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.
64 #include <dm/devres.h>
65 #include <linux/crc32.h>
66 #include <linux/slab.h>
67 #include <u-boot/crc.h>
69 #include <linux/compat.h>
70 #include <linux/err.h>
75 * ubifs_ro_mode - switch UBIFS to read read-only mode.
76 * @c: UBIFS file-system description object
77 * @err: error code which is the reason of switching to R/O mode
79 void ubifs_ro_mode(struct ubifs_info
*c
, int err
)
83 c
->no_chk_data_crc
= 0;
84 c
->vfs_sb
->s_flags
|= MS_RDONLY
;
85 ubifs_warn(c
, "switched to read-only mode, error %d", err
);
91 * Below are simple wrappers over UBI I/O functions which include some
92 * additional checks and UBIFS debugging stuff. See corresponding UBI function
93 * for more information.
96 int ubifs_leb_read(const struct ubifs_info
*c
, int lnum
, void *buf
, int offs
,
97 int len
, int even_ebadmsg
)
101 err
= ubi_read(c
->ubi
, lnum
, buf
, offs
, len
);
103 * In case of %-EBADMSG print the error message only if the
104 * @even_ebadmsg is true.
106 if (err
&& (err
!= -EBADMSG
|| even_ebadmsg
)) {
107 ubifs_err(c
, "reading %d bytes from LEB %d:%d failed, error %d",
108 len
, lnum
, offs
, err
);
114 int ubifs_leb_write(struct ubifs_info
*c
, int lnum
, const void *buf
, int offs
,
119 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
122 if (!dbg_is_tst_rcvry(c
))
123 err
= ubi_leb_write(c
->ubi
, lnum
, buf
, offs
, len
);
126 err
= dbg_leb_write(c
, lnum
, buf
, offs
, len
);
129 ubifs_err(c
, "writing %d bytes to LEB %d:%d failed, error %d",
130 len
, lnum
, offs
, err
);
131 ubifs_ro_mode(c
, err
);
137 int ubifs_leb_change(struct ubifs_info
*c
, int lnum
, const void *buf
, int len
)
141 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
144 if (!dbg_is_tst_rcvry(c
))
145 err
= ubi_leb_change(c
->ubi
, lnum
, buf
, len
);
148 err
= dbg_leb_change(c
, lnum
, buf
, len
);
151 ubifs_err(c
, "changing %d bytes in LEB %d failed, error %d",
153 ubifs_ro_mode(c
, err
);
159 int ubifs_leb_unmap(struct ubifs_info
*c
, int lnum
)
163 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
166 if (!dbg_is_tst_rcvry(c
))
167 err
= ubi_leb_unmap(c
->ubi
, lnum
);
170 err
= dbg_leb_unmap(c
, lnum
);
173 ubifs_err(c
, "unmap LEB %d failed, error %d", lnum
, err
);
174 ubifs_ro_mode(c
, err
);
180 int ubifs_leb_map(struct ubifs_info
*c
, int lnum
)
184 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
187 if (!dbg_is_tst_rcvry(c
))
188 err
= ubi_leb_map(c
->ubi
, lnum
);
191 err
= dbg_leb_map(c
, lnum
);
194 ubifs_err(c
, "mapping LEB %d failed, error %d", lnum
, err
);
195 ubifs_ro_mode(c
, err
);
201 int ubifs_is_mapped(const struct ubifs_info
*c
, int lnum
)
205 err
= ubi_is_mapped(c
->ubi
, lnum
);
207 ubifs_err(c
, "ubi_is_mapped failed for LEB %d, error %d",
215 * ubifs_check_node - check node.
216 * @c: UBIFS file-system description object
217 * @buf: node to check
218 * @lnum: logical eraseblock number
219 * @offs: offset within the logical eraseblock
220 * @quiet: print no messages
221 * @must_chk_crc: indicates whether to always check the CRC
223 * This function checks node magic number and CRC checksum. This function also
224 * validates node length to prevent UBIFS from becoming crazy when an attacker
225 * feeds it a file-system image with incorrect nodes. For example, too large
226 * node length in the common header could cause UBIFS to read memory outside of
227 * allocated buffer when checking the CRC checksum.
229 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
230 * true, which is controlled by corresponding UBIFS mount option. However, if
231 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
232 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
233 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
234 * is checked. This is because during mounting or re-mounting from R/O mode to
235 * R/W mode we may read journal nodes (when replying the journal or doing the
236 * recovery) and the journal nodes may potentially be corrupted, so checking is
239 * This function returns zero in case of success and %-EUCLEAN in case of bad
242 int ubifs_check_node(const struct ubifs_info
*c
, const void *buf
, int lnum
,
243 int offs
, int quiet
, int must_chk_crc
)
245 int err
= -EINVAL
, type
, node_len
;
246 uint32_t crc
, node_crc
, magic
;
247 const struct ubifs_ch
*ch
= buf
;
249 ubifs_assert(lnum
>= 0 && lnum
< c
->leb_cnt
&& offs
>= 0);
250 ubifs_assert(!(offs
& 7) && offs
< c
->leb_size
);
252 magic
= le32_to_cpu(ch
->magic
);
253 if (magic
!= UBIFS_NODE_MAGIC
) {
255 ubifs_err(c
, "bad magic %#08x, expected %#08x",
256 magic
, UBIFS_NODE_MAGIC
);
261 type
= ch
->node_type
;
262 if (type
< 0 || type
>= UBIFS_NODE_TYPES_CNT
) {
264 ubifs_err(c
, "bad node type %d", type
);
268 node_len
= le32_to_cpu(ch
->len
);
269 if (node_len
+ offs
> c
->leb_size
)
272 if (c
->ranges
[type
].max_len
== 0) {
273 if (node_len
!= c
->ranges
[type
].len
)
275 } else if (node_len
< c
->ranges
[type
].min_len
||
276 node_len
> c
->ranges
[type
].max_len
)
279 if (!must_chk_crc
&& type
== UBIFS_DATA_NODE
&& !c
->mounting
&&
280 !c
->remounting_rw
&& c
->no_chk_data_crc
)
283 crc
= crc32(UBIFS_CRC32_INIT
, buf
+ 8, node_len
- 8);
284 node_crc
= le32_to_cpu(ch
->crc
);
285 if (crc
!= node_crc
) {
287 ubifs_err(c
, "bad CRC: calculated %#08x, read %#08x",
297 ubifs_err(c
, "bad node length %d", node_len
);
300 ubifs_err(c
, "bad node at LEB %d:%d", lnum
, offs
);
301 ubifs_dump_node(c
, buf
);
308 * ubifs_pad - pad flash space.
309 * @c: UBIFS file-system description object
310 * @buf: buffer to put padding to
311 * @pad: how many bytes to pad
313 * The flash media obliges us to write only in chunks of %c->min_io_size and
314 * when we have to write less data we add padding node to the write-buffer and
315 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
316 * media is being scanned. If the amount of wasted space is not enough to fit a
317 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
318 * pattern (%UBIFS_PADDING_BYTE).
320 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
323 void ubifs_pad(const struct ubifs_info
*c
, void *buf
, int pad
)
327 ubifs_assert(pad
>= 0 && !(pad
& 7));
329 if (pad
>= UBIFS_PAD_NODE_SZ
) {
330 struct ubifs_ch
*ch
= buf
;
331 struct ubifs_pad_node
*pad_node
= buf
;
333 ch
->magic
= cpu_to_le32(UBIFS_NODE_MAGIC
);
334 ch
->node_type
= UBIFS_PAD_NODE
;
335 ch
->group_type
= UBIFS_NO_NODE_GROUP
;
336 ch
->padding
[0] = ch
->padding
[1] = 0;
338 ch
->len
= cpu_to_le32(UBIFS_PAD_NODE_SZ
);
339 pad
-= UBIFS_PAD_NODE_SZ
;
340 pad_node
->pad_len
= cpu_to_le32(pad
);
341 crc
= crc32(UBIFS_CRC32_INIT
, buf
+ 8, UBIFS_PAD_NODE_SZ
- 8);
342 ch
->crc
= cpu_to_le32(crc
);
343 memset(buf
+ UBIFS_PAD_NODE_SZ
, 0, pad
);
345 /* Too little space, padding node won't fit */
346 memset(buf
, UBIFS_PADDING_BYTE
, pad
);
350 * next_sqnum - get next sequence number.
351 * @c: UBIFS file-system description object
353 static unsigned long long next_sqnum(struct ubifs_info
*c
)
355 unsigned long long sqnum
;
357 spin_lock(&c
->cnt_lock
);
358 sqnum
= ++c
->max_sqnum
;
359 spin_unlock(&c
->cnt_lock
);
361 if (unlikely(sqnum
>= SQNUM_WARN_WATERMARK
)) {
362 if (sqnum
>= SQNUM_WATERMARK
) {
363 ubifs_err(c
, "sequence number overflow %llu, end of life",
365 ubifs_ro_mode(c
, -EINVAL
);
367 ubifs_warn(c
, "running out of sequence numbers, end of life soon");
374 * ubifs_prepare_node - prepare node to be written to flash.
375 * @c: UBIFS file-system description object
376 * @node: the node to pad
378 * @pad: if the buffer has to be padded
380 * This function prepares node at @node to be written to the media - it
381 * calculates node CRC, fills the common header, and adds proper padding up to
382 * the next minimum I/O unit if @pad is not zero.
384 void ubifs_prepare_node(struct ubifs_info
*c
, void *node
, int len
, int pad
)
387 struct ubifs_ch
*ch
= node
;
388 unsigned long long sqnum
= next_sqnum(c
);
390 ubifs_assert(len
>= UBIFS_CH_SZ
);
392 ch
->magic
= cpu_to_le32(UBIFS_NODE_MAGIC
);
393 ch
->len
= cpu_to_le32(len
);
394 ch
->group_type
= UBIFS_NO_NODE_GROUP
;
395 ch
->sqnum
= cpu_to_le64(sqnum
);
396 ch
->padding
[0] = ch
->padding
[1] = 0;
397 crc
= crc32(UBIFS_CRC32_INIT
, node
+ 8, len
- 8);
398 ch
->crc
= cpu_to_le32(crc
);
402 pad
= ALIGN(len
, c
->min_io_size
) - len
;
403 ubifs_pad(c
, node
+ len
, pad
);
408 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
409 * @c: UBIFS file-system description object
410 * @node: the node to pad
412 * @last: indicates the last node of the group
414 * This function prepares node at @node to be written to the media - it
415 * calculates node CRC and fills the common header.
417 void ubifs_prep_grp_node(struct ubifs_info
*c
, void *node
, int len
, int last
)
420 struct ubifs_ch
*ch
= node
;
421 unsigned long long sqnum
= next_sqnum(c
);
423 ubifs_assert(len
>= UBIFS_CH_SZ
);
425 ch
->magic
= cpu_to_le32(UBIFS_NODE_MAGIC
);
426 ch
->len
= cpu_to_le32(len
);
428 ch
->group_type
= UBIFS_LAST_OF_NODE_GROUP
;
430 ch
->group_type
= UBIFS_IN_NODE_GROUP
;
431 ch
->sqnum
= cpu_to_le64(sqnum
);
432 ch
->padding
[0] = ch
->padding
[1] = 0;
433 crc
= crc32(UBIFS_CRC32_INIT
, node
+ 8, len
- 8);
434 ch
->crc
= cpu_to_le32(crc
);
439 * wbuf_timer_callback - write-buffer timer callback function.
440 * @timer: timer data (write-buffer descriptor)
442 * This function is called when the write-buffer timer expires.
444 static enum hrtimer_restart
wbuf_timer_callback_nolock(struct hrtimer
*timer
)
446 struct ubifs_wbuf
*wbuf
= container_of(timer
, struct ubifs_wbuf
, timer
);
448 dbg_io("jhead %s", dbg_jhead(wbuf
->jhead
));
450 wbuf
->c
->need_wbuf_sync
= 1;
451 ubifs_wake_up_bgt(wbuf
->c
);
452 return HRTIMER_NORESTART
;
456 * new_wbuf_timer - start new write-buffer timer.
457 * @wbuf: write-buffer descriptor
459 static void new_wbuf_timer_nolock(struct ubifs_wbuf
*wbuf
)
461 ubifs_assert(!hrtimer_active(&wbuf
->timer
));
465 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
466 dbg_jhead(wbuf
->jhead
),
467 div_u64(ktime_to_ns(wbuf
->softlimit
), USEC_PER_SEC
),
468 div_u64(ktime_to_ns(wbuf
->softlimit
) + wbuf
->delta
,
470 hrtimer_start_range_ns(&wbuf
->timer
, wbuf
->softlimit
, wbuf
->delta
,
476 * cancel_wbuf_timer - cancel write-buffer timer.
477 * @wbuf: write-buffer descriptor
479 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf
*wbuf
)
485 hrtimer_cancel(&wbuf
->timer
);
490 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
491 * @wbuf: write-buffer to synchronize
493 * This function synchronizes write-buffer @buf and returns zero in case of
494 * success or a negative error code in case of failure.
496 * Note, although write-buffers are of @c->max_write_size, this function does
497 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
498 * if the write-buffer is only partially filled with data, only the used part
499 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
500 * This way we waste less space.
502 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf
*wbuf
)
504 struct ubifs_info
*c
= wbuf
->c
;
505 int err
, dirt
, sync_len
;
507 cancel_wbuf_timer_nolock(wbuf
);
508 if (!wbuf
->used
|| wbuf
->lnum
== -1)
509 /* Write-buffer is empty or not seeked */
512 dbg_io("LEB %d:%d, %d bytes, jhead %s",
513 wbuf
->lnum
, wbuf
->offs
, wbuf
->used
, dbg_jhead(wbuf
->jhead
));
514 ubifs_assert(!(wbuf
->avail
& 7));
515 ubifs_assert(wbuf
->offs
+ wbuf
->size
<= c
->leb_size
);
516 ubifs_assert(wbuf
->size
>= c
->min_io_size
);
517 ubifs_assert(wbuf
->size
<= c
->max_write_size
);
518 ubifs_assert(wbuf
->size
% c
->min_io_size
== 0);
519 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
520 if (c
->leb_size
- wbuf
->offs
>= c
->max_write_size
)
521 ubifs_assert(!((wbuf
->offs
+ wbuf
->size
) % c
->max_write_size
));
527 * Do not write whole write buffer but write only the minimum necessary
528 * amount of min. I/O units.
530 sync_len
= ALIGN(wbuf
->used
, c
->min_io_size
);
531 dirt
= sync_len
- wbuf
->used
;
533 ubifs_pad(c
, wbuf
->buf
+ wbuf
->used
, dirt
);
534 err
= ubifs_leb_write(c
, wbuf
->lnum
, wbuf
->buf
, wbuf
->offs
, sync_len
);
538 spin_lock(&wbuf
->lock
);
539 wbuf
->offs
+= sync_len
;
541 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
542 * But our goal is to optimize writes and make sure we write in
543 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
544 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
545 * sure that @wbuf->offs + @wbuf->size is aligned to
546 * @c->max_write_size. This way we make sure that after next
547 * write-buffer flush we are again at the optimal offset (aligned to
548 * @c->max_write_size).
550 if (c
->leb_size
- wbuf
->offs
< c
->max_write_size
)
551 wbuf
->size
= c
->leb_size
- wbuf
->offs
;
552 else if (wbuf
->offs
& (c
->max_write_size
- 1))
553 wbuf
->size
= ALIGN(wbuf
->offs
, c
->max_write_size
) - wbuf
->offs
;
555 wbuf
->size
= c
->max_write_size
;
556 wbuf
->avail
= wbuf
->size
;
559 spin_unlock(&wbuf
->lock
);
561 if (wbuf
->sync_callback
)
562 err
= wbuf
->sync_callback(c
, wbuf
->lnum
,
563 c
->leb_size
- wbuf
->offs
, dirt
);
568 * ubifs_wbuf_seek_nolock - seek write-buffer.
569 * @wbuf: write-buffer
570 * @lnum: logical eraseblock number to seek to
571 * @offs: logical eraseblock offset to seek to
573 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
574 * The write-buffer has to be empty. Returns zero in case of success and a
575 * negative error code in case of failure.
577 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf
*wbuf
, int lnum
, int offs
)
579 const struct ubifs_info
*c
= wbuf
->c
;
581 dbg_io("LEB %d:%d, jhead %s", lnum
, offs
, dbg_jhead(wbuf
->jhead
));
582 ubifs_assert(lnum
>= 0 && lnum
< c
->leb_cnt
);
583 ubifs_assert(offs
>= 0 && offs
<= c
->leb_size
);
584 ubifs_assert(offs
% c
->min_io_size
== 0 && !(offs
& 7));
585 ubifs_assert(lnum
!= wbuf
->lnum
);
586 ubifs_assert(wbuf
->used
== 0);
588 spin_lock(&wbuf
->lock
);
591 if (c
->leb_size
- wbuf
->offs
< c
->max_write_size
)
592 wbuf
->size
= c
->leb_size
- wbuf
->offs
;
593 else if (wbuf
->offs
& (c
->max_write_size
- 1))
594 wbuf
->size
= ALIGN(wbuf
->offs
, c
->max_write_size
) - wbuf
->offs
;
596 wbuf
->size
= c
->max_write_size
;
597 wbuf
->avail
= wbuf
->size
;
599 spin_unlock(&wbuf
->lock
);
606 * ubifs_bg_wbufs_sync - synchronize write-buffers.
607 * @c: UBIFS file-system description object
609 * This function is called by background thread to synchronize write-buffers.
610 * Returns zero in case of success and a negative error code in case of
613 int ubifs_bg_wbufs_sync(struct ubifs_info
*c
)
617 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
618 if (!c
->need_wbuf_sync
)
620 c
->need_wbuf_sync
= 0;
627 dbg_io("synchronize");
628 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
629 struct ubifs_wbuf
*wbuf
= &c
->jheads
[i
].wbuf
;
634 * If the mutex is locked then wbuf is being changed, so
635 * synchronization is not necessary.
637 if (mutex_is_locked(&wbuf
->io_mutex
))
640 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
641 if (!wbuf
->need_sync
) {
642 mutex_unlock(&wbuf
->io_mutex
);
646 err
= ubifs_wbuf_sync_nolock(wbuf
);
647 mutex_unlock(&wbuf
->io_mutex
);
649 ubifs_err(c
, "cannot sync write-buffer, error %d", err
);
650 ubifs_ro_mode(c
, err
);
658 /* Cancel all timers to prevent repeated errors */
659 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
660 struct ubifs_wbuf
*wbuf
= &c
->jheads
[i
].wbuf
;
662 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
663 cancel_wbuf_timer_nolock(wbuf
);
664 mutex_unlock(&wbuf
->io_mutex
);
670 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
671 * @wbuf: write-buffer
672 * @buf: node to write
675 * This function writes data to flash via write-buffer @wbuf. This means that
676 * the last piece of the node won't reach the flash media immediately if it
677 * does not take whole max. write unit (@c->max_write_size). Instead, the node
678 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
679 * because more data are appended to the write-buffer).
681 * This function returns zero in case of success and a negative error code in
682 * case of failure. If the node cannot be written because there is no more
683 * space in this logical eraseblock, %-ENOSPC is returned.
685 int ubifs_wbuf_write_nolock(struct ubifs_wbuf
*wbuf
, void *buf
, int len
)
687 struct ubifs_info
*c
= wbuf
->c
;
688 int err
, written
, n
, aligned_len
= ALIGN(len
, 8);
690 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len
,
691 dbg_ntype(((struct ubifs_ch
*)buf
)->node_type
),
692 dbg_jhead(wbuf
->jhead
), wbuf
->lnum
, wbuf
->offs
+ wbuf
->used
);
693 ubifs_assert(len
> 0 && wbuf
->lnum
>= 0 && wbuf
->lnum
< c
->leb_cnt
);
694 ubifs_assert(wbuf
->offs
>= 0 && wbuf
->offs
% c
->min_io_size
== 0);
695 ubifs_assert(!(wbuf
->offs
& 7) && wbuf
->offs
<= c
->leb_size
);
696 ubifs_assert(wbuf
->avail
> 0 && wbuf
->avail
<= wbuf
->size
);
697 ubifs_assert(wbuf
->size
>= c
->min_io_size
);
698 ubifs_assert(wbuf
->size
<= c
->max_write_size
);
699 ubifs_assert(wbuf
->size
% c
->min_io_size
== 0);
700 ubifs_assert(mutex_is_locked(&wbuf
->io_mutex
));
701 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
702 ubifs_assert(!c
->space_fixup
);
703 if (c
->leb_size
- wbuf
->offs
>= c
->max_write_size
)
704 ubifs_assert(!((wbuf
->offs
+ wbuf
->size
) % c
->max_write_size
));
706 if (c
->leb_size
- wbuf
->offs
- wbuf
->used
< aligned_len
) {
711 cancel_wbuf_timer_nolock(wbuf
);
716 if (aligned_len
<= wbuf
->avail
) {
718 * The node is not very large and fits entirely within
721 memcpy(wbuf
->buf
+ wbuf
->used
, buf
, len
);
723 if (aligned_len
== wbuf
->avail
) {
724 dbg_io("flush jhead %s wbuf to LEB %d:%d",
725 dbg_jhead(wbuf
->jhead
), wbuf
->lnum
, wbuf
->offs
);
726 err
= ubifs_leb_write(c
, wbuf
->lnum
, wbuf
->buf
,
727 wbuf
->offs
, wbuf
->size
);
731 spin_lock(&wbuf
->lock
);
732 wbuf
->offs
+= wbuf
->size
;
733 if (c
->leb_size
- wbuf
->offs
>= c
->max_write_size
)
734 wbuf
->size
= c
->max_write_size
;
736 wbuf
->size
= c
->leb_size
- wbuf
->offs
;
737 wbuf
->avail
= wbuf
->size
;
740 spin_unlock(&wbuf
->lock
);
742 spin_lock(&wbuf
->lock
);
743 wbuf
->avail
-= aligned_len
;
744 wbuf
->used
+= aligned_len
;
745 spin_unlock(&wbuf
->lock
);
755 * The node is large enough and does not fit entirely within
756 * current available space. We have to fill and flush
757 * write-buffer and switch to the next max. write unit.
759 dbg_io("flush jhead %s wbuf to LEB %d:%d",
760 dbg_jhead(wbuf
->jhead
), wbuf
->lnum
, wbuf
->offs
);
761 memcpy(wbuf
->buf
+ wbuf
->used
, buf
, wbuf
->avail
);
762 err
= ubifs_leb_write(c
, wbuf
->lnum
, wbuf
->buf
, wbuf
->offs
,
767 wbuf
->offs
+= wbuf
->size
;
769 aligned_len
-= wbuf
->avail
;
770 written
+= wbuf
->avail
;
771 } else if (wbuf
->offs
& (c
->max_write_size
- 1)) {
773 * The write-buffer offset is not aligned to
774 * @c->max_write_size and @wbuf->size is less than
775 * @c->max_write_size. Write @wbuf->size bytes to make sure the
776 * following writes are done in optimal @c->max_write_size
779 dbg_io("write %d bytes to LEB %d:%d",
780 wbuf
->size
, wbuf
->lnum
, wbuf
->offs
);
781 err
= ubifs_leb_write(c
, wbuf
->lnum
, buf
, wbuf
->offs
,
786 wbuf
->offs
+= wbuf
->size
;
788 aligned_len
-= wbuf
->size
;
789 written
+= wbuf
->size
;
793 * The remaining data may take more whole max. write units, so write the
794 * remains multiple to max. write unit size directly to the flash media.
795 * We align node length to 8-byte boundary because we anyway flash wbuf
796 * if the remaining space is less than 8 bytes.
798 n
= aligned_len
>> c
->max_write_shift
;
800 n
<<= c
->max_write_shift
;
801 dbg_io("write %d bytes to LEB %d:%d", n
, wbuf
->lnum
,
803 err
= ubifs_leb_write(c
, wbuf
->lnum
, buf
+ written
,
813 spin_lock(&wbuf
->lock
);
816 * And now we have what's left and what does not take whole
817 * max. write unit, so write it to the write-buffer and we are
820 memcpy(wbuf
->buf
, buf
+ written
, len
);
822 if (c
->leb_size
- wbuf
->offs
>= c
->max_write_size
)
823 wbuf
->size
= c
->max_write_size
;
825 wbuf
->size
= c
->leb_size
- wbuf
->offs
;
826 wbuf
->avail
= wbuf
->size
- aligned_len
;
827 wbuf
->used
= aligned_len
;
829 spin_unlock(&wbuf
->lock
);
832 if (wbuf
->sync_callback
) {
833 int free
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
835 err
= wbuf
->sync_callback(c
, wbuf
->lnum
, free
, 0);
841 new_wbuf_timer_nolock(wbuf
);
846 ubifs_err(c
, "cannot write %d bytes to LEB %d:%d, error %d",
847 len
, wbuf
->lnum
, wbuf
->offs
, err
);
848 ubifs_dump_node(c
, buf
);
850 ubifs_dump_leb(c
, wbuf
->lnum
);
855 * ubifs_write_node - write node to the media.
856 * @c: UBIFS file-system description object
857 * @buf: the node to write
859 * @lnum: logical eraseblock number
860 * @offs: offset within the logical eraseblock
862 * This function automatically fills node magic number, assigns sequence
863 * number, and calculates node CRC checksum. The length of the @buf buffer has
864 * to be aligned to the minimal I/O unit size. This function automatically
865 * appends padding node and padding bytes if needed. Returns zero in case of
866 * success and a negative error code in case of failure.
868 int ubifs_write_node(struct ubifs_info
*c
, void *buf
, int len
, int lnum
,
871 int err
, buf_len
= ALIGN(len
, c
->min_io_size
);
873 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
874 lnum
, offs
, dbg_ntype(((struct ubifs_ch
*)buf
)->node_type
), len
,
876 ubifs_assert(lnum
>= 0 && lnum
< c
->leb_cnt
&& offs
>= 0);
877 ubifs_assert(offs
% c
->min_io_size
== 0 && offs
< c
->leb_size
);
878 ubifs_assert(!c
->ro_media
&& !c
->ro_mount
);
879 ubifs_assert(!c
->space_fixup
);
884 ubifs_prepare_node(c
, buf
, len
, 1);
885 err
= ubifs_leb_write(c
, lnum
, buf
, offs
, buf_len
);
887 ubifs_dump_node(c
, buf
);
894 * ubifs_read_node_wbuf - read node from the media or write-buffer.
895 * @wbuf: wbuf to check for un-written data
896 * @buf: buffer to read to
899 * @lnum: logical eraseblock number
900 * @offs: offset within the logical eraseblock
902 * This function reads a node of known type and length, checks it and stores
903 * in @buf. If the node partially or fully sits in the write-buffer, this
904 * function takes data from the buffer, otherwise it reads the flash media.
905 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
906 * error code in case of failure.
908 int ubifs_read_node_wbuf(struct ubifs_wbuf
*wbuf
, void *buf
, int type
, int len
,
911 const struct ubifs_info
*c
= wbuf
->c
;
912 int err
, rlen
, overlap
;
913 struct ubifs_ch
*ch
= buf
;
915 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum
, offs
,
916 dbg_ntype(type
), len
, dbg_jhead(wbuf
->jhead
));
917 ubifs_assert(wbuf
&& lnum
>= 0 && lnum
< c
->leb_cnt
&& offs
>= 0);
918 ubifs_assert(!(offs
& 7) && offs
< c
->leb_size
);
919 ubifs_assert(type
>= 0 && type
< UBIFS_NODE_TYPES_CNT
);
921 spin_lock(&wbuf
->lock
);
922 overlap
= (lnum
== wbuf
->lnum
&& offs
+ len
> wbuf
->offs
);
924 /* We may safely unlock the write-buffer and read the data */
925 spin_unlock(&wbuf
->lock
);
926 return ubifs_read_node(c
, buf
, type
, len
, lnum
, offs
);
929 /* Don't read under wbuf */
930 rlen
= wbuf
->offs
- offs
;
934 /* Copy the rest from the write-buffer */
935 memcpy(buf
+ rlen
, wbuf
->buf
+ offs
+ rlen
- wbuf
->offs
, len
- rlen
);
936 spin_unlock(&wbuf
->lock
);
939 /* Read everything that goes before write-buffer */
940 err
= ubifs_leb_read(c
, lnum
, buf
, offs
, rlen
, 0);
941 if (err
&& err
!= -EBADMSG
)
945 if (type
!= ch
->node_type
) {
946 ubifs_err(c
, "bad node type (%d but expected %d)",
947 ch
->node_type
, type
);
951 err
= ubifs_check_node(c
, buf
, lnum
, offs
, 0, 0);
953 ubifs_err(c
, "expected node type %d", type
);
957 rlen
= le32_to_cpu(ch
->len
);
959 ubifs_err(c
, "bad node length %d, expected %d", rlen
, len
);
966 ubifs_err(c
, "bad node at LEB %d:%d", lnum
, offs
);
967 ubifs_dump_node(c
, buf
);
973 * ubifs_read_node - read node.
974 * @c: UBIFS file-system description object
975 * @buf: buffer to read to
977 * @len: node length (not aligned)
978 * @lnum: logical eraseblock number
979 * @offs: offset within the logical eraseblock
981 * This function reads a node of known type and and length, checks it and
982 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
983 * and a negative error code in case of failure.
985 int ubifs_read_node(const struct ubifs_info
*c
, void *buf
, int type
, int len
,
989 struct ubifs_ch
*ch
= buf
;
991 dbg_io("LEB %d:%d, %s, length %d", lnum
, offs
, dbg_ntype(type
), len
);
992 ubifs_assert(lnum
>= 0 && lnum
< c
->leb_cnt
&& offs
>= 0);
993 ubifs_assert(len
>= UBIFS_CH_SZ
&& offs
+ len
<= c
->leb_size
);
994 ubifs_assert(!(offs
& 7) && offs
< c
->leb_size
);
995 ubifs_assert(type
>= 0 && type
< UBIFS_NODE_TYPES_CNT
);
997 err
= ubifs_leb_read(c
, lnum
, buf
, offs
, len
, 0);
998 if (err
&& err
!= -EBADMSG
)
1001 if (type
!= ch
->node_type
) {
1002 ubifs_errc(c
, "bad node type (%d but expected %d)",
1003 ch
->node_type
, type
);
1007 err
= ubifs_check_node(c
, buf
, lnum
, offs
, 0, 0);
1009 ubifs_errc(c
, "expected node type %d", type
);
1013 l
= le32_to_cpu(ch
->len
);
1015 ubifs_errc(c
, "bad node length %d, expected %d", l
, len
);
1022 ubifs_errc(c
, "bad node at LEB %d:%d, LEB mapping status %d", lnum
,
1023 offs
, ubi_is_mapped(c
->ubi
, lnum
));
1025 ubifs_dump_node(c
, buf
);
1032 * ubifs_wbuf_init - initialize write-buffer.
1033 * @c: UBIFS file-system description object
1034 * @wbuf: write-buffer to initialize
1036 * This function initializes write-buffer. Returns zero in case of success
1037 * %-ENOMEM in case of failure.
1039 int ubifs_wbuf_init(struct ubifs_info
*c
, struct ubifs_wbuf
*wbuf
)
1043 wbuf
->buf
= kmalloc(c
->max_write_size
, GFP_KERNEL
);
1047 size
= (c
->max_write_size
/ UBIFS_CH_SZ
+ 1) * sizeof(ino_t
);
1048 wbuf
->inodes
= kmalloc(size
, GFP_KERNEL
);
1049 if (!wbuf
->inodes
) {
1056 wbuf
->lnum
= wbuf
->offs
= -1;
1058 * If the LEB starts at the max. write size aligned address, then
1059 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1060 * set it to something smaller so that it ends at the closest max.
1061 * write size boundary.
1063 size
= c
->max_write_size
- (c
->leb_start
% c
->max_write_size
);
1064 wbuf
->avail
= wbuf
->size
= size
;
1065 wbuf
->sync_callback
= NULL
;
1066 mutex_init(&wbuf
->io_mutex
);
1067 spin_lock_init(&wbuf
->lock
);
1072 hrtimer_init(&wbuf
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1073 wbuf
->timer
.function
= wbuf_timer_callback_nolock
;
1074 wbuf
->softlimit
= ktime_set(WBUF_TIMEOUT_SOFTLIMIT
, 0);
1075 wbuf
->delta
= WBUF_TIMEOUT_HARDLIMIT
- WBUF_TIMEOUT_SOFTLIMIT
;
1076 wbuf
->delta
*= 1000000000ULL;
1077 ubifs_assert(wbuf
->delta
<= ULONG_MAX
);
1083 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1084 * @wbuf: the write-buffer where to add
1085 * @inum: the inode number
1087 * This function adds an inode number to the inode array of the write-buffer.
1089 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf
*wbuf
, ino_t inum
)
1092 /* NOR flash or something similar */
1095 spin_lock(&wbuf
->lock
);
1097 wbuf
->inodes
[wbuf
->next_ino
++] = inum
;
1098 spin_unlock(&wbuf
->lock
);
1102 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1103 * @wbuf: the write-buffer
1104 * @inum: the inode number
1106 * This function returns with %1 if the write-buffer contains some data from the
1107 * given inode otherwise it returns with %0.
1109 static int wbuf_has_ino(struct ubifs_wbuf
*wbuf
, ino_t inum
)
1113 spin_lock(&wbuf
->lock
);
1114 for (i
= 0; i
< wbuf
->next_ino
; i
++)
1115 if (inum
== wbuf
->inodes
[i
]) {
1119 spin_unlock(&wbuf
->lock
);
1125 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1126 * @c: UBIFS file-system description object
1127 * @inode: inode to synchronize
1129 * This function synchronizes write-buffers which contain nodes belonging to
1130 * @inode. Returns zero in case of success and a negative error code in case of
1133 int ubifs_sync_wbufs_by_inode(struct ubifs_info
*c
, struct inode
*inode
)
1137 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
1138 struct ubifs_wbuf
*wbuf
= &c
->jheads
[i
].wbuf
;
1142 * GC head is special, do not look at it. Even if the
1143 * head contains something related to this inode, it is
1144 * a _copy_ of corresponding on-flash node which sits
1149 if (!wbuf_has_ino(wbuf
, inode
->i_ino
))
1152 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
1153 if (wbuf_has_ino(wbuf
, inode
->i_ino
))
1154 err
= ubifs_wbuf_sync_nolock(wbuf
);
1155 mutex_unlock(&wbuf
->io_mutex
);
1158 ubifs_ro_mode(c
, err
);