projects/linux·blob master · refs

1262 lines · 36.3 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This file is part of UBIFS.
4 *
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 * Copyright (C) 2006, 2007 University of Szeged, Hungary
7 *
8 * Authors: Artem Bityutskiy (Битюцкий Артём)
9 * Adrian Hunter
10 * Zoltan Sogor
11 */
12
13/*
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.
21 *
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.
31 *
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).
41 *
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.
47 *
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.
53 *
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.
56 *
57 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
58 * they are read from the flash media.
59 */
60
61#include <linux/crc32.h>
62#include <linux/slab.h>
63#include "ubifs.h"
64
65/**
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
69 */
70void ubifs_ro_mode(struct ubifs_info *c, int err)
71{
72 if (!c->ro_error) {
73 c->ro_error = 1;
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);
77 dump_stack();
78 }
79}
80
81/*
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.
85 */
86
87int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
88 int len, int even_ebadmsg)
89{
90 int err;
91
92 err = ubi_read(c->ubi, lnum, buf, offs, len);
93 /*
94 * In case of %-EBADMSG print the error message only if the
95 * @even_ebadmsg is true.
96 */
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);
100 dump_stack();
101 }
102 return err;
103}
104
105int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
106 int len)
107{
108 int err;
109
110 ubifs_assert(c, !c->ro_media && !c->ro_mount);
111 if (c->ro_error)
112 return -EROFS;
113 if (!dbg_is_tst_rcvry(c))
114 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
115 else
116 err = dbg_leb_write(c, lnum, buf, offs, len);
117 if (err) {
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);
121 dump_stack();
122 }
123 return err;
124}
125
126int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
127{
128 int err;
129
130 ubifs_assert(c, !c->ro_media && !c->ro_mount);
131 if (c->ro_error)
132 return -EROFS;
133 if (!dbg_is_tst_rcvry(c))
134 err = ubi_leb_change(c->ubi, lnum, buf, len);
135 else
136 err = dbg_leb_change(c, lnum, buf, len);
137 if (err) {
138 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
139 len, lnum, err);
140 ubifs_ro_mode(c, err);
141 dump_stack();
142 }
143 return err;
144}
145
146int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
147{
148 int err;
149
150 ubifs_assert(c, !c->ro_media && !c->ro_mount);
151 if (c->ro_error)
152 return -EROFS;
153 if (!dbg_is_tst_rcvry(c))
154 err = ubi_leb_unmap(c->ubi, lnum);
155 else
156 err = dbg_leb_unmap(c, lnum);
157 if (err) {
158 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
159 ubifs_ro_mode(c, err);
160 dump_stack();
161 }
162 return err;
163}
164
165int ubifs_leb_map(struct ubifs_info *c, int lnum)
166{
167 int err;
168
169 ubifs_assert(c, !c->ro_media && !c->ro_mount);
170 if (c->ro_error)
171 return -EROFS;
172 if (!dbg_is_tst_rcvry(c))
173 err = ubi_leb_map(c->ubi, lnum);
174 else
175 err = dbg_leb_map(c, lnum);
176 if (err) {
177 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
178 ubifs_ro_mode(c, err);
179 dump_stack();
180 }
181 return err;
182}
183
184int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
185{
186 int err;
187
188 err = ubi_is_mapped(c->ubi, lnum);
189 if (err < 0) {
190 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
191 lnum, err);
192 dump_stack();
193 }
194 return err;
195}
196
197static void record_magic_error(struct ubifs_stats_info *stats)
198{
199 if (stats)
200 stats->magic_errors++;
201}
202
203static void record_node_error(struct ubifs_stats_info *stats)
204{
205 if (stats)
206 stats->node_errors++;
207}
208
209static void record_crc_error(struct ubifs_stats_info *stats)
210{
211 if (stats)
212 stats->crc_errors++;
213}
214
215/**
216 * ubifs_check_node - check node.
217 * @c: UBIFS file-system description object
218 * @buf: node to check
219 * @len: node length
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
224 *
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.
230 *
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
239 * required.
240 *
241 * This function returns zero in case of success and %-EUCLEAN in case of bad
242 * CRC or magic.
243 */
244int ubifs_check_node(const struct ubifs_info *c, const void *buf, int len,
245 int lnum, int offs, int quiet, int must_chk_crc)
246{
247 int err = -EINVAL, type, node_len;
248 uint32_t crc, node_crc, magic;
249 const struct ubifs_ch *ch = buf;
250
251 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
252 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
253
254 magic = le32_to_cpu(ch->magic);
255 if (magic != UBIFS_NODE_MAGIC) {
256 if (!quiet)
257 ubifs_err(c, "bad magic %#08x, expected %#08x",
258 magic, UBIFS_NODE_MAGIC);
259 record_magic_error(c->stats);
260 err = -EUCLEAN;
261 goto out;
262 }
263
264 type = ch->node_type;
265 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
266 if (!quiet)
267 ubifs_err(c, "bad node type %d", type);
268 record_node_error(c->stats);
269 goto out;
270 }
271
272 node_len = le32_to_cpu(ch->len);
273 if (node_len + offs > c->leb_size)
274 goto out_len;
275
276 if (c->ranges[type].max_len == 0) {
277 if (node_len != c->ranges[type].len)
278 goto out_len;
279 } else if (node_len < c->ranges[type].min_len ||
280 node_len > c->ranges[type].max_len)
281 goto out_len;
282
283 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
284 !c->remounting_rw && c->no_chk_data_crc)
285 return 0;
286
287 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
288 node_crc = le32_to_cpu(ch->crc);
289 if (crc != node_crc) {
290 if (!quiet)
291 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
292 crc, node_crc);
293 record_crc_error(c->stats);
294 err = -EUCLEAN;
295 goto out;
296 }
297
298 return 0;
299
300out_len:
301 if (!quiet)
302 ubifs_err(c, "bad node length %d", node_len);
303out:
304 if (!quiet) {
305 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
306 ubifs_dump_node(c, buf, len);
307 dump_stack();
308 }
309 return err;
310}
311
312/**
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
317 *
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).
324 *
325 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
326 * used.
327 */
328void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
329{
330 ubifs_assert(c, pad >= 0);
331
332 if (pad >= UBIFS_PAD_NODE_SZ) {
333 struct ubifs_ch *ch = buf;
334 struct ubifs_pad_node *pad_node = buf;
335
336 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
337 ch->node_type = UBIFS_PAD_NODE;
338 ch->group_type = UBIFS_NO_NODE_GROUP;
339 ch->padding[0] = ch->padding[1] = 0;
340 ch->sqnum = 0;
341 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
342 pad -= UBIFS_PAD_NODE_SZ;
343 pad_node->pad_len = cpu_to_le32(pad);
344 ubifs_crc_node(buf, UBIFS_PAD_NODE_SZ);
345 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
346 } else if (pad > 0)
347 /* Too little space, padding node won't fit */
348 memset(buf, UBIFS_PADDING_BYTE, pad);
349}
350
351/**
352 * next_sqnum - get next sequence number.
353 * @c: UBIFS file-system description object
354 */
355static unsigned long long next_sqnum(struct ubifs_info *c)
356{
357 unsigned long long sqnum;
358
359 spin_lock(&c->cnt_lock);
360 sqnum = ++c->max_sqnum;
361 spin_unlock(&c->cnt_lock);
362
363 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
364 if (sqnum >= SQNUM_WATERMARK) {
365 ubifs_err(c, "sequence number overflow %llu, end of life",
366 sqnum);
367 ubifs_ro_mode(c, -EINVAL);
368 }
369 ubifs_warn(c, "running out of sequence numbers, end of life soon");
370 }
371
372 return sqnum;
373}
374
375void ubifs_init_node(struct ubifs_info *c, void *node, int len, int pad)
376{
377 struct ubifs_ch *ch = node;
378 unsigned long long sqnum = next_sqnum(c);
379
380 ubifs_assert(c, len >= UBIFS_CH_SZ);
381
382 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
383 ch->len = cpu_to_le32(len);
384 ch->group_type = UBIFS_NO_NODE_GROUP;
385 ch->sqnum = cpu_to_le64(sqnum);
386 ch->padding[0] = ch->padding[1] = 0;
387
388 if (pad) {
389 len = ALIGN(len, 8);
390 pad = ALIGN(len, c->min_io_size) - len;
391 ubifs_pad(c, node + len, pad);
392 }
393}
394
395void ubifs_crc_node(void *node, int len)
396{
397 struct ubifs_ch *ch = node;
398 uint32_t crc;
399
400 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
401 ch->crc = cpu_to_le32(crc);
402}
403
404/**
405 * ubifs_prepare_node_hmac - prepare node to be written to flash.
406 * @c: UBIFS file-system description object
407 * @node: the node to pad
408 * @len: node length
409 * @hmac_offs: offset of the HMAC in the node
410 * @pad: if the buffer has to be padded
411 *
412 * This function prepares node at @node to be written to the media - it
413 * calculates node CRC, fills the common header, and adds proper padding up to
414 * the next minimum I/O unit if @pad is not zero. if @hmac_offs is positive then
415 * a HMAC is inserted into the node at the given offset.
416 *
417 * This function returns 0 for success or a negative error code otherwise.
418 */
419int ubifs_prepare_node_hmac(struct ubifs_info *c, void *node, int len,
420 int hmac_offs, int pad)
421{
422 int err;
423
424 ubifs_init_node(c, node, len, pad);
425
426 if (hmac_offs > 0) {
427 err = ubifs_node_insert_hmac(c, node, len, hmac_offs);
428 if (err)
429 return err;
430 }
431
432 ubifs_crc_node(node, len);
433
434 return 0;
435}
436
437/**
438 * ubifs_prepare_node - prepare node to be written to flash.
439 * @c: UBIFS file-system description object
440 * @node: the node to pad
441 * @len: node length
442 * @pad: if the buffer has to be padded
443 *
444 * This function prepares node at @node to be written to the media - it
445 * calculates node CRC, fills the common header, and adds proper padding up to
446 * the next minimum I/O unit if @pad is not zero.
447 */
448void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
449{
450 /*
451 * Deliberately ignore return value since this function can only fail
452 * when a hmac offset is given.
453 */
454 ubifs_prepare_node_hmac(c, node, len, 0, pad);
455}
456
457/**
458 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
459 * @c: UBIFS file-system description object
460 * @node: the node to pad
461 * @len: node length
462 * @last: indicates the last node of the group
463 *
464 * This function prepares node at @node to be written to the media - it
465 * calculates node CRC and fills the common header.
466 */
467void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
468{
469 struct ubifs_ch *ch = node;
470 unsigned long long sqnum = next_sqnum(c);
471
472 ubifs_assert(c, len >= UBIFS_CH_SZ);
473
474 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
475 ch->len = cpu_to_le32(len);
476 if (last)
477 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
478 else
479 ch->group_type = UBIFS_IN_NODE_GROUP;
480 ch->sqnum = cpu_to_le64(sqnum);
481 ch->padding[0] = ch->padding[1] = 0;
482 ubifs_crc_node(node, len);
483}
484
485/**
486 * wbuf_timer_callback_nolock - write-buffer timer callback function.
487 * @timer: timer data (write-buffer descriptor)
488 *
489 * This function is called when the write-buffer timer expires.
490 */
491static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
492{
493 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
494
495 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
496 wbuf->need_sync = 1;
497 wbuf->c->need_wbuf_sync = 1;
498 ubifs_wake_up_bgt(wbuf->c);
499 return HRTIMER_NORESTART;
500}
501
502/**
503 * new_wbuf_timer_nolock - start new write-buffer timer.
504 * @c: UBIFS file-system description object
505 * @wbuf: write-buffer descriptor
506 */
507static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
508{
509 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
510 unsigned long long delta = dirty_writeback_interval;
511
512 /* centi to milli, milli to nano, then 10% */
513 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
514
515 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
516 ubifs_assert(c, delta <= ULONG_MAX);
517
518 if (wbuf->no_timer)
519 return;
520 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
521 dbg_jhead(wbuf->jhead),
522 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
523 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
524 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
525 HRTIMER_MODE_REL);
526}
527
528/**
529 * cancel_wbuf_timer_nolock - cancel write-buffer timer.
530 * @wbuf: write-buffer descriptor
531 */
532static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
533{
534 if (wbuf->no_timer)
535 return;
536 wbuf->need_sync = 0;
537 hrtimer_cancel(&wbuf->timer);
538}
539
540/**
541 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
542 * @wbuf: write-buffer to synchronize
543 *
544 * This function synchronizes write-buffer @buf and returns zero in case of
545 * success or a negative error code in case of failure.
546 *
547 * Note, although write-buffers are of @c->max_write_size, this function does
548 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
549 * if the write-buffer is only partially filled with data, only the used part
550 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
551 * This way we waste less space.
552 */
553int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
554{
555 struct ubifs_info *c = wbuf->c;
556 int err, dirt, sync_len;
557
558 cancel_wbuf_timer_nolock(wbuf);
559 if (!wbuf->used || wbuf->lnum == -1)
560 /* Write-buffer is empty or not seeked */
561 return 0;
562
563 dbg_io("LEB %d:%d, %d bytes, jhead %s",
564 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
565 ubifs_assert(c, !(wbuf->avail & 7));
566 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
567 ubifs_assert(c, wbuf->size >= c->min_io_size);
568 ubifs_assert(c, wbuf->size <= c->max_write_size);
569 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
570 ubifs_assert(c, !c->ro_media && !c->ro_mount);
571 if (c->leb_size - wbuf->offs >= c->max_write_size)
572 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
573
574 if (c->ro_error)
575 return -EROFS;
576
577 /*
578 * Do not write whole write buffer but write only the minimum necessary
579 * amount of min. I/O units.
580 */
581 sync_len = ALIGN(wbuf->used, c->min_io_size);
582 dirt = sync_len - wbuf->used;
583 if (dirt)
584 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
585 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
586 if (err)
587 return err;
588
589 spin_lock(&wbuf->lock);
590 wbuf->offs += sync_len;
591 /*
592 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
593 * But our goal is to optimize writes and make sure we write in
594 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
595 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
596 * sure that @wbuf->offs + @wbuf->size is aligned to
597 * @c->max_write_size. This way we make sure that after next
598 * write-buffer flush we are again at the optimal offset (aligned to
599 * @c->max_write_size).
600 */
601 if (c->leb_size - wbuf->offs < c->max_write_size)
602 wbuf->size = c->leb_size - wbuf->offs;
603 else if (wbuf->offs & (c->max_write_size - 1))
604 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
605 else
606 wbuf->size = c->max_write_size;
607 wbuf->avail = wbuf->size;
608 wbuf->used = 0;
609 wbuf->next_ino = 0;
610 spin_unlock(&wbuf->lock);
611
612 if (wbuf->sync_callback)
613 err = wbuf->sync_callback(c, wbuf->lnum,
614 c->leb_size - wbuf->offs, dirt);
615 return err;
616}
617
618/**
619 * ubifs_wbuf_seek_nolock - seek write-buffer.
620 * @wbuf: write-buffer
621 * @lnum: logical eraseblock number to seek to
622 * @offs: logical eraseblock offset to seek to
623 *
624 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
625 * The write-buffer has to be empty. Returns zero in case of success and a
626 * negative error code in case of failure.
627 */
628int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
629{
630 const struct ubifs_info *c = wbuf->c;
631
632 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
633 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
634 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
635 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
636 ubifs_assert(c, lnum != wbuf->lnum);
637 ubifs_assert(c, wbuf->used == 0);
638
639 spin_lock(&wbuf->lock);
640 wbuf->lnum = lnum;
641 wbuf->offs = offs;
642 if (c->leb_size - wbuf->offs < c->max_write_size)
643 wbuf->size = c->leb_size - wbuf->offs;
644 else if (wbuf->offs & (c->max_write_size - 1))
645 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
646 else
647 wbuf->size = c->max_write_size;
648 wbuf->avail = wbuf->size;
649 wbuf->used = 0;
650 spin_unlock(&wbuf->lock);
651
652 return 0;
653}
654
655/**
656 * ubifs_bg_wbufs_sync - synchronize write-buffers.
657 * @c: UBIFS file-system description object
658 *
659 * This function is called by background thread to synchronize write-buffers.
660 * Returns zero in case of success and a negative error code in case of
661 * failure.
662 */
663int ubifs_bg_wbufs_sync(struct ubifs_info *c)
664{
665 int err, i;
666
667 ubifs_assert(c, !c->ro_media && !c->ro_mount);
668 if (!c->need_wbuf_sync)
669 return 0;
670 c->need_wbuf_sync = 0;
671
672 if (c->ro_error) {
673 err = -EROFS;
674 goto out_timers;
675 }
676
677 dbg_io("synchronize");
678 for (i = 0; i < c->jhead_cnt; i++) {
679 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
680
681 cond_resched();
682
683 /*
684 * If the mutex is locked then wbuf is being changed, so
685 * synchronization is not necessary.
686 */
687 if (mutex_is_locked(&wbuf->io_mutex))
688 continue;
689
690 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
691 if (!wbuf->need_sync) {
692 mutex_unlock(&wbuf->io_mutex);
693 continue;
694 }
695
696 err = ubifs_wbuf_sync_nolock(wbuf);
697 mutex_unlock(&wbuf->io_mutex);
698 if (err) {
699 ubifs_err(c, "cannot sync write-buffer, error %d", err);
700 ubifs_ro_mode(c, err);
701 goto out_timers;
702 }
703 }
704
705 return 0;
706
707out_timers:
708 /* Cancel all timers to prevent repeated errors */
709 for (i = 0; i < c->jhead_cnt; i++) {
710 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
711
712 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
713 cancel_wbuf_timer_nolock(wbuf);
714 mutex_unlock(&wbuf->io_mutex);
715 }
716 return err;
717}
718
719/**
720 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
721 * @wbuf: write-buffer
722 * @buf: node to write
723 * @len: node length
724 *
725 * This function writes data to flash via write-buffer @wbuf. This means that
726 * the last piece of the node won't reach the flash media immediately if it
727 * does not take whole max. write unit (@c->max_write_size). Instead, the node
728 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
729 * because more data are appended to the write-buffer).
730 *
731 * This function returns zero in case of success and a negative error code in
732 * case of failure. If the node cannot be written because there is no more
733 * space in this logical eraseblock, %-ENOSPC is returned.
734 */
735int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
736{
737 struct ubifs_info *c = wbuf->c;
738 int err, n, written = 0, aligned_len = ALIGN(len, 8);
739
740 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
741 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
742 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
743 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
744 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
745 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
746 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
747 ubifs_assert(c, wbuf->size >= c->min_io_size);
748 ubifs_assert(c, wbuf->size <= c->max_write_size);
749 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
750 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
751 ubifs_assert(c, !c->ro_media && !c->ro_mount);
752 ubifs_assert(c, !c->space_fixup);
753 if (c->leb_size - wbuf->offs >= c->max_write_size)
754 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
755
756 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
757 err = -ENOSPC;
758 goto out;
759 }
760
761 cancel_wbuf_timer_nolock(wbuf);
762
763 if (c->ro_error)
764 return -EROFS;
765
766 if (aligned_len <= wbuf->avail) {
767 /*
768 * The node is not very large and fits entirely within
769 * write-buffer.
770 */
771 memcpy(wbuf->buf + wbuf->used, buf, len);
772 if (aligned_len > len) {
773 ubifs_assert(c, aligned_len - len < 8);
774 ubifs_pad(c, wbuf->buf + wbuf->used + len, aligned_len - len);
775 }
776
777 if (aligned_len == wbuf->avail) {
778 dbg_io("flush jhead %s wbuf to LEB %d:%d",
779 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
780 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
781 wbuf->offs, wbuf->size);
782 if (err)
783 goto out;
784
785 spin_lock(&wbuf->lock);
786 wbuf->offs += wbuf->size;
787 if (c->leb_size - wbuf->offs >= c->max_write_size)
788 wbuf->size = c->max_write_size;
789 else
790 wbuf->size = c->leb_size - wbuf->offs;
791 wbuf->avail = wbuf->size;
792 wbuf->used = 0;
793 wbuf->next_ino = 0;
794 spin_unlock(&wbuf->lock);
795 } else {
796 spin_lock(&wbuf->lock);
797 wbuf->avail -= aligned_len;
798 wbuf->used += aligned_len;
799 spin_unlock(&wbuf->lock);
800 }
801
802 goto exit;
803 }
804
805 if (wbuf->used) {
806 /*
807 * The node is large enough and does not fit entirely within
808 * current available space. We have to fill and flush
809 * write-buffer and switch to the next max. write unit.
810 */
811 dbg_io("flush jhead %s wbuf to LEB %d:%d",
812 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
813 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
814 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
815 wbuf->size);
816 if (err)
817 goto out;
818
819 wbuf->offs += wbuf->size;
820 len -= wbuf->avail;
821 aligned_len -= wbuf->avail;
822 written += wbuf->avail;
823 } else if (wbuf->offs & (c->max_write_size - 1)) {
824 /*
825 * The write-buffer offset is not aligned to
826 * @c->max_write_size and @wbuf->size is less than
827 * @c->max_write_size. Write @wbuf->size bytes to make sure the
828 * following writes are done in optimal @c->max_write_size
829 * chunks.
830 */
831 dbg_io("write %d bytes to LEB %d:%d",
832 wbuf->size, wbuf->lnum, wbuf->offs);
833 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
834 wbuf->size);
835 if (err)
836 goto out;
837
838 wbuf->offs += wbuf->size;
839 len -= wbuf->size;
840 aligned_len -= wbuf->size;
841 written += wbuf->size;
842 }
843
844 /*
845 * The remaining data may take more whole max. write units, so write the
846 * remains multiple to max. write unit size directly to the flash media.
847 * We align node length to 8-byte boundary because we anyway flash wbuf
848 * if the remaining space is less than 8 bytes.
849 */
850 n = aligned_len >> c->max_write_shift;
851 if (n) {
852 int m = n - 1;
853
854 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
855 wbuf->offs);
856
857 if (m) {
858 /* '(n-1)<<c->max_write_shift < len' is always true. */
859 m <<= c->max_write_shift;
860 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
861 wbuf->offs, m);
862 if (err)
863 goto out;
864 wbuf->offs += m;
865 aligned_len -= m;
866 len -= m;
867 written += m;
868 }
869
870 /*
871 * The non-written len of buf may be less than 'n' because
872 * parameter 'len' is not 8 bytes aligned, so here we read
873 * min(len, n) bytes from buf.
874 */
875 n = 1 << c->max_write_shift;
876 memcpy(wbuf->buf, buf + written, min(len, n));
877 if (n > len) {
878 ubifs_assert(c, n - len < 8);
879 ubifs_pad(c, wbuf->buf + len, n - len);
880 }
881
882 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, n);
883 if (err)
884 goto out;
885 wbuf->offs += n;
886 aligned_len -= n;
887 len -= min(len, n);
888 written += n;
889 }
890
891 spin_lock(&wbuf->lock);
892 if (aligned_len) {
893 /*
894 * And now we have what's left and what does not take whole
895 * max. write unit, so write it to the write-buffer and we are
896 * done.
897 */
898 memcpy(wbuf->buf, buf + written, len);
899 if (aligned_len > len) {
900 ubifs_assert(c, aligned_len - len < 8);
901 ubifs_pad(c, wbuf->buf + len, aligned_len - len);
902 }
903 }
904
905 if (c->leb_size - wbuf->offs >= c->max_write_size)
906 wbuf->size = c->max_write_size;
907 else
908 wbuf->size = c->leb_size - wbuf->offs;
909 wbuf->avail = wbuf->size - aligned_len;
910 wbuf->used = aligned_len;
911 wbuf->next_ino = 0;
912 spin_unlock(&wbuf->lock);
913
914exit:
915 if (wbuf->sync_callback) {
916 int free = c->leb_size - wbuf->offs - wbuf->used;
917
918 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
919 if (err)
920 goto out;
921 }
922
923 if (wbuf->used)
924 new_wbuf_timer_nolock(c, wbuf);
925
926 return 0;
927
928out:
929 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
930 len, wbuf->lnum, wbuf->offs, err);
931 ubifs_dump_node(c, buf, written + len);
932 dump_stack();
933 ubifs_dump_leb(c, wbuf->lnum);
934 return err;
935}
936
937/**
938 * ubifs_write_node_hmac - write node to the media.
939 * @c: UBIFS file-system description object
940 * @buf: the node to write
941 * @len: node length
942 * @lnum: logical eraseblock number
943 * @offs: offset within the logical eraseblock
944 * @hmac_offs: offset of the HMAC within the node
945 *
946 * This function automatically fills node magic number, assigns sequence
947 * number, and calculates node CRC checksum. The length of the @buf buffer has
948 * to be aligned to the minimal I/O unit size. This function automatically
949 * appends padding node and padding bytes if needed. Returns zero in case of
950 * success and a negative error code in case of failure.
951 */
952int ubifs_write_node_hmac(struct ubifs_info *c, void *buf, int len, int lnum,
953 int offs, int hmac_offs)
954{
955 int err, buf_len = ALIGN(len, c->min_io_size);
956
957 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
958 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
959 buf_len);
960 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
961 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
962 ubifs_assert(c, !c->ro_media && !c->ro_mount);
963 ubifs_assert(c, !c->space_fixup);
964
965 if (c->ro_error)
966 return -EROFS;
967
968 err = ubifs_prepare_node_hmac(c, buf, len, hmac_offs, 1);
969 if (err)
970 return err;
971
972 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
973 if (err)
974 ubifs_dump_node(c, buf, len);
975
976 return err;
977}
978
979/**
980 * ubifs_write_node - write node to the media.
981 * @c: UBIFS file-system description object
982 * @buf: the node to write
983 * @len: node length
984 * @lnum: logical eraseblock number
985 * @offs: offset within the logical eraseblock
986 *
987 * This function automatically fills node magic number, assigns sequence
988 * number, and calculates node CRC checksum. The length of the @buf buffer has
989 * to be aligned to the minimal I/O unit size. This function automatically
990 * appends padding node and padding bytes if needed. Returns zero in case of
991 * success and a negative error code in case of failure.
992 */
993int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
994 int offs)
995{
996 return ubifs_write_node_hmac(c, buf, len, lnum, offs, -1);
997}
998
999/**
1000 * ubifs_read_node_wbuf - read node from the media or write-buffer.
1001 * @wbuf: wbuf to check for un-written data
1002 * @buf: buffer to read to
1003 * @type: node type
1004 * @len: node length
1005 * @lnum: logical eraseblock number
1006 * @offs: offset within the logical eraseblock
1007 *
1008 * This function reads a node of known type and length, checks it and stores
1009 * in @buf. If the node partially or fully sits in the write-buffer, this
1010 * function takes data from the buffer, otherwise it reads the flash media.
1011 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
1012 * error code in case of failure.
1013 */
1014int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
1015 int lnum, int offs)
1016{
1017 const struct ubifs_info *c = wbuf->c;
1018 int err, rlen, overlap;
1019 struct ubifs_ch *ch = buf;
1020
1021 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
1022 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
1023 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1024 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1025 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1026
1027 spin_lock(&wbuf->lock);
1028 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1029 if (!overlap) {
1030 /* We may safely unlock the write-buffer and read the data */
1031 spin_unlock(&wbuf->lock);
1032 return ubifs_read_node(c, buf, type, len, lnum, offs);
1033 }
1034
1035 /* Don't read under wbuf */
1036 rlen = wbuf->offs - offs;
1037 if (rlen < 0)
1038 rlen = 0;
1039
1040 /* Copy the rest from the write-buffer */
1041 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1042 spin_unlock(&wbuf->lock);
1043
1044 if (rlen > 0) {
1045 /* Read everything that goes before write-buffer */
1046 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1047 if (err && err != -EBADMSG)
1048 return err;
1049 }
1050
1051 if (type != ch->node_type) {
1052 ubifs_err(c, "bad node type (%d but expected %d)",
1053 ch->node_type, type);
1054 goto out;
1055 }
1056
1057 err = ubifs_check_node(c, buf, len, lnum, offs, 0, 0);
1058 if (err) {
1059 ubifs_err(c, "expected node type %d", type);
1060 return err;
1061 }
1062
1063 rlen = le32_to_cpu(ch->len);
1064 if (rlen != len) {
1065 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
1066 goto out;
1067 }
1068
1069 return 0;
1070
1071out:
1072 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
1073 ubifs_dump_node(c, buf, len);
1074 dump_stack();
1075 return -EINVAL;
1076}
1077
1078/**
1079 * ubifs_read_node - read node.
1080 * @c: UBIFS file-system description object
1081 * @buf: buffer to read to
1082 * @type: node type
1083 * @len: node length (not aligned)
1084 * @lnum: logical eraseblock number
1085 * @offs: offset within the logical eraseblock
1086 *
1087 * This function reads a node of known type and length, checks it and
1088 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1089 * and a negative error code in case of failure.
1090 */
1091int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1092 int lnum, int offs)
1093{
1094 int err, l;
1095 struct ubifs_ch *ch = buf;
1096
1097 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1098 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1099 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1100 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1101 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1102
1103 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1104 if (err && err != -EBADMSG)
1105 return err;
1106
1107 if (type != ch->node_type) {
1108 ubifs_errc(c, "bad node type (%d but expected %d)",
1109 ch->node_type, type);
1110 goto out;
1111 }
1112
1113 err = ubifs_check_node(c, buf, len, lnum, offs, 0, 0);
1114 if (err) {
1115 ubifs_errc(c, "expected node type %d", type);
1116 return err;
1117 }
1118
1119 l = le32_to_cpu(ch->len);
1120 if (l != len) {
1121 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1122 goto out;
1123 }
1124
1125 return 0;
1126
1127out:
1128 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1129 offs, ubi_is_mapped(c->ubi, lnum));
1130 if (!c->probing) {
1131 ubifs_dump_node(c, buf, len);
1132 dump_stack();
1133 }
1134 return -EINVAL;
1135}
1136
1137/**
1138 * ubifs_wbuf_init - initialize write-buffer.
1139 * @c: UBIFS file-system description object
1140 * @wbuf: write-buffer to initialize
1141 *
1142 * This function initializes write-buffer. Returns zero in case of success
1143 * %-ENOMEM in case of failure.
1144 */
1145int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1146{
1147 size_t size;
1148
1149 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1150 if (!wbuf->buf)
1151 return -ENOMEM;
1152
1153 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1154 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1155 if (!wbuf->inodes) {
1156 kfree(wbuf->buf);
1157 wbuf->buf = NULL;
1158 return -ENOMEM;
1159 }
1160
1161 wbuf->used = 0;
1162 wbuf->lnum = wbuf->offs = -1;
1163 /*
1164 * If the LEB starts at the max. write size aligned address, then
1165 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1166 * set it to something smaller so that it ends at the closest max.
1167 * write size boundary.
1168 */
1169 size = c->max_write_size - (c->leb_start % c->max_write_size);
1170 wbuf->avail = wbuf->size = size;
1171 wbuf->sync_callback = NULL;
1172 mutex_init(&wbuf->io_mutex);
1173 spin_lock_init(&wbuf->lock);
1174 wbuf->c = c;
1175 wbuf->next_ino = 0;
1176
1177 hrtimer_setup(&wbuf->timer, wbuf_timer_callback_nolock, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1178 return 0;
1179}
1180
1181/**
1182 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1183 * @wbuf: the write-buffer where to add
1184 * @inum: the inode number
1185 *
1186 * This function adds an inode number to the inode array of the write-buffer.
1187 */
1188void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1189{
1190 if (!wbuf->buf)
1191 /* NOR flash or something similar */
1192 return;
1193
1194 spin_lock(&wbuf->lock);
1195 if (wbuf->used)
1196 wbuf->inodes[wbuf->next_ino++] = inum;
1197 spin_unlock(&wbuf->lock);
1198}
1199
1200/**
1201 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1202 * @wbuf: the write-buffer
1203 * @inum: the inode number
1204 *
1205 * This function returns with %1 if the write-buffer contains some data from the
1206 * given inode otherwise it returns with %0.
1207 */
1208static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1209{
1210 int i, ret = 0;
1211
1212 spin_lock(&wbuf->lock);
1213 for (i = 0; i < wbuf->next_ino; i++)
1214 if (inum == wbuf->inodes[i]) {
1215 ret = 1;
1216 break;
1217 }
1218 spin_unlock(&wbuf->lock);
1219
1220 return ret;
1221}
1222
1223/**
1224 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1225 * @c: UBIFS file-system description object
1226 * @inode: inode to synchronize
1227 *
1228 * This function synchronizes write-buffers which contain nodes belonging to
1229 * @inode. Returns zero in case of success and a negative error code in case of
1230 * failure.
1231 */
1232int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1233{
1234 int i, err = 0;
1235
1236 for (i = 0; i < c->jhead_cnt; i++) {
1237 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1238
1239 if (i == GCHD)
1240 /*
1241 * GC head is special, do not look at it. Even if the
1242 * head contains something related to this inode, it is
1243 * a _copy_ of corresponding on-flash node which sits
1244 * somewhere else.
1245 */
1246 continue;
1247
1248 if (!wbuf_has_ino(wbuf, inode->i_ino))
1249 continue;
1250
1251 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1252 if (wbuf_has_ino(wbuf, inode->i_ino))
1253 err = ubifs_wbuf_sync_nolock(wbuf);
1254 mutex_unlock(&wbuf->io_mutex);
1255
1256 if (err) {
1257 ubifs_ro_mode(c, err);
1258 return err;
1259 }
1260 }
1261 return 0;
1262}