2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 * Copyright (C) 2006, 2007 University of Szeged, Hungary
7 * SPDX-License-Identifier: GPL-2.0+
9 * Authors: Artem Bityutskiy (Битюцкий Артём)
15 * This file implements UBIFS I/O subsystem which provides various I/O-related
16 * helper functions (reading/writing/checking/validating nodes) and implements
17 * write-buffering support. Write buffers help to save space which otherwise
18 * would have been wasted for padding to the nearest minimal I/O unit boundary.
19 * Instead, data first goes to the write-buffer and is flushed when the
20 * buffer is full or when it is not used for some time (by timer). This is
21 * similar to the mechanism is used by JFFS2.
23 * UBIFS distinguishes between minimum write size (@c->min_io_size) and maximum
24 * write size (@c->max_write_size). The latter is the maximum amount of bytes
25 * the underlying flash is able to program at a time, and writing in
26 * @c->max_write_size units should presumably be faster. Obviously,
27 * @c->min_io_size <= @c->max_write_size. Write-buffers are of
28 * @c->max_write_size bytes in size for maximum performance. However, when a
29 * write-buffer is flushed, only the portion of it (aligned to @c->min_io_size
30 * boundary) which contains data is written, not the whole write-buffer,
31 * because this is more space-efficient.
33 * This optimization adds few complications to the code. Indeed, on the one
34 * hand, we want to write in optimal @c->max_write_size bytes chunks, which
35 * also means aligning writes at the @c->max_write_size bytes offsets. On the
36 * other hand, we do not want to waste space when synchronizing the write
37 * buffer, so during synchronization we writes in smaller chunks. And this makes
38 * the next write offset to be not aligned to @c->max_write_size bytes. So the
39 * have to make sure that the write-buffer offset (@wbuf->offs) becomes aligned
40 * to @c->max_write_size bytes again. We do this by temporarily shrinking
41 * write-buffer size (@wbuf->size).
43 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
44 * mutexes defined inside these objects. Since sometimes upper-level code
45 * has to lock the write-buffer (e.g. journal space reservation code), many
46 * functions related to write-buffers have "nolock" suffix which means that the
47 * caller has to lock the write-buffer before calling this function.
49 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
50 * aligned, UBIFS starts the next node from the aligned address, and the padded
51 * bytes may contain any rubbish. In other words, UBIFS does not put padding
52 * bytes in those small gaps. Common headers of nodes store real node lengths,
53 * not aligned lengths. Indexing nodes also store real lengths in branches.
55 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
56 * uses padding nodes or padding bytes, if the padding node does not fit.
58 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
59 * they are read from the flash media.
63 #include <linux/crc32.h>
64 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/err.h>
72 * ubifs_ro_mode - switch UBIFS to read read-only mode.
73 * @c: UBIFS file-system description object
74 * @err: error code which is the reason of switching to R/O mode
76 void ubifs_ro_mode(struct ubifs_info *c, int err)
80 c->no_chk_data_crc = 0;
81 c->vfs_sb->s_flags |= MS_RDONLY;
82 ubifs_warn(c, "switched to read-only mode, error %d", err);
88 * Below are simple wrappers over UBI I/O functions which include some
89 * additional checks and UBIFS debugging stuff. See corresponding UBI function
90 * for more information.
93 int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
94 int len, int even_ebadmsg)
98 err = ubi_read(c->ubi, lnum, buf, offs, len);
100 * In case of %-EBADMSG print the error message only if the
101 * @even_ebadmsg is true.
103 if (err && (err != -EBADMSG || even_ebadmsg)) {
104 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
105 len, lnum, offs, err);
111 int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
116 ubifs_assert(!c->ro_media && !c->ro_mount);
119 if (!dbg_is_tst_rcvry(c))
120 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
123 err = dbg_leb_write(c, lnum, buf, offs, len);
126 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
127 len, lnum, offs, err);
128 ubifs_ro_mode(c, err);
134 int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
138 ubifs_assert(!c->ro_media && !c->ro_mount);
141 if (!dbg_is_tst_rcvry(c))
142 err = ubi_leb_change(c->ubi, lnum, buf, len);
145 err = dbg_leb_change(c, lnum, buf, len);
148 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
150 ubifs_ro_mode(c, err);
156 int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
160 ubifs_assert(!c->ro_media && !c->ro_mount);
163 if (!dbg_is_tst_rcvry(c))
164 err = ubi_leb_unmap(c->ubi, lnum);
167 err = dbg_leb_unmap(c, lnum);
170 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
171 ubifs_ro_mode(c, err);
177 int ubifs_leb_map(struct ubifs_info *c, int lnum)
181 ubifs_assert(!c->ro_media && !c->ro_mount);
184 if (!dbg_is_tst_rcvry(c))
185 err = ubi_leb_map(c->ubi, lnum);
188 err = dbg_leb_map(c, lnum);
191 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
192 ubifs_ro_mode(c, err);
198 int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
202 err = ubi_is_mapped(c->ubi, lnum);
204 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
212 * ubifs_check_node - check node.
213 * @c: UBIFS file-system description object
214 * @buf: node to check
215 * @lnum: logical eraseblock number
216 * @offs: offset within the logical eraseblock
217 * @quiet: print no messages
218 * @must_chk_crc: indicates whether to always check the CRC
220 * This function checks node magic number and CRC checksum. This function also
221 * validates node length to prevent UBIFS from becoming crazy when an attacker
222 * feeds it a file-system image with incorrect nodes. For example, too large
223 * node length in the common header could cause UBIFS to read memory outside of
224 * allocated buffer when checking the CRC checksum.
226 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
227 * true, which is controlled by corresponding UBIFS mount option. However, if
228 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
229 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
230 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
231 * is checked. This is because during mounting or re-mounting from R/O mode to
232 * R/W mode we may read journal nodes (when replying the journal or doing the
233 * recovery) and the journal nodes may potentially be corrupted, so checking is
236 * This function returns zero in case of success and %-EUCLEAN in case of bad
239 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
240 int offs, int quiet, int must_chk_crc)
242 int err = -EINVAL, type, node_len;
243 uint32_t crc, node_crc, magic;
244 const struct ubifs_ch *ch = buf;
246 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
247 ubifs_assert(!(offs & 7) && offs < c->leb_size);
249 magic = le32_to_cpu(ch->magic);
250 if (magic != UBIFS_NODE_MAGIC) {
252 ubifs_err(c, "bad magic %#08x, expected %#08x",
253 magic, UBIFS_NODE_MAGIC);
258 type = ch->node_type;
259 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
261 ubifs_err(c, "bad node type %d", type);
265 node_len = le32_to_cpu(ch->len);
266 if (node_len + offs > c->leb_size)
269 if (c->ranges[type].max_len == 0) {
270 if (node_len != c->ranges[type].len)
272 } else if (node_len < c->ranges[type].min_len ||
273 node_len > c->ranges[type].max_len)
276 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
277 !c->remounting_rw && c->no_chk_data_crc)
280 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
281 node_crc = le32_to_cpu(ch->crc);
282 if (crc != node_crc) {
284 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
294 ubifs_err(c, "bad node length %d", node_len);
297 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
298 ubifs_dump_node(c, buf);
305 * ubifs_pad - pad flash space.
306 * @c: UBIFS file-system description object
307 * @buf: buffer to put padding to
308 * @pad: how many bytes to pad
310 * The flash media obliges us to write only in chunks of %c->min_io_size and
311 * when we have to write less data we add padding node to the write-buffer and
312 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
313 * media is being scanned. If the amount of wasted space is not enough to fit a
314 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
315 * pattern (%UBIFS_PADDING_BYTE).
317 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
320 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
324 ubifs_assert(pad >= 0 && !(pad & 7));
326 if (pad >= UBIFS_PAD_NODE_SZ) {
327 struct ubifs_ch *ch = buf;
328 struct ubifs_pad_node *pad_node = buf;
330 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
331 ch->node_type = UBIFS_PAD_NODE;
332 ch->group_type = UBIFS_NO_NODE_GROUP;
333 ch->padding[0] = ch->padding[1] = 0;
335 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
336 pad -= UBIFS_PAD_NODE_SZ;
337 pad_node->pad_len = cpu_to_le32(pad);
338 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
339 ch->crc = cpu_to_le32(crc);
340 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
342 /* Too little space, padding node won't fit */
343 memset(buf, UBIFS_PADDING_BYTE, pad);
347 * next_sqnum - get next sequence number.
348 * @c: UBIFS file-system description object
350 static unsigned long long next_sqnum(struct ubifs_info *c)
352 unsigned long long sqnum;
354 spin_lock(&c->cnt_lock);
355 sqnum = ++c->max_sqnum;
356 spin_unlock(&c->cnt_lock);
358 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
359 if (sqnum >= SQNUM_WATERMARK) {
360 ubifs_err(c, "sequence number overflow %llu, end of life",
362 ubifs_ro_mode(c, -EINVAL);
364 ubifs_warn(c, "running out of sequence numbers, end of life soon");
371 * ubifs_prepare_node - prepare node to be written to flash.
372 * @c: UBIFS file-system description object
373 * @node: the node to pad
375 * @pad: if the buffer has to be padded
377 * This function prepares node at @node to be written to the media - it
378 * calculates node CRC, fills the common header, and adds proper padding up to
379 * the next minimum I/O unit if @pad is not zero.
381 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
384 struct ubifs_ch *ch = node;
385 unsigned long long sqnum = next_sqnum(c);
387 ubifs_assert(len >= UBIFS_CH_SZ);
389 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
390 ch->len = cpu_to_le32(len);
391 ch->group_type = UBIFS_NO_NODE_GROUP;
392 ch->sqnum = cpu_to_le64(sqnum);
393 ch->padding[0] = ch->padding[1] = 0;
394 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
395 ch->crc = cpu_to_le32(crc);
399 pad = ALIGN(len, c->min_io_size) - len;
400 ubifs_pad(c, node + len, pad);
405 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
406 * @c: UBIFS file-system description object
407 * @node: the node to pad
409 * @last: indicates the last node of the group
411 * This function prepares node at @node to be written to the media - it
412 * calculates node CRC and fills the common header.
414 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
417 struct ubifs_ch *ch = node;
418 unsigned long long sqnum = next_sqnum(c);
420 ubifs_assert(len >= UBIFS_CH_SZ);
422 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
423 ch->len = cpu_to_le32(len);
425 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
427 ch->group_type = UBIFS_IN_NODE_GROUP;
428 ch->sqnum = cpu_to_le64(sqnum);
429 ch->padding[0] = ch->padding[1] = 0;
430 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
431 ch->crc = cpu_to_le32(crc);
436 * wbuf_timer_callback - write-buffer timer callback function.
437 * @timer: timer data (write-buffer descriptor)
439 * This function is called when the write-buffer timer expires.
441 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
443 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
445 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
447 wbuf->c->need_wbuf_sync = 1;
448 ubifs_wake_up_bgt(wbuf->c);
449 return HRTIMER_NORESTART;
453 * new_wbuf_timer - start new write-buffer timer.
454 * @wbuf: write-buffer descriptor
456 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
458 ubifs_assert(!hrtimer_active(&wbuf->timer));
462 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
463 dbg_jhead(wbuf->jhead),
464 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
465 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
467 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
473 * cancel_wbuf_timer - cancel write-buffer timer.
474 * @wbuf: write-buffer descriptor
476 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
482 hrtimer_cancel(&wbuf->timer);
487 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
488 * @wbuf: write-buffer to synchronize
490 * This function synchronizes write-buffer @buf and returns zero in case of
491 * success or a negative error code in case of failure.
493 * Note, although write-buffers are of @c->max_write_size, this function does
494 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
495 * if the write-buffer is only partially filled with data, only the used part
496 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
497 * This way we waste less space.
499 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
501 struct ubifs_info *c = wbuf->c;
502 int err, dirt, sync_len;
504 cancel_wbuf_timer_nolock(wbuf);
505 if (!wbuf->used || wbuf->lnum == -1)
506 /* Write-buffer is empty or not seeked */
509 dbg_io("LEB %d:%d, %d bytes, jhead %s",
510 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
511 ubifs_assert(!(wbuf->avail & 7));
512 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
513 ubifs_assert(wbuf->size >= c->min_io_size);
514 ubifs_assert(wbuf->size <= c->max_write_size);
515 ubifs_assert(wbuf->size % c->min_io_size == 0);
516 ubifs_assert(!c->ro_media && !c->ro_mount);
517 if (c->leb_size - wbuf->offs >= c->max_write_size)
518 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
524 * Do not write whole write buffer but write only the minimum necessary
525 * amount of min. I/O units.
527 sync_len = ALIGN(wbuf->used, c->min_io_size);
528 dirt = sync_len - wbuf->used;
530 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
531 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
535 spin_lock(&wbuf->lock);
536 wbuf->offs += sync_len;
538 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
539 * But our goal is to optimize writes and make sure we write in
540 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
541 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
542 * sure that @wbuf->offs + @wbuf->size is aligned to
543 * @c->max_write_size. This way we make sure that after next
544 * write-buffer flush we are again at the optimal offset (aligned to
545 * @c->max_write_size).
547 if (c->leb_size - wbuf->offs < c->max_write_size)
548 wbuf->size = c->leb_size - wbuf->offs;
549 else if (wbuf->offs & (c->max_write_size - 1))
550 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
552 wbuf->size = c->max_write_size;
553 wbuf->avail = wbuf->size;
556 spin_unlock(&wbuf->lock);
558 if (wbuf->sync_callback)
559 err = wbuf->sync_callback(c, wbuf->lnum,
560 c->leb_size - wbuf->offs, dirt);
565 * ubifs_wbuf_seek_nolock - seek write-buffer.
566 * @wbuf: write-buffer
567 * @lnum: logical eraseblock number to seek to
568 * @offs: logical eraseblock offset to seek to
570 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
571 * The write-buffer has to be empty. Returns zero in case of success and a
572 * negative error code in case of failure.
574 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
576 const struct ubifs_info *c = wbuf->c;
578 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
579 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
580 ubifs_assert(offs >= 0 && offs <= c->leb_size);
581 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
582 ubifs_assert(lnum != wbuf->lnum);
583 ubifs_assert(wbuf->used == 0);
585 spin_lock(&wbuf->lock);
588 if (c->leb_size - wbuf->offs < c->max_write_size)
589 wbuf->size = c->leb_size - wbuf->offs;
590 else if (wbuf->offs & (c->max_write_size - 1))
591 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
593 wbuf->size = c->max_write_size;
594 wbuf->avail = wbuf->size;
596 spin_unlock(&wbuf->lock);
603 * ubifs_bg_wbufs_sync - synchronize write-buffers.
604 * @c: UBIFS file-system description object
606 * This function is called by background thread to synchronize write-buffers.
607 * Returns zero in case of success and a negative error code in case of
610 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
614 ubifs_assert(!c->ro_media && !c->ro_mount);
615 if (!c->need_wbuf_sync)
617 c->need_wbuf_sync = 0;
624 dbg_io("synchronize");
625 for (i = 0; i < c->jhead_cnt; i++) {
626 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
631 * If the mutex is locked then wbuf is being changed, so
632 * synchronization is not necessary.
634 if (mutex_is_locked(&wbuf->io_mutex))
637 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
638 if (!wbuf->need_sync) {
639 mutex_unlock(&wbuf->io_mutex);
643 err = ubifs_wbuf_sync_nolock(wbuf);
644 mutex_unlock(&wbuf->io_mutex);
646 ubifs_err(c, "cannot sync write-buffer, error %d", err);
647 ubifs_ro_mode(c, err);
655 /* Cancel all timers to prevent repeated errors */
656 for (i = 0; i < c->jhead_cnt; i++) {
657 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
659 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
660 cancel_wbuf_timer_nolock(wbuf);
661 mutex_unlock(&wbuf->io_mutex);
667 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
668 * @wbuf: write-buffer
669 * @buf: node to write
672 * This function writes data to flash via write-buffer @wbuf. This means that
673 * the last piece of the node won't reach the flash media immediately if it
674 * does not take whole max. write unit (@c->max_write_size). Instead, the node
675 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
676 * because more data are appended to the write-buffer).
678 * This function returns zero in case of success and a negative error code in
679 * case of failure. If the node cannot be written because there is no more
680 * space in this logical eraseblock, %-ENOSPC is returned.
682 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
684 struct ubifs_info *c = wbuf->c;
685 int err, written, n, aligned_len = ALIGN(len, 8);
687 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
688 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
689 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
690 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
691 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
692 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
693 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
694 ubifs_assert(wbuf->size >= c->min_io_size);
695 ubifs_assert(wbuf->size <= c->max_write_size);
696 ubifs_assert(wbuf->size % c->min_io_size == 0);
697 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
698 ubifs_assert(!c->ro_media && !c->ro_mount);
699 ubifs_assert(!c->space_fixup);
700 if (c->leb_size - wbuf->offs >= c->max_write_size)
701 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
703 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
708 cancel_wbuf_timer_nolock(wbuf);
713 if (aligned_len <= wbuf->avail) {
715 * The node is not very large and fits entirely within
718 memcpy(wbuf->buf + wbuf->used, buf, len);
720 if (aligned_len == wbuf->avail) {
721 dbg_io("flush jhead %s wbuf to LEB %d:%d",
722 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
723 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
724 wbuf->offs, wbuf->size);
728 spin_lock(&wbuf->lock);
729 wbuf->offs += wbuf->size;
730 if (c->leb_size - wbuf->offs >= c->max_write_size)
731 wbuf->size = c->max_write_size;
733 wbuf->size = c->leb_size - wbuf->offs;
734 wbuf->avail = wbuf->size;
737 spin_unlock(&wbuf->lock);
739 spin_lock(&wbuf->lock);
740 wbuf->avail -= aligned_len;
741 wbuf->used += aligned_len;
742 spin_unlock(&wbuf->lock);
752 * The node is large enough and does not fit entirely within
753 * current available space. We have to fill and flush
754 * write-buffer and switch to the next max. write unit.
756 dbg_io("flush jhead %s wbuf to LEB %d:%d",
757 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
758 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
759 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
764 wbuf->offs += wbuf->size;
766 aligned_len -= wbuf->avail;
767 written += wbuf->avail;
768 } else if (wbuf->offs & (c->max_write_size - 1)) {
770 * The write-buffer offset is not aligned to
771 * @c->max_write_size and @wbuf->size is less than
772 * @c->max_write_size. Write @wbuf->size bytes to make sure the
773 * following writes are done in optimal @c->max_write_size
776 dbg_io("write %d bytes to LEB %d:%d",
777 wbuf->size, wbuf->lnum, wbuf->offs);
778 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
783 wbuf->offs += wbuf->size;
785 aligned_len -= wbuf->size;
786 written += wbuf->size;
790 * The remaining data may take more whole max. write units, so write the
791 * remains multiple to max. write unit size directly to the flash media.
792 * We align node length to 8-byte boundary because we anyway flash wbuf
793 * if the remaining space is less than 8 bytes.
795 n = aligned_len >> c->max_write_shift;
797 n <<= c->max_write_shift;
798 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
800 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
810 spin_lock(&wbuf->lock);
813 * And now we have what's left and what does not take whole
814 * max. write unit, so write it to the write-buffer and we are
817 memcpy(wbuf->buf, buf + written, len);
819 if (c->leb_size - wbuf->offs >= c->max_write_size)
820 wbuf->size = c->max_write_size;
822 wbuf->size = c->leb_size - wbuf->offs;
823 wbuf->avail = wbuf->size - aligned_len;
824 wbuf->used = aligned_len;
826 spin_unlock(&wbuf->lock);
829 if (wbuf->sync_callback) {
830 int free = c->leb_size - wbuf->offs - wbuf->used;
832 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
838 new_wbuf_timer_nolock(wbuf);
843 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
844 len, wbuf->lnum, wbuf->offs, err);
845 ubifs_dump_node(c, buf);
847 ubifs_dump_leb(c, wbuf->lnum);
852 * ubifs_write_node - write node to the media.
853 * @c: UBIFS file-system description object
854 * @buf: the node to write
856 * @lnum: logical eraseblock number
857 * @offs: offset within the logical eraseblock
859 * This function automatically fills node magic number, assigns sequence
860 * number, and calculates node CRC checksum. The length of the @buf buffer has
861 * to be aligned to the minimal I/O unit size. This function automatically
862 * appends padding node and padding bytes if needed. Returns zero in case of
863 * success and a negative error code in case of failure.
865 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
868 int err, buf_len = ALIGN(len, c->min_io_size);
870 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
871 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
873 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
874 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
875 ubifs_assert(!c->ro_media && !c->ro_mount);
876 ubifs_assert(!c->space_fixup);
881 ubifs_prepare_node(c, buf, len, 1);
882 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
884 ubifs_dump_node(c, buf);
891 * ubifs_read_node_wbuf - read node from the media or write-buffer.
892 * @wbuf: wbuf to check for un-written data
893 * @buf: buffer to read to
896 * @lnum: logical eraseblock number
897 * @offs: offset within the logical eraseblock
899 * This function reads a node of known type and length, checks it and stores
900 * in @buf. If the node partially or fully sits in the write-buffer, this
901 * function takes data from the buffer, otherwise it reads the flash media.
902 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
903 * error code in case of failure.
905 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
908 const struct ubifs_info *c = wbuf->c;
909 int err, rlen, overlap;
910 struct ubifs_ch *ch = buf;
912 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
913 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
914 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
915 ubifs_assert(!(offs & 7) && offs < c->leb_size);
916 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
918 spin_lock(&wbuf->lock);
919 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
921 /* We may safely unlock the write-buffer and read the data */
922 spin_unlock(&wbuf->lock);
923 return ubifs_read_node(c, buf, type, len, lnum, offs);
926 /* Don't read under wbuf */
927 rlen = wbuf->offs - offs;
931 /* Copy the rest from the write-buffer */
932 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
933 spin_unlock(&wbuf->lock);
936 /* Read everything that goes before write-buffer */
937 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
938 if (err && err != -EBADMSG)
942 if (type != ch->node_type) {
943 ubifs_err(c, "bad node type (%d but expected %d)",
944 ch->node_type, type);
948 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
950 ubifs_err(c, "expected node type %d", type);
954 rlen = le32_to_cpu(ch->len);
956 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
963 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
964 ubifs_dump_node(c, buf);
970 * ubifs_read_node - read node.
971 * @c: UBIFS file-system description object
972 * @buf: buffer to read to
974 * @len: node length (not aligned)
975 * @lnum: logical eraseblock number
976 * @offs: offset within the logical eraseblock
978 * This function reads a node of known type and and length, checks it and
979 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
980 * and a negative error code in case of failure.
982 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
986 struct ubifs_ch *ch = buf;
988 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
989 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
990 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
991 ubifs_assert(!(offs & 7) && offs < c->leb_size);
992 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
994 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
995 if (err && err != -EBADMSG)
998 if (type != ch->node_type) {
999 ubifs_errc(c, "bad node type (%d but expected %d)",
1000 ch->node_type, type);
1004 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1006 ubifs_errc(c, "expected node type %d", type);
1010 l = le32_to_cpu(ch->len);
1012 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1019 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1020 offs, ubi_is_mapped(c->ubi, lnum));
1022 ubifs_dump_node(c, buf);
1029 * ubifs_wbuf_init - initialize write-buffer.
1030 * @c: UBIFS file-system description object
1031 * @wbuf: write-buffer to initialize
1033 * This function initializes write-buffer. Returns zero in case of success
1034 * %-ENOMEM in case of failure.
1036 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1040 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1044 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1045 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1046 if (!wbuf->inodes) {
1053 wbuf->lnum = wbuf->offs = -1;
1055 * If the LEB starts at the max. write size aligned address, then
1056 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1057 * set it to something smaller so that it ends at the closest max.
1058 * write size boundary.
1060 size = c->max_write_size - (c->leb_start % c->max_write_size);
1061 wbuf->avail = wbuf->size = size;
1062 wbuf->sync_callback = NULL;
1063 mutex_init(&wbuf->io_mutex);
1064 spin_lock_init(&wbuf->lock);
1069 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1070 wbuf->timer.function = wbuf_timer_callback_nolock;
1071 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1072 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1073 wbuf->delta *= 1000000000ULL;
1074 ubifs_assert(wbuf->delta <= ULONG_MAX);
1080 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1081 * @wbuf: the write-buffer where to add
1082 * @inum: the inode number
1084 * This function adds an inode number to the inode array of the write-buffer.
1086 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1089 /* NOR flash or something similar */
1092 spin_lock(&wbuf->lock);
1094 wbuf->inodes[wbuf->next_ino++] = inum;
1095 spin_unlock(&wbuf->lock);
1099 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1100 * @wbuf: the write-buffer
1101 * @inum: the inode number
1103 * This function returns with %1 if the write-buffer contains some data from the
1104 * given inode otherwise it returns with %0.
1106 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1110 spin_lock(&wbuf->lock);
1111 for (i = 0; i < wbuf->next_ino; i++)
1112 if (inum == wbuf->inodes[i]) {
1116 spin_unlock(&wbuf->lock);
1122 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1123 * @c: UBIFS file-system description object
1124 * @inode: inode to synchronize
1126 * This function synchronizes write-buffers which contain nodes belonging to
1127 * @inode. Returns zero in case of success and a negative error code in case of
1130 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1134 for (i = 0; i < c->jhead_cnt; i++) {
1135 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1139 * GC head is special, do not look at it. Even if the
1140 * head contains something related to this inode, it is
1141 * a _copy_ of corresponding on-flash node which sits
1146 if (!wbuf_has_ino(wbuf, inode->i_ino))
1149 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1150 if (wbuf_has_ino(wbuf, inode->i_ino))
1151 err = ubifs_wbuf_sync_nolock(wbuf);
1152 mutex_unlock(&wbuf->io_mutex);
1155 ubifs_ro_mode(c, err);