blob: 01682713af69e9e4f20b9648cfc4f74380da2da8 [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
32 * similarto the mechanism is used by JFFS2.
33 *
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
Adrian Hunter2953e732008-09-04 16:26:00 +030078 * @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 *
86 * This function returns zero in case of success %-EUCLEAN in case of bad CRC
87 * or magic.
88 */
89int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
Adrian Hunter2953e732008-09-04 16:26:00 +030090 int offs, int quiet, int chk_crc)
Artem Bityutskiy1e517642008-07-14 19:08:37 +030091{
92 int err = -EINVAL, type, node_len;
93 uint32_t crc, node_crc, magic;
94 const struct ubifs_ch *ch = buf;
95
96 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
97 ubifs_assert(!(offs & 7) && offs < c->leb_size);
98
99 magic = le32_to_cpu(ch->magic);
100 if (magic != UBIFS_NODE_MAGIC) {
101 if (!quiet)
102 ubifs_err("bad magic %#08x, expected %#08x",
103 magic, UBIFS_NODE_MAGIC);
104 err = -EUCLEAN;
105 goto out;
106 }
107
108 type = ch->node_type;
109 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
110 if (!quiet)
111 ubifs_err("bad node type %d", type);
112 goto out;
113 }
114
115 node_len = le32_to_cpu(ch->len);
116 if (node_len + offs > c->leb_size)
117 goto out_len;
118
119 if (c->ranges[type].max_len == 0) {
120 if (node_len != c->ranges[type].len)
121 goto out_len;
122 } else if (node_len < c->ranges[type].min_len ||
123 node_len > c->ranges[type].max_len)
124 goto out_len;
125
Adrian Hunter2953e732008-09-04 16:26:00 +0300126 if (!chk_crc && type == UBIFS_DATA_NODE && !c->always_chk_crc)
127 if (c->no_chk_data_crc)
128 return 0;
129
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300130 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
131 node_crc = le32_to_cpu(ch->crc);
132 if (crc != node_crc) {
133 if (!quiet)
134 ubifs_err("bad CRC: calculated %#08x, read %#08x",
135 crc, node_crc);
136 err = -EUCLEAN;
137 goto out;
138 }
139
140 return 0;
141
142out_len:
143 if (!quiet)
144 ubifs_err("bad node length %d", node_len);
145out:
146 if (!quiet) {
147 ubifs_err("bad node at LEB %d:%d", lnum, offs);
148 dbg_dump_node(c, buf);
149 dbg_dump_stack();
150 }
151 return err;
152}
153
154/**
155 * ubifs_pad - pad flash space.
156 * @c: UBIFS file-system description object
157 * @buf: buffer to put padding to
158 * @pad: how many bytes to pad
159 *
160 * The flash media obliges us to write only in chunks of %c->min_io_size and
161 * when we have to write less data we add padding node to the write-buffer and
162 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
163 * media is being scanned. If the amount of wasted space is not enough to fit a
164 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
165 * pattern (%UBIFS_PADDING_BYTE).
166 *
167 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
168 * used.
169 */
170void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
171{
172 uint32_t crc;
173
174 ubifs_assert(pad >= 0 && !(pad & 7));
175
176 if (pad >= UBIFS_PAD_NODE_SZ) {
177 struct ubifs_ch *ch = buf;
178 struct ubifs_pad_node *pad_node = buf;
179
180 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
181 ch->node_type = UBIFS_PAD_NODE;
182 ch->group_type = UBIFS_NO_NODE_GROUP;
183 ch->padding[0] = ch->padding[1] = 0;
184 ch->sqnum = 0;
185 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
186 pad -= UBIFS_PAD_NODE_SZ;
187 pad_node->pad_len = cpu_to_le32(pad);
188 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
189 ch->crc = cpu_to_le32(crc);
190 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
191 } else if (pad > 0)
192 /* Too little space, padding node won't fit */
193 memset(buf, UBIFS_PADDING_BYTE, pad);
194}
195
196/**
197 * next_sqnum - get next sequence number.
198 * @c: UBIFS file-system description object
199 */
200static unsigned long long next_sqnum(struct ubifs_info *c)
201{
202 unsigned long long sqnum;
203
204 spin_lock(&c->cnt_lock);
205 sqnum = ++c->max_sqnum;
206 spin_unlock(&c->cnt_lock);
207
208 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
209 if (sqnum >= SQNUM_WATERMARK) {
210 ubifs_err("sequence number overflow %llu, end of life",
211 sqnum);
212 ubifs_ro_mode(c, -EINVAL);
213 }
214 ubifs_warn("running out of sequence numbers, end of life soon");
215 }
216
217 return sqnum;
218}
219
220/**
221 * ubifs_prepare_node - prepare node to be written to flash.
222 * @c: UBIFS file-system description object
223 * @node: the node to pad
224 * @len: node length
225 * @pad: if the buffer has to be padded
226 *
227 * This function prepares node at @node to be written to the media - it
228 * calculates node CRC, fills the common header, and adds proper padding up to
229 * the next minimum I/O unit if @pad is not zero.
230 */
231void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
232{
233 uint32_t crc;
234 struct ubifs_ch *ch = node;
235 unsigned long long sqnum = next_sqnum(c);
236
237 ubifs_assert(len >= UBIFS_CH_SZ);
238
239 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
240 ch->len = cpu_to_le32(len);
241 ch->group_type = UBIFS_NO_NODE_GROUP;
242 ch->sqnum = cpu_to_le64(sqnum);
243 ch->padding[0] = ch->padding[1] = 0;
244 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
245 ch->crc = cpu_to_le32(crc);
246
247 if (pad) {
248 len = ALIGN(len, 8);
249 pad = ALIGN(len, c->min_io_size) - len;
250 ubifs_pad(c, node + len, pad);
251 }
252}
253
254/**
255 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
256 * @c: UBIFS file-system description object
257 * @node: the node to pad
258 * @len: node length
259 * @last: indicates the last node of the group
260 *
261 * This function prepares node at @node to be written to the media - it
262 * calculates node CRC and fills the common header.
263 */
264void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
265{
266 uint32_t crc;
267 struct ubifs_ch *ch = node;
268 unsigned long long sqnum = next_sqnum(c);
269
270 ubifs_assert(len >= UBIFS_CH_SZ);
271
272 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
273 ch->len = cpu_to_le32(len);
274 if (last)
275 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
276 else
277 ch->group_type = UBIFS_IN_NODE_GROUP;
278 ch->sqnum = cpu_to_le64(sqnum);
279 ch->padding[0] = ch->padding[1] = 0;
280 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
281 ch->crc = cpu_to_le32(crc);
282}
283
284/**
285 * wbuf_timer_callback - write-buffer timer callback function.
286 * @data: timer data (write-buffer descriptor)
287 *
288 * This function is called when the write-buffer timer expires.
289 */
290static void wbuf_timer_callback_nolock(unsigned long data)
291{
292 struct ubifs_wbuf *wbuf = (struct ubifs_wbuf *)data;
293
294 wbuf->need_sync = 1;
295 wbuf->c->need_wbuf_sync = 1;
296 ubifs_wake_up_bgt(wbuf->c);
297}
298
299/**
300 * new_wbuf_timer - start new write-buffer timer.
301 * @wbuf: write-buffer descriptor
302 */
303static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
304{
305 ubifs_assert(!timer_pending(&wbuf->timer));
306
307 if (!wbuf->timeout)
308 return;
309
310 wbuf->timer.expires = jiffies + wbuf->timeout;
311 add_timer(&wbuf->timer);
312}
313
314/**
315 * cancel_wbuf_timer - cancel write-buffer timer.
316 * @wbuf: write-buffer descriptor
317 */
318static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
319{
320 /*
321 * If the syncer is waiting for the lock (from the background thread's
322 * context) and another task is changing write-buffer then the syncing
323 * should be canceled.
324 */
325 wbuf->need_sync = 0;
326 del_timer(&wbuf->timer);
327}
328
329/**
330 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
331 * @wbuf: write-buffer to synchronize
332 *
333 * This function synchronizes write-buffer @buf and returns zero in case of
334 * success or a negative error code in case of failure.
335 */
336int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
337{
338 struct ubifs_info *c = wbuf->c;
339 int err, dirt;
340
341 cancel_wbuf_timer_nolock(wbuf);
342 if (!wbuf->used || wbuf->lnum == -1)
343 /* Write-buffer is empty or not seeked */
344 return 0;
345
346 dbg_io("LEB %d:%d, %d bytes",
347 wbuf->lnum, wbuf->offs, wbuf->used);
348 ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY));
349 ubifs_assert(!(wbuf->avail & 7));
350 ubifs_assert(wbuf->offs + c->min_io_size <= c->leb_size);
351
352 if (c->ro_media)
353 return -EROFS;
354
355 ubifs_pad(c, wbuf->buf + wbuf->used, wbuf->avail);
356 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
357 c->min_io_size, wbuf->dtype);
358 if (err) {
359 ubifs_err("cannot write %d bytes to LEB %d:%d",
360 c->min_io_size, wbuf->lnum, wbuf->offs);
361 dbg_dump_stack();
362 return err;
363 }
364
365 dirt = wbuf->avail;
366
367 spin_lock(&wbuf->lock);
368 wbuf->offs += c->min_io_size;
369 wbuf->avail = c->min_io_size;
370 wbuf->used = 0;
371 wbuf->next_ino = 0;
372 spin_unlock(&wbuf->lock);
373
374 if (wbuf->sync_callback)
375 err = wbuf->sync_callback(c, wbuf->lnum,
376 c->leb_size - wbuf->offs, dirt);
377 return err;
378}
379
380/**
381 * ubifs_wbuf_seek_nolock - seek write-buffer.
382 * @wbuf: write-buffer
383 * @lnum: logical eraseblock number to seek to
384 * @offs: logical eraseblock offset to seek to
385 * @dtype: data type
386 *
387 * This function targets the write buffer to logical eraseblock @lnum:@offs.
388 * The write-buffer is synchronized if it is not empty. Returns zero in case of
389 * success and a negative error code in case of failure.
390 */
391int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs,
392 int dtype)
393{
394 const struct ubifs_info *c = wbuf->c;
395
396 dbg_io("LEB %d:%d", lnum, offs);
397 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
398 ubifs_assert(offs >= 0 && offs <= c->leb_size);
399 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
400 ubifs_assert(lnum != wbuf->lnum);
401
402 if (wbuf->used > 0) {
403 int err = ubifs_wbuf_sync_nolock(wbuf);
404
405 if (err)
406 return err;
407 }
408
409 spin_lock(&wbuf->lock);
410 wbuf->lnum = lnum;
411 wbuf->offs = offs;
412 wbuf->avail = c->min_io_size;
413 wbuf->used = 0;
414 spin_unlock(&wbuf->lock);
415 wbuf->dtype = dtype;
416
417 return 0;
418}
419
420/**
421 * ubifs_bg_wbufs_sync - synchronize write-buffers.
422 * @c: UBIFS file-system description object
423 *
424 * This function is called by background thread to synchronize write-buffers.
425 * Returns zero in case of success and a negative error code in case of
426 * failure.
427 */
428int ubifs_bg_wbufs_sync(struct ubifs_info *c)
429{
430 int err, i;
431
432 if (!c->need_wbuf_sync)
433 return 0;
434 c->need_wbuf_sync = 0;
435
436 if (c->ro_media) {
437 err = -EROFS;
438 goto out_timers;
439 }
440
441 dbg_io("synchronize");
442 for (i = 0; i < c->jhead_cnt; i++) {
443 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
444
445 cond_resched();
446
447 /*
448 * If the mutex is locked then wbuf is being changed, so
449 * synchronization is not necessary.
450 */
451 if (mutex_is_locked(&wbuf->io_mutex))
452 continue;
453
454 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
455 if (!wbuf->need_sync) {
456 mutex_unlock(&wbuf->io_mutex);
457 continue;
458 }
459
460 err = ubifs_wbuf_sync_nolock(wbuf);
461 mutex_unlock(&wbuf->io_mutex);
462 if (err) {
463 ubifs_err("cannot sync write-buffer, error %d", err);
464 ubifs_ro_mode(c, err);
465 goto out_timers;
466 }
467 }
468
469 return 0;
470
471out_timers:
472 /* Cancel all timers to prevent repeated errors */
473 for (i = 0; i < c->jhead_cnt; i++) {
474 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
475
476 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
477 cancel_wbuf_timer_nolock(wbuf);
478 mutex_unlock(&wbuf->io_mutex);
479 }
480 return err;
481}
482
483/**
484 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
485 * @wbuf: write-buffer
486 * @buf: node to write
487 * @len: node length
488 *
489 * This function writes data to flash via write-buffer @wbuf. This means that
490 * the last piece of the node won't reach the flash media immediately if it
491 * does not take whole minimal I/O unit. Instead, the node will sit in RAM
492 * until the write-buffer is synchronized (e.g., by timer).
493 *
494 * This function returns zero in case of success and a negative error code in
495 * case of failure. If the node cannot be written because there is no more
496 * space in this logical eraseblock, %-ENOSPC is returned.
497 */
498int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
499{
500 struct ubifs_info *c = wbuf->c;
501 int err, written, n, aligned_len = ALIGN(len, 8), offs;
502
503 dbg_io("%d bytes (%s) to wbuf at LEB %d:%d", len,
504 dbg_ntype(((struct ubifs_ch *)buf)->node_type), wbuf->lnum,
505 wbuf->offs + wbuf->used);
506 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
507 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
508 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
509 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= c->min_io_size);
510 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
511
512 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
513 err = -ENOSPC;
514 goto out;
515 }
516
517 cancel_wbuf_timer_nolock(wbuf);
518
519 if (c->ro_media)
520 return -EROFS;
521
522 if (aligned_len <= wbuf->avail) {
523 /*
524 * The node is not very large and fits entirely within
525 * write-buffer.
526 */
527 memcpy(wbuf->buf + wbuf->used, buf, len);
528
529 if (aligned_len == wbuf->avail) {
530 dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum,
531 wbuf->offs);
532 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf,
533 wbuf->offs, c->min_io_size,
534 wbuf->dtype);
535 if (err)
536 goto out;
537
538 spin_lock(&wbuf->lock);
539 wbuf->offs += c->min_io_size;
540 wbuf->avail = c->min_io_size;
541 wbuf->used = 0;
542 wbuf->next_ino = 0;
543 spin_unlock(&wbuf->lock);
544 } else {
545 spin_lock(&wbuf->lock);
546 wbuf->avail -= aligned_len;
547 wbuf->used += aligned_len;
548 spin_unlock(&wbuf->lock);
549 }
550
551 goto exit;
552 }
553
554 /*
555 * The node is large enough and does not fit entirely within current
556 * minimal I/O unit. We have to fill and flush write-buffer and switch
557 * to the next min. I/O unit.
558 */
559 dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum, wbuf->offs);
560 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
561 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
562 c->min_io_size, wbuf->dtype);
563 if (err)
564 goto out;
565
566 offs = wbuf->offs + c->min_io_size;
567 len -= wbuf->avail;
568 aligned_len -= wbuf->avail;
569 written = wbuf->avail;
570
571 /*
572 * The remaining data may take more whole min. I/O units, so write the
573 * remains multiple to min. I/O unit size directly to the flash media.
574 * We align node length to 8-byte boundary because we anyway flash wbuf
575 * if the remaining space is less than 8 bytes.
576 */
577 n = aligned_len >> c->min_io_shift;
578 if (n) {
579 n <<= c->min_io_shift;
580 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum, offs);
581 err = ubi_leb_write(c->ubi, wbuf->lnum, buf + written, offs, n,
582 wbuf->dtype);
583 if (err)
584 goto out;
585 offs += n;
586 aligned_len -= n;
587 len -= n;
588 written += n;
589 }
590
591 spin_lock(&wbuf->lock);
592 if (aligned_len)
593 /*
594 * And now we have what's left and what does not take whole
595 * min. I/O unit, so write it to the write-buffer and we are
596 * done.
597 */
598 memcpy(wbuf->buf, buf + written, len);
599
600 wbuf->offs = offs;
601 wbuf->used = aligned_len;
602 wbuf->avail = c->min_io_size - aligned_len;
603 wbuf->next_ino = 0;
604 spin_unlock(&wbuf->lock);
605
606exit:
607 if (wbuf->sync_callback) {
608 int free = c->leb_size - wbuf->offs - wbuf->used;
609
610 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
611 if (err)
612 goto out;
613 }
614
615 if (wbuf->used)
616 new_wbuf_timer_nolock(wbuf);
617
618 return 0;
619
620out:
621 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
622 len, wbuf->lnum, wbuf->offs, err);
623 dbg_dump_node(c, buf);
624 dbg_dump_stack();
625 dbg_dump_leb(c, wbuf->lnum);
626 return err;
627}
628
629/**
630 * ubifs_write_node - write node to the media.
631 * @c: UBIFS file-system description object
632 * @buf: the node to write
633 * @len: node length
634 * @lnum: logical eraseblock number
635 * @offs: offset within the logical eraseblock
636 * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
637 *
638 * This function automatically fills node magic number, assigns sequence
639 * number, and calculates node CRC checksum. The length of the @buf buffer has
640 * to be aligned to the minimal I/O unit size. This function automatically
641 * appends padding node and padding bytes if needed. Returns zero in case of
642 * success and a negative error code in case of failure.
643 */
644int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
645 int offs, int dtype)
646{
647 int err, buf_len = ALIGN(len, c->min_io_size);
648
649 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
650 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
651 buf_len);
652 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
653 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
654
655 if (c->ro_media)
656 return -EROFS;
657
658 ubifs_prepare_node(c, buf, len, 1);
659 err = ubi_leb_write(c->ubi, lnum, buf, offs, buf_len, dtype);
660 if (err) {
661 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
662 buf_len, lnum, offs, err);
663 dbg_dump_node(c, buf);
664 dbg_dump_stack();
665 }
666
667 return err;
668}
669
670/**
671 * ubifs_read_node_wbuf - read node from the media or write-buffer.
672 * @wbuf: wbuf to check for un-written data
673 * @buf: buffer to read to
674 * @type: node type
675 * @len: node length
676 * @lnum: logical eraseblock number
677 * @offs: offset within the logical eraseblock
678 *
679 * This function reads a node of known type and length, checks it and stores
680 * in @buf. If the node partially or fully sits in the write-buffer, this
681 * function takes data from the buffer, otherwise it reads the flash media.
682 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
683 * error code in case of failure.
684 */
685int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
686 int lnum, int offs)
687{
688 const struct ubifs_info *c = wbuf->c;
689 int err, rlen, overlap;
690 struct ubifs_ch *ch = buf;
691
692 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
693 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
694 ubifs_assert(!(offs & 7) && offs < c->leb_size);
695 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
696
697 spin_lock(&wbuf->lock);
698 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
699 if (!overlap) {
700 /* We may safely unlock the write-buffer and read the data */
701 spin_unlock(&wbuf->lock);
702 return ubifs_read_node(c, buf, type, len, lnum, offs);
703 }
704
705 /* Don't read under wbuf */
706 rlen = wbuf->offs - offs;
707 if (rlen < 0)
708 rlen = 0;
709
710 /* Copy the rest from the write-buffer */
711 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
712 spin_unlock(&wbuf->lock);
713
714 if (rlen > 0) {
715 /* Read everything that goes before write-buffer */
716 err = ubi_read(c->ubi, lnum, buf, offs, rlen);
717 if (err && err != -EBADMSG) {
718 ubifs_err("failed to read node %d from LEB %d:%d, "
719 "error %d", type, lnum, offs, err);
720 dbg_dump_stack();
721 return err;
722 }
723 }
724
725 if (type != ch->node_type) {
726 ubifs_err("bad node type (%d but expected %d)",
727 ch->node_type, type);
728 goto out;
729 }
730
Adrian Hunter2953e732008-09-04 16:26:00 +0300731 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300732 if (err) {
733 ubifs_err("expected node type %d", type);
734 return err;
735 }
736
737 rlen = le32_to_cpu(ch->len);
738 if (rlen != len) {
739 ubifs_err("bad node length %d, expected %d", rlen, len);
740 goto out;
741 }
742
743 return 0;
744
745out:
746 ubifs_err("bad node at LEB %d:%d", lnum, offs);
747 dbg_dump_node(c, buf);
748 dbg_dump_stack();
749 return -EINVAL;
750}
751
752/**
753 * ubifs_read_node - read node.
754 * @c: UBIFS file-system description object
755 * @buf: buffer to read to
756 * @type: node type
757 * @len: node length (not aligned)
758 * @lnum: logical eraseblock number
759 * @offs: offset within the logical eraseblock
760 *
761 * This function reads a node of known type and and length, checks it and
762 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
763 * and a negative error code in case of failure.
764 */
765int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
766 int lnum, int offs)
767{
768 int err, l;
769 struct ubifs_ch *ch = buf;
770
771 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
772 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
773 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
774 ubifs_assert(!(offs & 7) && offs < c->leb_size);
775 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
776
777 err = ubi_read(c->ubi, lnum, buf, offs, len);
778 if (err && err != -EBADMSG) {
779 ubifs_err("cannot read node %d from LEB %d:%d, error %d",
780 type, lnum, offs, err);
781 return err;
782 }
783
784 if (type != ch->node_type) {
785 ubifs_err("bad node type (%d but expected %d)",
786 ch->node_type, type);
787 goto out;
788 }
789
Adrian Hunter2953e732008-09-04 16:26:00 +0300790 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300791 if (err) {
792 ubifs_err("expected node type %d", type);
793 return err;
794 }
795
796 l = le32_to_cpu(ch->len);
797 if (l != len) {
798 ubifs_err("bad node length %d, expected %d", l, len);
799 goto out;
800 }
801
802 return 0;
803
804out:
805 ubifs_err("bad node at LEB %d:%d", lnum, offs);
806 dbg_dump_node(c, buf);
807 dbg_dump_stack();
808 return -EINVAL;
809}
810
811/**
812 * ubifs_wbuf_init - initialize write-buffer.
813 * @c: UBIFS file-system description object
814 * @wbuf: write-buffer to initialize
815 *
816 * This function initializes write buffer. Returns zero in case of success
817 * %-ENOMEM in case of failure.
818 */
819int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
820{
821 size_t size;
822
823 wbuf->buf = kmalloc(c->min_io_size, GFP_KERNEL);
824 if (!wbuf->buf)
825 return -ENOMEM;
826
827 size = (c->min_io_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
828 wbuf->inodes = kmalloc(size, GFP_KERNEL);
829 if (!wbuf->inodes) {
830 kfree(wbuf->buf);
831 wbuf->buf = NULL;
832 return -ENOMEM;
833 }
834
835 wbuf->used = 0;
836 wbuf->lnum = wbuf->offs = -1;
837 wbuf->avail = c->min_io_size;
838 wbuf->dtype = UBI_UNKNOWN;
839 wbuf->sync_callback = NULL;
840 mutex_init(&wbuf->io_mutex);
841 spin_lock_init(&wbuf->lock);
842
843 wbuf->c = c;
844 init_timer(&wbuf->timer);
845 wbuf->timer.function = wbuf_timer_callback_nolock;
846 wbuf->timer.data = (unsigned long)wbuf;
847 wbuf->timeout = DEFAULT_WBUF_TIMEOUT;
848 wbuf->next_ino = 0;
849
850 return 0;
851}
852
853/**
854 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
855 * @wbuf: the write-buffer whereto add
856 * @inum: the inode number
857 *
858 * This function adds an inode number to the inode array of the write-buffer.
859 */
860void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
861{
862 if (!wbuf->buf)
863 /* NOR flash or something similar */
864 return;
865
866 spin_lock(&wbuf->lock);
867 if (wbuf->used)
868 wbuf->inodes[wbuf->next_ino++] = inum;
869 spin_unlock(&wbuf->lock);
870}
871
872/**
873 * wbuf_has_ino - returns if the wbuf contains data from the inode.
874 * @wbuf: the write-buffer
875 * @inum: the inode number
876 *
877 * This function returns with %1 if the write-buffer contains some data from the
878 * given inode otherwise it returns with %0.
879 */
880static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
881{
882 int i, ret = 0;
883
884 spin_lock(&wbuf->lock);
885 for (i = 0; i < wbuf->next_ino; i++)
886 if (inum == wbuf->inodes[i]) {
887 ret = 1;
888 break;
889 }
890 spin_unlock(&wbuf->lock);
891
892 return ret;
893}
894
895/**
896 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
897 * @c: UBIFS file-system description object
898 * @inode: inode to synchronize
899 *
900 * This function synchronizes write-buffers which contain nodes belonging to
901 * @inode. Returns zero in case of success and a negative error code in case of
902 * failure.
903 */
904int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
905{
906 int i, err = 0;
907
908 for (i = 0; i < c->jhead_cnt; i++) {
909 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
910
911 if (i == GCHD)
912 /*
913 * GC head is special, do not look at it. Even if the
914 * head contains something related to this inode, it is
915 * a _copy_ of corresponding on-flash node which sits
916 * somewhere else.
917 */
918 continue;
919
920 if (!wbuf_has_ino(wbuf, inode->i_ino))
921 continue;
922
923 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
924 if (wbuf_has_ino(wbuf, inode->i_ino))
925 err = ubifs_wbuf_sync_nolock(wbuf);
926 mutex_unlock(&wbuf->io_mutex);
927
928 if (err) {
929 ubifs_ro_mode(c, err);
930 return err;
931 }
932 }
933 return 0;
934}