf. If the node partially or fully sits in the write-buffer, this
902 * function takes data from the buffer, otherwise it reads the flash media.
903 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
904 * error code in case of failure.
906 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
909 const struct ubifs_info *c = wbuf->c;
910 int err, rlen, overlap;
911 struct ubifs_ch *ch = buf;
913 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
914 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
915 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
916 ubifs_assert(!(offs & 7) && offs < c->leb_size);
917 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
919 spin_lock(&wbuf->lock);
920 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
922 /* We may safely unlock the write-buffer and read the data */
923 spin_unlock(&wbuf->lock);
924 return ubifs_read_node(c, buf, type, len, lnum, offs);
927 /* Don't read under wbuf */
928 rlen = wbuf->offs - offs;
932 /* Copy the rest from the write-buffer */
933 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
934 spin_unlock(&wbuf->lock);
937 /* Read everything that goes before write-buffer */
938 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
939 if (err && err != -EBADMSG)
943 if (type != ch->node_type) {
944 ubifs_err(c, "bad node type (%d but expected %d)",
945 ch->node_type, type);
949 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
951 ubifs_err(c, "expected node type %d", type);
955 rlen = le32_to_cpu(ch->len);
957 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
964 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
965 ubifs_dump_node(c, buf);
971 * ubifs_read_node - read node.
972 * @c: UBIFS file-system description object
973 * @buf: buffer to read to
975 * @len: node length (not aligned)
976 * @lnum: logical eraseblock number
977 * @offs: offset within the logical eraseblock
979 * This function reads a node of known type and and length, checks it and
980 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
981 * and a negative error code in case of failure.
983 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
987 struct ubifs_ch *ch = buf;
989 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
990 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
991 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
992 ubifs_assert(!(offs & 7) && offs < c->leb_size);
993 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
995 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
996 if (err && err != -EBADMSG)
999 if (type != ch->node_type) {
1000 ubifs_errc(c, "bad node type (%d but expected %d)",
1001 ch->node_type, type);
1005 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1007 ubifs_errc(c, "expected node type %d", type);
1011 l = le32_to_cpu(ch->len);
1013 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1020 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1021 offs, ubi_is_mapped(c->ubi, lnum));
1023 ubifs_dump_node(c, buf);
1030 * ubifs_wbuf_init - initialize write-buffer.
1031 * @c: UBIFS file-system description object
1032 * @wbuf: write-buffer to initialize
1034 * This function initializes write-buffer. Returns zero in case of success
1035 * %-ENOMEM in case of failure.
1037 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1041 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1045 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1046 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1047 if (!wbuf->inodes) {
1054 wbuf->lnum = wbuf->offs = -1;
1056 * If the LEB starts at the max. write size aligned address, then
1057 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1058 * set it to something smaller so that it ends at the closest max.
1059 * write size boundary.
1061 size = c->max_write_size - (c->leb_start % c->max_write_size);
1062 wbuf->avail = wbuf->size = size;
1063 wbuf->sync_callback = NULL;
1064 mutex_init(&wbuf->io_mutex);
1065 spin_lock_init(&wbuf->lock);
1070 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1071 wbuf->timer.function = wbuf_timer_callback_nolock;
1072 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1073 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1074 wbuf->delta *= 1000000000ULL;
1075 ubifs_assert(wbuf->delta <= ULONG_MAX);
1081 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1082 * @wbuf: the write-buffer where to add
1083 * @inum: the inode number
1085 * This function adds an inode number to the inode array of the write-buffer.
1087 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1090 /* NOR flash or something similar */
1093 spin_lock(&wbuf->lock);
1095 wbuf->inodes[wbuf->next_ino++] = inum;
1096 spin_unlock(&wbuf->lock);
1100 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1101 * @wbuf: the write-buffer
1102 * @inum: the inode number
1104 * This function returns with %1 if the write-buffer contains some data from the
1105 * given inode otherwise it returns with %0.
1107 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1111 spin_lock(&wbuf->lock);
1112 for (i = 0; i < wbuf->next_ino; i++)
1113 if (inum == wbuf->inodes[i]) {
1117 spin_unlock(&wbuf->lock);
1123 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1124 * @c: UBIFS file-system description object
1125 * @inode: inode to synchronize
1127 * This function synchronizes write-buffers which contain nodes belonging to
1128 * @inode. Returns zero in case of success and a negative error code in case of
1131 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1135 for (i = 0; i < c->jhead_cnt; i++) {
1136 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1140 * GC head is special, do not look at it. Even if the
1141 * head contains something related to this inode, it is
1142 * a _copy_ of corresponding on-flash node which sits
1147 if (!wbuf_has_ino(wbuf, inode->i_ino))
1150 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1151 if (wbuf_has_ino(wbuf, inode->i_ino))
1152 err = ubifs_wbuf_sync_nolock(wbuf);
1153 mutex_unlock(&wbuf->io_mutex);
1156 ubifs_ro_mode(c, err);