1 // SPDX-License-Identifier: GPL-2.0-only
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.
61 #include <linux/crc32.h>
62 #include <linux/slab.h>
66 * ubifs_ro_mode - switch UBIFS to read read-only mode.
67 * @c: UBIFS file-system description object
68 * @err: error code which is the reason of switching to R/O mode
70 void ubifs_ro_mode(struct ubifs_info *c, int err)
74 c->no_chk_data_crc = 0;
75 c->vfs_sb->s_flags |= SB_RDONLY;
76 ubifs_warn(c, "switched to read-only mode, error %d", err);
82 * Below are simple wrappers over UBI I/O functions which include some
83 * additional checks and UBIFS debugging stuff. See corresponding UBI function
84 * for more information.
87 int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
88 int len, int even_ebadmsg)
92 err = ubi_read(c->ubi, lnum, buf, offs, len);
94 * In case of %-EBADMSG print the error message only if the
95 * @even_ebadmsg is true.
97 if (err && (err != -EBADMSG || even_ebadmsg)) {
98 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
99 len, lnum, offs, err);
105 int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
110 ubifs_assert(c, !c->ro_media && !c->ro_mount);
113 if (!dbg_is_tst_rcvry(c))
114 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
116 err = dbg_leb_write(c, lnum, buf, offs, len);
118 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
119 len, lnum, offs, err);
120 ubifs_ro_mode(c, err);
126 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
130 ubifs_assert(c, !c->ro_media && !c->ro_mount);
133 if (!dbg_is_tst_rcvry(c))
134 err = ubi_leb_change(c->ubi, lnum, buf, len);
136 err = dbg_leb_change(c, lnum, buf, len);
138 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
140 ubifs_ro_mode(c, err);
146 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
150 ubifs_assert(c, !c->ro_media && !c->ro_mount);
153 if (!dbg_is_tst_rcvry(c))
154 err = ubi_leb_unmap(c->ubi, lnum);
156 err = dbg_leb_unmap(c, lnum);
158 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
159 ubifs_ro_mode(c, err);
165 int ubifs_leb_map(struct ubifs_info *c, int lnum)
169 ubifs_assert(c, !c->ro_media && !c->ro_mount);
172 if (!dbg_is_tst_rcvry(c))
173 err = ubi_leb_map(c->ubi, lnum);
175 err = dbg_leb_map(c, lnum);
177 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
178 ubifs_ro_mode(c, err);
184 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
188 err = ubi_is_mapped(c->ubi, lnum);
190 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
197 static void record_magic_error(struct ubifs_stats_info *stats)
200 stats->magic_errors++;
203 static void record_node_error(struct ubifs_stats_info *stats)
206 stats->node_errors++;
209 static void record_crc_error(struct ubifs_stats_info *stats)
216 * ubifs_check_node - check node.
217 * @c: UBIFS file-system description object
218 * @buf: node to check
220 * @lnum: logical eraseblock number
221 * @offs: offset within the logical eraseblock
222 * @quiet: print no messages
223 * @must_chk_crc: indicates whether to always check the CRC
225 * This function checks node magic number and CRC checksum. This function also
226 * validates node length to prevent UBIFS from becoming crazy when an attacker
227 * feeds it a file-system image with incorrect nodes. For example, too large
228 * node length in the common header could cause UBIFS to read memory outside of
229 * allocated buffer when checking the CRC checksum.
231 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
232 * true, which is controlled by corresponding UBIFS mount option. However, if
233 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
234 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
235 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
236 * is checked. This is because during mounting or re-mounting from R/O mode to
237 * R/W mode we may read journal nodes (when replying the journal or doing the
238 * recovery) and the journal nodes may potentially be corrupted, so checking is
241 * This function returns zero in case of success and %-EUCLEAN in case of bad
244 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int len,
245 int lnum, int offs, int quiet, int must_chk_crc)
247 int err = -EINVAL, type, node_len;
248 uint32_t crc, node_crc, magic;
249 const struct ubifs_ch *ch = buf;
251 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
252 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
254 magic = le32_to_cpu(ch->magic);
255 if (magic != UBIFS_NODE_MAGIC) {
257 ubifs_err(c, "bad magic %#08x, expected %#08x",
258 magic, UBIFS_NODE_MAGIC);
259 record_magic_error(c->stats);
264 type = ch->node_type;
265 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
267 ubifs_err(c, "bad node type %d", type);
268 record_node_error(c->stats);
272 node_len = le32_to_cpu(ch->len);
273 if (node_len + offs > c->leb_size)
276 if (c->ranges[type].max_len == 0) {
277 if (node_len != c->ranges[type].len)
279 } else if (node_len < c->ranges[type].min_len ||
280 node_len > c->ranges[type].max_len)
283 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
284 !c->remounting_rw && c->no_chk_data_crc)
287 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
288 node_crc = le32_to_cpu(ch->crc);
289 if (crc != node_crc) {
291 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
293 record_crc_error(c->stats);
302 ubifs_err(c, "bad node length %d", node_len);
305 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
306 ubifs_dump_node(c, buf, len);
313 * ubifs_pad - pad flash space.
314 * @c: UBIFS file-system description object
315 * @buf: buffer to put padding to
316 * @pad: how many bytes to pad
318 * The flash media obliges us to write only in chunks of %c->min_io_size and
319 * when we have to write less data we add padding node to the write-buffer and
320 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
321 * media is being scanned. If the amount of wasted space is not enough to fit a
322 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
323 * pattern (%UBIFS_PADDING_BYTE).
325 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
328 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
332 ubifs_assert(c, pad >= 0);
334 if (pad >= UBIFS_PAD_NODE_SZ) {
335 struct ubifs_ch *ch = buf;
336 struct ubifs_pad_node *pad_node = buf;
338 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
339 ch->node_type = UBIFS_PAD_NODE;
340 ch->group_type = UBIFS_NO_NODE_GROUP;
341 ch->padding[0] = ch->padding[1] = 0;
343 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
344 pad -= UBIFS_PAD_NODE_SZ;
345 pad_node->pad_len = cpu_to_le32(pad);
346 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
347 ch->crc = cpu_to_le32(crc);
348 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
350 /* Too little space, padding node won't fit */
351 memset(buf, UBIFS_PADDING_BYTE, pad);
355 * next_sqnum - get next sequence number.
356 * @c: UBIFS file-system description object
358 static unsigned long long next_sqnum(struct ubifs_info *c)
360 unsigned long long sqnum;
362 spin_lock(&c->cnt_lock);
363 sqnum = ++c->max_sqnum;
364 spin_unlock(&c->cnt_lock);
366 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
367 if (sqnum >= SQNUM_WATERMARK) {
368 ubifs_err(c, "sequence number overflow %llu, end of life",
370 ubifs_ro_mode(c, -EINVAL);
372 ubifs_warn(c, "running out of sequence numbers, end of life soon");
378 void ubifs_init_node(struct ubifs_info *c, void *node, int len, int pad)
380 struct ubifs_ch *ch = node;
381 unsigned long long sqnum = next_sqnum(c);
383 ubifs_assert(c, len >= UBIFS_CH_SZ);
385 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
386 ch->len = cpu_to_le32(len);
387 ch->group_type = UBIFS_NO_NODE_GROUP;
388 ch->sqnum = cpu_to_le64(sqnum);
389 ch->padding[0] = ch->padding[1] = 0;
393 pad = ALIGN(len, c->min_io_size) - len;
394 ubifs_pad(c, node + len, pad);
398 void ubifs_crc_node(struct ubifs_info *c, void *node, int len)
400 struct ubifs_ch *ch = node;
403 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
404 ch->crc = cpu_to_le32(crc);
408 * ubifs_prepare_node_hmac - prepare node to be written to flash.
409 * @c: UBIFS file-system description object
410 * @node: the node to pad
412 * @hmac_offs: offset of the HMAC in the node
413 * @pad: if the buffer has to be padded
415 * This function prepares node at @node to be written to the media - it
416 * calculates node CRC, fills the common header, and adds proper padding up to
417 * the next minimum I/O unit if @pad is not zero. if @hmac_offs is positive then
418 * a HMAC is inserted into the node at the given offset.
420 * This function returns 0 for success or a negative error code otherwise.
422 int ubifs_prepare_node_hmac(struct ubifs_info *c, void *node, int len,
423 int hmac_offs, int pad)
427 ubifs_init_node(c, node, len, pad);
430 err = ubifs_node_insert_hmac(c, node, len, hmac_offs);
435 ubifs_crc_node(c, node, len);
441 * ubifs_prepare_node - prepare node to be written to flash.
442 * @c: UBIFS file-system description object
443 * @node: the node to pad
445 * @pad: if the buffer has to be padded
447 * This function prepares node at @node to be written to the media - it
448 * calculates node CRC, fills the common header, and adds proper padding up to
449 * the next minimum I/O unit if @pad is not zero.
451 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
454 * Deliberately ignore return value since this function can only fail
455 * when a hmac offset is given.
457 ubifs_prepare_node_hmac(c, node, len, 0, pad);
461 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
462 * @c: UBIFS file-system description object
463 * @node: the node to pad
465 * @last: indicates the last node of the group
467 * This function prepares node at @node to be written to the media - it
468 * calculates node CRC and fills the common header.
470 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
473 struct ubifs_ch *ch = node;
474 unsigned long long sqnum = next_sqnum(c);
476 ubifs_assert(c, len >= UBIFS_CH_SZ);
478 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
479 ch->len = cpu_to_le32(len);
481 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
483 ch->group_type = UBIFS_IN_NODE_GROUP;
484 ch->sqnum = cpu_to_le64(sqnum);
485 ch->padding[0] = ch->padding[1] = 0;
486 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
487 ch->crc = cpu_to_le32(crc);
491 * wbuf_timer_callback - write-buffer timer callback function.
492 * @timer: timer data (write-buffer descriptor)
494 * This function is called when the write-buffer timer expires.
496 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
498 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
500 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
502 wbuf->c->need_wbuf_sync = 1;
503 ubifs_wake_up_bgt(wbuf->c);
504 return HRTIMER_NORESTART;
508 * new_wbuf_timer - start new write-buffer timer.
509 * @c: UBIFS file-system description object
510 * @wbuf: write-buffer descriptor
512 static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
514 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
515 unsigned long long delta = dirty_writeback_interval;
517 /* centi to milli, milli to nano, then 10% */
518 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
520 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
521 ubifs_assert(c, delta <= ULONG_MAX);
525 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
526 dbg_jhead(wbuf->jhead),
527 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
528 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
529 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
534 * cancel_wbuf_timer - cancel write-buffer timer.
535 * @wbuf: write-buffer descriptor
537 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
542 hrtimer_cancel(&wbuf->timer);
546 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
547 * @wbuf: write-buffer to synchronize
549 * This function synchronizes write-buffer @buf and returns zero in case of
550 * success or a negative error code in case of failure.
552 * Note, although write-buffers are of @c->max_write_size, this function does
553 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
554 * if the write-buffer is only partially filled with data, only the used part
555 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
556 * This way we waste less space.
558 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
560 struct ubifs_info *c = wbuf->c;
561 int err, dirt, sync_len;
563 cancel_wbuf_timer_nolock(wbuf);
564 if (!wbuf->used || wbuf->lnum == -1)
565 /* Write-buffer is empty or not seeked */
568 dbg_io("LEB %d:%d, %d bytes, jhead %s",
569 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
570 ubifs_assert(c, !(wbuf->avail & 7));
571 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
572 ubifs_assert(c, wbuf->size >= c->min_io_size);
573 ubifs_assert(c, wbuf->size <= c->max_write_size);
574 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
575 ubifs_assert(c, !c->ro_media && !c->ro_mount);
576 if (c->leb_size - wbuf->offs >= c->max_write_size)
577 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
583 * Do not write whole write buffer but write only the minimum necessary
584 * amount of min. I/O units.
586 sync_len = ALIGN(wbuf->used, c->min_io_size);
587 dirt = sync_len - wbuf->used;
589 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
590 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
594 spin_lock(&wbuf->lock);
595 wbuf->offs += sync_len;
597 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
598 * But our goal is to optimize writes and make sure we write in
599 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
600 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
601 * sure that @wbuf->offs + @wbuf->size is aligned to
602 * @c->max_write_size. This way we make sure that after next
603 * write-buffer flush we are again at the optimal offset (aligned to
604 * @c->max_write_size).
606 if (c->leb_size - wbuf->offs < c->max_write_size)
607 wbuf->size = c->leb_size - wbuf->offs;
608 else if (wbuf->offs & (c->max_write_size - 1))
609 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
611 wbuf->size = c->max_write_size;
612 wbuf->avail = wbuf->size;
615 spin_unlock(&wbuf->lock);
617 if (wbuf->sync_callback)
618 err = wbuf->sync_callback(c, wbuf->lnum,
619 c->leb_size - wbuf->offs, dirt);
624 * ubifs_wbuf_seek_nolock - seek write-buffer.
625 * @wbuf: write-buffer
626 * @lnum: logical eraseblock number to seek to
627 * @offs: logical eraseblock offset to seek to
629 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
630 * The write-buffer has to be empty. Returns zero in case of success and a
631 * negative error code in case of failure.
633 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
635 const struct ubifs_info *c = wbuf->c;
637 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
638 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
639 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
640 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
641 ubifs_assert(c, lnum != wbuf->lnum);
642 ubifs_assert(c, wbuf->used == 0);
644 spin_lock(&wbuf->lock);
647 if (c->leb_size - wbuf->offs < c->max_write_size)
648 wbuf->size = c->leb_size - wbuf->offs;
649 else if (wbuf->offs & (c->max_write_size - 1))
650 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
652 wbuf->size = c->max_write_size;
653 wbuf->avail = wbuf->size;
655 spin_unlock(&wbuf->lock);
661 * ubifs_bg_wbufs_sync - synchronize write-buffers.
662 * @c: UBIFS file-system description object
664 * This function is called by background thread to synchronize write-buffers.
665 * Returns zero in case of success and a negative error code in case of
668 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
672 ubifs_assert(c, !c->ro_media && !c->ro_mount);
673 if (!c->need_wbuf_sync)
675 c->need_wbuf_sync = 0;
682 dbg_io("synchronize");
683 for (i = 0; i < c->jhead_cnt; i++) {
684 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
689 * If the mutex is locked then wbuf is being changed, so
690 * synchronization is not necessary.
692 if (mutex_is_locked(&wbuf->io_mutex))
695 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
696 if (!wbuf->need_sync) {
697 mutex_unlock(&wbuf->io_mutex);
701 err = ubifs_wbuf_sync_nolock(wbuf);
702 mutex_unlock(&wbuf->io_mutex);
704 ubifs_err(c, "cannot sync write-buffer, error %d", err);
705 ubifs_ro_mode(c, err);
713 /* Cancel all timers to prevent repeated errors */
714 for (i = 0; i < c->jhead_cnt; i++) {
715 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
717 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
718 cancel_wbuf_timer_nolock(wbuf);
719 mutex_unlock(&wbuf->io_mutex);
725 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
726 * @wbuf: write-buffer
727 * @buf: node to write
730 * This function writes data to flash via write-buffer @wbuf. This means that
731 * the last piece of the node won't reach the flash media immediately if it
732 * does not take whole max. write unit (@c->max_write_size). Instead, the node
733 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
734 * because more data are appended to the write-buffer).
736 * This function returns zero in case of success and a negative error code in
737 * case of failure. If the node cannot be written because there is no more
738 * space in this logical eraseblock, %-ENOSPC is returned.
740 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
742 struct ubifs_info *c = wbuf->c;
743 int err, n, written = 0, aligned_len = ALIGN(len, 8);
745 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
746 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
747 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
748 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
749 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
750 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
751 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
752 ubifs_assert(c, wbuf->size >= c->min_io_size);
753 ubifs_assert(c, wbuf->size <= c->max_write_size);
754 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
755 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
756 ubifs_assert(c, !c->ro_media && !c->ro_mount);
757 ubifs_assert(c, !c->space_fixup);
758 if (c->leb_size - wbuf->offs >= c->max_write_size)
759 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
761 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
766 cancel_wbuf_timer_nolock(wbuf);
771 if (aligned_len <= wbuf->avail) {
773 * The node is not very large and fits entirely within
776 memcpy(wbuf->buf + wbuf->used, buf, len);
777 if (aligned_len > len) {
778 ubifs_assert(c, aligned_len - len < 8);
779 ubifs_pad(c, wbuf->buf + wbuf->used + len, aligned_len - len);
782 if (aligned_len == wbuf->avail) {
783 dbg_io("flush jhead %s wbuf to LEB %d:%d",
784 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
785 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
786 wbuf->offs, wbuf->size);
790 spin_lock(&wbuf->lock);
791 wbuf->offs += wbuf->size;
792 if (c->leb_size - wbuf->offs >= c->max_write_size)
793 wbuf->size = c->max_write_size;
795 wbuf->size = c->leb_size - wbuf->offs;
796 wbuf->avail = wbuf->size;
799 spin_unlock(&wbuf->lock);
801 spin_lock(&wbuf->lock);
802 wbuf->avail -= aligned_len;
803 wbuf->used += aligned_len;
804 spin_unlock(&wbuf->lock);
812 * The node is large enough and does not fit entirely within
813 * current available space. We have to fill and flush
814 * write-buffer and switch to the next max. write unit.
816 dbg_io("flush jhead %s wbuf to LEB %d:%d",
817 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
818 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
819 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
824 wbuf->offs += wbuf->size;
826 aligned_len -= wbuf->avail;
827 written += wbuf->avail;
828 } else if (wbuf->offs & (c->max_write_size - 1)) {
830 * The write-buffer offset is not aligned to
831 * @c->max_write_size and @wbuf->size is less than
832 * @c->max_write_size. Write @wbuf->size bytes to make sure the
833 * following writes are done in optimal @c->max_write_size
836 dbg_io("write %d bytes to LEB %d:%d",
837 wbuf->size, wbuf->lnum, wbuf->offs);
838 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
843 wbuf->offs += wbuf->size;
845 aligned_len -= wbuf->size;
846 written += wbuf->size;
850 * The remaining data may take more whole max. write units, so write the
851 * remains multiple to max. write unit size directly to the flash media.
852 * We align node length to 8-byte boundary because we anyway flash wbuf
853 * if the remaining space is less than 8 bytes.
855 n = aligned_len >> c->max_write_shift;
859 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
863 /* '(n-1)<<c->max_write_shift < len' is always true. */
864 m <<= c->max_write_shift;
865 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
876 * The non-written len of buf may be less than 'n' because
877 * parameter 'len' is not 8 bytes aligned, so here we read
878 * min(len, n) bytes from buf.
880 n = 1 << c->max_write_shift;
881 memcpy(wbuf->buf, buf + written, min(len, n));
883 ubifs_assert(c, n - len < 8);
884 ubifs_pad(c, wbuf->buf + len, n - len);
887 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, n);
896 spin_lock(&wbuf->lock);
899 * And now we have what's left and what does not take whole
900 * max. write unit, so write it to the write-buffer and we are
903 memcpy(wbuf->buf, buf + written, len);
904 if (aligned_len > len) {
905 ubifs_assert(c, aligned_len - len < 8);
906 ubifs_pad(c, wbuf->buf + len, aligned_len - len);
910 if (c->leb_size - wbuf->offs >= c->max_write_size)
911 wbuf->size = c->max_write_size;
913 wbuf->size = c->leb_size - wbuf->offs;
914 wbuf->avail = wbuf->size - aligned_len;
915 wbuf->used = aligned_len;
917 spin_unlock(&wbuf->lock);
920 if (wbuf->sync_callback) {
921 int free = c->leb_size - wbuf->offs - wbuf->used;
923 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
929 new_wbuf_timer_nolock(c, wbuf);
934 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
935 len, wbuf->lnum, wbuf->offs, err);
936 ubifs_dump_node(c, buf, written + len);
938 ubifs_dump_leb(c, wbuf->lnum);
943 * ubifs_write_node_hmac - write node to the media.
944 * @c: UBIFS file-system description object
945 * @buf: the node to write
947 * @lnum: logical eraseblock number
948 * @offs: offset within the logical eraseblock
949 * @hmac_offs: offset of the HMAC within the node
951 * This function automatically fills node magic number, assigns sequence
952 * number, and calculates node CRC checksum. The length of the @buf buffer has
953 * to be aligned to the minimal I/O unit size. This function automatically
954 * appends padding node and padding bytes if needed. Returns zero in case of
955 * success and a negative error code in case of failure.
957 int ubifs_write_node_hmac(struct ubifs_info *c, void *buf, int len, int lnum,
958 int offs, int hmac_offs)
960 int err, buf_len = ALIGN(len, c->min_io_size);
962 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
963 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
965 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
966 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
967 ubifs_assert(c, !c->ro_media && !c->ro_mount);
968 ubifs_assert(c, !c->space_fixup);
973 err = ubifs_prepare_node_hmac(c, buf, len, hmac_offs, 1);
977 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
979 ubifs_dump_node(c, buf, len);
985 * ubifs_write_node - write node to the media.
986 * @c: UBIFS file-system description object
987 * @buf: the node to write
989 * @lnum: logical eraseblock number
990 * @offs: offset within the logical eraseblock
992 * This function automatically fills node magic number, assigns sequence
993 * number, and calculates node CRC checksum. The length of the @buf buffer has
994 * to be aligned to the minimal I/O unit size. This function automatically
995 * appends padding node and padding bytes if needed. Returns zero in case of
996 * success and a negative error code in case of failure.
998 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
1001 return ubifs_write_node_hmac(c, buf, len, lnum, offs, -1);
1005 * ubifs_read_node_wbuf - read node from the media or write-buffer.
1006 * @wbuf: wbuf to check for un-written data
1007 * @buf: buffer to read to
1010 * @lnum: logical eraseblock number
1011 * @offs: offset within the logical eraseblock
1013 * This function reads a node of known type and length, checks it and stores
1014 * in @buf. If the node partially or fully sits in the write-buffer, this
1015 * function takes data from the buffer, otherwise it reads the flash media.
1016 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
1017 * error code in case of failure.
1019 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
1022 const struct ubifs_info *c = wbuf->c;
1023 int err, rlen, overlap;
1024 struct ubifs_ch *ch = buf;
1026 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
1027 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
1028 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1029 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1030 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1032 spin_lock(&wbuf->lock);
1033 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1035 /* We may safely unlock the write-buffer and read the data */
1036 spin_unlock(&wbuf->lock);
1037 return ubifs_read_node(c, buf, type, len, lnum, offs);
1040 /* Don't read under wbuf */
1041 rlen = wbuf->offs - offs;
1045 /* Copy the rest from the write-buffer */
1046 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1047 spin_unlock(&wbuf->lock);
1050 /* Read everything that goes before write-buffer */
1051 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1052 if (err && err != -EBADMSG)
1056 if (type != ch->node_type) {
1057 ubifs_err(c, "bad node type (%d but expected %d)",
1058 ch->node_type, type);
1062 err = ubifs_check_node(c, buf, len, lnum, offs, 0, 0);
1064 ubifs_err(c, "expected node type %d", type);
1068 rlen = le32_to_cpu(ch->len);
1070 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
1077 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
1078 ubifs_dump_node(c, buf, len);
1084 * ubifs_read_node - read node.
1085 * @c: UBIFS file-system description object
1086 * @buf: buffer to read to
1088 * @len: node length (not aligned)
1089 * @lnum: logical eraseblock number
1090 * @offs: offset within the logical eraseblock
1092 * This function reads a node of known type and length, checks it and
1093 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1094 * and a negative error code in case of failure.
1096 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1100 struct ubifs_ch *ch = buf;
1102 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1103 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1104 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1105 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1106 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1108 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1109 if (err && err != -EBADMSG)
1112 if (type != ch->node_type) {
1113 ubifs_errc(c, "bad node type (%d but expected %d)",
1114 ch->node_type, type);
1118 err = ubifs_check_node(c, buf, len, lnum, offs, 0, 0);
1120 ubifs_errc(c, "expected node type %d", type);
1124 l = le32_to_cpu(ch->len);
1126 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1133 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1134 offs, ubi_is_mapped(c->ubi, lnum));
1136 ubifs_dump_node(c, buf, len);
1143 * ubifs_wbuf_init - initialize write-buffer.
1144 * @c: UBIFS file-system description object
1145 * @wbuf: write-buffer to initialize
1147 * This function initializes write-buffer. Returns zero in case of success
1148 * %-ENOMEM in case of failure.
1150 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1154 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1158 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1159 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1160 if (!wbuf->inodes) {
1167 wbuf->lnum = wbuf->offs = -1;
1169 * If the LEB starts at the max. write size aligned address, then
1170 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1171 * set it to something smaller so that it ends at the closest max.
1172 * write size boundary.
1174 size = c->max_write_size - (c->leb_start % c->max_write_size);
1175 wbuf->avail = wbuf->size = size;
1176 wbuf->sync_callback = NULL;
1177 mutex_init(&wbuf->io_mutex);
1178 spin_lock_init(&wbuf->lock);
1182 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1183 wbuf->timer.function = wbuf_timer_callback_nolock;
1188 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1189 * @wbuf: the write-buffer where to add
1190 * @inum: the inode number
1192 * This function adds an inode number to the inode array of the write-buffer.
1194 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1197 /* NOR flash or something similar */
1200 spin_lock(&wbuf->lock);
1202 wbuf->inodes[wbuf->next_ino++] = inum;
1203 spin_unlock(&wbuf->lock);
1207 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1208 * @wbuf: the write-buffer
1209 * @inum: the inode number
1211 * This function returns with %1 if the write-buffer contains some data from the
1212 * given inode otherwise it returns with %0.
1214 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1218 spin_lock(&wbuf->lock);
1219 for (i = 0; i < wbuf->next_ino; i++)
1220 if (inum == wbuf->inodes[i]) {
1224 spin_unlock(&wbuf->lock);
1230 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1231 * @c: UBIFS file-system description object
1232 * @inode: inode to synchronize
1234 * This function synchronizes write-buffers which contain nodes belonging to
1235 * @inode. Returns zero in case of success and a negative error code in case of
1238 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1242 for (i = 0; i < c->jhead_cnt; i++) {
1243 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1247 * GC head is special, do not look at it. Even if the
1248 * head contains something related to this inode, it is
1249 * a _copy_ of corresponding on-flash node which sits
1254 if (!wbuf_has_ino(wbuf, inode->i_ino))
1257 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1258 if (wbuf_has_ino(wbuf, inode->i_ino))
1259 err = ubifs_wbuf_sync_nolock(wbuf);
1260 mutex_unlock(&wbuf->io_mutex);
1263 ubifs_ro_mode(c, err);