blob: e8e632a1dcdf1ea52834bf4a0441108d72e6a4a4 [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
2 * This file is part of UBIFS.
3 *
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 * Copyright (C) 2006, 2007 University of Szeged, Hungary
6 *
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 as published by
9 * the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along with
17 * this program; if not, write to the Free Software Foundation, Inc., 51
18 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Authors: Artem Bityutskiy (Битюцкий Артём)
21 * Adrian Hunter
22 * Zoltan Sogor
23 */
24
25/*
26 * This file implements UBIFS I/O subsystem which provides various I/O-related
27 * helper functions (reading/writing/checking/validating nodes) and implements
28 * write-buffering support. Write buffers help to save space which otherwise
29 * would have been wasted for padding to the nearest minimal I/O unit boundary.
30 * Instead, data first goes to the write-buffer and is flushed when the
31 * buffer is full or when it is not used for some time (by timer). This is
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +020032 * similar to the mechanism is used by JFFS2.
Artem Bityutskiy1e517642008-07-14 19:08:37 +030033 *
34 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
35 * mutexes defined inside these objects. Since sometimes upper-level code
36 * has to lock the write-buffer (e.g. journal space reservation code), many
37 * functions related to write-buffers have "nolock" suffix which means that the
38 * caller has to lock the write-buffer before calling this function.
39 *
40 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
41 * aligned, UBIFS starts the next node from the aligned address, and the padded
42 * bytes may contain any rubbish. In other words, UBIFS does not put padding
43 * bytes in those small gaps. Common headers of nodes store real node lengths,
44 * not aligned lengths. Indexing nodes also store real lengths in branches.
45 *
46 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
47 * uses padding nodes or padding bytes, if the padding node does not fit.
48 *
49 * All UBIFS nodes are protected by CRC checksums and UBIFS checks all nodes
50 * every time they are read from the flash media.
51 */
52
53#include <linux/crc32.h>
54#include "ubifs.h"
55
56/**
Adrian Hunterff46d7b2008-07-21 15:39:05 +030057 * ubifs_ro_mode - switch UBIFS to read read-only mode.
58 * @c: UBIFS file-system description object
59 * @err: error code which is the reason of switching to R/O mode
60 */
61void ubifs_ro_mode(struct ubifs_info *c, int err)
62{
63 if (!c->ro_media) {
64 c->ro_media = 1;
Artem Bityutskiyccb3eba2008-09-08 16:07:01 +030065 c->no_chk_data_crc = 0;
Adrian Hunterff46d7b2008-07-21 15:39:05 +030066 ubifs_warn("switched to read-only mode, error %d", err);
67 dbg_dump_stack();
68 }
69}
70
71/**
Artem Bityutskiy1e517642008-07-14 19:08:37 +030072 * ubifs_check_node - check node.
73 * @c: UBIFS file-system description object
74 * @buf: node to check
75 * @lnum: logical eraseblock number
76 * @offs: offset within the logical eraseblock
77 * @quiet: print no messages
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +020078 * @must_chk_crc: indicates whether to always check the CRC
Artem Bityutskiy1e517642008-07-14 19:08:37 +030079 *
80 * This function checks node magic number and CRC checksum. This function also
81 * validates node length to prevent UBIFS from becoming crazy when an attacker
82 * feeds it a file-system image with incorrect nodes. For example, too large
83 * node length in the common header could cause UBIFS to read memory outside of
84 * allocated buffer when checking the CRC checksum.
85 *
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +020086 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
87 * true, which is controlled by corresponding UBIFS mount option. However, if
88 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
89 * checked. Similarly, if @c->always_chk_crc is true, @c->no_chk_data_crc is
90 * ignored and CRC is checked.
91 *
92 * This function returns zero in case of success and %-EUCLEAN in case of bad
93 * CRC or magic.
Artem Bityutskiy1e517642008-07-14 19:08:37 +030094 */
95int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +020096 int offs, int quiet, int must_chk_crc)
Artem Bityutskiy1e517642008-07-14 19:08:37 +030097{
98 int err = -EINVAL, type, node_len;
99 uint32_t crc, node_crc, magic;
100 const struct ubifs_ch *ch = buf;
101
102 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
103 ubifs_assert(!(offs & 7) && offs < c->leb_size);
104
105 magic = le32_to_cpu(ch->magic);
106 if (magic != UBIFS_NODE_MAGIC) {
107 if (!quiet)
108 ubifs_err("bad magic %#08x, expected %#08x",
109 magic, UBIFS_NODE_MAGIC);
110 err = -EUCLEAN;
111 goto out;
112 }
113
114 type = ch->node_type;
115 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
116 if (!quiet)
117 ubifs_err("bad node type %d", type);
118 goto out;
119 }
120
121 node_len = le32_to_cpu(ch->len);
122 if (node_len + offs > c->leb_size)
123 goto out_len;
124
125 if (c->ranges[type].max_len == 0) {
126 if (node_len != c->ranges[type].len)
127 goto out_len;
128 } else if (node_len < c->ranges[type].min_len ||
129 node_len > c->ranges[type].max_len)
130 goto out_len;
131
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +0200132 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->always_chk_crc &&
133 c->no_chk_data_crc)
134 return 0;
Adrian Hunter2953e732008-09-04 16:26:00 +0300135
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300136 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
137 node_crc = le32_to_cpu(ch->crc);
138 if (crc != node_crc) {
139 if (!quiet)
140 ubifs_err("bad CRC: calculated %#08x, read %#08x",
141 crc, node_crc);
142 err = -EUCLEAN;
143 goto out;
144 }
145
146 return 0;
147
148out_len:
149 if (!quiet)
150 ubifs_err("bad node length %d", node_len);
151out:
152 if (!quiet) {
153 ubifs_err("bad node at LEB %d:%d", lnum, offs);
154 dbg_dump_node(c, buf);
155 dbg_dump_stack();
156 }
157 return err;
158}
159
160/**
161 * ubifs_pad - pad flash space.
162 * @c: UBIFS file-system description object
163 * @buf: buffer to put padding to
164 * @pad: how many bytes to pad
165 *
166 * The flash media obliges us to write only in chunks of %c->min_io_size and
167 * when we have to write less data we add padding node to the write-buffer and
168 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
169 * media is being scanned. If the amount of wasted space is not enough to fit a
170 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
171 * pattern (%UBIFS_PADDING_BYTE).
172 *
173 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
174 * used.
175 */
176void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
177{
178 uint32_t crc;
179
180 ubifs_assert(pad >= 0 && !(pad & 7));
181
182 if (pad >= UBIFS_PAD_NODE_SZ) {
183 struct ubifs_ch *ch = buf;
184 struct ubifs_pad_node *pad_node = buf;
185
186 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
187 ch->node_type = UBIFS_PAD_NODE;
188 ch->group_type = UBIFS_NO_NODE_GROUP;
189 ch->padding[0] = ch->padding[1] = 0;
190 ch->sqnum = 0;
191 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
192 pad -= UBIFS_PAD_NODE_SZ;
193 pad_node->pad_len = cpu_to_le32(pad);
194 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
195 ch->crc = cpu_to_le32(crc);
196 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
197 } else if (pad > 0)
198 /* Too little space, padding node won't fit */
199 memset(buf, UBIFS_PADDING_BYTE, pad);
200}
201
202/**
203 * next_sqnum - get next sequence number.
204 * @c: UBIFS file-system description object
205 */
206static unsigned long long next_sqnum(struct ubifs_info *c)
207{
208 unsigned long long sqnum;
209
210 spin_lock(&c->cnt_lock);
211 sqnum = ++c->max_sqnum;
212 spin_unlock(&c->cnt_lock);
213
214 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
215 if (sqnum >= SQNUM_WATERMARK) {
216 ubifs_err("sequence number overflow %llu, end of life",
217 sqnum);
218 ubifs_ro_mode(c, -EINVAL);
219 }
220 ubifs_warn("running out of sequence numbers, end of life soon");
221 }
222
223 return sqnum;
224}
225
226/**
227 * ubifs_prepare_node - prepare node to be written to flash.
228 * @c: UBIFS file-system description object
229 * @node: the node to pad
230 * @len: node length
231 * @pad: if the buffer has to be padded
232 *
233 * This function prepares node at @node to be written to the media - it
234 * calculates node CRC, fills the common header, and adds proper padding up to
235 * the next minimum I/O unit if @pad is not zero.
236 */
237void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
238{
239 uint32_t crc;
240 struct ubifs_ch *ch = node;
241 unsigned long long sqnum = next_sqnum(c);
242
243 ubifs_assert(len >= UBIFS_CH_SZ);
244
245 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
246 ch->len = cpu_to_le32(len);
247 ch->group_type = UBIFS_NO_NODE_GROUP;
248 ch->sqnum = cpu_to_le64(sqnum);
249 ch->padding[0] = ch->padding[1] = 0;
250 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
251 ch->crc = cpu_to_le32(crc);
252
253 if (pad) {
254 len = ALIGN(len, 8);
255 pad = ALIGN(len, c->min_io_size) - len;
256 ubifs_pad(c, node + len, pad);
257 }
258}
259
260/**
261 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
262 * @c: UBIFS file-system description object
263 * @node: the node to pad
264 * @len: node length
265 * @last: indicates the last node of the group
266 *
267 * This function prepares node at @node to be written to the media - it
268 * calculates node CRC and fills the common header.
269 */
270void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
271{
272 uint32_t crc;
273 struct ubifs_ch *ch = node;
274 unsigned long long sqnum = next_sqnum(c);
275
276 ubifs_assert(len >= UBIFS_CH_SZ);
277
278 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
279 ch->len = cpu_to_le32(len);
280 if (last)
281 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
282 else
283 ch->group_type = UBIFS_IN_NODE_GROUP;
284 ch->sqnum = cpu_to_le64(sqnum);
285 ch->padding[0] = ch->padding[1] = 0;
286 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
287 ch->crc = cpu_to_le32(crc);
288}
289
290/**
291 * wbuf_timer_callback - write-buffer timer callback function.
292 * @data: timer data (write-buffer descriptor)
293 *
294 * This function is called when the write-buffer timer expires.
295 */
296static void wbuf_timer_callback_nolock(unsigned long data)
297{
298 struct ubifs_wbuf *wbuf = (struct ubifs_wbuf *)data;
299
300 wbuf->need_sync = 1;
301 wbuf->c->need_wbuf_sync = 1;
302 ubifs_wake_up_bgt(wbuf->c);
303}
304
305/**
306 * new_wbuf_timer - start new write-buffer timer.
307 * @wbuf: write-buffer descriptor
308 */
309static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
310{
311 ubifs_assert(!timer_pending(&wbuf->timer));
312
313 if (!wbuf->timeout)
314 return;
315
316 wbuf->timer.expires = jiffies + wbuf->timeout;
317 add_timer(&wbuf->timer);
318}
319
320/**
321 * cancel_wbuf_timer - cancel write-buffer timer.
322 * @wbuf: write-buffer descriptor
323 */
324static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
325{
326 /*
327 * If the syncer is waiting for the lock (from the background thread's
328 * context) and another task is changing write-buffer then the syncing
329 * should be canceled.
330 */
331 wbuf->need_sync = 0;
332 del_timer(&wbuf->timer);
333}
334
335/**
336 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
337 * @wbuf: write-buffer to synchronize
338 *
339 * This function synchronizes write-buffer @buf and returns zero in case of
340 * success or a negative error code in case of failure.
341 */
342int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
343{
344 struct ubifs_info *c = wbuf->c;
345 int err, dirt;
346
347 cancel_wbuf_timer_nolock(wbuf);
348 if (!wbuf->used || wbuf->lnum == -1)
349 /* Write-buffer is empty or not seeked */
350 return 0;
351
352 dbg_io("LEB %d:%d, %d bytes",
353 wbuf->lnum, wbuf->offs, wbuf->used);
354 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY));
355 ubifs_assert(!(wbuf->avail & 7));
356 ubifs_assert(wbuf->offs + c->min_io_size <= c->leb_size);
357
358 if (c->ro_media)
359 return -EROFS;
360
361 ubifs_pad(c, wbuf->buf + wbuf->used, wbuf->avail);
362 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
363 c->min_io_size, wbuf->dtype);
364 if (err) {
365 ubifs_err("cannot write %d bytes to LEB %d:%d",
366 c->min_io_size, wbuf->lnum, wbuf->offs);
367 dbg_dump_stack();
368 return err;
369 }
370
371 dirt = wbuf->avail;
372
373 spin_lock(&wbuf->lock);
374 wbuf->offs += c->min_io_size;
375 wbuf->avail = c->min_io_size;
376 wbuf->used = 0;
377 wbuf->next_ino = 0;
378 spin_unlock(&wbuf->lock);
379
380 if (wbuf->sync_callback)
381 err = wbuf->sync_callback(c, wbuf->lnum,
382 c->leb_size - wbuf->offs, dirt);
383 return err;
384}
385
386/**
387 * ubifs_wbuf_seek_nolock - seek write-buffer.
388 * @wbuf: write-buffer
389 * @lnum: logical eraseblock number to seek to
390 * @offs: logical eraseblock offset to seek to
391 * @dtype: data type
392 *
393 * This function targets the write buffer to logical eraseblock @lnum:@offs.
394 * The write-buffer is synchronized if it is not empty. Returns zero in case of
395 * success and a negative error code in case of failure.
396 */
397int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs,
398 int dtype)
399{
400 const struct ubifs_info *c = wbuf->c;
401
402 dbg_io("LEB %d:%d", lnum, offs);
403 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
404 ubifs_assert(offs >= 0 && offs <= c->leb_size);
405 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
406 ubifs_assert(lnum != wbuf->lnum);
407
408 if (wbuf->used > 0) {
409 int err = ubifs_wbuf_sync_nolock(wbuf);
410
411 if (err)
412 return err;
413 }
414
415 spin_lock(&wbuf->lock);
416 wbuf->lnum = lnum;
417 wbuf->offs = offs;
418 wbuf->avail = c->min_io_size;
419 wbuf->used = 0;
420 spin_unlock(&wbuf->lock);
421 wbuf->dtype = dtype;
422
423 return 0;
424}
425
426/**
427 * ubifs_bg_wbufs_sync - synchronize write-buffers.
428 * @c: UBIFS file-system description object
429 *
430 * This function is called by background thread to synchronize write-buffers.
431 * Returns zero in case of success and a negative error code in case of
432 * failure.
433 */
434int ubifs_bg_wbufs_sync(struct ubifs_info *c)
435{
436 int err, i;
437
438 if (!c->need_wbuf_sync)
439 return 0;
440 c->need_wbuf_sync = 0;
441
442 if (c->ro_media) {
443 err = -EROFS;
444 goto out_timers;
445 }
446
447 dbg_io("synchronize");
448 for (i = 0; i < c->jhead_cnt; i++) {
449 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
450
451 cond_resched();
452
453 /*
454 * If the mutex is locked then wbuf is being changed, so
455 * synchronization is not necessary.
456 */
457 if (mutex_is_locked(&wbuf->io_mutex))
458 continue;
459
460 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
461 if (!wbuf->need_sync) {
462 mutex_unlock(&wbuf->io_mutex);
463 continue;
464 }
465
466 err = ubifs_wbuf_sync_nolock(wbuf);
467 mutex_unlock(&wbuf->io_mutex);
468 if (err) {
469 ubifs_err("cannot sync write-buffer, error %d", err);
470 ubifs_ro_mode(c, err);
471 goto out_timers;
472 }
473 }
474
475 return 0;
476
477out_timers:
478 /* Cancel all timers to prevent repeated errors */
479 for (i = 0; i < c->jhead_cnt; i++) {
480 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
481
482 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
483 cancel_wbuf_timer_nolock(wbuf);
484 mutex_unlock(&wbuf->io_mutex);
485 }
486 return err;
487}
488
489/**
490 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
491 * @wbuf: write-buffer
492 * @buf: node to write
493 * @len: node length
494 *
495 * This function writes data to flash via write-buffer @wbuf. This means that
496 * the last piece of the node won't reach the flash media immediately if it
497 * does not take whole minimal I/O unit. Instead, the node will sit in RAM
498 * until the write-buffer is synchronized (e.g., by timer).
499 *
500 * This function returns zero in case of success and a negative error code in
501 * case of failure. If the node cannot be written because there is no more
502 * space in this logical eraseblock, %-ENOSPC is returned.
503 */
504int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
505{
506 struct ubifs_info *c = wbuf->c;
507 int err, written, n, aligned_len = ALIGN(len, 8), offs;
508
509 dbg_io("%d bytes (%s) to wbuf at LEB %d:%d", len,
510 dbg_ntype(((struct ubifs_ch *)buf)->node_type), wbuf->lnum,
511 wbuf->offs + wbuf->used);
512 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
513 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
514 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
515 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= c->min_io_size);
516 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
517
518 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
519 err = -ENOSPC;
520 goto out;
521 }
522
523 cancel_wbuf_timer_nolock(wbuf);
524
525 if (c->ro_media)
526 return -EROFS;
527
528 if (aligned_len <= wbuf->avail) {
529 /*
530 * The node is not very large and fits entirely within
531 * write-buffer.
532 */
533 memcpy(wbuf->buf + wbuf->used, buf, len);
534
535 if (aligned_len == wbuf->avail) {
536 dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum,
537 wbuf->offs);
538 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf,
539 wbuf->offs, c->min_io_size,
540 wbuf->dtype);
541 if (err)
542 goto out;
543
544 spin_lock(&wbuf->lock);
545 wbuf->offs += c->min_io_size;
546 wbuf->avail = c->min_io_size;
547 wbuf->used = 0;
548 wbuf->next_ino = 0;
549 spin_unlock(&wbuf->lock);
550 } else {
551 spin_lock(&wbuf->lock);
552 wbuf->avail -= aligned_len;
553 wbuf->used += aligned_len;
554 spin_unlock(&wbuf->lock);
555 }
556
557 goto exit;
558 }
559
560 /*
561 * The node is large enough and does not fit entirely within current
562 * minimal I/O unit. We have to fill and flush write-buffer and switch
563 * to the next min. I/O unit.
564 */
565 dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum, wbuf->offs);
566 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
567 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
568 c->min_io_size, wbuf->dtype);
569 if (err)
570 goto out;
571
572 offs = wbuf->offs + c->min_io_size;
573 len -= wbuf->avail;
574 aligned_len -= wbuf->avail;
575 written = wbuf->avail;
576
577 /*
578 * The remaining data may take more whole min. I/O units, so write the
579 * remains multiple to min. I/O unit size directly to the flash media.
580 * We align node length to 8-byte boundary because we anyway flash wbuf
581 * if the remaining space is less than 8 bytes.
582 */
583 n = aligned_len >> c->min_io_shift;
584 if (n) {
585 n <<= c->min_io_shift;
586 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum, offs);
587 err = ubi_leb_write(c->ubi, wbuf->lnum, buf + written, offs, n,
588 wbuf->dtype);
589 if (err)
590 goto out;
591 offs += n;
592 aligned_len -= n;
593 len -= n;
594 written += n;
595 }
596
597 spin_lock(&wbuf->lock);
598 if (aligned_len)
599 /*
600 * And now we have what's left and what does not take whole
601 * min. I/O unit, so write it to the write-buffer and we are
602 * done.
603 */
604 memcpy(wbuf->buf, buf + written, len);
605
606 wbuf->offs = offs;
607 wbuf->used = aligned_len;
608 wbuf->avail = c->min_io_size - aligned_len;
609 wbuf->next_ino = 0;
610 spin_unlock(&wbuf->lock);
611
612exit:
613 if (wbuf->sync_callback) {
614 int free = c->leb_size - wbuf->offs - wbuf->used;
615
616 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
617 if (err)
618 goto out;
619 }
620
621 if (wbuf->used)
622 new_wbuf_timer_nolock(wbuf);
623
624 return 0;
625
626out:
627 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
628 len, wbuf->lnum, wbuf->offs, err);
629 dbg_dump_node(c, buf);
630 dbg_dump_stack();
631 dbg_dump_leb(c, wbuf->lnum);
632 return err;
633}
634
635/**
636 * ubifs_write_node - write node to the media.
637 * @c: UBIFS file-system description object
638 * @buf: the node to write
639 * @len: node length
640 * @lnum: logical eraseblock number
641 * @offs: offset within the logical eraseblock
642 * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
643 *
644 * This function automatically fills node magic number, assigns sequence
645 * number, and calculates node CRC checksum. The length of the @buf buffer has
646 * to be aligned to the minimal I/O unit size. This function automatically
647 * appends padding node and padding bytes if needed. Returns zero in case of
648 * success and a negative error code in case of failure.
649 */
650int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
651 int offs, int dtype)
652{
653 int err, buf_len = ALIGN(len, c->min_io_size);
654
655 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
656 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
657 buf_len);
658 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
659 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
660
661 if (c->ro_media)
662 return -EROFS;
663
664 ubifs_prepare_node(c, buf, len, 1);
665 err = ubi_leb_write(c->ubi, lnum, buf, offs, buf_len, dtype);
666 if (err) {
667 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
668 buf_len, lnum, offs, err);
669 dbg_dump_node(c, buf);
670 dbg_dump_stack();
671 }
672
673 return err;
674}
675
676/**
677 * ubifs_read_node_wbuf - read node from the media or write-buffer.
678 * @wbuf: wbuf to check for un-written data
679 * @buf: buffer to read to
680 * @type: node type
681 * @len: node length
682 * @lnum: logical eraseblock number
683 * @offs: offset within the logical eraseblock
684 *
685 * This function reads a node of known type and length, checks it and stores
686 * in @buf. If the node partially or fully sits in the write-buffer, this
687 * function takes data from the buffer, otherwise it reads the flash media.
688 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
689 * error code in case of failure.
690 */
691int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
692 int lnum, int offs)
693{
694 const struct ubifs_info *c = wbuf->c;
695 int err, rlen, overlap;
696 struct ubifs_ch *ch = buf;
697
698 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
699 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
700 ubifs_assert(!(offs & 7) && offs < c->leb_size);
701 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
702
703 spin_lock(&wbuf->lock);
704 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
705 if (!overlap) {
706 /* We may safely unlock the write-buffer and read the data */
707 spin_unlock(&wbuf->lock);
708 return ubifs_read_node(c, buf, type, len, lnum, offs);
709 }
710
711 /* Don't read under wbuf */
712 rlen = wbuf->offs - offs;
713 if (rlen < 0)
714 rlen = 0;
715
716 /* Copy the rest from the write-buffer */
717 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
718 spin_unlock(&wbuf->lock);
719
720 if (rlen > 0) {
721 /* Read everything that goes before write-buffer */
722 err = ubi_read(c->ubi, lnum, buf, offs, rlen);
723 if (err && err != -EBADMSG) {
724 ubifs_err("failed to read node %d from LEB %d:%d, "
725 "error %d", type, lnum, offs, err);
726 dbg_dump_stack();
727 return err;
728 }
729 }
730
731 if (type != ch->node_type) {
732 ubifs_err("bad node type (%d but expected %d)",
733 ch->node_type, type);
734 goto out;
735 }
736
Adrian Hunter2953e732008-09-04 16:26:00 +0300737 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300738 if (err) {
739 ubifs_err("expected node type %d", type);
740 return err;
741 }
742
743 rlen = le32_to_cpu(ch->len);
744 if (rlen != len) {
745 ubifs_err("bad node length %d, expected %d", rlen, len);
746 goto out;
747 }
748
749 return 0;
750
751out:
752 ubifs_err("bad node at LEB %d:%d", lnum, offs);
753 dbg_dump_node(c, buf);
754 dbg_dump_stack();
755 return -EINVAL;
756}
757
758/**
759 * ubifs_read_node - read node.
760 * @c: UBIFS file-system description object
761 * @buf: buffer to read to
762 * @type: node type
763 * @len: node length (not aligned)
764 * @lnum: logical eraseblock number
765 * @offs: offset within the logical eraseblock
766 *
767 * This function reads a node of known type and and length, checks it and
768 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
769 * and a negative error code in case of failure.
770 */
771int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
772 int lnum, int offs)
773{
774 int err, l;
775 struct ubifs_ch *ch = buf;
776
777 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
778 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
779 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
780 ubifs_assert(!(offs & 7) && offs < c->leb_size);
781 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
782
783 err = ubi_read(c->ubi, lnum, buf, offs, len);
784 if (err && err != -EBADMSG) {
785 ubifs_err("cannot read node %d from LEB %d:%d, error %d",
786 type, lnum, offs, err);
787 return err;
788 }
789
790 if (type != ch->node_type) {
791 ubifs_err("bad node type (%d but expected %d)",
792 ch->node_type, type);
793 goto out;
794 }
795
Adrian Hunter2953e732008-09-04 16:26:00 +0300796 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300797 if (err) {
798 ubifs_err("expected node type %d", type);
799 return err;
800 }
801
802 l = le32_to_cpu(ch->len);
803 if (l != len) {
804 ubifs_err("bad node length %d, expected %d", l, len);
805 goto out;
806 }
807
808 return 0;
809
810out:
811 ubifs_err("bad node at LEB %d:%d", lnum, offs);
812 dbg_dump_node(c, buf);
813 dbg_dump_stack();
814 return -EINVAL;
815}
816
817/**
818 * ubifs_wbuf_init - initialize write-buffer.
819 * @c: UBIFS file-system description object
820 * @wbuf: write-buffer to initialize
821 *
822 * This function initializes write buffer. Returns zero in case of success
823 * %-ENOMEM in case of failure.
824 */
825int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
826{
827 size_t size;
828
829 wbuf->buf = kmalloc(c->min_io_size, GFP_KERNEL);
830 if (!wbuf->buf)
831 return -ENOMEM;
832
833 size = (c->min_io_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
834 wbuf->inodes = kmalloc(size, GFP_KERNEL);
835 if (!wbuf->inodes) {
836 kfree(wbuf->buf);
837 wbuf->buf = NULL;
838 return -ENOMEM;
839 }
840
841 wbuf->used = 0;
842 wbuf->lnum = wbuf->offs = -1;
843 wbuf->avail = c->min_io_size;
844 wbuf->dtype = UBI_UNKNOWN;
845 wbuf->sync_callback = NULL;
846 mutex_init(&wbuf->io_mutex);
847 spin_lock_init(&wbuf->lock);
848
849 wbuf->c = c;
850 init_timer(&wbuf->timer);
851 wbuf->timer.function = wbuf_timer_callback_nolock;
852 wbuf->timer.data = (unsigned long)wbuf;
853 wbuf->timeout = DEFAULT_WBUF_TIMEOUT;
854 wbuf->next_ino = 0;
855
856 return 0;
857}
858
859/**
860 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
861 * @wbuf: the write-buffer whereto add
862 * @inum: the inode number
863 *
864 * This function adds an inode number to the inode array of the write-buffer.
865 */
866void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
867{
868 if (!wbuf->buf)
869 /* NOR flash or something similar */
870 return;
871
872 spin_lock(&wbuf->lock);
873 if (wbuf->used)
874 wbuf->inodes[wbuf->next_ino++] = inum;
875 spin_unlock(&wbuf->lock);
876}
877
878/**
879 * wbuf_has_ino - returns if the wbuf contains data from the inode.
880 * @wbuf: the write-buffer
881 * @inum: the inode number
882 *
883 * This function returns with %1 if the write-buffer contains some data from the
884 * given inode otherwise it returns with %0.
885 */
886static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
887{
888 int i, ret = 0;
889
890 spin_lock(&wbuf->lock);
891 for (i = 0; i < wbuf->next_ino; i++)
892 if (inum == wbuf->inodes[i]) {
893 ret = 1;
894 break;
895 }
896 spin_unlock(&wbuf->lock);
897
898 return ret;
899}
900
901/**
902 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
903 * @c: UBIFS file-system description object
904 * @inode: inode to synchronize
905 *
906 * This function synchronizes write-buffers which contain nodes belonging to
907 * @inode. Returns zero in case of success and a negative error code in case of
908 * failure.
909 */
910int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
911{
912 int i, err = 0;
913
914 for (i = 0; i < c->jhead_cnt; i++) {
915 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
916
917 if (i == GCHD)
918 /*
919 * GC head is special, do not look at it. Even if the
920 * head contains something related to this inode, it is
921 * a _copy_ of corresponding on-flash node which sits
922 * somewhere else.
923 */
924 continue;
925
926 if (!wbuf_has_ino(wbuf, inode->i_ino))
927 continue;
928
929 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
930 if (wbuf_has_ino(wbuf, inode->i_ino))
931 err = ubifs_wbuf_sync_nolock(wbuf);
932 mutex_unlock(&wbuf->io_mutex);
933
934 if (err) {
935 ubifs_ro_mode(c, err);
936 return err;
937 }
938 }
939 return 0;
940}