Loading...
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 uint32_t crc;
331
332 ubifs_assert(c, pad >= 0);
333
334 if (pad >= UBIFS_PAD_NODE_SZ) {
335 struct ubifs_ch *ch = buf;
336 struct ubifs_pad_node *pad_node = buf;
337
338 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
339 ch->node_type = UBIFS_PAD_NODE;
340 ch->group_type = UBIFS_NO_NODE_GROUP;
341 ch->padding[0] = ch->padding[1] = 0;
342 ch->sqnum = 0;
343 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
344 pad -= UBIFS_PAD_NODE_SZ;
345 pad_node->pad_len = cpu_to_le32(pad);
346 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
347 ch->crc = cpu_to_le32(crc);
348 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
349 } else if (pad > 0)
350 /* Too little space, padding node won't fit */
351 memset(buf, UBIFS_PADDING_BYTE, pad);
352}
353
354/**
355 * next_sqnum - get next sequence number.
356 * @c: UBIFS file-system description object
357 */
358static unsigned long long next_sqnum(struct ubifs_info *c)
359{
360 unsigned long long sqnum;
361
362 spin_lock(&c->cnt_lock);
363 sqnum = ++c->max_sqnum;
364 spin_unlock(&c->cnt_lock);
365
366 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
367 if (sqnum >= SQNUM_WATERMARK) {
368 ubifs_err(c, "sequence number overflow %llu, end of life",
369 sqnum);
370 ubifs_ro_mode(c, -EINVAL);
371 }
372 ubifs_warn(c, "running out of sequence numbers, end of life soon");
373 }
374
375 return sqnum;
376}
377
378void ubifs_init_node(struct ubifs_info *c, void *node, int len, int pad)
379{
380 struct ubifs_ch *ch = node;
381 unsigned long long sqnum = next_sqnum(c);
382
383 ubifs_assert(c, len >= UBIFS_CH_SZ);
384
385 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
386 ch->len = cpu_to_le32(len);
387 ch->group_type = UBIFS_NO_NODE_GROUP;
388 ch->sqnum = cpu_to_le64(sqnum);
389 ch->padding[0] = ch->padding[1] = 0;
390
391 if (pad) {
392 len = ALIGN(len, 8);
393 pad = ALIGN(len, c->min_io_size) - len;
394 ubifs_pad(c, node + len, pad);
395 }
396}
397
398void ubifs_crc_node(struct ubifs_info *c, void *node, int len)
399{
400 struct ubifs_ch *ch = node;
401 uint32_t crc;
402
403 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
404 ch->crc = cpu_to_le32(crc);
405}
406
407/**
408 * ubifs_prepare_node_hmac - prepare node to be written to flash.
409 * @c: UBIFS file-system description object
410 * @node: the node to pad
411 * @len: node length
412 * @hmac_offs: offset of the HMAC in the node
413 * @pad: if the buffer has to be padded
414 *
415 * This function prepares node at @node to be written to the media - it
416 * calculates node CRC, fills the common header, and adds proper padding up to
417 * the next minimum I/O unit if @pad is not zero. if @hmac_offs is positive then
418 * a HMAC is inserted into the node at the given offset.
419 *
420 * This function returns 0 for success or a negative error code otherwise.
421 */
422int ubifs_prepare_node_hmac(struct ubifs_info *c, void *node, int len,
423 int hmac_offs, int pad)
424{
425 int err;
426
427 ubifs_init_node(c, node, len, pad);
428
429 if (hmac_offs > 0) {
430 err = ubifs_node_insert_hmac(c, node, len, hmac_offs);
431 if (err)
432 return err;
433 }
434
435 ubifs_crc_node(c, node, len);
436
437 return 0;
438}
439
440/**
441 * ubifs_prepare_node - prepare node to be written to flash.
442 * @c: UBIFS file-system description object
443 * @node: the node to pad
444 * @len: node length
445 * @pad: if the buffer has to be padded
446 *
447 * This function prepares node at @node to be written to the media - it
448 * calculates node CRC, fills the common header, and adds proper padding up to
449 * the next minimum I/O unit if @pad is not zero.
450 */
451void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
452{
453 /*
454 * Deliberately ignore return value since this function can only fail
455 * when a hmac offset is given.
456 */
457 ubifs_prepare_node_hmac(c, node, len, 0, pad);
458}
459
460/**
461 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
462 * @c: UBIFS file-system description object
463 * @node: the node to pad
464 * @len: node length
465 * @last: indicates the last node of the group
466 *
467 * This function prepares node at @node to be written to the media - it
468 * calculates node CRC and fills the common header.
469 */
470void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
471{
472 uint32_t crc;
473 struct ubifs_ch *ch = node;
474 unsigned long long sqnum = next_sqnum(c);
475
476 ubifs_assert(c, len >= UBIFS_CH_SZ);
477
478 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
479 ch->len = cpu_to_le32(len);
480 if (last)
481 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
482 else
483 ch->group_type = UBIFS_IN_NODE_GROUP;
484 ch->sqnum = cpu_to_le64(sqnum);
485 ch->padding[0] = ch->padding[1] = 0;
486 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
487 ch->crc = cpu_to_le32(crc);
488}
489
490/**
491 * wbuf_timer_callback_nolock - write-buffer timer callback function.
492 * @timer: timer data (write-buffer descriptor)
493 *
494 * This function is called when the write-buffer timer expires.
495 */
496static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
497{
498 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
499
500 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
501 wbuf->need_sync = 1;
502 wbuf->c->need_wbuf_sync = 1;
503 ubifs_wake_up_bgt(wbuf->c);
504 return HRTIMER_NORESTART;
505}
506
507/**
508 * new_wbuf_timer_nolock - start new write-buffer timer.
509 * @c: UBIFS file-system description object
510 * @wbuf: write-buffer descriptor
511 */
512static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
513{
514 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
515 unsigned long long delta = dirty_writeback_interval;
516
517 /* centi to milli, milli to nano, then 10% */
518 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
519
520 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
521 ubifs_assert(c, delta <= ULONG_MAX);
522
523 if (wbuf->no_timer)
524 return;
525 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
526 dbg_jhead(wbuf->jhead),
527 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
528 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
529 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
530 HRTIMER_MODE_REL);
531}
532
533/**
534 * cancel_wbuf_timer_nolock - cancel write-buffer timer.
535 * @wbuf: write-buffer descriptor
536 */
537static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
538{
539 if (wbuf->no_timer)
540 return;
541 wbuf->need_sync = 0;
542 hrtimer_cancel(&wbuf->timer);
543}
544
545/**
546 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
547 * @wbuf: write-buffer to synchronize
548 *
549 * This function synchronizes write-buffer @buf and returns zero in case of
550 * success or a negative error code in case of failure.
551 *
552 * Note, although write-buffers are of @c->max_write_size, this function does
553 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
554 * if the write-buffer is only partially filled with data, only the used part
555 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
556 * This way we waste less space.
557 */
558int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
559{
560 struct ubifs_info *c = wbuf->c;
561 int err, dirt, sync_len;
562
563 cancel_wbuf_timer_nolock(wbuf);
564 if (!wbuf->used || wbuf->lnum == -1)
565 /* Write-buffer is empty or not seeked */
566 return 0;
567
568 dbg_io("LEB %d:%d, %d bytes, jhead %s",
569 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
570 ubifs_assert(c, !(wbuf->avail & 7));
571 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
572 ubifs_assert(c, wbuf->size >= c->min_io_size);
573 ubifs_assert(c, wbuf->size <= c->max_write_size);
574 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
575 ubifs_assert(c, !c->ro_media && !c->ro_mount);
576 if (c->leb_size - wbuf->offs >= c->max_write_size)
577 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
578
579 if (c->ro_error)
580 return -EROFS;
581
582 /*
583 * Do not write whole write buffer but write only the minimum necessary
584 * amount of min. I/O units.
585 */
586 sync_len = ALIGN(wbuf->used, c->min_io_size);
587 dirt = sync_len - wbuf->used;
588 if (dirt)
589 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
590 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
591 if (err)
592 return err;
593
594 spin_lock(&wbuf->lock);
595 wbuf->offs += sync_len;
596 /*
597 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
598 * But our goal is to optimize writes and make sure we write in
599 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
600 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
601 * sure that @wbuf->offs + @wbuf->size is aligned to
602 * @c->max_write_size. This way we make sure that after next
603 * write-buffer flush we are again at the optimal offset (aligned to
604 * @c->max_write_size).
605 */
606 if (c->leb_size - wbuf->offs < c->max_write_size)
607 wbuf->size = c->leb_size - wbuf->offs;
608 else if (wbuf->offs & (c->max_write_size - 1))
609 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
610 else
611 wbuf->size = c->max_write_size;
612 wbuf->avail = wbuf->size;
613 wbuf->used = 0;
614 wbuf->next_ino = 0;
615 spin_unlock(&wbuf->lock);
616
617 if (wbuf->sync_callback)
618 err = wbuf->sync_callback(c, wbuf->lnum,
619 c->leb_size - wbuf->offs, dirt);
620 return err;
621}
622
623/**
624 * ubifs_wbuf_seek_nolock - seek write-buffer.
625 * @wbuf: write-buffer
626 * @lnum: logical eraseblock number to seek to
627 * @offs: logical eraseblock offset to seek to
628 *
629 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
630 * The write-buffer has to be empty. Returns zero in case of success and a
631 * negative error code in case of failure.
632 */
633int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
634{
635 const struct ubifs_info *c = wbuf->c;
636
637 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
638 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
639 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
640 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
641 ubifs_assert(c, lnum != wbuf->lnum);
642 ubifs_assert(c, wbuf->used == 0);
643
644 spin_lock(&wbuf->lock);
645 wbuf->lnum = lnum;
646 wbuf->offs = offs;
647 if (c->leb_size - wbuf->offs < c->max_write_size)
648 wbuf->size = c->leb_size - wbuf->offs;
649 else if (wbuf->offs & (c->max_write_size - 1))
650 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
651 else
652 wbuf->size = c->max_write_size;
653 wbuf->avail = wbuf->size;
654 wbuf->used = 0;
655 spin_unlock(&wbuf->lock);
656
657 return 0;
658}
659
660/**
661 * ubifs_bg_wbufs_sync - synchronize write-buffers.
662 * @c: UBIFS file-system description object
663 *
664 * This function is called by background thread to synchronize write-buffers.
665 * Returns zero in case of success and a negative error code in case of
666 * failure.
667 */
668int ubifs_bg_wbufs_sync(struct ubifs_info *c)
669{
670 int err, i;
671
672 ubifs_assert(c, !c->ro_media && !c->ro_mount);
673 if (!c->need_wbuf_sync)
674 return 0;
675 c->need_wbuf_sync = 0;
676
677 if (c->ro_error) {
678 err = -EROFS;
679 goto out_timers;
680 }
681
682 dbg_io("synchronize");
683 for (i = 0; i < c->jhead_cnt; i++) {
684 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
685
686 cond_resched();
687
688 /*
689 * If the mutex is locked then wbuf is being changed, so
690 * synchronization is not necessary.
691 */
692 if (mutex_is_locked(&wbuf->io_mutex))
693 continue;
694
695 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
696 if (!wbuf->need_sync) {
697 mutex_unlock(&wbuf->io_mutex);
698 continue;
699 }
700
701 err = ubifs_wbuf_sync_nolock(wbuf);
702 mutex_unlock(&wbuf->io_mutex);
703 if (err) {
704 ubifs_err(c, "cannot sync write-buffer, error %d", err);
705 ubifs_ro_mode(c, err);
706 goto out_timers;
707 }
708 }
709
710 return 0;
711
712out_timers:
713 /* Cancel all timers to prevent repeated errors */
714 for (i = 0; i < c->jhead_cnt; i++) {
715 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
716
717 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
718 cancel_wbuf_timer_nolock(wbuf);
719 mutex_unlock(&wbuf->io_mutex);
720 }
721 return err;
722}
723
724/**
725 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
726 * @wbuf: write-buffer
727 * @buf: node to write
728 * @len: node length
729 *
730 * This function writes data to flash via write-buffer @wbuf. This means that
731 * the last piece of the node won't reach the flash media immediately if it
732 * does not take whole max. write unit (@c->max_write_size). Instead, the node
733 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
734 * because more data are appended to the write-buffer).
735 *
736 * This function returns zero in case of success and a negative error code in
737 * case of failure. If the node cannot be written because there is no more
738 * space in this logical eraseblock, %-ENOSPC is returned.
739 */
740int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
741{
742 struct ubifs_info *c = wbuf->c;
743 int err, n, written = 0, aligned_len = ALIGN(len, 8);
744
745 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
746 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
747 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
748 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
749 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
750 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
751 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
752 ubifs_assert(c, wbuf->size >= c->min_io_size);
753 ubifs_assert(c, wbuf->size <= c->max_write_size);
754 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
755 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
756 ubifs_assert(c, !c->ro_media && !c->ro_mount);
757 ubifs_assert(c, !c->space_fixup);
758 if (c->leb_size - wbuf->offs >= c->max_write_size)
759 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
760
761 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
762 err = -ENOSPC;
763 goto out;
764 }
765
766 cancel_wbuf_timer_nolock(wbuf);
767
768 if (c->ro_error)
769 return -EROFS;
770
771 if (aligned_len <= wbuf->avail) {
772 /*
773 * The node is not very large and fits entirely within
774 * write-buffer.
775 */
776 memcpy(wbuf->buf + wbuf->used, buf, len);
777 if (aligned_len > len) {
778 ubifs_assert(c, aligned_len - len < 8);
779 ubifs_pad(c, wbuf->buf + wbuf->used + len, aligned_len - len);
780 }
781
782 if (aligned_len == wbuf->avail) {
783 dbg_io("flush jhead %s wbuf to LEB %d:%d",
784 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
785 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
786 wbuf->offs, wbuf->size);
787 if (err)
788 goto out;
789
790 spin_lock(&wbuf->lock);
791 wbuf->offs += wbuf->size;
792 if (c->leb_size - wbuf->offs >= c->max_write_size)
793 wbuf->size = c->max_write_size;
794 else
795 wbuf->size = c->leb_size - wbuf->offs;
796 wbuf->avail = wbuf->size;
797 wbuf->used = 0;
798 wbuf->next_ino = 0;
799 spin_unlock(&wbuf->lock);
800 } else {
801 spin_lock(&wbuf->lock);
802 wbuf->avail -= aligned_len;
803 wbuf->used += aligned_len;
804 spin_unlock(&wbuf->lock);
805 }
806
807 goto exit;
808 }
809
810 if (wbuf->used) {
811 /*
812 * The node is large enough and does not fit entirely within
813 * current available space. We have to fill and flush
814 * write-buffer and switch to the next max. write unit.
815 */
816 dbg_io("flush jhead %s wbuf to LEB %d:%d",
817 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
818 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
819 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
820 wbuf->size);
821 if (err)
822 goto out;
823
824 wbuf->offs += wbuf->size;
825 len -= wbuf->avail;
826 aligned_len -= wbuf->avail;
827 written += wbuf->avail;
828 } else if (wbuf->offs & (c->max_write_size - 1)) {
829 /*
830 * The write-buffer offset is not aligned to
831 * @c->max_write_size and @wbuf->size is less than
832 * @c->max_write_size. Write @wbuf->size bytes to make sure the
833 * following writes are done in optimal @c->max_write_size
834 * chunks.
835 */
836 dbg_io("write %d bytes to LEB %d:%d",
837 wbuf->size, wbuf->lnum, wbuf->offs);
838 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
839 wbuf->size);
840 if (err)
841 goto out;
842
843 wbuf->offs += wbuf->size;
844 len -= wbuf->size;
845 aligned_len -= wbuf->size;
846 written += wbuf->size;
847 }
848
849 /*
850 * The remaining data may take more whole max. write units, so write the
851 * remains multiple to max. write unit size directly to the flash media.
852 * We align node length to 8-byte boundary because we anyway flash wbuf
853 * if the remaining space is less than 8 bytes.
854 */
855 n = aligned_len >> c->max_write_shift;
856 if (n) {
857 int m = n - 1;
858
859 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
860 wbuf->offs);
861
862 if (m) {
863 /* '(n-1)<<c->max_write_shift < len' is always true. */
864 m <<= c->max_write_shift;
865 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
866 wbuf->offs, m);
867 if (err)
868 goto out;
869 wbuf->offs += m;
870 aligned_len -= m;
871 len -= m;
872 written += m;
873 }
874
875 /*
876 * The non-written len of buf may be less than 'n' because
877 * parameter 'len' is not 8 bytes aligned, so here we read
878 * min(len, n) bytes from buf.
879 */
880 n = 1 << c->max_write_shift;
881 memcpy(wbuf->buf, buf + written, min(len, n));
882 if (n > len) {
883 ubifs_assert(c, n - len < 8);
884 ubifs_pad(c, wbuf->buf + len, n - len);
885 }
886
887 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, n);
888 if (err)
889 goto out;
890 wbuf->offs += n;
891 aligned_len -= n;
892 len -= min(len, n);
893 written += n;
894 }
895
896 spin_lock(&wbuf->lock);
897 if (aligned_len) {
898 /*
899 * And now we have what's left and what does not take whole
900 * max. write unit, so write it to the write-buffer and we are
901 * done.
902 */
903 memcpy(wbuf->buf, buf + written, len);
904 if (aligned_len > len) {
905 ubifs_assert(c, aligned_len - len < 8);
906 ubifs_pad(c, wbuf->buf + len, aligned_len - len);
907 }
908 }
909
910 if (c->leb_size - wbuf->offs >= c->max_write_size)
911 wbuf->size = c->max_write_size;
912 else
913 wbuf->size = c->leb_size - wbuf->offs;
914 wbuf->avail = wbuf->size - aligned_len;
915 wbuf->used = aligned_len;
916 wbuf->next_ino = 0;
917 spin_unlock(&wbuf->lock);
918
919exit:
920 if (wbuf->sync_callback) {
921 int free = c->leb_size - wbuf->offs - wbuf->used;
922
923 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
924 if (err)
925 goto out;
926 }
927
928 if (wbuf->used)
929 new_wbuf_timer_nolock(c, wbuf);
930
931 return 0;
932
933out:
934 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
935 len, wbuf->lnum, wbuf->offs, err);
936 ubifs_dump_node(c, buf, written + len);
937 dump_stack();
938 ubifs_dump_leb(c, wbuf->lnum);
939 return err;
940}
941
942/**
943 * ubifs_write_node_hmac - write node to the media.
944 * @c: UBIFS file-system description object
945 * @buf: the node to write
946 * @len: node length
947 * @lnum: logical eraseblock number
948 * @offs: offset within the logical eraseblock
949 * @hmac_offs: offset of the HMAC within the node
950 *
951 * This function automatically fills node magic number, assigns sequence
952 * number, and calculates node CRC checksum. The length of the @buf buffer has
953 * to be aligned to the minimal I/O unit size. This function automatically
954 * appends padding node and padding bytes if needed. Returns zero in case of
955 * success and a negative error code in case of failure.
956 */
957int ubifs_write_node_hmac(struct ubifs_info *c, void *buf, int len, int lnum,
958 int offs, int hmac_offs)
959{
960 int err, buf_len = ALIGN(len, c->min_io_size);
961
962 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
963 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
964 buf_len);
965 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
966 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
967 ubifs_assert(c, !c->ro_media && !c->ro_mount);
968 ubifs_assert(c, !c->space_fixup);
969
970 if (c->ro_error)
971 return -EROFS;
972
973 err = ubifs_prepare_node_hmac(c, buf, len, hmac_offs, 1);
974 if (err)
975 return err;
976
977 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
978 if (err)
979 ubifs_dump_node(c, buf, len);
980
981 return err;
982}
983
984/**
985 * ubifs_write_node - write node to the media.
986 * @c: UBIFS file-system description object
987 * @buf: the node to write
988 * @len: node length
989 * @lnum: logical eraseblock number
990 * @offs: offset within the logical eraseblock
991 *
992 * This function automatically fills node magic number, assigns sequence
993 * number, and calculates node CRC checksum. The length of the @buf buffer has
994 * to be aligned to the minimal I/O unit size. This function automatically
995 * appends padding node and padding bytes if needed. Returns zero in case of
996 * success and a negative error code in case of failure.
997 */
998int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
999 int offs)
1000{
1001 return ubifs_write_node_hmac(c, buf, len, lnum, offs, -1);
1002}
1003
1004/**
1005 * ubifs_read_node_wbuf - read node from the media or write-buffer.
1006 * @wbuf: wbuf to check for un-written data
1007 * @buf: buffer to read to
1008 * @type: node type
1009 * @len: node length
1010 * @lnum: logical eraseblock number
1011 * @offs: offset within the logical eraseblock
1012 *
1013 * This function reads a node of known type and length, checks it and stores
1014 * in @buf. If the node partially or fully sits in the write-buffer, this
1015 * function takes data from the buffer, otherwise it reads the flash media.
1016 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
1017 * error code in case of failure.
1018 */
1019int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
1020 int lnum, int offs)
1021{
1022 const struct ubifs_info *c = wbuf->c;
1023 int err, rlen, overlap;
1024 struct ubifs_ch *ch = buf;
1025
1026 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
1027 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
1028 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1029 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1030 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1031
1032 spin_lock(&wbuf->lock);
1033 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1034 if (!overlap) {
1035 /* We may safely unlock the write-buffer and read the data */
1036 spin_unlock(&wbuf->lock);
1037 return ubifs_read_node(c, buf, type, len, lnum, offs);
1038 }
1039
1040 /* Don't read under wbuf */
1041 rlen = wbuf->offs - offs;
1042 if (rlen < 0)
1043 rlen = 0;
1044
1045 /* Copy the rest from the write-buffer */
1046 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1047 spin_unlock(&wbuf->lock);
1048
1049 if (rlen > 0) {
1050 /* Read everything that goes before write-buffer */
1051 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1052 if (err && err != -EBADMSG)
1053 return err;
1054 }
1055
1056 if (type != ch->node_type) {
1057 ubifs_err(c, "bad node type (%d but expected %d)",
1058 ch->node_type, type);
1059 goto out;
1060 }
1061
1062 err = ubifs_check_node(c, buf, len, lnum, offs, 0, 0);
1063 if (err) {
1064 ubifs_err(c, "expected node type %d", type);
1065 return err;
1066 }
1067
1068 rlen = le32_to_cpu(ch->len);
1069 if (rlen != len) {
1070 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
1071 goto out;
1072 }
1073
1074 return 0;
1075
1076out:
1077 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
1078 ubifs_dump_node(c, buf, len);
1079 dump_stack();
1080 return -EINVAL;
1081}
1082
1083/**
1084 * ubifs_read_node - read node.
1085 * @c: UBIFS file-system description object
1086 * @buf: buffer to read to
1087 * @type: node type
1088 * @len: node length (not aligned)
1089 * @lnum: logical eraseblock number
1090 * @offs: offset within the logical eraseblock
1091 *
1092 * This function reads a node of known type and length, checks it and
1093 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1094 * and a negative error code in case of failure.
1095 */
1096int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1097 int lnum, int offs)
1098{
1099 int err, l;
1100 struct ubifs_ch *ch = buf;
1101
1102 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1103 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1104 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1105 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1106 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1107
1108 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1109 if (err && err != -EBADMSG)
1110 return err;
1111
1112 if (type != ch->node_type) {
1113 ubifs_errc(c, "bad node type (%d but expected %d)",
1114 ch->node_type, type);
1115 goto out;
1116 }
1117
1118 err = ubifs_check_node(c, buf, len, lnum, offs, 0, 0);
1119 if (err) {
1120 ubifs_errc(c, "expected node type %d", type);
1121 return err;
1122 }
1123
1124 l = le32_to_cpu(ch->len);
1125 if (l != len) {
1126 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1127 goto out;
1128 }
1129
1130 return 0;
1131
1132out:
1133 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1134 offs, ubi_is_mapped(c->ubi, lnum));
1135 if (!c->probing) {
1136 ubifs_dump_node(c, buf, len);
1137 dump_stack();
1138 }
1139 return -EINVAL;
1140}
1141
1142/**
1143 * ubifs_wbuf_init - initialize write-buffer.
1144 * @c: UBIFS file-system description object
1145 * @wbuf: write-buffer to initialize
1146 *
1147 * This function initializes write-buffer. Returns zero in case of success
1148 * %-ENOMEM in case of failure.
1149 */
1150int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1151{
1152 size_t size;
1153
1154 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1155 if (!wbuf->buf)
1156 return -ENOMEM;
1157
1158 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1159 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1160 if (!wbuf->inodes) {
1161 kfree(wbuf->buf);
1162 wbuf->buf = NULL;
1163 return -ENOMEM;
1164 }
1165
1166 wbuf->used = 0;
1167 wbuf->lnum = wbuf->offs = -1;
1168 /*
1169 * If the LEB starts at the max. write size aligned address, then
1170 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1171 * set it to something smaller so that it ends at the closest max.
1172 * write size boundary.
1173 */
1174 size = c->max_write_size - (c->leb_start % c->max_write_size);
1175 wbuf->avail = wbuf->size = size;
1176 wbuf->sync_callback = NULL;
1177 mutex_init(&wbuf->io_mutex);
1178 spin_lock_init(&wbuf->lock);
1179 wbuf->c = c;
1180 wbuf->next_ino = 0;
1181
1182 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1183 wbuf->timer.function = wbuf_timer_callback_nolock;
1184 return 0;
1185}
1186
1187/**
1188 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1189 * @wbuf: the write-buffer where to add
1190 * @inum: the inode number
1191 *
1192 * This function adds an inode number to the inode array of the write-buffer.
1193 */
1194void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1195{
1196 if (!wbuf->buf)
1197 /* NOR flash or something similar */
1198 return;
1199
1200 spin_lock(&wbuf->lock);
1201 if (wbuf->used)
1202 wbuf->inodes[wbuf->next_ino++] = inum;
1203 spin_unlock(&wbuf->lock);
1204}
1205
1206/**
1207 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1208 * @wbuf: the write-buffer
1209 * @inum: the inode number
1210 *
1211 * This function returns with %1 if the write-buffer contains some data from the
1212 * given inode otherwise it returns with %0.
1213 */
1214static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1215{
1216 int i, ret = 0;
1217
1218 spin_lock(&wbuf->lock);
1219 for (i = 0; i < wbuf->next_ino; i++)
1220 if (inum == wbuf->inodes[i]) {
1221 ret = 1;
1222 break;
1223 }
1224 spin_unlock(&wbuf->lock);
1225
1226 return ret;
1227}
1228
1229/**
1230 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1231 * @c: UBIFS file-system description object
1232 * @inode: inode to synchronize
1233 *
1234 * This function synchronizes write-buffers which contain nodes belonging to
1235 * @inode. Returns zero in case of success and a negative error code in case of
1236 * failure.
1237 */
1238int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1239{
1240 int i, err = 0;
1241
1242 for (i = 0; i < c->jhead_cnt; i++) {
1243 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1244
1245 if (i == GCHD)
1246 /*
1247 * GC head is special, do not look at it. Even if the
1248 * head contains something related to this inode, it is
1249 * a _copy_ of corresponding on-flash node which sits
1250 * somewhere else.
1251 */
1252 continue;
1253
1254 if (!wbuf_has_ino(wbuf, inode->i_ino))
1255 continue;
1256
1257 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1258 if (wbuf_has_ino(wbuf, inode->i_ino))
1259 err = ubifs_wbuf_sync_nolock(wbuf);
1260 mutex_unlock(&wbuf->io_mutex);
1261
1262 if (err) {
1263 ubifs_ro_mode(c, err);
1264 return err;
1265 }
1266 }
1267 return 0;
1268}
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
197/**
198 * ubifs_check_node - check node.
199 * @c: UBIFS file-system description object
200 * @buf: node to check
201 * @lnum: logical eraseblock number
202 * @offs: offset within the logical eraseblock
203 * @quiet: print no messages
204 * @must_chk_crc: indicates whether to always check the CRC
205 *
206 * This function checks node magic number and CRC checksum. This function also
207 * validates node length to prevent UBIFS from becoming crazy when an attacker
208 * feeds it a file-system image with incorrect nodes. For example, too large
209 * node length in the common header could cause UBIFS to read memory outside of
210 * allocated buffer when checking the CRC checksum.
211 *
212 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
213 * true, which is controlled by corresponding UBIFS mount option. However, if
214 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
215 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
216 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
217 * is checked. This is because during mounting or re-mounting from R/O mode to
218 * R/W mode we may read journal nodes (when replying the journal or doing the
219 * recovery) and the journal nodes may potentially be corrupted, so checking is
220 * required.
221 *
222 * This function returns zero in case of success and %-EUCLEAN in case of bad
223 * CRC or magic.
224 */
225int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
226 int offs, int quiet, int must_chk_crc)
227{
228 int err = -EINVAL, type, node_len, dump_node = 1;
229 uint32_t crc, node_crc, magic;
230 const struct ubifs_ch *ch = buf;
231
232 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
233 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
234
235 magic = le32_to_cpu(ch->magic);
236 if (magic != UBIFS_NODE_MAGIC) {
237 if (!quiet)
238 ubifs_err(c, "bad magic %#08x, expected %#08x",
239 magic, UBIFS_NODE_MAGIC);
240 err = -EUCLEAN;
241 goto out;
242 }
243
244 type = ch->node_type;
245 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
246 if (!quiet)
247 ubifs_err(c, "bad node type %d", type);
248 goto out;
249 }
250
251 node_len = le32_to_cpu(ch->len);
252 if (node_len + offs > c->leb_size)
253 goto out_len;
254
255 if (c->ranges[type].max_len == 0) {
256 if (node_len != c->ranges[type].len)
257 goto out_len;
258 } else if (node_len < c->ranges[type].min_len ||
259 node_len > c->ranges[type].max_len)
260 goto out_len;
261
262 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
263 !c->remounting_rw && c->no_chk_data_crc)
264 return 0;
265
266 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
267 node_crc = le32_to_cpu(ch->crc);
268 if (crc != node_crc) {
269 if (!quiet)
270 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
271 crc, node_crc);
272 err = -EUCLEAN;
273 goto out;
274 }
275
276 return 0;
277
278out_len:
279 if (!quiet)
280 ubifs_err(c, "bad node length %d", node_len);
281 if (type == UBIFS_DATA_NODE && node_len > UBIFS_DATA_NODE_SZ)
282 dump_node = 0;
283out:
284 if (!quiet) {
285 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
286 if (dump_node) {
287 ubifs_dump_node(c, buf);
288 } else {
289 int safe_len = min3(node_len, c->leb_size - offs,
290 (int)UBIFS_MAX_DATA_NODE_SZ);
291 pr_err("\tprevent out-of-bounds memory access\n");
292 pr_err("\ttruncated data node length %d\n", safe_len);
293 pr_err("\tcorrupted data node:\n");
294 print_hex_dump(KERN_ERR, "\t", DUMP_PREFIX_OFFSET, 32, 1,
295 buf, safe_len, 0);
296 }
297 dump_stack();
298 }
299 return err;
300}
301
302/**
303 * ubifs_pad - pad flash space.
304 * @c: UBIFS file-system description object
305 * @buf: buffer to put padding to
306 * @pad: how many bytes to pad
307 *
308 * The flash media obliges us to write only in chunks of %c->min_io_size and
309 * when we have to write less data we add padding node to the write-buffer and
310 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
311 * media is being scanned. If the amount of wasted space is not enough to fit a
312 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
313 * pattern (%UBIFS_PADDING_BYTE).
314 *
315 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
316 * used.
317 */
318void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
319{
320 uint32_t crc;
321
322 ubifs_assert(c, pad >= 0 && !(pad & 7));
323
324 if (pad >= UBIFS_PAD_NODE_SZ) {
325 struct ubifs_ch *ch = buf;
326 struct ubifs_pad_node *pad_node = buf;
327
328 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
329 ch->node_type = UBIFS_PAD_NODE;
330 ch->group_type = UBIFS_NO_NODE_GROUP;
331 ch->padding[0] = ch->padding[1] = 0;
332 ch->sqnum = 0;
333 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
334 pad -= UBIFS_PAD_NODE_SZ;
335 pad_node->pad_len = cpu_to_le32(pad);
336 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
337 ch->crc = cpu_to_le32(crc);
338 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
339 } else if (pad > 0)
340 /* Too little space, padding node won't fit */
341 memset(buf, UBIFS_PADDING_BYTE, pad);
342}
343
344/**
345 * next_sqnum - get next sequence number.
346 * @c: UBIFS file-system description object
347 */
348static unsigned long long next_sqnum(struct ubifs_info *c)
349{
350 unsigned long long sqnum;
351
352 spin_lock(&c->cnt_lock);
353 sqnum = ++c->max_sqnum;
354 spin_unlock(&c->cnt_lock);
355
356 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
357 if (sqnum >= SQNUM_WATERMARK) {
358 ubifs_err(c, "sequence number overflow %llu, end of life",
359 sqnum);
360 ubifs_ro_mode(c, -EINVAL);
361 }
362 ubifs_warn(c, "running out of sequence numbers, end of life soon");
363 }
364
365 return sqnum;
366}
367
368void ubifs_init_node(struct ubifs_info *c, void *node, int len, int pad)
369{
370 struct ubifs_ch *ch = node;
371 unsigned long long sqnum = next_sqnum(c);
372
373 ubifs_assert(c, len >= UBIFS_CH_SZ);
374
375 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
376 ch->len = cpu_to_le32(len);
377 ch->group_type = UBIFS_NO_NODE_GROUP;
378 ch->sqnum = cpu_to_le64(sqnum);
379 ch->padding[0] = ch->padding[1] = 0;
380
381 if (pad) {
382 len = ALIGN(len, 8);
383 pad = ALIGN(len, c->min_io_size) - len;
384 ubifs_pad(c, node + len, pad);
385 }
386}
387
388void ubifs_crc_node(struct ubifs_info *c, void *node, int len)
389{
390 struct ubifs_ch *ch = node;
391 uint32_t crc;
392
393 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
394 ch->crc = cpu_to_le32(crc);
395}
396
397/**
398 * ubifs_prepare_node_hmac - prepare node to be written to flash.
399 * @c: UBIFS file-system description object
400 * @node: the node to pad
401 * @len: node length
402 * @hmac_offs: offset of the HMAC in the node
403 * @pad: if the buffer has to be padded
404 *
405 * This function prepares node at @node to be written to the media - it
406 * calculates node CRC, fills the common header, and adds proper padding up to
407 * the next minimum I/O unit if @pad is not zero. if @hmac_offs is positive then
408 * a HMAC is inserted into the node at the given offset.
409 *
410 * This function returns 0 for success or a negative error code otherwise.
411 */
412int ubifs_prepare_node_hmac(struct ubifs_info *c, void *node, int len,
413 int hmac_offs, int pad)
414{
415 int err;
416
417 ubifs_init_node(c, node, len, pad);
418
419 if (hmac_offs > 0) {
420 err = ubifs_node_insert_hmac(c, node, len, hmac_offs);
421 if (err)
422 return err;
423 }
424
425 ubifs_crc_node(c, node, len);
426
427 return 0;
428}
429
430/**
431 * ubifs_prepare_node - prepare node to be written to flash.
432 * @c: UBIFS file-system description object
433 * @node: the node to pad
434 * @len: node length
435 * @pad: if the buffer has to be padded
436 *
437 * This function prepares node at @node to be written to the media - it
438 * calculates node CRC, fills the common header, and adds proper padding up to
439 * the next minimum I/O unit if @pad is not zero.
440 */
441void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
442{
443 /*
444 * Deliberately ignore return value since this function can only fail
445 * when a hmac offset is given.
446 */
447 ubifs_prepare_node_hmac(c, node, len, 0, pad);
448}
449
450/**
451 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
452 * @c: UBIFS file-system description object
453 * @node: the node to pad
454 * @len: node length
455 * @last: indicates the last node of the group
456 *
457 * This function prepares node at @node to be written to the media - it
458 * calculates node CRC and fills the common header.
459 */
460void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
461{
462 uint32_t crc;
463 struct ubifs_ch *ch = node;
464 unsigned long long sqnum = next_sqnum(c);
465
466 ubifs_assert(c, len >= UBIFS_CH_SZ);
467
468 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
469 ch->len = cpu_to_le32(len);
470 if (last)
471 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
472 else
473 ch->group_type = UBIFS_IN_NODE_GROUP;
474 ch->sqnum = cpu_to_le64(sqnum);
475 ch->padding[0] = ch->padding[1] = 0;
476 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
477 ch->crc = cpu_to_le32(crc);
478}
479
480/**
481 * wbuf_timer_callback - write-buffer timer callback function.
482 * @timer: timer data (write-buffer descriptor)
483 *
484 * This function is called when the write-buffer timer expires.
485 */
486static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
487{
488 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
489
490 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
491 wbuf->need_sync = 1;
492 wbuf->c->need_wbuf_sync = 1;
493 ubifs_wake_up_bgt(wbuf->c);
494 return HRTIMER_NORESTART;
495}
496
497/**
498 * new_wbuf_timer - start new write-buffer timer.
499 * @c: UBIFS file-system description object
500 * @wbuf: write-buffer descriptor
501 */
502static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
503{
504 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
505 unsigned long long delta = dirty_writeback_interval;
506
507 /* centi to milli, milli to nano, then 10% */
508 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
509
510 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
511 ubifs_assert(c, delta <= ULONG_MAX);
512
513 if (wbuf->no_timer)
514 return;
515 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
516 dbg_jhead(wbuf->jhead),
517 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
518 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
519 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
520 HRTIMER_MODE_REL);
521}
522
523/**
524 * cancel_wbuf_timer - cancel write-buffer timer.
525 * @wbuf: write-buffer descriptor
526 */
527static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
528{
529 if (wbuf->no_timer)
530 return;
531 wbuf->need_sync = 0;
532 hrtimer_cancel(&wbuf->timer);
533}
534
535/**
536 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
537 * @wbuf: write-buffer to synchronize
538 *
539 * This function synchronizes write-buffer @buf and returns zero in case of
540 * success or a negative error code in case of failure.
541 *
542 * Note, although write-buffers are of @c->max_write_size, this function does
543 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
544 * if the write-buffer is only partially filled with data, only the used part
545 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
546 * This way we waste less space.
547 */
548int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
549{
550 struct ubifs_info *c = wbuf->c;
551 int err, dirt, sync_len;
552
553 cancel_wbuf_timer_nolock(wbuf);
554 if (!wbuf->used || wbuf->lnum == -1)
555 /* Write-buffer is empty or not seeked */
556 return 0;
557
558 dbg_io("LEB %d:%d, %d bytes, jhead %s",
559 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
560 ubifs_assert(c, !(wbuf->avail & 7));
561 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
562 ubifs_assert(c, wbuf->size >= c->min_io_size);
563 ubifs_assert(c, wbuf->size <= c->max_write_size);
564 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
565 ubifs_assert(c, !c->ro_media && !c->ro_mount);
566 if (c->leb_size - wbuf->offs >= c->max_write_size)
567 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
568
569 if (c->ro_error)
570 return -EROFS;
571
572 /*
573 * Do not write whole write buffer but write only the minimum necessary
574 * amount of min. I/O units.
575 */
576 sync_len = ALIGN(wbuf->used, c->min_io_size);
577 dirt = sync_len - wbuf->used;
578 if (dirt)
579 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
580 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
581 if (err)
582 return err;
583
584 spin_lock(&wbuf->lock);
585 wbuf->offs += sync_len;
586 /*
587 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
588 * But our goal is to optimize writes and make sure we write in
589 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
590 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
591 * sure that @wbuf->offs + @wbuf->size is aligned to
592 * @c->max_write_size. This way we make sure that after next
593 * write-buffer flush we are again at the optimal offset (aligned to
594 * @c->max_write_size).
595 */
596 if (c->leb_size - wbuf->offs < c->max_write_size)
597 wbuf->size = c->leb_size - wbuf->offs;
598 else if (wbuf->offs & (c->max_write_size - 1))
599 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
600 else
601 wbuf->size = c->max_write_size;
602 wbuf->avail = wbuf->size;
603 wbuf->used = 0;
604 wbuf->next_ino = 0;
605 spin_unlock(&wbuf->lock);
606
607 if (wbuf->sync_callback)
608 err = wbuf->sync_callback(c, wbuf->lnum,
609 c->leb_size - wbuf->offs, dirt);
610 return err;
611}
612
613/**
614 * ubifs_wbuf_seek_nolock - seek write-buffer.
615 * @wbuf: write-buffer
616 * @lnum: logical eraseblock number to seek to
617 * @offs: logical eraseblock offset to seek to
618 *
619 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
620 * The write-buffer has to be empty. Returns zero in case of success and a
621 * negative error code in case of failure.
622 */
623int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
624{
625 const struct ubifs_info *c = wbuf->c;
626
627 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
628 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
629 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
630 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
631 ubifs_assert(c, lnum != wbuf->lnum);
632 ubifs_assert(c, wbuf->used == 0);
633
634 spin_lock(&wbuf->lock);
635 wbuf->lnum = lnum;
636 wbuf->offs = offs;
637 if (c->leb_size - wbuf->offs < c->max_write_size)
638 wbuf->size = c->leb_size - wbuf->offs;
639 else if (wbuf->offs & (c->max_write_size - 1))
640 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
641 else
642 wbuf->size = c->max_write_size;
643 wbuf->avail = wbuf->size;
644 wbuf->used = 0;
645 spin_unlock(&wbuf->lock);
646
647 return 0;
648}
649
650/**
651 * ubifs_bg_wbufs_sync - synchronize write-buffers.
652 * @c: UBIFS file-system description object
653 *
654 * This function is called by background thread to synchronize write-buffers.
655 * Returns zero in case of success and a negative error code in case of
656 * failure.
657 */
658int ubifs_bg_wbufs_sync(struct ubifs_info *c)
659{
660 int err, i;
661
662 ubifs_assert(c, !c->ro_media && !c->ro_mount);
663 if (!c->need_wbuf_sync)
664 return 0;
665 c->need_wbuf_sync = 0;
666
667 if (c->ro_error) {
668 err = -EROFS;
669 goto out_timers;
670 }
671
672 dbg_io("synchronize");
673 for (i = 0; i < c->jhead_cnt; i++) {
674 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
675
676 cond_resched();
677
678 /*
679 * If the mutex is locked then wbuf is being changed, so
680 * synchronization is not necessary.
681 */
682 if (mutex_is_locked(&wbuf->io_mutex))
683 continue;
684
685 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
686 if (!wbuf->need_sync) {
687 mutex_unlock(&wbuf->io_mutex);
688 continue;
689 }
690
691 err = ubifs_wbuf_sync_nolock(wbuf);
692 mutex_unlock(&wbuf->io_mutex);
693 if (err) {
694 ubifs_err(c, "cannot sync write-buffer, error %d", err);
695 ubifs_ro_mode(c, err);
696 goto out_timers;
697 }
698 }
699
700 return 0;
701
702out_timers:
703 /* Cancel all timers to prevent repeated errors */
704 for (i = 0; i < c->jhead_cnt; i++) {
705 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
706
707 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
708 cancel_wbuf_timer_nolock(wbuf);
709 mutex_unlock(&wbuf->io_mutex);
710 }
711 return err;
712}
713
714/**
715 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
716 * @wbuf: write-buffer
717 * @buf: node to write
718 * @len: node length
719 *
720 * This function writes data to flash via write-buffer @wbuf. This means that
721 * the last piece of the node won't reach the flash media immediately if it
722 * does not take whole max. write unit (@c->max_write_size). Instead, the node
723 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
724 * because more data are appended to the write-buffer).
725 *
726 * This function returns zero in case of success and a negative error code in
727 * case of failure. If the node cannot be written because there is no more
728 * space in this logical eraseblock, %-ENOSPC is returned.
729 */
730int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
731{
732 struct ubifs_info *c = wbuf->c;
733 int err, written, n, aligned_len = ALIGN(len, 8);
734
735 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
736 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
737 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
738 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
739 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
740 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
741 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
742 ubifs_assert(c, wbuf->size >= c->min_io_size);
743 ubifs_assert(c, wbuf->size <= c->max_write_size);
744 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
745 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
746 ubifs_assert(c, !c->ro_media && !c->ro_mount);
747 ubifs_assert(c, !c->space_fixup);
748 if (c->leb_size - wbuf->offs >= c->max_write_size)
749 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
750
751 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
752 err = -ENOSPC;
753 goto out;
754 }
755
756 cancel_wbuf_timer_nolock(wbuf);
757
758 if (c->ro_error)
759 return -EROFS;
760
761 if (aligned_len <= wbuf->avail) {
762 /*
763 * The node is not very large and fits entirely within
764 * write-buffer.
765 */
766 memcpy(wbuf->buf + wbuf->used, buf, len);
767
768 if (aligned_len == wbuf->avail) {
769 dbg_io("flush jhead %s wbuf to LEB %d:%d",
770 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
771 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
772 wbuf->offs, wbuf->size);
773 if (err)
774 goto out;
775
776 spin_lock(&wbuf->lock);
777 wbuf->offs += wbuf->size;
778 if (c->leb_size - wbuf->offs >= c->max_write_size)
779 wbuf->size = c->max_write_size;
780 else
781 wbuf->size = c->leb_size - wbuf->offs;
782 wbuf->avail = wbuf->size;
783 wbuf->used = 0;
784 wbuf->next_ino = 0;
785 spin_unlock(&wbuf->lock);
786 } else {
787 spin_lock(&wbuf->lock);
788 wbuf->avail -= aligned_len;
789 wbuf->used += aligned_len;
790 spin_unlock(&wbuf->lock);
791 }
792
793 goto exit;
794 }
795
796 written = 0;
797
798 if (wbuf->used) {
799 /*
800 * The node is large enough and does not fit entirely within
801 * current available space. We have to fill and flush
802 * write-buffer and switch to the next max. write unit.
803 */
804 dbg_io("flush jhead %s wbuf to LEB %d:%d",
805 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
806 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
807 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
808 wbuf->size);
809 if (err)
810 goto out;
811
812 wbuf->offs += wbuf->size;
813 len -= wbuf->avail;
814 aligned_len -= wbuf->avail;
815 written += wbuf->avail;
816 } else if (wbuf->offs & (c->max_write_size - 1)) {
817 /*
818 * The write-buffer offset is not aligned to
819 * @c->max_write_size and @wbuf->size is less than
820 * @c->max_write_size. Write @wbuf->size bytes to make sure the
821 * following writes are done in optimal @c->max_write_size
822 * chunks.
823 */
824 dbg_io("write %d bytes to LEB %d:%d",
825 wbuf->size, wbuf->lnum, wbuf->offs);
826 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
827 wbuf->size);
828 if (err)
829 goto out;
830
831 wbuf->offs += wbuf->size;
832 len -= wbuf->size;
833 aligned_len -= wbuf->size;
834 written += wbuf->size;
835 }
836
837 /*
838 * The remaining data may take more whole max. write units, so write the
839 * remains multiple to max. write unit size directly to the flash media.
840 * We align node length to 8-byte boundary because we anyway flash wbuf
841 * if the remaining space is less than 8 bytes.
842 */
843 n = aligned_len >> c->max_write_shift;
844 if (n) {
845 n <<= c->max_write_shift;
846 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
847 wbuf->offs);
848 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
849 wbuf->offs, n);
850 if (err)
851 goto out;
852 wbuf->offs += n;
853 aligned_len -= n;
854 len -= n;
855 written += n;
856 }
857
858 spin_lock(&wbuf->lock);
859 if (aligned_len)
860 /*
861 * And now we have what's left and what does not take whole
862 * max. write unit, so write it to the write-buffer and we are
863 * done.
864 */
865 memcpy(wbuf->buf, buf + written, len);
866
867 if (c->leb_size - wbuf->offs >= c->max_write_size)
868 wbuf->size = c->max_write_size;
869 else
870 wbuf->size = c->leb_size - wbuf->offs;
871 wbuf->avail = wbuf->size - aligned_len;
872 wbuf->used = aligned_len;
873 wbuf->next_ino = 0;
874 spin_unlock(&wbuf->lock);
875
876exit:
877 if (wbuf->sync_callback) {
878 int free = c->leb_size - wbuf->offs - wbuf->used;
879
880 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
881 if (err)
882 goto out;
883 }
884
885 if (wbuf->used)
886 new_wbuf_timer_nolock(c, wbuf);
887
888 return 0;
889
890out:
891 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
892 len, wbuf->lnum, wbuf->offs, err);
893 ubifs_dump_node(c, buf);
894 dump_stack();
895 ubifs_dump_leb(c, wbuf->lnum);
896 return err;
897}
898
899/**
900 * ubifs_write_node_hmac - write node to the media.
901 * @c: UBIFS file-system description object
902 * @buf: the node to write
903 * @len: node length
904 * @lnum: logical eraseblock number
905 * @offs: offset within the logical eraseblock
906 * @hmac_offs: offset of the HMAC within the node
907 *
908 * This function automatically fills node magic number, assigns sequence
909 * number, and calculates node CRC checksum. The length of the @buf buffer has
910 * to be aligned to the minimal I/O unit size. This function automatically
911 * appends padding node and padding bytes if needed. Returns zero in case of
912 * success and a negative error code in case of failure.
913 */
914int ubifs_write_node_hmac(struct ubifs_info *c, void *buf, int len, int lnum,
915 int offs, int hmac_offs)
916{
917 int err, buf_len = ALIGN(len, c->min_io_size);
918
919 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
920 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
921 buf_len);
922 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
923 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
924 ubifs_assert(c, !c->ro_media && !c->ro_mount);
925 ubifs_assert(c, !c->space_fixup);
926
927 if (c->ro_error)
928 return -EROFS;
929
930 err = ubifs_prepare_node_hmac(c, buf, len, hmac_offs, 1);
931 if (err)
932 return err;
933
934 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
935 if (err)
936 ubifs_dump_node(c, buf);
937
938 return err;
939}
940
941/**
942 * ubifs_write_node - write node to the media.
943 * @c: UBIFS file-system description object
944 * @buf: the node to write
945 * @len: node length
946 * @lnum: logical eraseblock number
947 * @offs: offset within the logical eraseblock
948 *
949 * This function automatically fills node magic number, assigns sequence
950 * number, and calculates node CRC checksum. The length of the @buf buffer has
951 * to be aligned to the minimal I/O unit size. This function automatically
952 * appends padding node and padding bytes if needed. Returns zero in case of
953 * success and a negative error code in case of failure.
954 */
955int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
956 int offs)
957{
958 return ubifs_write_node_hmac(c, buf, len, lnum, offs, -1);
959}
960
961/**
962 * ubifs_read_node_wbuf - read node from the media or write-buffer.
963 * @wbuf: wbuf to check for un-written data
964 * @buf: buffer to read to
965 * @type: node type
966 * @len: node length
967 * @lnum: logical eraseblock number
968 * @offs: offset within the logical eraseblock
969 *
970 * This function reads a node of known type and length, checks it and stores
971 * in @buf. If the node partially or fully sits in the write-buffer, this
972 * function takes data from the buffer, otherwise it reads the flash media.
973 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
974 * error code in case of failure.
975 */
976int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
977 int lnum, int offs)
978{
979 const struct ubifs_info *c = wbuf->c;
980 int err, rlen, overlap;
981 struct ubifs_ch *ch = buf;
982
983 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
984 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
985 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
986 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
987 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
988
989 spin_lock(&wbuf->lock);
990 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
991 if (!overlap) {
992 /* We may safely unlock the write-buffer and read the data */
993 spin_unlock(&wbuf->lock);
994 return ubifs_read_node(c, buf, type, len, lnum, offs);
995 }
996
997 /* Don't read under wbuf */
998 rlen = wbuf->offs - offs;
999 if (rlen < 0)
1000 rlen = 0;
1001
1002 /* Copy the rest from the write-buffer */
1003 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1004 spin_unlock(&wbuf->lock);
1005
1006 if (rlen > 0) {
1007 /* Read everything that goes before write-buffer */
1008 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1009 if (err && err != -EBADMSG)
1010 return err;
1011 }
1012
1013 if (type != ch->node_type) {
1014 ubifs_err(c, "bad node type (%d but expected %d)",
1015 ch->node_type, type);
1016 goto out;
1017 }
1018
1019 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1020 if (err) {
1021 ubifs_err(c, "expected node type %d", type);
1022 return err;
1023 }
1024
1025 rlen = le32_to_cpu(ch->len);
1026 if (rlen != len) {
1027 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
1028 goto out;
1029 }
1030
1031 return 0;
1032
1033out:
1034 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
1035 ubifs_dump_node(c, buf);
1036 dump_stack();
1037 return -EINVAL;
1038}
1039
1040/**
1041 * ubifs_read_node - read node.
1042 * @c: UBIFS file-system description object
1043 * @buf: buffer to read to
1044 * @type: node type
1045 * @len: node length (not aligned)
1046 * @lnum: logical eraseblock number
1047 * @offs: offset within the logical eraseblock
1048 *
1049 * This function reads a node of known type and and length, checks it and
1050 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1051 * and a negative error code in case of failure.
1052 */
1053int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1054 int lnum, int offs)
1055{
1056 int err, l;
1057 struct ubifs_ch *ch = buf;
1058
1059 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1060 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1061 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1062 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1063 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1064
1065 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1066 if (err && err != -EBADMSG)
1067 return err;
1068
1069 if (type != ch->node_type) {
1070 ubifs_errc(c, "bad node type (%d but expected %d)",
1071 ch->node_type, type);
1072 goto out;
1073 }
1074
1075 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1076 if (err) {
1077 ubifs_errc(c, "expected node type %d", type);
1078 return err;
1079 }
1080
1081 l = le32_to_cpu(ch->len);
1082 if (l != len) {
1083 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1084 goto out;
1085 }
1086
1087 return 0;
1088
1089out:
1090 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1091 offs, ubi_is_mapped(c->ubi, lnum));
1092 if (!c->probing) {
1093 ubifs_dump_node(c, buf);
1094 dump_stack();
1095 }
1096 return -EINVAL;
1097}
1098
1099/**
1100 * ubifs_wbuf_init - initialize write-buffer.
1101 * @c: UBIFS file-system description object
1102 * @wbuf: write-buffer to initialize
1103 *
1104 * This function initializes write-buffer. Returns zero in case of success
1105 * %-ENOMEM in case of failure.
1106 */
1107int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1108{
1109 size_t size;
1110
1111 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1112 if (!wbuf->buf)
1113 return -ENOMEM;
1114
1115 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1116 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1117 if (!wbuf->inodes) {
1118 kfree(wbuf->buf);
1119 wbuf->buf = NULL;
1120 return -ENOMEM;
1121 }
1122
1123 wbuf->used = 0;
1124 wbuf->lnum = wbuf->offs = -1;
1125 /*
1126 * If the LEB starts at the max. write size aligned address, then
1127 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1128 * set it to something smaller so that it ends at the closest max.
1129 * write size boundary.
1130 */
1131 size = c->max_write_size - (c->leb_start % c->max_write_size);
1132 wbuf->avail = wbuf->size = size;
1133 wbuf->sync_callback = NULL;
1134 mutex_init(&wbuf->io_mutex);
1135 spin_lock_init(&wbuf->lock);
1136 wbuf->c = c;
1137 wbuf->next_ino = 0;
1138
1139 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1140 wbuf->timer.function = wbuf_timer_callback_nolock;
1141 return 0;
1142}
1143
1144/**
1145 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1146 * @wbuf: the write-buffer where to add
1147 * @inum: the inode number
1148 *
1149 * This function adds an inode number to the inode array of the write-buffer.
1150 */
1151void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1152{
1153 if (!wbuf->buf)
1154 /* NOR flash or something similar */
1155 return;
1156
1157 spin_lock(&wbuf->lock);
1158 if (wbuf->used)
1159 wbuf->inodes[wbuf->next_ino++] = inum;
1160 spin_unlock(&wbuf->lock);
1161}
1162
1163/**
1164 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1165 * @wbuf: the write-buffer
1166 * @inum: the inode number
1167 *
1168 * This function returns with %1 if the write-buffer contains some data from the
1169 * given inode otherwise it returns with %0.
1170 */
1171static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1172{
1173 int i, ret = 0;
1174
1175 spin_lock(&wbuf->lock);
1176 for (i = 0; i < wbuf->next_ino; i++)
1177 if (inum == wbuf->inodes[i]) {
1178 ret = 1;
1179 break;
1180 }
1181 spin_unlock(&wbuf->lock);
1182
1183 return ret;
1184}
1185
1186/**
1187 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1188 * @c: UBIFS file-system description object
1189 * @inode: inode to synchronize
1190 *
1191 * This function synchronizes write-buffers which contain nodes belonging to
1192 * @inode. Returns zero in case of success and a negative error code in case of
1193 * failure.
1194 */
1195int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1196{
1197 int i, err = 0;
1198
1199 for (i = 0; i < c->jhead_cnt; i++) {
1200 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1201
1202 if (i == GCHD)
1203 /*
1204 * GC head is special, do not look at it. Even if the
1205 * head contains something related to this inode, it is
1206 * a _copy_ of corresponding on-flash node which sits
1207 * somewhere else.
1208 */
1209 continue;
1210
1211 if (!wbuf_has_ino(wbuf, inode->i_ino))
1212 continue;
1213
1214 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1215 if (wbuf_has_ino(wbuf, inode->i_ino))
1216 err = ubifs_wbuf_sync_nolock(wbuf);
1217 mutex_unlock(&wbuf->io_mutex);
1218
1219 if (err) {
1220 ubifs_ro_mode(c, err);
1221 return err;
1222 }
1223 }
1224 return 0;
1225}