Commit | Line | Data |
---|---|---|
1e51764a AB |
1 | /* |
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 | ||
ff46d7b3 AH |
56 | /** |
57 | * 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 | */ | |
61 | void ubifs_ro_mode(struct ubifs_info *c, int err) | |
62 | { | |
63 | if (!c->ro_media) { | |
64 | c->ro_media = 1; | |
65 | ubifs_warn("switched to read-only mode, error %d", err); | |
66 | dbg_dump_stack(); | |
67 | } | |
68 | } | |
69 | ||
1e51764a AB |
70 | /** |
71 | * ubifs_check_node - check node. | |
72 | * @c: UBIFS file-system description object | |
73 | * @buf: node to check | |
74 | * @lnum: logical eraseblock number | |
75 | * @offs: offset within the logical eraseblock | |
76 | * @quiet: print no messages | |
77 | * | |
78 | * This function checks node magic number and CRC checksum. This function also | |
79 | * validates node length to prevent UBIFS from becoming crazy when an attacker | |
80 | * feeds it a file-system image with incorrect nodes. For example, too large | |
81 | * node length in the common header could cause UBIFS to read memory outside of | |
82 | * allocated buffer when checking the CRC checksum. | |
83 | * | |
84 | * This function returns zero in case of success %-EUCLEAN in case of bad CRC | |
85 | * or magic. | |
86 | */ | |
87 | int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum, | |
88 | int offs, int quiet) | |
89 | { | |
90 | int err = -EINVAL, type, node_len; | |
91 | uint32_t crc, node_crc, magic; | |
92 | const struct ubifs_ch *ch = buf; | |
93 | ||
94 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
95 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
96 | ||
97 | magic = le32_to_cpu(ch->magic); | |
98 | if (magic != UBIFS_NODE_MAGIC) { | |
99 | if (!quiet) | |
100 | ubifs_err("bad magic %#08x, expected %#08x", | |
101 | magic, UBIFS_NODE_MAGIC); | |
102 | err = -EUCLEAN; | |
103 | goto out; | |
104 | } | |
105 | ||
106 | type = ch->node_type; | |
107 | if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) { | |
108 | if (!quiet) | |
109 | ubifs_err("bad node type %d", type); | |
110 | goto out; | |
111 | } | |
112 | ||
113 | node_len = le32_to_cpu(ch->len); | |
114 | if (node_len + offs > c->leb_size) | |
115 | goto out_len; | |
116 | ||
117 | if (c->ranges[type].max_len == 0) { | |
118 | if (node_len != c->ranges[type].len) | |
119 | goto out_len; | |
120 | } else if (node_len < c->ranges[type].min_len || | |
121 | node_len > c->ranges[type].max_len) | |
122 | goto out_len; | |
123 | ||
124 | crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8); | |
125 | node_crc = le32_to_cpu(ch->crc); | |
126 | if (crc != node_crc) { | |
127 | if (!quiet) | |
128 | ubifs_err("bad CRC: calculated %#08x, read %#08x", | |
129 | crc, node_crc); | |
130 | err = -EUCLEAN; | |
131 | goto out; | |
132 | } | |
133 | ||
134 | return 0; | |
135 | ||
136 | out_len: | |
137 | if (!quiet) | |
138 | ubifs_err("bad node length %d", node_len); | |
139 | out: | |
140 | if (!quiet) { | |
141 | ubifs_err("bad node at LEB %d:%d", lnum, offs); | |
142 | dbg_dump_node(c, buf); | |
143 | dbg_dump_stack(); | |
144 | } | |
145 | return err; | |
146 | } | |
147 | ||
148 | /** | |
149 | * ubifs_pad - pad flash space. | |
150 | * @c: UBIFS file-system description object | |
151 | * @buf: buffer to put padding to | |
152 | * @pad: how many bytes to pad | |
153 | * | |
154 | * The flash media obliges us to write only in chunks of %c->min_io_size and | |
155 | * when we have to write less data we add padding node to the write-buffer and | |
156 | * pad it to the next minimal I/O unit's boundary. Padding nodes help when the | |
157 | * media is being scanned. If the amount of wasted space is not enough to fit a | |
158 | * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes | |
159 | * pattern (%UBIFS_PADDING_BYTE). | |
160 | * | |
161 | * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is | |
162 | * used. | |
163 | */ | |
164 | void ubifs_pad(const struct ubifs_info *c, void *buf, int pad) | |
165 | { | |
166 | uint32_t crc; | |
167 | ||
168 | ubifs_assert(pad >= 0 && !(pad & 7)); | |
169 | ||
170 | if (pad >= UBIFS_PAD_NODE_SZ) { | |
171 | struct ubifs_ch *ch = buf; | |
172 | struct ubifs_pad_node *pad_node = buf; | |
173 | ||
174 | ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC); | |
175 | ch->node_type = UBIFS_PAD_NODE; | |
176 | ch->group_type = UBIFS_NO_NODE_GROUP; | |
177 | ch->padding[0] = ch->padding[1] = 0; | |
178 | ch->sqnum = 0; | |
179 | ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ); | |
180 | pad -= UBIFS_PAD_NODE_SZ; | |
181 | pad_node->pad_len = cpu_to_le32(pad); | |
182 | crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8); | |
183 | ch->crc = cpu_to_le32(crc); | |
184 | memset(buf + UBIFS_PAD_NODE_SZ, 0, pad); | |
185 | } else if (pad > 0) | |
186 | /* Too little space, padding node won't fit */ | |
187 | memset(buf, UBIFS_PADDING_BYTE, pad); | |
188 | } | |
189 | ||
190 | /** | |
191 | * next_sqnum - get next sequence number. | |
192 | * @c: UBIFS file-system description object | |
193 | */ | |
194 | static unsigned long long next_sqnum(struct ubifs_info *c) | |
195 | { | |
196 | unsigned long long sqnum; | |
197 | ||
198 | spin_lock(&c->cnt_lock); | |
199 | sqnum = ++c->max_sqnum; | |
200 | spin_unlock(&c->cnt_lock); | |
201 | ||
202 | if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) { | |
203 | if (sqnum >= SQNUM_WATERMARK) { | |
204 | ubifs_err("sequence number overflow %llu, end of life", | |
205 | sqnum); | |
206 | ubifs_ro_mode(c, -EINVAL); | |
207 | } | |
208 | ubifs_warn("running out of sequence numbers, end of life soon"); | |
209 | } | |
210 | ||
211 | return sqnum; | |
212 | } | |
213 | ||
214 | /** | |
215 | * ubifs_prepare_node - prepare node to be written to flash. | |
216 | * @c: UBIFS file-system description object | |
217 | * @node: the node to pad | |
218 | * @len: node length | |
219 | * @pad: if the buffer has to be padded | |
220 | * | |
221 | * This function prepares node at @node to be written to the media - it | |
222 | * calculates node CRC, fills the common header, and adds proper padding up to | |
223 | * the next minimum I/O unit if @pad is not zero. | |
224 | */ | |
225 | void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad) | |
226 | { | |
227 | uint32_t crc; | |
228 | struct ubifs_ch *ch = node; | |
229 | unsigned long long sqnum = next_sqnum(c); | |
230 | ||
231 | ubifs_assert(len >= UBIFS_CH_SZ); | |
232 | ||
233 | ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC); | |
234 | ch->len = cpu_to_le32(len); | |
235 | ch->group_type = UBIFS_NO_NODE_GROUP; | |
236 | ch->sqnum = cpu_to_le64(sqnum); | |
237 | ch->padding[0] = ch->padding[1] = 0; | |
238 | crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8); | |
239 | ch->crc = cpu_to_le32(crc); | |
240 | ||
241 | if (pad) { | |
242 | len = ALIGN(len, 8); | |
243 | pad = ALIGN(len, c->min_io_size) - len; | |
244 | ubifs_pad(c, node + len, pad); | |
245 | } | |
246 | } | |
247 | ||
248 | /** | |
249 | * ubifs_prep_grp_node - prepare node of a group to be written to flash. | |
250 | * @c: UBIFS file-system description object | |
251 | * @node: the node to pad | |
252 | * @len: node length | |
253 | * @last: indicates the last node of the group | |
254 | * | |
255 | * This function prepares node at @node to be written to the media - it | |
256 | * calculates node CRC and fills the common header. | |
257 | */ | |
258 | void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last) | |
259 | { | |
260 | uint32_t crc; | |
261 | struct ubifs_ch *ch = node; | |
262 | unsigned long long sqnum = next_sqnum(c); | |
263 | ||
264 | ubifs_assert(len >= UBIFS_CH_SZ); | |
265 | ||
266 | ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC); | |
267 | ch->len = cpu_to_le32(len); | |
268 | if (last) | |
269 | ch->group_type = UBIFS_LAST_OF_NODE_GROUP; | |
270 | else | |
271 | ch->group_type = UBIFS_IN_NODE_GROUP; | |
272 | ch->sqnum = cpu_to_le64(sqnum); | |
273 | ch->padding[0] = ch->padding[1] = 0; | |
274 | crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8); | |
275 | ch->crc = cpu_to_le32(crc); | |
276 | } | |
277 | ||
278 | /** | |
279 | * wbuf_timer_callback - write-buffer timer callback function. | |
280 | * @data: timer data (write-buffer descriptor) | |
281 | * | |
282 | * This function is called when the write-buffer timer expires. | |
283 | */ | |
284 | static void wbuf_timer_callback_nolock(unsigned long data) | |
285 | { | |
286 | struct ubifs_wbuf *wbuf = (struct ubifs_wbuf *)data; | |
287 | ||
288 | wbuf->need_sync = 1; | |
289 | wbuf->c->need_wbuf_sync = 1; | |
290 | ubifs_wake_up_bgt(wbuf->c); | |
291 | } | |
292 | ||
293 | /** | |
294 | * new_wbuf_timer - start new write-buffer timer. | |
295 | * @wbuf: write-buffer descriptor | |
296 | */ | |
297 | static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf) | |
298 | { | |
299 | ubifs_assert(!timer_pending(&wbuf->timer)); | |
300 | ||
301 | if (!wbuf->timeout) | |
302 | return; | |
303 | ||
304 | wbuf->timer.expires = jiffies + wbuf->timeout; | |
305 | add_timer(&wbuf->timer); | |
306 | } | |
307 | ||
308 | /** | |
309 | * cancel_wbuf_timer - cancel write-buffer timer. | |
310 | * @wbuf: write-buffer descriptor | |
311 | */ | |
312 | static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf) | |
313 | { | |
314 | /* | |
315 | * If the syncer is waiting for the lock (from the background thread's | |
316 | * context) and another task is changing write-buffer then the syncing | |
317 | * should be canceled. | |
318 | */ | |
319 | wbuf->need_sync = 0; | |
320 | del_timer(&wbuf->timer); | |
321 | } | |
322 | ||
323 | /** | |
324 | * ubifs_wbuf_sync_nolock - synchronize write-buffer. | |
325 | * @wbuf: write-buffer to synchronize | |
326 | * | |
327 | * This function synchronizes write-buffer @buf and returns zero in case of | |
328 | * success or a negative error code in case of failure. | |
329 | */ | |
330 | int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf) | |
331 | { | |
332 | struct ubifs_info *c = wbuf->c; | |
333 | int err, dirt; | |
334 | ||
335 | cancel_wbuf_timer_nolock(wbuf); | |
336 | if (!wbuf->used || wbuf->lnum == -1) | |
337 | /* Write-buffer is empty or not seeked */ | |
338 | return 0; | |
339 | ||
340 | dbg_io("LEB %d:%d, %d bytes", | |
341 | wbuf->lnum, wbuf->offs, wbuf->used); | |
342 | ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY)); | |
343 | ubifs_assert(!(wbuf->avail & 7)); | |
344 | ubifs_assert(wbuf->offs + c->min_io_size <= c->leb_size); | |
345 | ||
346 | if (c->ro_media) | |
347 | return -EROFS; | |
348 | ||
349 | ubifs_pad(c, wbuf->buf + wbuf->used, wbuf->avail); | |
350 | err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs, | |
351 | c->min_io_size, wbuf->dtype); | |
352 | if (err) { | |
353 | ubifs_err("cannot write %d bytes to LEB %d:%d", | |
354 | c->min_io_size, wbuf->lnum, wbuf->offs); | |
355 | dbg_dump_stack(); | |
356 | return err; | |
357 | } | |
358 | ||
359 | dirt = wbuf->avail; | |
360 | ||
361 | spin_lock(&wbuf->lock); | |
362 | wbuf->offs += c->min_io_size; | |
363 | wbuf->avail = c->min_io_size; | |
364 | wbuf->used = 0; | |
365 | wbuf->next_ino = 0; | |
366 | spin_unlock(&wbuf->lock); | |
367 | ||
368 | if (wbuf->sync_callback) | |
369 | err = wbuf->sync_callback(c, wbuf->lnum, | |
370 | c->leb_size - wbuf->offs, dirt); | |
371 | return err; | |
372 | } | |
373 | ||
374 | /** | |
375 | * ubifs_wbuf_seek_nolock - seek write-buffer. | |
376 | * @wbuf: write-buffer | |
377 | * @lnum: logical eraseblock number to seek to | |
378 | * @offs: logical eraseblock offset to seek to | |
379 | * @dtype: data type | |
380 | * | |
381 | * This function targets the write buffer to logical eraseblock @lnum:@offs. | |
382 | * The write-buffer is synchronized if it is not empty. Returns zero in case of | |
383 | * success and a negative error code in case of failure. | |
384 | */ | |
385 | int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs, | |
386 | int dtype) | |
387 | { | |
388 | const struct ubifs_info *c = wbuf->c; | |
389 | ||
390 | dbg_io("LEB %d:%d", lnum, offs); | |
391 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt); | |
392 | ubifs_assert(offs >= 0 && offs <= c->leb_size); | |
393 | ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7)); | |
394 | ubifs_assert(lnum != wbuf->lnum); | |
395 | ||
396 | if (wbuf->used > 0) { | |
397 | int err = ubifs_wbuf_sync_nolock(wbuf); | |
398 | ||
399 | if (err) | |
400 | return err; | |
401 | } | |
402 | ||
403 | spin_lock(&wbuf->lock); | |
404 | wbuf->lnum = lnum; | |
405 | wbuf->offs = offs; | |
406 | wbuf->avail = c->min_io_size; | |
407 | wbuf->used = 0; | |
408 | spin_unlock(&wbuf->lock); | |
409 | wbuf->dtype = dtype; | |
410 | ||
411 | return 0; | |
412 | } | |
413 | ||
414 | /** | |
415 | * ubifs_bg_wbufs_sync - synchronize write-buffers. | |
416 | * @c: UBIFS file-system description object | |
417 | * | |
418 | * This function is called by background thread to synchronize write-buffers. | |
419 | * Returns zero in case of success and a negative error code in case of | |
420 | * failure. | |
421 | */ | |
422 | int ubifs_bg_wbufs_sync(struct ubifs_info *c) | |
423 | { | |
424 | int err, i; | |
425 | ||
426 | if (!c->need_wbuf_sync) | |
427 | return 0; | |
428 | c->need_wbuf_sync = 0; | |
429 | ||
430 | if (c->ro_media) { | |
431 | err = -EROFS; | |
432 | goto out_timers; | |
433 | } | |
434 | ||
435 | dbg_io("synchronize"); | |
436 | for (i = 0; i < c->jhead_cnt; i++) { | |
437 | struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf; | |
438 | ||
439 | cond_resched(); | |
440 | ||
441 | /* | |
442 | * If the mutex is locked then wbuf is being changed, so | |
443 | * synchronization is not necessary. | |
444 | */ | |
445 | if (mutex_is_locked(&wbuf->io_mutex)) | |
446 | continue; | |
447 | ||
448 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
449 | if (!wbuf->need_sync) { | |
450 | mutex_unlock(&wbuf->io_mutex); | |
451 | continue; | |
452 | } | |
453 | ||
454 | err = ubifs_wbuf_sync_nolock(wbuf); | |
455 | mutex_unlock(&wbuf->io_mutex); | |
456 | if (err) { | |
457 | ubifs_err("cannot sync write-buffer, error %d", err); | |
458 | ubifs_ro_mode(c, err); | |
459 | goto out_timers; | |
460 | } | |
461 | } | |
462 | ||
463 | return 0; | |
464 | ||
465 | out_timers: | |
466 | /* Cancel all timers to prevent repeated errors */ | |
467 | for (i = 0; i < c->jhead_cnt; i++) { | |
468 | struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf; | |
469 | ||
470 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
471 | cancel_wbuf_timer_nolock(wbuf); | |
472 | mutex_unlock(&wbuf->io_mutex); | |
473 | } | |
474 | return err; | |
475 | } | |
476 | ||
477 | /** | |
478 | * ubifs_wbuf_write_nolock - write data to flash via write-buffer. | |
479 | * @wbuf: write-buffer | |
480 | * @buf: node to write | |
481 | * @len: node length | |
482 | * | |
483 | * This function writes data to flash via write-buffer @wbuf. This means that | |
484 | * the last piece of the node won't reach the flash media immediately if it | |
485 | * does not take whole minimal I/O unit. Instead, the node will sit in RAM | |
486 | * until the write-buffer is synchronized (e.g., by timer). | |
487 | * | |
488 | * This function returns zero in case of success and a negative error code in | |
489 | * case of failure. If the node cannot be written because there is no more | |
490 | * space in this logical eraseblock, %-ENOSPC is returned. | |
491 | */ | |
492 | int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len) | |
493 | { | |
494 | struct ubifs_info *c = wbuf->c; | |
495 | int err, written, n, aligned_len = ALIGN(len, 8), offs; | |
496 | ||
497 | dbg_io("%d bytes (%s) to wbuf at LEB %d:%d", len, | |
498 | dbg_ntype(((struct ubifs_ch *)buf)->node_type), wbuf->lnum, | |
499 | wbuf->offs + wbuf->used); | |
500 | ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt); | |
501 | ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0); | |
502 | ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size); | |
503 | ubifs_assert(wbuf->avail > 0 && wbuf->avail <= c->min_io_size); | |
504 | ubifs_assert(mutex_is_locked(&wbuf->io_mutex)); | |
505 | ||
506 | if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) { | |
507 | err = -ENOSPC; | |
508 | goto out; | |
509 | } | |
510 | ||
511 | cancel_wbuf_timer_nolock(wbuf); | |
512 | ||
513 | if (c->ro_media) | |
514 | return -EROFS; | |
515 | ||
516 | if (aligned_len <= wbuf->avail) { | |
517 | /* | |
518 | * The node is not very large and fits entirely within | |
519 | * write-buffer. | |
520 | */ | |
521 | memcpy(wbuf->buf + wbuf->used, buf, len); | |
522 | ||
523 | if (aligned_len == wbuf->avail) { | |
524 | dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum, | |
525 | wbuf->offs); | |
526 | err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, | |
527 | wbuf->offs, c->min_io_size, | |
528 | wbuf->dtype); | |
529 | if (err) | |
530 | goto out; | |
531 | ||
532 | spin_lock(&wbuf->lock); | |
533 | wbuf->offs += c->min_io_size; | |
534 | wbuf->avail = c->min_io_size; | |
535 | wbuf->used = 0; | |
536 | wbuf->next_ino = 0; | |
537 | spin_unlock(&wbuf->lock); | |
538 | } else { | |
539 | spin_lock(&wbuf->lock); | |
540 | wbuf->avail -= aligned_len; | |
541 | wbuf->used += aligned_len; | |
542 | spin_unlock(&wbuf->lock); | |
543 | } | |
544 | ||
545 | goto exit; | |
546 | } | |
547 | ||
548 | /* | |
549 | * The node is large enough and does not fit entirely within current | |
550 | * minimal I/O unit. We have to fill and flush write-buffer and switch | |
551 | * to the next min. I/O unit. | |
552 | */ | |
553 | dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum, wbuf->offs); | |
554 | memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail); | |
555 | err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs, | |
556 | c->min_io_size, wbuf->dtype); | |
557 | if (err) | |
558 | goto out; | |
559 | ||
560 | offs = wbuf->offs + c->min_io_size; | |
561 | len -= wbuf->avail; | |
562 | aligned_len -= wbuf->avail; | |
563 | written = wbuf->avail; | |
564 | ||
565 | /* | |
566 | * The remaining data may take more whole min. I/O units, so write the | |
567 | * remains multiple to min. I/O unit size directly to the flash media. | |
568 | * We align node length to 8-byte boundary because we anyway flash wbuf | |
569 | * if the remaining space is less than 8 bytes. | |
570 | */ | |
571 | n = aligned_len >> c->min_io_shift; | |
572 | if (n) { | |
573 | n <<= c->min_io_shift; | |
574 | dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum, offs); | |
575 | err = ubi_leb_write(c->ubi, wbuf->lnum, buf + written, offs, n, | |
576 | wbuf->dtype); | |
577 | if (err) | |
578 | goto out; | |
579 | offs += n; | |
580 | aligned_len -= n; | |
581 | len -= n; | |
582 | written += n; | |
583 | } | |
584 | ||
585 | spin_lock(&wbuf->lock); | |
586 | if (aligned_len) | |
587 | /* | |
588 | * And now we have what's left and what does not take whole | |
589 | * min. I/O unit, so write it to the write-buffer and we are | |
590 | * done. | |
591 | */ | |
592 | memcpy(wbuf->buf, buf + written, len); | |
593 | ||
594 | wbuf->offs = offs; | |
595 | wbuf->used = aligned_len; | |
596 | wbuf->avail = c->min_io_size - aligned_len; | |
597 | wbuf->next_ino = 0; | |
598 | spin_unlock(&wbuf->lock); | |
599 | ||
600 | exit: | |
601 | if (wbuf->sync_callback) { | |
602 | int free = c->leb_size - wbuf->offs - wbuf->used; | |
603 | ||
604 | err = wbuf->sync_callback(c, wbuf->lnum, free, 0); | |
605 | if (err) | |
606 | goto out; | |
607 | } | |
608 | ||
609 | if (wbuf->used) | |
610 | new_wbuf_timer_nolock(wbuf); | |
611 | ||
612 | return 0; | |
613 | ||
614 | out: | |
615 | ubifs_err("cannot write %d bytes to LEB %d:%d, error %d", | |
616 | len, wbuf->lnum, wbuf->offs, err); | |
617 | dbg_dump_node(c, buf); | |
618 | dbg_dump_stack(); | |
619 | dbg_dump_leb(c, wbuf->lnum); | |
620 | return err; | |
621 | } | |
622 | ||
623 | /** | |
624 | * ubifs_write_node - write node to the media. | |
625 | * @c: UBIFS file-system description object | |
626 | * @buf: the node to write | |
627 | * @len: node length | |
628 | * @lnum: logical eraseblock number | |
629 | * @offs: offset within the logical eraseblock | |
630 | * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN) | |
631 | * | |
632 | * This function automatically fills node magic number, assigns sequence | |
633 | * number, and calculates node CRC checksum. The length of the @buf buffer has | |
634 | * to be aligned to the minimal I/O unit size. This function automatically | |
635 | * appends padding node and padding bytes if needed. Returns zero in case of | |
636 | * success and a negative error code in case of failure. | |
637 | */ | |
638 | int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum, | |
639 | int offs, int dtype) | |
640 | { | |
641 | int err, buf_len = ALIGN(len, c->min_io_size); | |
642 | ||
643 | dbg_io("LEB %d:%d, %s, length %d (aligned %d)", | |
644 | lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len, | |
645 | buf_len); | |
646 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
647 | ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size); | |
648 | ||
649 | if (c->ro_media) | |
650 | return -EROFS; | |
651 | ||
652 | ubifs_prepare_node(c, buf, len, 1); | |
653 | err = ubi_leb_write(c->ubi, lnum, buf, offs, buf_len, dtype); | |
654 | if (err) { | |
655 | ubifs_err("cannot write %d bytes to LEB %d:%d, error %d", | |
656 | buf_len, lnum, offs, err); | |
657 | dbg_dump_node(c, buf); | |
658 | dbg_dump_stack(); | |
659 | } | |
660 | ||
661 | return err; | |
662 | } | |
663 | ||
664 | /** | |
665 | * ubifs_read_node_wbuf - read node from the media or write-buffer. | |
666 | * @wbuf: wbuf to check for un-written data | |
667 | * @buf: buffer to read to | |
668 | * @type: node type | |
669 | * @len: node length | |
670 | * @lnum: logical eraseblock number | |
671 | * @offs: offset within the logical eraseblock | |
672 | * | |
673 | * This function reads a node of known type and length, checks it and stores | |
674 | * in @buf. If the node partially or fully sits in the write-buffer, this | |
675 | * function takes data from the buffer, otherwise it reads the flash media. | |
676 | * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative | |
677 | * error code in case of failure. | |
678 | */ | |
679 | int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len, | |
680 | int lnum, int offs) | |
681 | { | |
682 | const struct ubifs_info *c = wbuf->c; | |
683 | int err, rlen, overlap; | |
684 | struct ubifs_ch *ch = buf; | |
685 | ||
686 | dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); | |
687 | ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
688 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
689 | ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT); | |
690 | ||
691 | spin_lock(&wbuf->lock); | |
692 | overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs); | |
693 | if (!overlap) { | |
694 | /* We may safely unlock the write-buffer and read the data */ | |
695 | spin_unlock(&wbuf->lock); | |
696 | return ubifs_read_node(c, buf, type, len, lnum, offs); | |
697 | } | |
698 | ||
699 | /* Don't read under wbuf */ | |
700 | rlen = wbuf->offs - offs; | |
701 | if (rlen < 0) | |
702 | rlen = 0; | |
703 | ||
704 | /* Copy the rest from the write-buffer */ | |
705 | memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen); | |
706 | spin_unlock(&wbuf->lock); | |
707 | ||
708 | if (rlen > 0) { | |
709 | /* Read everything that goes before write-buffer */ | |
710 | err = ubi_read(c->ubi, lnum, buf, offs, rlen); | |
711 | if (err && err != -EBADMSG) { | |
712 | ubifs_err("failed to read node %d from LEB %d:%d, " | |
713 | "error %d", type, lnum, offs, err); | |
714 | dbg_dump_stack(); | |
715 | return err; | |
716 | } | |
717 | } | |
718 | ||
719 | if (type != ch->node_type) { | |
720 | ubifs_err("bad node type (%d but expected %d)", | |
721 | ch->node_type, type); | |
722 | goto out; | |
723 | } | |
724 | ||
725 | err = ubifs_check_node(c, buf, lnum, offs, 0); | |
726 | if (err) { | |
727 | ubifs_err("expected node type %d", type); | |
728 | return err; | |
729 | } | |
730 | ||
731 | rlen = le32_to_cpu(ch->len); | |
732 | if (rlen != len) { | |
733 | ubifs_err("bad node length %d, expected %d", rlen, len); | |
734 | goto out; | |
735 | } | |
736 | ||
737 | return 0; | |
738 | ||
739 | out: | |
740 | ubifs_err("bad node at LEB %d:%d", lnum, offs); | |
741 | dbg_dump_node(c, buf); | |
742 | dbg_dump_stack(); | |
743 | return -EINVAL; | |
744 | } | |
745 | ||
746 | /** | |
747 | * ubifs_read_node - read node. | |
748 | * @c: UBIFS file-system description object | |
749 | * @buf: buffer to read to | |
750 | * @type: node type | |
751 | * @len: node length (not aligned) | |
752 | * @lnum: logical eraseblock number | |
753 | * @offs: offset within the logical eraseblock | |
754 | * | |
755 | * This function reads a node of known type and and length, checks it and | |
756 | * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched | |
757 | * and a negative error code in case of failure. | |
758 | */ | |
759 | int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len, | |
760 | int lnum, int offs) | |
761 | { | |
762 | int err, l; | |
763 | struct ubifs_ch *ch = buf; | |
764 | ||
765 | dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); | |
766 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
767 | ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size); | |
768 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
769 | ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT); | |
770 | ||
771 | err = ubi_read(c->ubi, lnum, buf, offs, len); | |
772 | if (err && err != -EBADMSG) { | |
773 | ubifs_err("cannot read node %d from LEB %d:%d, error %d", | |
774 | type, lnum, offs, err); | |
775 | return err; | |
776 | } | |
777 | ||
778 | if (type != ch->node_type) { | |
779 | ubifs_err("bad node type (%d but expected %d)", | |
780 | ch->node_type, type); | |
781 | goto out; | |
782 | } | |
783 | ||
784 | err = ubifs_check_node(c, buf, lnum, offs, 0); | |
785 | if (err) { | |
786 | ubifs_err("expected node type %d", type); | |
787 | return err; | |
788 | } | |
789 | ||
790 | l = le32_to_cpu(ch->len); | |
791 | if (l != len) { | |
792 | ubifs_err("bad node length %d, expected %d", l, len); | |
793 | goto out; | |
794 | } | |
795 | ||
796 | return 0; | |
797 | ||
798 | out: | |
799 | ubifs_err("bad node at LEB %d:%d", lnum, offs); | |
800 | dbg_dump_node(c, buf); | |
801 | dbg_dump_stack(); | |
802 | return -EINVAL; | |
803 | } | |
804 | ||
805 | /** | |
806 | * ubifs_wbuf_init - initialize write-buffer. | |
807 | * @c: UBIFS file-system description object | |
808 | * @wbuf: write-buffer to initialize | |
809 | * | |
810 | * This function initializes write buffer. Returns zero in case of success | |
811 | * %-ENOMEM in case of failure. | |
812 | */ | |
813 | int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf) | |
814 | { | |
815 | size_t size; | |
816 | ||
817 | wbuf->buf = kmalloc(c->min_io_size, GFP_KERNEL); | |
818 | if (!wbuf->buf) | |
819 | return -ENOMEM; | |
820 | ||
821 | size = (c->min_io_size / UBIFS_CH_SZ + 1) * sizeof(ino_t); | |
822 | wbuf->inodes = kmalloc(size, GFP_KERNEL); | |
823 | if (!wbuf->inodes) { | |
824 | kfree(wbuf->buf); | |
825 | wbuf->buf = NULL; | |
826 | return -ENOMEM; | |
827 | } | |
828 | ||
829 | wbuf->used = 0; | |
830 | wbuf->lnum = wbuf->offs = -1; | |
831 | wbuf->avail = c->min_io_size; | |
832 | wbuf->dtype = UBI_UNKNOWN; | |
833 | wbuf->sync_callback = NULL; | |
834 | mutex_init(&wbuf->io_mutex); | |
835 | spin_lock_init(&wbuf->lock); | |
836 | ||
837 | wbuf->c = c; | |
838 | init_timer(&wbuf->timer); | |
839 | wbuf->timer.function = wbuf_timer_callback_nolock; | |
840 | wbuf->timer.data = (unsigned long)wbuf; | |
841 | wbuf->timeout = DEFAULT_WBUF_TIMEOUT; | |
842 | wbuf->next_ino = 0; | |
843 | ||
844 | return 0; | |
845 | } | |
846 | ||
847 | /** | |
848 | * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array. | |
849 | * @wbuf: the write-buffer whereto add | |
850 | * @inum: the inode number | |
851 | * | |
852 | * This function adds an inode number to the inode array of the write-buffer. | |
853 | */ | |
854 | void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum) | |
855 | { | |
856 | if (!wbuf->buf) | |
857 | /* NOR flash or something similar */ | |
858 | return; | |
859 | ||
860 | spin_lock(&wbuf->lock); | |
861 | if (wbuf->used) | |
862 | wbuf->inodes[wbuf->next_ino++] = inum; | |
863 | spin_unlock(&wbuf->lock); | |
864 | } | |
865 | ||
866 | /** | |
867 | * wbuf_has_ino - returns if the wbuf contains data from the inode. | |
868 | * @wbuf: the write-buffer | |
869 | * @inum: the inode number | |
870 | * | |
871 | * This function returns with %1 if the write-buffer contains some data from the | |
872 | * given inode otherwise it returns with %0. | |
873 | */ | |
874 | static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum) | |
875 | { | |
876 | int i, ret = 0; | |
877 | ||
878 | spin_lock(&wbuf->lock); | |
879 | for (i = 0; i < wbuf->next_ino; i++) | |
880 | if (inum == wbuf->inodes[i]) { | |
881 | ret = 1; | |
882 | break; | |
883 | } | |
884 | spin_unlock(&wbuf->lock); | |
885 | ||
886 | return ret; | |
887 | } | |
888 | ||
889 | /** | |
890 | * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode. | |
891 | * @c: UBIFS file-system description object | |
892 | * @inode: inode to synchronize | |
893 | * | |
894 | * This function synchronizes write-buffers which contain nodes belonging to | |
895 | * @inode. Returns zero in case of success and a negative error code in case of | |
896 | * failure. | |
897 | */ | |
898 | int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode) | |
899 | { | |
900 | int i, err = 0; | |
901 | ||
902 | for (i = 0; i < c->jhead_cnt; i++) { | |
903 | struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf; | |
904 | ||
905 | if (i == GCHD) | |
906 | /* | |
907 | * GC head is special, do not look at it. Even if the | |
908 | * head contains something related to this inode, it is | |
909 | * a _copy_ of corresponding on-flash node which sits | |
910 | * somewhere else. | |
911 | */ | |
912 | continue; | |
913 | ||
914 | if (!wbuf_has_ino(wbuf, inode->i_ino)) | |
915 | continue; | |
916 | ||
917 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
918 | if (wbuf_has_ino(wbuf, inode->i_ino)) | |
919 | err = ubifs_wbuf_sync_nolock(wbuf); | |
920 | mutex_unlock(&wbuf->io_mutex); | |
921 | ||
922 | if (err) { | |
923 | ubifs_ro_mode(c, err); | |
924 | return err; | |
925 | } | |
926 | } | |
927 | return 0; | |
928 | } |