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 | |
2953e73f | 77 | * @chk_crc: indicates whether to always check the CRC |
1e51764a AB |
78 | * |
79 | * This function checks node magic number and CRC checksum. This function also | |
80 | * validates node length to prevent UBIFS from becoming crazy when an attacker | |
81 | * feeds it a file-system image with incorrect nodes. For example, too large | |
82 | * node length in the common header could cause UBIFS to read memory outside of | |
83 | * allocated buffer when checking the CRC checksum. | |
84 | * | |
85 | * This function returns zero in case of success %-EUCLEAN in case of bad CRC | |
86 | * or magic. | |
87 | */ | |
88 | int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum, | |
2953e73f | 89 | int offs, int quiet, int chk_crc) |
1e51764a AB |
90 | { |
91 | int err = -EINVAL, type, node_len; | |
92 | uint32_t crc, node_crc, magic; | |
93 | const struct ubifs_ch *ch = buf; | |
94 | ||
95 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
96 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
97 | ||
98 | magic = le32_to_cpu(ch->magic); | |
99 | if (magic != UBIFS_NODE_MAGIC) { | |
100 | if (!quiet) | |
101 | ubifs_err("bad magic %#08x, expected %#08x", | |
102 | magic, UBIFS_NODE_MAGIC); | |
103 | err = -EUCLEAN; | |
104 | goto out; | |
105 | } | |
106 | ||
107 | type = ch->node_type; | |
108 | if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) { | |
109 | if (!quiet) | |
110 | ubifs_err("bad node type %d", type); | |
111 | goto out; | |
112 | } | |
113 | ||
114 | node_len = le32_to_cpu(ch->len); | |
115 | if (node_len + offs > c->leb_size) | |
116 | goto out_len; | |
117 | ||
118 | if (c->ranges[type].max_len == 0) { | |
119 | if (node_len != c->ranges[type].len) | |
120 | goto out_len; | |
121 | } else if (node_len < c->ranges[type].min_len || | |
122 | node_len > c->ranges[type].max_len) | |
123 | goto out_len; | |
124 | ||
2953e73f AH |
125 | if (!chk_crc && type == UBIFS_DATA_NODE && !c->always_chk_crc) |
126 | if (c->no_chk_data_crc) | |
127 | return 0; | |
128 | ||
1e51764a AB |
129 | crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8); |
130 | node_crc = le32_to_cpu(ch->crc); | |
131 | if (crc != node_crc) { | |
132 | if (!quiet) | |
133 | ubifs_err("bad CRC: calculated %#08x, read %#08x", | |
134 | crc, node_crc); | |
135 | err = -EUCLEAN; | |
136 | goto out; | |
137 | } | |
138 | ||
139 | return 0; | |
140 | ||
141 | out_len: | |
142 | if (!quiet) | |
143 | ubifs_err("bad node length %d", node_len); | |
144 | out: | |
145 | if (!quiet) { | |
146 | ubifs_err("bad node at LEB %d:%d", lnum, offs); | |
147 | dbg_dump_node(c, buf); | |
148 | dbg_dump_stack(); | |
149 | } | |
150 | return err; | |
151 | } | |
152 | ||
153 | /** | |
154 | * ubifs_pad - pad flash space. | |
155 | * @c: UBIFS file-system description object | |
156 | * @buf: buffer to put padding to | |
157 | * @pad: how many bytes to pad | |
158 | * | |
159 | * The flash media obliges us to write only in chunks of %c->min_io_size and | |
160 | * when we have to write less data we add padding node to the write-buffer and | |
161 | * pad it to the next minimal I/O unit's boundary. Padding nodes help when the | |
162 | * media is being scanned. If the amount of wasted space is not enough to fit a | |
163 | * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes | |
164 | * pattern (%UBIFS_PADDING_BYTE). | |
165 | * | |
166 | * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is | |
167 | * used. | |
168 | */ | |
169 | void ubifs_pad(const struct ubifs_info *c, void *buf, int pad) | |
170 | { | |
171 | uint32_t crc; | |
172 | ||
173 | ubifs_assert(pad >= 0 && !(pad & 7)); | |
174 | ||
175 | if (pad >= UBIFS_PAD_NODE_SZ) { | |
176 | struct ubifs_ch *ch = buf; | |
177 | struct ubifs_pad_node *pad_node = buf; | |
178 | ||
179 | ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC); | |
180 | ch->node_type = UBIFS_PAD_NODE; | |
181 | ch->group_type = UBIFS_NO_NODE_GROUP; | |
182 | ch->padding[0] = ch->padding[1] = 0; | |
183 | ch->sqnum = 0; | |
184 | ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ); | |
185 | pad -= UBIFS_PAD_NODE_SZ; | |
186 | pad_node->pad_len = cpu_to_le32(pad); | |
187 | crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8); | |
188 | ch->crc = cpu_to_le32(crc); | |
189 | memset(buf + UBIFS_PAD_NODE_SZ, 0, pad); | |
190 | } else if (pad > 0) | |
191 | /* Too little space, padding node won't fit */ | |
192 | memset(buf, UBIFS_PADDING_BYTE, pad); | |
193 | } | |
194 | ||
195 | /** | |
196 | * next_sqnum - get next sequence number. | |
197 | * @c: UBIFS file-system description object | |
198 | */ | |
199 | static unsigned long long next_sqnum(struct ubifs_info *c) | |
200 | { | |
201 | unsigned long long sqnum; | |
202 | ||
203 | spin_lock(&c->cnt_lock); | |
204 | sqnum = ++c->max_sqnum; | |
205 | spin_unlock(&c->cnt_lock); | |
206 | ||
207 | if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) { | |
208 | if (sqnum >= SQNUM_WATERMARK) { | |
209 | ubifs_err("sequence number overflow %llu, end of life", | |
210 | sqnum); | |
211 | ubifs_ro_mode(c, -EINVAL); | |
212 | } | |
213 | ubifs_warn("running out of sequence numbers, end of life soon"); | |
214 | } | |
215 | ||
216 | return sqnum; | |
217 | } | |
218 | ||
219 | /** | |
220 | * ubifs_prepare_node - prepare node to be written to flash. | |
221 | * @c: UBIFS file-system description object | |
222 | * @node: the node to pad | |
223 | * @len: node length | |
224 | * @pad: if the buffer has to be padded | |
225 | * | |
226 | * This function prepares node at @node to be written to the media - it | |
227 | * calculates node CRC, fills the common header, and adds proper padding up to | |
228 | * the next minimum I/O unit if @pad is not zero. | |
229 | */ | |
230 | void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad) | |
231 | { | |
232 | uint32_t crc; | |
233 | struct ubifs_ch *ch = node; | |
234 | unsigned long long sqnum = next_sqnum(c); | |
235 | ||
236 | ubifs_assert(len >= UBIFS_CH_SZ); | |
237 | ||
238 | ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC); | |
239 | ch->len = cpu_to_le32(len); | |
240 | ch->group_type = UBIFS_NO_NODE_GROUP; | |
241 | ch->sqnum = cpu_to_le64(sqnum); | |
242 | ch->padding[0] = ch->padding[1] = 0; | |
243 | crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8); | |
244 | ch->crc = cpu_to_le32(crc); | |
245 | ||
246 | if (pad) { | |
247 | len = ALIGN(len, 8); | |
248 | pad = ALIGN(len, c->min_io_size) - len; | |
249 | ubifs_pad(c, node + len, pad); | |
250 | } | |
251 | } | |
252 | ||
253 | /** | |
254 | * ubifs_prep_grp_node - prepare node of a group to be written to flash. | |
255 | * @c: UBIFS file-system description object | |
256 | * @node: the node to pad | |
257 | * @len: node length | |
258 | * @last: indicates the last node of the group | |
259 | * | |
260 | * This function prepares node at @node to be written to the media - it | |
261 | * calculates node CRC and fills the common header. | |
262 | */ | |
263 | void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last) | |
264 | { | |
265 | uint32_t crc; | |
266 | struct ubifs_ch *ch = node; | |
267 | unsigned long long sqnum = next_sqnum(c); | |
268 | ||
269 | ubifs_assert(len >= UBIFS_CH_SZ); | |
270 | ||
271 | ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC); | |
272 | ch->len = cpu_to_le32(len); | |
273 | if (last) | |
274 | ch->group_type = UBIFS_LAST_OF_NODE_GROUP; | |
275 | else | |
276 | ch->group_type = UBIFS_IN_NODE_GROUP; | |
277 | ch->sqnum = cpu_to_le64(sqnum); | |
278 | ch->padding[0] = ch->padding[1] = 0; | |
279 | crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8); | |
280 | ch->crc = cpu_to_le32(crc); | |
281 | } | |
282 | ||
283 | /** | |
284 | * wbuf_timer_callback - write-buffer timer callback function. | |
285 | * @data: timer data (write-buffer descriptor) | |
286 | * | |
287 | * This function is called when the write-buffer timer expires. | |
288 | */ | |
289 | static void wbuf_timer_callback_nolock(unsigned long data) | |
290 | { | |
291 | struct ubifs_wbuf *wbuf = (struct ubifs_wbuf *)data; | |
292 | ||
293 | wbuf->need_sync = 1; | |
294 | wbuf->c->need_wbuf_sync = 1; | |
295 | ubifs_wake_up_bgt(wbuf->c); | |
296 | } | |
297 | ||
298 | /** | |
299 | * new_wbuf_timer - start new write-buffer timer. | |
300 | * @wbuf: write-buffer descriptor | |
301 | */ | |
302 | static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf) | |
303 | { | |
304 | ubifs_assert(!timer_pending(&wbuf->timer)); | |
305 | ||
306 | if (!wbuf->timeout) | |
307 | return; | |
308 | ||
309 | wbuf->timer.expires = jiffies + wbuf->timeout; | |
310 | add_timer(&wbuf->timer); | |
311 | } | |
312 | ||
313 | /** | |
314 | * cancel_wbuf_timer - cancel write-buffer timer. | |
315 | * @wbuf: write-buffer descriptor | |
316 | */ | |
317 | static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf) | |
318 | { | |
319 | /* | |
320 | * If the syncer is waiting for the lock (from the background thread's | |
321 | * context) and another task is changing write-buffer then the syncing | |
322 | * should be canceled. | |
323 | */ | |
324 | wbuf->need_sync = 0; | |
325 | del_timer(&wbuf->timer); | |
326 | } | |
327 | ||
328 | /** | |
329 | * ubifs_wbuf_sync_nolock - synchronize write-buffer. | |
330 | * @wbuf: write-buffer to synchronize | |
331 | * | |
332 | * This function synchronizes write-buffer @buf and returns zero in case of | |
333 | * success or a negative error code in case of failure. | |
334 | */ | |
335 | int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf) | |
336 | { | |
337 | struct ubifs_info *c = wbuf->c; | |
338 | int err, dirt; | |
339 | ||
340 | cancel_wbuf_timer_nolock(wbuf); | |
341 | if (!wbuf->used || wbuf->lnum == -1) | |
342 | /* Write-buffer is empty or not seeked */ | |
343 | return 0; | |
344 | ||
345 | dbg_io("LEB %d:%d, %d bytes", | |
346 | wbuf->lnum, wbuf->offs, wbuf->used); | |
347 | ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY)); | |
348 | ubifs_assert(!(wbuf->avail & 7)); | |
349 | ubifs_assert(wbuf->offs + c->min_io_size <= c->leb_size); | |
350 | ||
351 | if (c->ro_media) | |
352 | return -EROFS; | |
353 | ||
354 | ubifs_pad(c, wbuf->buf + wbuf->used, wbuf->avail); | |
355 | err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs, | |
356 | c->min_io_size, wbuf->dtype); | |
357 | if (err) { | |
358 | ubifs_err("cannot write %d bytes to LEB %d:%d", | |
359 | c->min_io_size, wbuf->lnum, wbuf->offs); | |
360 | dbg_dump_stack(); | |
361 | return err; | |
362 | } | |
363 | ||
364 | dirt = wbuf->avail; | |
365 | ||
366 | spin_lock(&wbuf->lock); | |
367 | wbuf->offs += c->min_io_size; | |
368 | wbuf->avail = c->min_io_size; | |
369 | wbuf->used = 0; | |
370 | wbuf->next_ino = 0; | |
371 | spin_unlock(&wbuf->lock); | |
372 | ||
373 | if (wbuf->sync_callback) | |
374 | err = wbuf->sync_callback(c, wbuf->lnum, | |
375 | c->leb_size - wbuf->offs, dirt); | |
376 | return err; | |
377 | } | |
378 | ||
379 | /** | |
380 | * ubifs_wbuf_seek_nolock - seek write-buffer. | |
381 | * @wbuf: write-buffer | |
382 | * @lnum: logical eraseblock number to seek to | |
383 | * @offs: logical eraseblock offset to seek to | |
384 | * @dtype: data type | |
385 | * | |
386 | * This function targets the write buffer to logical eraseblock @lnum:@offs. | |
387 | * The write-buffer is synchronized if it is not empty. Returns zero in case of | |
388 | * success and a negative error code in case of failure. | |
389 | */ | |
390 | int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs, | |
391 | int dtype) | |
392 | { | |
393 | const struct ubifs_info *c = wbuf->c; | |
394 | ||
395 | dbg_io("LEB %d:%d", lnum, offs); | |
396 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt); | |
397 | ubifs_assert(offs >= 0 && offs <= c->leb_size); | |
398 | ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7)); | |
399 | ubifs_assert(lnum != wbuf->lnum); | |
400 | ||
401 | if (wbuf->used > 0) { | |
402 | int err = ubifs_wbuf_sync_nolock(wbuf); | |
403 | ||
404 | if (err) | |
405 | return err; | |
406 | } | |
407 | ||
408 | spin_lock(&wbuf->lock); | |
409 | wbuf->lnum = lnum; | |
410 | wbuf->offs = offs; | |
411 | wbuf->avail = c->min_io_size; | |
412 | wbuf->used = 0; | |
413 | spin_unlock(&wbuf->lock); | |
414 | wbuf->dtype = dtype; | |
415 | ||
416 | return 0; | |
417 | } | |
418 | ||
419 | /** | |
420 | * ubifs_bg_wbufs_sync - synchronize write-buffers. | |
421 | * @c: UBIFS file-system description object | |
422 | * | |
423 | * This function is called by background thread to synchronize write-buffers. | |
424 | * Returns zero in case of success and a negative error code in case of | |
425 | * failure. | |
426 | */ | |
427 | int ubifs_bg_wbufs_sync(struct ubifs_info *c) | |
428 | { | |
429 | int err, i; | |
430 | ||
431 | if (!c->need_wbuf_sync) | |
432 | return 0; | |
433 | c->need_wbuf_sync = 0; | |
434 | ||
435 | if (c->ro_media) { | |
436 | err = -EROFS; | |
437 | goto out_timers; | |
438 | } | |
439 | ||
440 | dbg_io("synchronize"); | |
441 | for (i = 0; i < c->jhead_cnt; i++) { | |
442 | struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf; | |
443 | ||
444 | cond_resched(); | |
445 | ||
446 | /* | |
447 | * If the mutex is locked then wbuf is being changed, so | |
448 | * synchronization is not necessary. | |
449 | */ | |
450 | if (mutex_is_locked(&wbuf->io_mutex)) | |
451 | continue; | |
452 | ||
453 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
454 | if (!wbuf->need_sync) { | |
455 | mutex_unlock(&wbuf->io_mutex); | |
456 | continue; | |
457 | } | |
458 | ||
459 | err = ubifs_wbuf_sync_nolock(wbuf); | |
460 | mutex_unlock(&wbuf->io_mutex); | |
461 | if (err) { | |
462 | ubifs_err("cannot sync write-buffer, error %d", err); | |
463 | ubifs_ro_mode(c, err); | |
464 | goto out_timers; | |
465 | } | |
466 | } | |
467 | ||
468 | return 0; | |
469 | ||
470 | out_timers: | |
471 | /* Cancel all timers to prevent repeated errors */ | |
472 | for (i = 0; i < c->jhead_cnt; i++) { | |
473 | struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf; | |
474 | ||
475 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
476 | cancel_wbuf_timer_nolock(wbuf); | |
477 | mutex_unlock(&wbuf->io_mutex); | |
478 | } | |
479 | return err; | |
480 | } | |
481 | ||
482 | /** | |
483 | * ubifs_wbuf_write_nolock - write data to flash via write-buffer. | |
484 | * @wbuf: write-buffer | |
485 | * @buf: node to write | |
486 | * @len: node length | |
487 | * | |
488 | * This function writes data to flash via write-buffer @wbuf. This means that | |
489 | * the last piece of the node won't reach the flash media immediately if it | |
490 | * does not take whole minimal I/O unit. Instead, the node will sit in RAM | |
491 | * until the write-buffer is synchronized (e.g., by timer). | |
492 | * | |
493 | * This function returns zero in case of success and a negative error code in | |
494 | * case of failure. If the node cannot be written because there is no more | |
495 | * space in this logical eraseblock, %-ENOSPC is returned. | |
496 | */ | |
497 | int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len) | |
498 | { | |
499 | struct ubifs_info *c = wbuf->c; | |
500 | int err, written, n, aligned_len = ALIGN(len, 8), offs; | |
501 | ||
502 | dbg_io("%d bytes (%s) to wbuf at LEB %d:%d", len, | |
503 | dbg_ntype(((struct ubifs_ch *)buf)->node_type), wbuf->lnum, | |
504 | wbuf->offs + wbuf->used); | |
505 | ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt); | |
506 | ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0); | |
507 | ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size); | |
508 | ubifs_assert(wbuf->avail > 0 && wbuf->avail <= c->min_io_size); | |
509 | ubifs_assert(mutex_is_locked(&wbuf->io_mutex)); | |
510 | ||
511 | if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) { | |
512 | err = -ENOSPC; | |
513 | goto out; | |
514 | } | |
515 | ||
516 | cancel_wbuf_timer_nolock(wbuf); | |
517 | ||
518 | if (c->ro_media) | |
519 | return -EROFS; | |
520 | ||
521 | if (aligned_len <= wbuf->avail) { | |
522 | /* | |
523 | * The node is not very large and fits entirely within | |
524 | * write-buffer. | |
525 | */ | |
526 | memcpy(wbuf->buf + wbuf->used, buf, len); | |
527 | ||
528 | if (aligned_len == wbuf->avail) { | |
529 | dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum, | |
530 | wbuf->offs); | |
531 | err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, | |
532 | wbuf->offs, c->min_io_size, | |
533 | wbuf->dtype); | |
534 | if (err) | |
535 | goto out; | |
536 | ||
537 | spin_lock(&wbuf->lock); | |
538 | wbuf->offs += c->min_io_size; | |
539 | wbuf->avail = c->min_io_size; | |
540 | wbuf->used = 0; | |
541 | wbuf->next_ino = 0; | |
542 | spin_unlock(&wbuf->lock); | |
543 | } else { | |
544 | spin_lock(&wbuf->lock); | |
545 | wbuf->avail -= aligned_len; | |
546 | wbuf->used += aligned_len; | |
547 | spin_unlock(&wbuf->lock); | |
548 | } | |
549 | ||
550 | goto exit; | |
551 | } | |
552 | ||
553 | /* | |
554 | * The node is large enough and does not fit entirely within current | |
555 | * minimal I/O unit. We have to fill and flush write-buffer and switch | |
556 | * to the next min. I/O unit. | |
557 | */ | |
558 | dbg_io("flush wbuf to LEB %d:%d", wbuf->lnum, wbuf->offs); | |
559 | memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail); | |
560 | err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs, | |
561 | c->min_io_size, wbuf->dtype); | |
562 | if (err) | |
563 | goto out; | |
564 | ||
565 | offs = wbuf->offs + c->min_io_size; | |
566 | len -= wbuf->avail; | |
567 | aligned_len -= wbuf->avail; | |
568 | written = wbuf->avail; | |
569 | ||
570 | /* | |
571 | * The remaining data may take more whole min. I/O units, so write the | |
572 | * remains multiple to min. I/O unit size directly to the flash media. | |
573 | * We align node length to 8-byte boundary because we anyway flash wbuf | |
574 | * if the remaining space is less than 8 bytes. | |
575 | */ | |
576 | n = aligned_len >> c->min_io_shift; | |
577 | if (n) { | |
578 | n <<= c->min_io_shift; | |
579 | dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum, offs); | |
580 | err = ubi_leb_write(c->ubi, wbuf->lnum, buf + written, offs, n, | |
581 | wbuf->dtype); | |
582 | if (err) | |
583 | goto out; | |
584 | offs += n; | |
585 | aligned_len -= n; | |
586 | len -= n; | |
587 | written += n; | |
588 | } | |
589 | ||
590 | spin_lock(&wbuf->lock); | |
591 | if (aligned_len) | |
592 | /* | |
593 | * And now we have what's left and what does not take whole | |
594 | * min. I/O unit, so write it to the write-buffer and we are | |
595 | * done. | |
596 | */ | |
597 | memcpy(wbuf->buf, buf + written, len); | |
598 | ||
599 | wbuf->offs = offs; | |
600 | wbuf->used = aligned_len; | |
601 | wbuf->avail = c->min_io_size - aligned_len; | |
602 | wbuf->next_ino = 0; | |
603 | spin_unlock(&wbuf->lock); | |
604 | ||
605 | exit: | |
606 | if (wbuf->sync_callback) { | |
607 | int free = c->leb_size - wbuf->offs - wbuf->used; | |
608 | ||
609 | err = wbuf->sync_callback(c, wbuf->lnum, free, 0); | |
610 | if (err) | |
611 | goto out; | |
612 | } | |
613 | ||
614 | if (wbuf->used) | |
615 | new_wbuf_timer_nolock(wbuf); | |
616 | ||
617 | return 0; | |
618 | ||
619 | out: | |
620 | ubifs_err("cannot write %d bytes to LEB %d:%d, error %d", | |
621 | len, wbuf->lnum, wbuf->offs, err); | |
622 | dbg_dump_node(c, buf); | |
623 | dbg_dump_stack(); | |
624 | dbg_dump_leb(c, wbuf->lnum); | |
625 | return err; | |
626 | } | |
627 | ||
628 | /** | |
629 | * ubifs_write_node - write node to the media. | |
630 | * @c: UBIFS file-system description object | |
631 | * @buf: the node to write | |
632 | * @len: node length | |
633 | * @lnum: logical eraseblock number | |
634 | * @offs: offset within the logical eraseblock | |
635 | * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN) | |
636 | * | |
637 | * This function automatically fills node magic number, assigns sequence | |
638 | * number, and calculates node CRC checksum. The length of the @buf buffer has | |
639 | * to be aligned to the minimal I/O unit size. This function automatically | |
640 | * appends padding node and padding bytes if needed. Returns zero in case of | |
641 | * success and a negative error code in case of failure. | |
642 | */ | |
643 | int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum, | |
644 | int offs, int dtype) | |
645 | { | |
646 | int err, buf_len = ALIGN(len, c->min_io_size); | |
647 | ||
648 | dbg_io("LEB %d:%d, %s, length %d (aligned %d)", | |
649 | lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len, | |
650 | buf_len); | |
651 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
652 | ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size); | |
653 | ||
654 | if (c->ro_media) | |
655 | return -EROFS; | |
656 | ||
657 | ubifs_prepare_node(c, buf, len, 1); | |
658 | err = ubi_leb_write(c->ubi, lnum, buf, offs, buf_len, dtype); | |
659 | if (err) { | |
660 | ubifs_err("cannot write %d bytes to LEB %d:%d, error %d", | |
661 | buf_len, lnum, offs, err); | |
662 | dbg_dump_node(c, buf); | |
663 | dbg_dump_stack(); | |
664 | } | |
665 | ||
666 | return err; | |
667 | } | |
668 | ||
669 | /** | |
670 | * ubifs_read_node_wbuf - read node from the media or write-buffer. | |
671 | * @wbuf: wbuf to check for un-written data | |
672 | * @buf: buffer to read to | |
673 | * @type: node type | |
674 | * @len: node length | |
675 | * @lnum: logical eraseblock number | |
676 | * @offs: offset within the logical eraseblock | |
677 | * | |
678 | * This function reads a node of known type and length, checks it and stores | |
679 | * in @buf. If the node partially or fully sits in the write-buffer, this | |
680 | * function takes data from the buffer, otherwise it reads the flash media. | |
681 | * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative | |
682 | * error code in case of failure. | |
683 | */ | |
684 | int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len, | |
685 | int lnum, int offs) | |
686 | { | |
687 | const struct ubifs_info *c = wbuf->c; | |
688 | int err, rlen, overlap; | |
689 | struct ubifs_ch *ch = buf; | |
690 | ||
691 | dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); | |
692 | ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
693 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
694 | ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT); | |
695 | ||
696 | spin_lock(&wbuf->lock); | |
697 | overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs); | |
698 | if (!overlap) { | |
699 | /* We may safely unlock the write-buffer and read the data */ | |
700 | spin_unlock(&wbuf->lock); | |
701 | return ubifs_read_node(c, buf, type, len, lnum, offs); | |
702 | } | |
703 | ||
704 | /* Don't read under wbuf */ | |
705 | rlen = wbuf->offs - offs; | |
706 | if (rlen < 0) | |
707 | rlen = 0; | |
708 | ||
709 | /* Copy the rest from the write-buffer */ | |
710 | memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen); | |
711 | spin_unlock(&wbuf->lock); | |
712 | ||
713 | if (rlen > 0) { | |
714 | /* Read everything that goes before write-buffer */ | |
715 | err = ubi_read(c->ubi, lnum, buf, offs, rlen); | |
716 | if (err && err != -EBADMSG) { | |
717 | ubifs_err("failed to read node %d from LEB %d:%d, " | |
718 | "error %d", type, lnum, offs, err); | |
719 | dbg_dump_stack(); | |
720 | return err; | |
721 | } | |
722 | } | |
723 | ||
724 | if (type != ch->node_type) { | |
725 | ubifs_err("bad node type (%d but expected %d)", | |
726 | ch->node_type, type); | |
727 | goto out; | |
728 | } | |
729 | ||
2953e73f | 730 | err = ubifs_check_node(c, buf, lnum, offs, 0, 0); |
1e51764a AB |
731 | if (err) { |
732 | ubifs_err("expected node type %d", type); | |
733 | return err; | |
734 | } | |
735 | ||
736 | rlen = le32_to_cpu(ch->len); | |
737 | if (rlen != len) { | |
738 | ubifs_err("bad node length %d, expected %d", rlen, len); | |
739 | goto out; | |
740 | } | |
741 | ||
742 | return 0; | |
743 | ||
744 | out: | |
745 | ubifs_err("bad node at LEB %d:%d", lnum, offs); | |
746 | dbg_dump_node(c, buf); | |
747 | dbg_dump_stack(); | |
748 | return -EINVAL; | |
749 | } | |
750 | ||
751 | /** | |
752 | * ubifs_read_node - read node. | |
753 | * @c: UBIFS file-system description object | |
754 | * @buf: buffer to read to | |
755 | * @type: node type | |
756 | * @len: node length (not aligned) | |
757 | * @lnum: logical eraseblock number | |
758 | * @offs: offset within the logical eraseblock | |
759 | * | |
760 | * This function reads a node of known type and and length, checks it and | |
761 | * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched | |
762 | * and a negative error code in case of failure. | |
763 | */ | |
764 | int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len, | |
765 | int lnum, int offs) | |
766 | { | |
767 | int err, l; | |
768 | struct ubifs_ch *ch = buf; | |
769 | ||
770 | dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); | |
771 | ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
772 | ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size); | |
773 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
774 | ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT); | |
775 | ||
776 | err = ubi_read(c->ubi, lnum, buf, offs, len); | |
777 | if (err && err != -EBADMSG) { | |
778 | ubifs_err("cannot read node %d from LEB %d:%d, error %d", | |
779 | type, lnum, offs, err); | |
780 | return err; | |
781 | } | |
782 | ||
783 | if (type != ch->node_type) { | |
784 | ubifs_err("bad node type (%d but expected %d)", | |
785 | ch->node_type, type); | |
786 | goto out; | |
787 | } | |
788 | ||
2953e73f | 789 | err = ubifs_check_node(c, buf, lnum, offs, 0, 0); |
1e51764a AB |
790 | if (err) { |
791 | ubifs_err("expected node type %d", type); | |
792 | return err; | |
793 | } | |
794 | ||
795 | l = le32_to_cpu(ch->len); | |
796 | if (l != len) { | |
797 | ubifs_err("bad node length %d, expected %d", l, len); | |
798 | goto out; | |
799 | } | |
800 | ||
801 | return 0; | |
802 | ||
803 | out: | |
804 | ubifs_err("bad node at LEB %d:%d", lnum, offs); | |
805 | dbg_dump_node(c, buf); | |
806 | dbg_dump_stack(); | |
807 | return -EINVAL; | |
808 | } | |
809 | ||
810 | /** | |
811 | * ubifs_wbuf_init - initialize write-buffer. | |
812 | * @c: UBIFS file-system description object | |
813 | * @wbuf: write-buffer to initialize | |
814 | * | |
815 | * This function initializes write buffer. Returns zero in case of success | |
816 | * %-ENOMEM in case of failure. | |
817 | */ | |
818 | int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf) | |
819 | { | |
820 | size_t size; | |
821 | ||
822 | wbuf->buf = kmalloc(c->min_io_size, GFP_KERNEL); | |
823 | if (!wbuf->buf) | |
824 | return -ENOMEM; | |
825 | ||
826 | size = (c->min_io_size / UBIFS_CH_SZ + 1) * sizeof(ino_t); | |
827 | wbuf->inodes = kmalloc(size, GFP_KERNEL); | |
828 | if (!wbuf->inodes) { | |
829 | kfree(wbuf->buf); | |
830 | wbuf->buf = NULL; | |
831 | return -ENOMEM; | |
832 | } | |
833 | ||
834 | wbuf->used = 0; | |
835 | wbuf->lnum = wbuf->offs = -1; | |
836 | wbuf->avail = c->min_io_size; | |
837 | wbuf->dtype = UBI_UNKNOWN; | |
838 | wbuf->sync_callback = NULL; | |
839 | mutex_init(&wbuf->io_mutex); | |
840 | spin_lock_init(&wbuf->lock); | |
841 | ||
842 | wbuf->c = c; | |
843 | init_timer(&wbuf->timer); | |
844 | wbuf->timer.function = wbuf_timer_callback_nolock; | |
845 | wbuf->timer.data = (unsigned long)wbuf; | |
846 | wbuf->timeout = DEFAULT_WBUF_TIMEOUT; | |
847 | wbuf->next_ino = 0; | |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
852 | /** | |
853 | * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array. | |
854 | * @wbuf: the write-buffer whereto add | |
855 | * @inum: the inode number | |
856 | * | |
857 | * This function adds an inode number to the inode array of the write-buffer. | |
858 | */ | |
859 | void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum) | |
860 | { | |
861 | if (!wbuf->buf) | |
862 | /* NOR flash or something similar */ | |
863 | return; | |
864 | ||
865 | spin_lock(&wbuf->lock); | |
866 | if (wbuf->used) | |
867 | wbuf->inodes[wbuf->next_ino++] = inum; | |
868 | spin_unlock(&wbuf->lock); | |
869 | } | |
870 | ||
871 | /** | |
872 | * wbuf_has_ino - returns if the wbuf contains data from the inode. | |
873 | * @wbuf: the write-buffer | |
874 | * @inum: the inode number | |
875 | * | |
876 | * This function returns with %1 if the write-buffer contains some data from the | |
877 | * given inode otherwise it returns with %0. | |
878 | */ | |
879 | static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum) | |
880 | { | |
881 | int i, ret = 0; | |
882 | ||
883 | spin_lock(&wbuf->lock); | |
884 | for (i = 0; i < wbuf->next_ino; i++) | |
885 | if (inum == wbuf->inodes[i]) { | |
886 | ret = 1; | |
887 | break; | |
888 | } | |
889 | spin_unlock(&wbuf->lock); | |
890 | ||
891 | return ret; | |
892 | } | |
893 | ||
894 | /** | |
895 | * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode. | |
896 | * @c: UBIFS file-system description object | |
897 | * @inode: inode to synchronize | |
898 | * | |
899 | * This function synchronizes write-buffers which contain nodes belonging to | |
900 | * @inode. Returns zero in case of success and a negative error code in case of | |
901 | * failure. | |
902 | */ | |
903 | int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode) | |
904 | { | |
905 | int i, err = 0; | |
906 | ||
907 | for (i = 0; i < c->jhead_cnt; i++) { | |
908 | struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf; | |
909 | ||
910 | if (i == GCHD) | |
911 | /* | |
912 | * GC head is special, do not look at it. Even if the | |
913 | * head contains something related to this inode, it is | |
914 | * a _copy_ of corresponding on-flash node which sits | |
915 | * somewhere else. | |
916 | */ | |
917 | continue; | |
918 | ||
919 | if (!wbuf_has_ino(wbuf, inode->i_ino)) | |
920 | continue; | |
921 | ||
922 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
923 | if (wbuf_has_ino(wbuf, inode->i_ino)) | |
924 | err = ubifs_wbuf_sync_nolock(wbuf); | |
925 | mutex_unlock(&wbuf->io_mutex); | |
926 | ||
927 | if (err) { | |
928 | ubifs_ro_mode(c, err); | |
929 | return err; | |
930 | } | |
931 | } | |
932 | return 0; | |
933 | } |