Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This file is part of UBIFS.
4 *
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 * Copyright (C) 2006, 2007 University of Szeged, Hungary
7 *
8 * Authors: Artem Bityutskiy (Битюцкий Артём)
9 * Adrian Hunter
10 * Zoltan Sogor
11 */
12
13/*
14 * This file implements UBIFS I/O subsystem which provides various I/O-related
15 * helper functions (reading/writing/checking/validating nodes) and implements
16 * write-buffering support. Write buffers help to save space which otherwise
17 * would have been wasted for padding to the nearest minimal I/O unit boundary.
18 * Instead, data first goes to the write-buffer and is flushed when the
19 * buffer is full or when it is not used for some time (by timer). This is
20 * similar to the mechanism is used by JFFS2.
21 *
22 * UBIFS distinguishes between minimum write size (@c->min_io_size) and maximum
23 * write size (@c->max_write_size). The latter is the maximum amount of bytes
24 * the underlying flash is able to program at a time, and writing in
25 * @c->max_write_size units should presumably be faster. Obviously,
26 * @c->min_io_size <= @c->max_write_size. Write-buffers are of
27 * @c->max_write_size bytes in size for maximum performance. However, when a
28 * write-buffer is flushed, only the portion of it (aligned to @c->min_io_size
29 * boundary) which contains data is written, not the whole write-buffer,
30 * because this is more space-efficient.
31 *
32 * This optimization adds few complications to the code. Indeed, on the one
33 * hand, we want to write in optimal @c->max_write_size bytes chunks, which
34 * also means aligning writes at the @c->max_write_size bytes offsets. On the
35 * other hand, we do not want to waste space when synchronizing the write
36 * buffer, so during synchronization we writes in smaller chunks. And this makes
37 * the next write offset to be not aligned to @c->max_write_size bytes. So the
38 * have to make sure that the write-buffer offset (@wbuf->offs) becomes aligned
39 * to @c->max_write_size bytes again. We do this by temporarily shrinking
40 * write-buffer size (@wbuf->size).
41 *
42 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
43 * mutexes defined inside these objects. Since sometimes upper-level code
44 * has to lock the write-buffer (e.g. journal space reservation code), many
45 * functions related to write-buffers have "nolock" suffix which means that the
46 * caller has to lock the write-buffer before calling this function.
47 *
48 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
49 * aligned, UBIFS starts the next node from the aligned address, and the padded
50 * bytes may contain any rubbish. In other words, UBIFS does not put padding
51 * bytes in those small gaps. Common headers of nodes store real node lengths,
52 * not aligned lengths. Indexing nodes also store real lengths in branches.
53 *
54 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
55 * uses padding nodes or padding bytes, if the padding node does not fit.
56 *
57 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
58 * they are read from the flash media.
59 */
60
61#include <linux/crc32.h>
62#include <linux/slab.h>
63#include "ubifs.h"
64
65/**
66 * ubifs_ro_mode - switch UBIFS to read read-only mode.
67 * @c: UBIFS file-system description object
68 * @err: error code which is the reason of switching to R/O mode
69 */
70void ubifs_ro_mode(struct ubifs_info *c, int err)
71{
72 if (!c->ro_error) {
73 c->ro_error = 1;
74 c->no_chk_data_crc = 0;
75 c->vfs_sb->s_flags |= SB_RDONLY;
76 ubifs_warn(c, "switched to read-only mode, error %d", err);
77 dump_stack();
78 }
79}
80
81/*
82 * Below are simple wrappers over UBI I/O functions which include some
83 * additional checks and UBIFS debugging stuff. See corresponding UBI function
84 * for more information.
85 */
86
87int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
88 int len, int even_ebadmsg)
89{
90 int err;
91
92 err = ubi_read(c->ubi, lnum, buf, offs, len);
93 /*
94 * In case of %-EBADMSG print the error message only if the
95 * @even_ebadmsg is true.
96 */
97 if (err && (err != -EBADMSG || even_ebadmsg)) {
98 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
99 len, lnum, offs, err);
100 dump_stack();
101 }
102 return err;
103}
104
105int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
106 int len)
107{
108 int err;
109
110 ubifs_assert(c, !c->ro_media && !c->ro_mount);
111 if (c->ro_error)
112 return -EROFS;
113 if (!dbg_is_tst_rcvry(c))
114 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
115 else
116 err = dbg_leb_write(c, lnum, buf, offs, len);
117 if (err) {
118 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
119 len, lnum, offs, err);
120 ubifs_ro_mode(c, err);
121 dump_stack();
122 }
123 return err;
124}
125
126int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
127{
128 int err;
129
130 ubifs_assert(c, !c->ro_media && !c->ro_mount);
131 if (c->ro_error)
132 return -EROFS;
133 if (!dbg_is_tst_rcvry(c))
134 err = ubi_leb_change(c->ubi, lnum, buf, len);
135 else
136 err = dbg_leb_change(c, lnum, buf, len);
137 if (err) {
138 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
139 len, lnum, err);
140 ubifs_ro_mode(c, err);
141 dump_stack();
142 }
143 return err;
144}
145
146int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
147{
148 int err;
149
150 ubifs_assert(c, !c->ro_media && !c->ro_mount);
151 if (c->ro_error)
152 return -EROFS;
153 if (!dbg_is_tst_rcvry(c))
154 err = ubi_leb_unmap(c->ubi, lnum);
155 else
156 err = dbg_leb_unmap(c, lnum);
157 if (err) {
158 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
159 ubifs_ro_mode(c, err);
160 dump_stack();
161 }
162 return err;
163}
164
165int ubifs_leb_map(struct ubifs_info *c, int lnum)
166{
167 int err;
168
169 ubifs_assert(c, !c->ro_media && !c->ro_mount);
170 if (c->ro_error)
171 return -EROFS;
172 if (!dbg_is_tst_rcvry(c))
173 err = ubi_leb_map(c->ubi, lnum);
174 else
175 err = dbg_leb_map(c, lnum);
176 if (err) {
177 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
178 ubifs_ro_mode(c, err);
179 dump_stack();
180 }
181 return err;
182}
183
184int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
185{
186 int err;
187
188 err = ubi_is_mapped(c->ubi, lnum);
189 if (err < 0) {
190 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
191 lnum, err);
192 dump_stack();
193 }
194 return err;
195}
196
197/**
198 * ubifs_check_node - check node.
199 * @c: UBIFS file-system description object
200 * @buf: node to check
201 * @lnum: logical eraseblock number
202 * @offs: offset within the logical eraseblock
203 * @quiet: print no messages
204 * @must_chk_crc: indicates whether to always check the CRC
205 *
206 * This function checks node magic number and CRC checksum. This function also
207 * validates node length to prevent UBIFS from becoming crazy when an attacker
208 * feeds it a file-system image with incorrect nodes. For example, too large
209 * node length in the common header could cause UBIFS to read memory outside of
210 * allocated buffer when checking the CRC checksum.
211 *
212 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
213 * true, which is controlled by corresponding UBIFS mount option. However, if
214 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
215 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
216 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
217 * is checked. This is because during mounting or re-mounting from R/O mode to
218 * R/W mode we may read journal nodes (when replying the journal or doing the
219 * recovery) and the journal nodes may potentially be corrupted, so checking is
220 * required.
221 *
222 * This function returns zero in case of success and %-EUCLEAN in case of bad
223 * CRC or magic.
224 */
225int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
226 int offs, int quiet, int must_chk_crc)
227{
228 int err = -EINVAL, type, node_len, dump_node = 1;
229 uint32_t crc, node_crc, magic;
230 const struct ubifs_ch *ch = buf;
231
232 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
233 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
234
235 magic = le32_to_cpu(ch->magic);
236 if (magic != UBIFS_NODE_MAGIC) {
237 if (!quiet)
238 ubifs_err(c, "bad magic %#08x, expected %#08x",
239 magic, UBIFS_NODE_MAGIC);
240 err = -EUCLEAN;
241 goto out;
242 }
243
244 type = ch->node_type;
245 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
246 if (!quiet)
247 ubifs_err(c, "bad node type %d", type);
248 goto out;
249 }
250
251 node_len = le32_to_cpu(ch->len);
252 if (node_len + offs > c->leb_size)
253 goto out_len;
254
255 if (c->ranges[type].max_len == 0) {
256 if (node_len != c->ranges[type].len)
257 goto out_len;
258 } else if (node_len < c->ranges[type].min_len ||
259 node_len > c->ranges[type].max_len)
260 goto out_len;
261
262 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
263 !c->remounting_rw && c->no_chk_data_crc)
264 return 0;
265
266 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
267 node_crc = le32_to_cpu(ch->crc);
268 if (crc != node_crc) {
269 if (!quiet)
270 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
271 crc, node_crc);
272 err = -EUCLEAN;
273 goto out;
274 }
275
276 return 0;
277
278out_len:
279 if (!quiet)
280 ubifs_err(c, "bad node length %d", node_len);
281 if (type == UBIFS_DATA_NODE && node_len > UBIFS_DATA_NODE_SZ)
282 dump_node = 0;
283out:
284 if (!quiet) {
285 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
286 if (dump_node) {
287 ubifs_dump_node(c, buf);
288 } else {
289 int safe_len = min3(node_len, c->leb_size - offs,
290 (int)UBIFS_MAX_DATA_NODE_SZ);
291 pr_err("\tprevent out-of-bounds memory access\n");
292 pr_err("\ttruncated data node length %d\n", safe_len);
293 pr_err("\tcorrupted data node:\n");
294 print_hex_dump(KERN_ERR, "\t", DUMP_PREFIX_OFFSET, 32, 1,
295 buf, safe_len, 0);
296 }
297 dump_stack();
298 }
299 return err;
300}
301
302/**
303 * ubifs_pad - pad flash space.
304 * @c: UBIFS file-system description object
305 * @buf: buffer to put padding to
306 * @pad: how many bytes to pad
307 *
308 * The flash media obliges us to write only in chunks of %c->min_io_size and
309 * when we have to write less data we add padding node to the write-buffer and
310 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
311 * media is being scanned. If the amount of wasted space is not enough to fit a
312 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
313 * pattern (%UBIFS_PADDING_BYTE).
314 *
315 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
316 * used.
317 */
318void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
319{
320 uint32_t crc;
321
322 ubifs_assert(c, pad >= 0 && !(pad & 7));
323
324 if (pad >= UBIFS_PAD_NODE_SZ) {
325 struct ubifs_ch *ch = buf;
326 struct ubifs_pad_node *pad_node = buf;
327
328 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
329 ch->node_type = UBIFS_PAD_NODE;
330 ch->group_type = UBIFS_NO_NODE_GROUP;
331 ch->padding[0] = ch->padding[1] = 0;
332 ch->sqnum = 0;
333 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
334 pad -= UBIFS_PAD_NODE_SZ;
335 pad_node->pad_len = cpu_to_le32(pad);
336 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
337 ch->crc = cpu_to_le32(crc);
338 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
339 } else if (pad > 0)
340 /* Too little space, padding node won't fit */
341 memset(buf, UBIFS_PADDING_BYTE, pad);
342}
343
344/**
345 * next_sqnum - get next sequence number.
346 * @c: UBIFS file-system description object
347 */
348static unsigned long long next_sqnum(struct ubifs_info *c)
349{
350 unsigned long long sqnum;
351
352 spin_lock(&c->cnt_lock);
353 sqnum = ++c->max_sqnum;
354 spin_unlock(&c->cnt_lock);
355
356 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
357 if (sqnum >= SQNUM_WATERMARK) {
358 ubifs_err(c, "sequence number overflow %llu, end of life",
359 sqnum);
360 ubifs_ro_mode(c, -EINVAL);
361 }
362 ubifs_warn(c, "running out of sequence numbers, end of life soon");
363 }
364
365 return sqnum;
366}
367
368void ubifs_init_node(struct ubifs_info *c, void *node, int len, int pad)
369{
370 struct ubifs_ch *ch = node;
371 unsigned long long sqnum = next_sqnum(c);
372
373 ubifs_assert(c, len >= UBIFS_CH_SZ);
374
375 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
376 ch->len = cpu_to_le32(len);
377 ch->group_type = UBIFS_NO_NODE_GROUP;
378 ch->sqnum = cpu_to_le64(sqnum);
379 ch->padding[0] = ch->padding[1] = 0;
380
381 if (pad) {
382 len = ALIGN(len, 8);
383 pad = ALIGN(len, c->min_io_size) - len;
384 ubifs_pad(c, node + len, pad);
385 }
386}
387
388void ubifs_crc_node(struct ubifs_info *c, void *node, int len)
389{
390 struct ubifs_ch *ch = node;
391 uint32_t crc;
392
393 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
394 ch->crc = cpu_to_le32(crc);
395}
396
397/**
398 * ubifs_prepare_node_hmac - prepare node to be written to flash.
399 * @c: UBIFS file-system description object
400 * @node: the node to pad
401 * @len: node length
402 * @hmac_offs: offset of the HMAC in the node
403 * @pad: if the buffer has to be padded
404 *
405 * This function prepares node at @node to be written to the media - it
406 * calculates node CRC, fills the common header, and adds proper padding up to
407 * the next minimum I/O unit if @pad is not zero. if @hmac_offs is positive then
408 * a HMAC is inserted into the node at the given offset.
409 *
410 * This function returns 0 for success or a negative error code otherwise.
411 */
412int ubifs_prepare_node_hmac(struct ubifs_info *c, void *node, int len,
413 int hmac_offs, int pad)
414{
415 int err;
416
417 ubifs_init_node(c, node, len, pad);
418
419 if (hmac_offs > 0) {
420 err = ubifs_node_insert_hmac(c, node, len, hmac_offs);
421 if (err)
422 return err;
423 }
424
425 ubifs_crc_node(c, node, len);
426
427 return 0;
428}
429
430/**
431 * ubifs_prepare_node - prepare node to be written to flash.
432 * @c: UBIFS file-system description object
433 * @node: the node to pad
434 * @len: node length
435 * @pad: if the buffer has to be padded
436 *
437 * This function prepares node at @node to be written to the media - it
438 * calculates node CRC, fills the common header, and adds proper padding up to
439 * the next minimum I/O unit if @pad is not zero.
440 */
441void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
442{
443 /*
444 * Deliberately ignore return value since this function can only fail
445 * when a hmac offset is given.
446 */
447 ubifs_prepare_node_hmac(c, node, len, 0, pad);
448}
449
450/**
451 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
452 * @c: UBIFS file-system description object
453 * @node: the node to pad
454 * @len: node length
455 * @last: indicates the last node of the group
456 *
457 * This function prepares node at @node to be written to the media - it
458 * calculates node CRC and fills the common header.
459 */
460void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
461{
462 uint32_t crc;
463 struct ubifs_ch *ch = node;
464 unsigned long long sqnum = next_sqnum(c);
465
466 ubifs_assert(c, len >= UBIFS_CH_SZ);
467
468 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
469 ch->len = cpu_to_le32(len);
470 if (last)
471 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
472 else
473 ch->group_type = UBIFS_IN_NODE_GROUP;
474 ch->sqnum = cpu_to_le64(sqnum);
475 ch->padding[0] = ch->padding[1] = 0;
476 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
477 ch->crc = cpu_to_le32(crc);
478}
479
480/**
481 * wbuf_timer_callback - write-buffer timer callback function.
482 * @timer: timer data (write-buffer descriptor)
483 *
484 * This function is called when the write-buffer timer expires.
485 */
486static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
487{
488 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
489
490 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
491 wbuf->need_sync = 1;
492 wbuf->c->need_wbuf_sync = 1;
493 ubifs_wake_up_bgt(wbuf->c);
494 return HRTIMER_NORESTART;
495}
496
497/**
498 * new_wbuf_timer - start new write-buffer timer.
499 * @c: UBIFS file-system description object
500 * @wbuf: write-buffer descriptor
501 */
502static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
503{
504 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
505 unsigned long long delta = dirty_writeback_interval;
506
507 /* centi to milli, milli to nano, then 10% */
508 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
509
510 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
511 ubifs_assert(c, delta <= ULONG_MAX);
512
513 if (wbuf->no_timer)
514 return;
515 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
516 dbg_jhead(wbuf->jhead),
517 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
518 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
519 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
520 HRTIMER_MODE_REL);
521}
522
523/**
524 * cancel_wbuf_timer - cancel write-buffer timer.
525 * @wbuf: write-buffer descriptor
526 */
527static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
528{
529 if (wbuf->no_timer)
530 return;
531 wbuf->need_sync = 0;
532 hrtimer_cancel(&wbuf->timer);
533}
534
535/**
536 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
537 * @wbuf: write-buffer to synchronize
538 *
539 * This function synchronizes write-buffer @buf and returns zero in case of
540 * success or a negative error code in case of failure.
541 *
542 * Note, although write-buffers are of @c->max_write_size, this function does
543 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
544 * if the write-buffer is only partially filled with data, only the used part
545 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
546 * This way we waste less space.
547 */
548int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
549{
550 struct ubifs_info *c = wbuf->c;
551 int err, dirt, sync_len;
552
553 cancel_wbuf_timer_nolock(wbuf);
554 if (!wbuf->used || wbuf->lnum == -1)
555 /* Write-buffer is empty or not seeked */
556 return 0;
557
558 dbg_io("LEB %d:%d, %d bytes, jhead %s",
559 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
560 ubifs_assert(c, !(wbuf->avail & 7));
561 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
562 ubifs_assert(c, wbuf->size >= c->min_io_size);
563 ubifs_assert(c, wbuf->size <= c->max_write_size);
564 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
565 ubifs_assert(c, !c->ro_media && !c->ro_mount);
566 if (c->leb_size - wbuf->offs >= c->max_write_size)
567 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
568
569 if (c->ro_error)
570 return -EROFS;
571
572 /*
573 * Do not write whole write buffer but write only the minimum necessary
574 * amount of min. I/O units.
575 */
576 sync_len = ALIGN(wbuf->used, c->min_io_size);
577 dirt = sync_len - wbuf->used;
578 if (dirt)
579 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
580 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
581 if (err)
582 return err;
583
584 spin_lock(&wbuf->lock);
585 wbuf->offs += sync_len;
586 /*
587 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
588 * But our goal is to optimize writes and make sure we write in
589 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
590 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
591 * sure that @wbuf->offs + @wbuf->size is aligned to
592 * @c->max_write_size. This way we make sure that after next
593 * write-buffer flush we are again at the optimal offset (aligned to
594 * @c->max_write_size).
595 */
596 if (c->leb_size - wbuf->offs < c->max_write_size)
597 wbuf->size = c->leb_size - wbuf->offs;
598 else if (wbuf->offs & (c->max_write_size - 1))
599 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
600 else
601 wbuf->size = c->max_write_size;
602 wbuf->avail = wbuf->size;
603 wbuf->used = 0;
604 wbuf->next_ino = 0;
605 spin_unlock(&wbuf->lock);
606
607 if (wbuf->sync_callback)
608 err = wbuf->sync_callback(c, wbuf->lnum,
609 c->leb_size - wbuf->offs, dirt);
610 return err;
611}
612
613/**
614 * ubifs_wbuf_seek_nolock - seek write-buffer.
615 * @wbuf: write-buffer
616 * @lnum: logical eraseblock number to seek to
617 * @offs: logical eraseblock offset to seek to
618 *
619 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
620 * The write-buffer has to be empty. Returns zero in case of success and a
621 * negative error code in case of failure.
622 */
623int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
624{
625 const struct ubifs_info *c = wbuf->c;
626
627 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
628 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
629 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
630 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
631 ubifs_assert(c, lnum != wbuf->lnum);
632 ubifs_assert(c, wbuf->used == 0);
633
634 spin_lock(&wbuf->lock);
635 wbuf->lnum = lnum;
636 wbuf->offs = offs;
637 if (c->leb_size - wbuf->offs < c->max_write_size)
638 wbuf->size = c->leb_size - wbuf->offs;
639 else if (wbuf->offs & (c->max_write_size - 1))
640 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
641 else
642 wbuf->size = c->max_write_size;
643 wbuf->avail = wbuf->size;
644 wbuf->used = 0;
645 spin_unlock(&wbuf->lock);
646
647 return 0;
648}
649
650/**
651 * ubifs_bg_wbufs_sync - synchronize write-buffers.
652 * @c: UBIFS file-system description object
653 *
654 * This function is called by background thread to synchronize write-buffers.
655 * Returns zero in case of success and a negative error code in case of
656 * failure.
657 */
658int ubifs_bg_wbufs_sync(struct ubifs_info *c)
659{
660 int err, i;
661
662 ubifs_assert(c, !c->ro_media && !c->ro_mount);
663 if (!c->need_wbuf_sync)
664 return 0;
665 c->need_wbuf_sync = 0;
666
667 if (c->ro_error) {
668 err = -EROFS;
669 goto out_timers;
670 }
671
672 dbg_io("synchronize");
673 for (i = 0; i < c->jhead_cnt; i++) {
674 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
675
676 cond_resched();
677
678 /*
679 * If the mutex is locked then wbuf is being changed, so
680 * synchronization is not necessary.
681 */
682 if (mutex_is_locked(&wbuf->io_mutex))
683 continue;
684
685 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
686 if (!wbuf->need_sync) {
687 mutex_unlock(&wbuf->io_mutex);
688 continue;
689 }
690
691 err = ubifs_wbuf_sync_nolock(wbuf);
692 mutex_unlock(&wbuf->io_mutex);
693 if (err) {
694 ubifs_err(c, "cannot sync write-buffer, error %d", err);
695 ubifs_ro_mode(c, err);
696 goto out_timers;
697 }
698 }
699
700 return 0;
701
702out_timers:
703 /* Cancel all timers to prevent repeated errors */
704 for (i = 0; i < c->jhead_cnt; i++) {
705 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
706
707 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
708 cancel_wbuf_timer_nolock(wbuf);
709 mutex_unlock(&wbuf->io_mutex);
710 }
711 return err;
712}
713
714/**
715 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
716 * @wbuf: write-buffer
717 * @buf: node to write
718 * @len: node length
719 *
720 * This function writes data to flash via write-buffer @wbuf. This means that
721 * the last piece of the node won't reach the flash media immediately if it
722 * does not take whole max. write unit (@c->max_write_size). Instead, the node
723 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
724 * because more data are appended to the write-buffer).
725 *
726 * This function returns zero in case of success and a negative error code in
727 * case of failure. If the node cannot be written because there is no more
728 * space in this logical eraseblock, %-ENOSPC is returned.
729 */
730int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
731{
732 struct ubifs_info *c = wbuf->c;
733 int err, written, n, aligned_len = ALIGN(len, 8);
734
735 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
736 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
737 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
738 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
739 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
740 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
741 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
742 ubifs_assert(c, wbuf->size >= c->min_io_size);
743 ubifs_assert(c, wbuf->size <= c->max_write_size);
744 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
745 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
746 ubifs_assert(c, !c->ro_media && !c->ro_mount);
747 ubifs_assert(c, !c->space_fixup);
748 if (c->leb_size - wbuf->offs >= c->max_write_size)
749 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
750
751 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
752 err = -ENOSPC;
753 goto out;
754 }
755
756 cancel_wbuf_timer_nolock(wbuf);
757
758 if (c->ro_error)
759 return -EROFS;
760
761 if (aligned_len <= wbuf->avail) {
762 /*
763 * The node is not very large and fits entirely within
764 * write-buffer.
765 */
766 memcpy(wbuf->buf + wbuf->used, buf, len);
767
768 if (aligned_len == wbuf->avail) {
769 dbg_io("flush jhead %s wbuf to LEB %d:%d",
770 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
771 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
772 wbuf->offs, wbuf->size);
773 if (err)
774 goto out;
775
776 spin_lock(&wbuf->lock);
777 wbuf->offs += wbuf->size;
778 if (c->leb_size - wbuf->offs >= c->max_write_size)
779 wbuf->size = c->max_write_size;
780 else
781 wbuf->size = c->leb_size - wbuf->offs;
782 wbuf->avail = wbuf->size;
783 wbuf->used = 0;
784 wbuf->next_ino = 0;
785 spin_unlock(&wbuf->lock);
786 } else {
787 spin_lock(&wbuf->lock);
788 wbuf->avail -= aligned_len;
789 wbuf->used += aligned_len;
790 spin_unlock(&wbuf->lock);
791 }
792
793 goto exit;
794 }
795
796 written = 0;
797
798 if (wbuf->used) {
799 /*
800 * The node is large enough and does not fit entirely within
801 * current available space. We have to fill and flush
802 * write-buffer and switch to the next max. write unit.
803 */
804 dbg_io("flush jhead %s wbuf to LEB %d:%d",
805 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
806 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
807 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
808 wbuf->size);
809 if (err)
810 goto out;
811
812 wbuf->offs += wbuf->size;
813 len -= wbuf->avail;
814 aligned_len -= wbuf->avail;
815 written += wbuf->avail;
816 } else if (wbuf->offs & (c->max_write_size - 1)) {
817 /*
818 * The write-buffer offset is not aligned to
819 * @c->max_write_size and @wbuf->size is less than
820 * @c->max_write_size. Write @wbuf->size bytes to make sure the
821 * following writes are done in optimal @c->max_write_size
822 * chunks.
823 */
824 dbg_io("write %d bytes to LEB %d:%d",
825 wbuf->size, wbuf->lnum, wbuf->offs);
826 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
827 wbuf->size);
828 if (err)
829 goto out;
830
831 wbuf->offs += wbuf->size;
832 len -= wbuf->size;
833 aligned_len -= wbuf->size;
834 written += wbuf->size;
835 }
836
837 /*
838 * The remaining data may take more whole max. write units, so write the
839 * remains multiple to max. write unit size directly to the flash media.
840 * We align node length to 8-byte boundary because we anyway flash wbuf
841 * if the remaining space is less than 8 bytes.
842 */
843 n = aligned_len >> c->max_write_shift;
844 if (n) {
845 n <<= c->max_write_shift;
846 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
847 wbuf->offs);
848 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
849 wbuf->offs, n);
850 if (err)
851 goto out;
852 wbuf->offs += n;
853 aligned_len -= n;
854 len -= n;
855 written += n;
856 }
857
858 spin_lock(&wbuf->lock);
859 if (aligned_len)
860 /*
861 * And now we have what's left and what does not take whole
862 * max. write unit, so write it to the write-buffer and we are
863 * done.
864 */
865 memcpy(wbuf->buf, buf + written, len);
866
867 if (c->leb_size - wbuf->offs >= c->max_write_size)
868 wbuf->size = c->max_write_size;
869 else
870 wbuf->size = c->leb_size - wbuf->offs;
871 wbuf->avail = wbuf->size - aligned_len;
872 wbuf->used = aligned_len;
873 wbuf->next_ino = 0;
874 spin_unlock(&wbuf->lock);
875
876exit:
877 if (wbuf->sync_callback) {
878 int free = c->leb_size - wbuf->offs - wbuf->used;
879
880 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
881 if (err)
882 goto out;
883 }
884
885 if (wbuf->used)
886 new_wbuf_timer_nolock(c, wbuf);
887
888 return 0;
889
890out:
891 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
892 len, wbuf->lnum, wbuf->offs, err);
893 ubifs_dump_node(c, buf);
894 dump_stack();
895 ubifs_dump_leb(c, wbuf->lnum);
896 return err;
897}
898
899/**
900 * ubifs_write_node_hmac - write node to the media.
901 * @c: UBIFS file-system description object
902 * @buf: the node to write
903 * @len: node length
904 * @lnum: logical eraseblock number
905 * @offs: offset within the logical eraseblock
906 * @hmac_offs: offset of the HMAC within the node
907 *
908 * This function automatically fills node magic number, assigns sequence
909 * number, and calculates node CRC checksum. The length of the @buf buffer has
910 * to be aligned to the minimal I/O unit size. This function automatically
911 * appends padding node and padding bytes if needed. Returns zero in case of
912 * success and a negative error code in case of failure.
913 */
914int ubifs_write_node_hmac(struct ubifs_info *c, void *buf, int len, int lnum,
915 int offs, int hmac_offs)
916{
917 int err, buf_len = ALIGN(len, c->min_io_size);
918
919 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
920 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
921 buf_len);
922 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
923 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
924 ubifs_assert(c, !c->ro_media && !c->ro_mount);
925 ubifs_assert(c, !c->space_fixup);
926
927 if (c->ro_error)
928 return -EROFS;
929
930 err = ubifs_prepare_node_hmac(c, buf, len, hmac_offs, 1);
931 if (err)
932 return err;
933
934 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
935 if (err)
936 ubifs_dump_node(c, buf);
937
938 return err;
939}
940
941/**
942 * ubifs_write_node - write node to the media.
943 * @c: UBIFS file-system description object
944 * @buf: the node to write
945 * @len: node length
946 * @lnum: logical eraseblock number
947 * @offs: offset within the logical eraseblock
948 *
949 * This function automatically fills node magic number, assigns sequence
950 * number, and calculates node CRC checksum. The length of the @buf buffer has
951 * to be aligned to the minimal I/O unit size. This function automatically
952 * appends padding node and padding bytes if needed. Returns zero in case of
953 * success and a negative error code in case of failure.
954 */
955int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
956 int offs)
957{
958 return ubifs_write_node_hmac(c, buf, len, lnum, offs, -1);
959}
960
961/**
962 * ubifs_read_node_wbuf - read node from the media or write-buffer.
963 * @wbuf: wbuf to check for un-written data
964 * @buf: buffer to read to
965 * @type: node type
966 * @len: node length
967 * @lnum: logical eraseblock number
968 * @offs: offset within the logical eraseblock
969 *
970 * This function reads a node of known type and length, checks it and stores
971 * in @buf. If the node partially or fully sits in the write-buffer, this
972 * function takes data from the buffer, otherwise it reads the flash media.
973 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
974 * error code in case of failure.
975 */
976int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
977 int lnum, int offs)
978{
979 const struct ubifs_info *c = wbuf->c;
980 int err, rlen, overlap;
981 struct ubifs_ch *ch = buf;
982
983 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
984 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
985 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
986 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
987 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
988
989 spin_lock(&wbuf->lock);
990 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
991 if (!overlap) {
992 /* We may safely unlock the write-buffer and read the data */
993 spin_unlock(&wbuf->lock);
994 return ubifs_read_node(c, buf, type, len, lnum, offs);
995 }
996
997 /* Don't read under wbuf */
998 rlen = wbuf->offs - offs;
999 if (rlen < 0)
1000 rlen = 0;
1001
1002 /* Copy the rest from the write-buffer */
1003 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1004 spin_unlock(&wbuf->lock);
1005
1006 if (rlen > 0) {
1007 /* Read everything that goes before write-buffer */
1008 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1009 if (err && err != -EBADMSG)
1010 return err;
1011 }
1012
1013 if (type != ch->node_type) {
1014 ubifs_err(c, "bad node type (%d but expected %d)",
1015 ch->node_type, type);
1016 goto out;
1017 }
1018
1019 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1020 if (err) {
1021 ubifs_err(c, "expected node type %d", type);
1022 return err;
1023 }
1024
1025 rlen = le32_to_cpu(ch->len);
1026 if (rlen != len) {
1027 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
1028 goto out;
1029 }
1030
1031 return 0;
1032
1033out:
1034 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
1035 ubifs_dump_node(c, buf);
1036 dump_stack();
1037 return -EINVAL;
1038}
1039
1040/**
1041 * ubifs_read_node - read node.
1042 * @c: UBIFS file-system description object
1043 * @buf: buffer to read to
1044 * @type: node type
1045 * @len: node length (not aligned)
1046 * @lnum: logical eraseblock number
1047 * @offs: offset within the logical eraseblock
1048 *
1049 * This function reads a node of known type and and length, checks it and
1050 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1051 * and a negative error code in case of failure.
1052 */
1053int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1054 int lnum, int offs)
1055{
1056 int err, l;
1057 struct ubifs_ch *ch = buf;
1058
1059 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1060 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1061 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1062 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1063 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1064
1065 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1066 if (err && err != -EBADMSG)
1067 return err;
1068
1069 if (type != ch->node_type) {
1070 ubifs_errc(c, "bad node type (%d but expected %d)",
1071 ch->node_type, type);
1072 goto out;
1073 }
1074
1075 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1076 if (err) {
1077 ubifs_errc(c, "expected node type %d", type);
1078 return err;
1079 }
1080
1081 l = le32_to_cpu(ch->len);
1082 if (l != len) {
1083 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1084 goto out;
1085 }
1086
1087 return 0;
1088
1089out:
1090 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1091 offs, ubi_is_mapped(c->ubi, lnum));
1092 if (!c->probing) {
1093 ubifs_dump_node(c, buf);
1094 dump_stack();
1095 }
1096 return -EINVAL;
1097}
1098
1099/**
1100 * ubifs_wbuf_init - initialize write-buffer.
1101 * @c: UBIFS file-system description object
1102 * @wbuf: write-buffer to initialize
1103 *
1104 * This function initializes write-buffer. Returns zero in case of success
1105 * %-ENOMEM in case of failure.
1106 */
1107int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1108{
1109 size_t size;
1110
1111 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1112 if (!wbuf->buf)
1113 return -ENOMEM;
1114
1115 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1116 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1117 if (!wbuf->inodes) {
1118 kfree(wbuf->buf);
1119 wbuf->buf = NULL;
1120 return -ENOMEM;
1121 }
1122
1123 wbuf->used = 0;
1124 wbuf->lnum = wbuf->offs = -1;
1125 /*
1126 * If the LEB starts at the max. write size aligned address, then
1127 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1128 * set it to something smaller so that it ends at the closest max.
1129 * write size boundary.
1130 */
1131 size = c->max_write_size - (c->leb_start % c->max_write_size);
1132 wbuf->avail = wbuf->size = size;
1133 wbuf->sync_callback = NULL;
1134 mutex_init(&wbuf->io_mutex);
1135 spin_lock_init(&wbuf->lock);
1136 wbuf->c = c;
1137 wbuf->next_ino = 0;
1138
1139 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1140 wbuf->timer.function = wbuf_timer_callback_nolock;
1141 return 0;
1142}
1143
1144/**
1145 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1146 * @wbuf: the write-buffer where to add
1147 * @inum: the inode number
1148 *
1149 * This function adds an inode number to the inode array of the write-buffer.
1150 */
1151void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1152{
1153 if (!wbuf->buf)
1154 /* NOR flash or something similar */
1155 return;
1156
1157 spin_lock(&wbuf->lock);
1158 if (wbuf->used)
1159 wbuf->inodes[wbuf->next_ino++] = inum;
1160 spin_unlock(&wbuf->lock);
1161}
1162
1163/**
1164 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1165 * @wbuf: the write-buffer
1166 * @inum: the inode number
1167 *
1168 * This function returns with %1 if the write-buffer contains some data from the
1169 * given inode otherwise it returns with %0.
1170 */
1171static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1172{
1173 int i, ret = 0;
1174
1175 spin_lock(&wbuf->lock);
1176 for (i = 0; i < wbuf->next_ino; i++)
1177 if (inum == wbuf->inodes[i]) {
1178 ret = 1;
1179 break;
1180 }
1181 spin_unlock(&wbuf->lock);
1182
1183 return ret;
1184}
1185
1186/**
1187 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1188 * @c: UBIFS file-system description object
1189 * @inode: inode to synchronize
1190 *
1191 * This function synchronizes write-buffers which contain nodes belonging to
1192 * @inode. Returns zero in case of success and a negative error code in case of
1193 * failure.
1194 */
1195int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1196{
1197 int i, err = 0;
1198
1199 for (i = 0; i < c->jhead_cnt; i++) {
1200 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1201
1202 if (i == GCHD)
1203 /*
1204 * GC head is special, do not look at it. Even if the
1205 * head contains something related to this inode, it is
1206 * a _copy_ of corresponding on-flash node which sits
1207 * somewhere else.
1208 */
1209 continue;
1210
1211 if (!wbuf_has_ino(wbuf, inode->i_ino))
1212 continue;
1213
1214 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1215 if (wbuf_has_ino(wbuf, inode->i_ino))
1216 err = ubifs_wbuf_sync_nolock(wbuf);
1217 mutex_unlock(&wbuf->io_mutex);
1218
1219 if (err) {
1220 ubifs_ro_mode(c, err);
1221 return err;
1222 }
1223 }
1224 return 0;
1225}
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 * similar to the mechanism is used by JFFS2.
33 *
34 * UBIFS distinguishes between minimum write size (@c->min_io_size) and maximum
35 * write size (@c->max_write_size). The latter is the maximum amount of bytes
36 * the underlying flash is able to program at a time, and writing in
37 * @c->max_write_size units should presumably be faster. Obviously,
38 * @c->min_io_size <= @c->max_write_size. Write-buffers are of
39 * @c->max_write_size bytes in size for maximum performance. However, when a
40 * write-buffer is flushed, only the portion of it (aligned to @c->min_io_size
41 * boundary) which contains data is written, not the whole write-buffer,
42 * because this is more space-efficient.
43 *
44 * This optimization adds few complications to the code. Indeed, on the one
45 * hand, we want to write in optimal @c->max_write_size bytes chunks, which
46 * also means aligning writes at the @c->max_write_size bytes offsets. On the
47 * other hand, we do not want to waste space when synchronizing the write
48 * buffer, so during synchronization we writes in smaller chunks. And this makes
49 * the next write offset to be not aligned to @c->max_write_size bytes. So the
50 * have to make sure that the write-buffer offset (@wbuf->offs) becomes aligned
51 * to @c->max_write_size bytes again. We do this by temporarily shrinking
52 * write-buffer size (@wbuf->size).
53 *
54 * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
55 * mutexes defined inside these objects. Since sometimes upper-level code
56 * has to lock the write-buffer (e.g. journal space reservation code), many
57 * functions related to write-buffers have "nolock" suffix which means that the
58 * caller has to lock the write-buffer before calling this function.
59 *
60 * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
61 * aligned, UBIFS starts the next node from the aligned address, and the padded
62 * bytes may contain any rubbish. In other words, UBIFS does not put padding
63 * bytes in those small gaps. Common headers of nodes store real node lengths,
64 * not aligned lengths. Indexing nodes also store real lengths in branches.
65 *
66 * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
67 * uses padding nodes or padding bytes, if the padding node does not fit.
68 *
69 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
70 * they are read from the flash media.
71 */
72
73#include <linux/crc32.h>
74#include <linux/slab.h>
75#include "ubifs.h"
76
77/**
78 * ubifs_ro_mode - switch UBIFS to read read-only mode.
79 * @c: UBIFS file-system description object
80 * @err: error code which is the reason of switching to R/O mode
81 */
82void ubifs_ro_mode(struct ubifs_info *c, int err)
83{
84 if (!c->ro_error) {
85 c->ro_error = 1;
86 c->no_chk_data_crc = 0;
87 c->vfs_sb->s_flags |= SB_RDONLY;
88 ubifs_warn(c, "switched to read-only mode, error %d", err);
89 dump_stack();
90 }
91}
92
93/*
94 * Below are simple wrappers over UBI I/O functions which include some
95 * additional checks and UBIFS debugging stuff. See corresponding UBI function
96 * for more information.
97 */
98
99int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
100 int len, int even_ebadmsg)
101{
102 int err;
103
104 err = ubi_read(c->ubi, lnum, buf, offs, len);
105 /*
106 * In case of %-EBADMSG print the error message only if the
107 * @even_ebadmsg is true.
108 */
109 if (err && (err != -EBADMSG || even_ebadmsg)) {
110 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
111 len, lnum, offs, err);
112 dump_stack();
113 }
114 return err;
115}
116
117int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
118 int len)
119{
120 int err;
121
122 ubifs_assert(!c->ro_media && !c->ro_mount);
123 if (c->ro_error)
124 return -EROFS;
125 if (!dbg_is_tst_rcvry(c))
126 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
127 else
128 err = dbg_leb_write(c, lnum, buf, offs, len);
129 if (err) {
130 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
131 len, lnum, offs, err);
132 ubifs_ro_mode(c, err);
133 dump_stack();
134 }
135 return err;
136}
137
138int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
139{
140 int err;
141
142 ubifs_assert(!c->ro_media && !c->ro_mount);
143 if (c->ro_error)
144 return -EROFS;
145 if (!dbg_is_tst_rcvry(c))
146 err = ubi_leb_change(c->ubi, lnum, buf, len);
147 else
148 err = dbg_leb_change(c, lnum, buf, len);
149 if (err) {
150 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
151 len, lnum, err);
152 ubifs_ro_mode(c, err);
153 dump_stack();
154 }
155 return err;
156}
157
158int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
159{
160 int err;
161
162 ubifs_assert(!c->ro_media && !c->ro_mount);
163 if (c->ro_error)
164 return -EROFS;
165 if (!dbg_is_tst_rcvry(c))
166 err = ubi_leb_unmap(c->ubi, lnum);
167 else
168 err = dbg_leb_unmap(c, lnum);
169 if (err) {
170 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
171 ubifs_ro_mode(c, err);
172 dump_stack();
173 }
174 return err;
175}
176
177int ubifs_leb_map(struct ubifs_info *c, int lnum)
178{
179 int err;
180
181 ubifs_assert(!c->ro_media && !c->ro_mount);
182 if (c->ro_error)
183 return -EROFS;
184 if (!dbg_is_tst_rcvry(c))
185 err = ubi_leb_map(c->ubi, lnum);
186 else
187 err = dbg_leb_map(c, lnum);
188 if (err) {
189 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
190 ubifs_ro_mode(c, err);
191 dump_stack();
192 }
193 return err;
194}
195
196int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
197{
198 int err;
199
200 err = ubi_is_mapped(c->ubi, lnum);
201 if (err < 0) {
202 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
203 lnum, err);
204 dump_stack();
205 }
206 return err;
207}
208
209/**
210 * ubifs_check_node - check node.
211 * @c: UBIFS file-system description object
212 * @buf: node to check
213 * @lnum: logical eraseblock number
214 * @offs: offset within the logical eraseblock
215 * @quiet: print no messages
216 * @must_chk_crc: indicates whether to always check the CRC
217 *
218 * This function checks node magic number and CRC checksum. This function also
219 * validates node length to prevent UBIFS from becoming crazy when an attacker
220 * feeds it a file-system image with incorrect nodes. For example, too large
221 * node length in the common header could cause UBIFS to read memory outside of
222 * allocated buffer when checking the CRC checksum.
223 *
224 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
225 * true, which is controlled by corresponding UBIFS mount option. However, if
226 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
227 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
228 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
229 * is checked. This is because during mounting or re-mounting from R/O mode to
230 * R/W mode we may read journal nodes (when replying the journal or doing the
231 * recovery) and the journal nodes may potentially be corrupted, so checking is
232 * required.
233 *
234 * This function returns zero in case of success and %-EUCLEAN in case of bad
235 * CRC or magic.
236 */
237int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
238 int offs, int quiet, int must_chk_crc)
239{
240 int err = -EINVAL, type, node_len;
241 uint32_t crc, node_crc, magic;
242 const struct ubifs_ch *ch = buf;
243
244 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
245 ubifs_assert(!(offs & 7) && offs < c->leb_size);
246
247 magic = le32_to_cpu(ch->magic);
248 if (magic != UBIFS_NODE_MAGIC) {
249 if (!quiet)
250 ubifs_err(c, "bad magic %#08x, expected %#08x",
251 magic, UBIFS_NODE_MAGIC);
252 err = -EUCLEAN;
253 goto out;
254 }
255
256 type = ch->node_type;
257 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
258 if (!quiet)
259 ubifs_err(c, "bad node type %d", type);
260 goto out;
261 }
262
263 node_len = le32_to_cpu(ch->len);
264 if (node_len + offs > c->leb_size)
265 goto out_len;
266
267 if (c->ranges[type].max_len == 0) {
268 if (node_len != c->ranges[type].len)
269 goto out_len;
270 } else if (node_len < c->ranges[type].min_len ||
271 node_len > c->ranges[type].max_len)
272 goto out_len;
273
274 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
275 !c->remounting_rw && c->no_chk_data_crc)
276 return 0;
277
278 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
279 node_crc = le32_to_cpu(ch->crc);
280 if (crc != node_crc) {
281 if (!quiet)
282 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
283 crc, node_crc);
284 err = -EUCLEAN;
285 goto out;
286 }
287
288 return 0;
289
290out_len:
291 if (!quiet)
292 ubifs_err(c, "bad node length %d", node_len);
293out:
294 if (!quiet) {
295 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
296 ubifs_dump_node(c, buf);
297 dump_stack();
298 }
299 return err;
300}
301
302/**
303 * ubifs_pad - pad flash space.
304 * @c: UBIFS file-system description object
305 * @buf: buffer to put padding to
306 * @pad: how many bytes to pad
307 *
308 * The flash media obliges us to write only in chunks of %c->min_io_size and
309 * when we have to write less data we add padding node to the write-buffer and
310 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
311 * media is being scanned. If the amount of wasted space is not enough to fit a
312 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
313 * pattern (%UBIFS_PADDING_BYTE).
314 *
315 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
316 * used.
317 */
318void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
319{
320 uint32_t crc;
321
322 ubifs_assert(pad >= 0 && !(pad & 7));
323
324 if (pad >= UBIFS_PAD_NODE_SZ) {
325 struct ubifs_ch *ch = buf;
326 struct ubifs_pad_node *pad_node = buf;
327
328 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
329 ch->node_type = UBIFS_PAD_NODE;
330 ch->group_type = UBIFS_NO_NODE_GROUP;
331 ch->padding[0] = ch->padding[1] = 0;
332 ch->sqnum = 0;
333 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
334 pad -= UBIFS_PAD_NODE_SZ;
335 pad_node->pad_len = cpu_to_le32(pad);
336 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
337 ch->crc = cpu_to_le32(crc);
338 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
339 } else if (pad > 0)
340 /* Too little space, padding node won't fit */
341 memset(buf, UBIFS_PADDING_BYTE, pad);
342}
343
344/**
345 * next_sqnum - get next sequence number.
346 * @c: UBIFS file-system description object
347 */
348static unsigned long long next_sqnum(struct ubifs_info *c)
349{
350 unsigned long long sqnum;
351
352 spin_lock(&c->cnt_lock);
353 sqnum = ++c->max_sqnum;
354 spin_unlock(&c->cnt_lock);
355
356 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
357 if (sqnum >= SQNUM_WATERMARK) {
358 ubifs_err(c, "sequence number overflow %llu, end of life",
359 sqnum);
360 ubifs_ro_mode(c, -EINVAL);
361 }
362 ubifs_warn(c, "running out of sequence numbers, end of life soon");
363 }
364
365 return sqnum;
366}
367
368/**
369 * ubifs_prepare_node - prepare node to be written to flash.
370 * @c: UBIFS file-system description object
371 * @node: the node to pad
372 * @len: node length
373 * @pad: if the buffer has to be padded
374 *
375 * This function prepares node at @node to be written to the media - it
376 * calculates node CRC, fills the common header, and adds proper padding up to
377 * the next minimum I/O unit if @pad is not zero.
378 */
379void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
380{
381 uint32_t crc;
382 struct ubifs_ch *ch = node;
383 unsigned long long sqnum = next_sqnum(c);
384
385 ubifs_assert(len >= UBIFS_CH_SZ);
386
387 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
388 ch->len = cpu_to_le32(len);
389 ch->group_type = UBIFS_NO_NODE_GROUP;
390 ch->sqnum = cpu_to_le64(sqnum);
391 ch->padding[0] = ch->padding[1] = 0;
392 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
393 ch->crc = cpu_to_le32(crc);
394
395 if (pad) {
396 len = ALIGN(len, 8);
397 pad = ALIGN(len, c->min_io_size) - len;
398 ubifs_pad(c, node + len, pad);
399 }
400}
401
402/**
403 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
404 * @c: UBIFS file-system description object
405 * @node: the node to pad
406 * @len: node length
407 * @last: indicates the last node of the group
408 *
409 * This function prepares node at @node to be written to the media - it
410 * calculates node CRC and fills the common header.
411 */
412void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
413{
414 uint32_t crc;
415 struct ubifs_ch *ch = node;
416 unsigned long long sqnum = next_sqnum(c);
417
418 ubifs_assert(len >= UBIFS_CH_SZ);
419
420 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
421 ch->len = cpu_to_le32(len);
422 if (last)
423 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
424 else
425 ch->group_type = UBIFS_IN_NODE_GROUP;
426 ch->sqnum = cpu_to_le64(sqnum);
427 ch->padding[0] = ch->padding[1] = 0;
428 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
429 ch->crc = cpu_to_le32(crc);
430}
431
432/**
433 * wbuf_timer_callback - write-buffer timer callback function.
434 * @timer: timer data (write-buffer descriptor)
435 *
436 * This function is called when the write-buffer timer expires.
437 */
438static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
439{
440 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
441
442 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
443 wbuf->need_sync = 1;
444 wbuf->c->need_wbuf_sync = 1;
445 ubifs_wake_up_bgt(wbuf->c);
446 return HRTIMER_NORESTART;
447}
448
449/**
450 * new_wbuf_timer - start new write-buffer timer.
451 * @wbuf: write-buffer descriptor
452 */
453static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
454{
455 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
456 unsigned long long delta = dirty_writeback_interval;
457
458 /* centi to milli, milli to nano, then 10% */
459 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
460
461 ubifs_assert(!hrtimer_active(&wbuf->timer));
462 ubifs_assert(delta <= ULONG_MAX);
463
464 if (wbuf->no_timer)
465 return;
466 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
467 dbg_jhead(wbuf->jhead),
468 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
469 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
470 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
471 HRTIMER_MODE_REL);
472}
473
474/**
475 * cancel_wbuf_timer - cancel write-buffer timer.
476 * @wbuf: write-buffer descriptor
477 */
478static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
479{
480 if (wbuf->no_timer)
481 return;
482 wbuf->need_sync = 0;
483 hrtimer_cancel(&wbuf->timer);
484}
485
486/**
487 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
488 * @wbuf: write-buffer to synchronize
489 *
490 * This function synchronizes write-buffer @buf and returns zero in case of
491 * success or a negative error code in case of failure.
492 *
493 * Note, although write-buffers are of @c->max_write_size, this function does
494 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
495 * if the write-buffer is only partially filled with data, only the used part
496 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
497 * This way we waste less space.
498 */
499int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
500{
501 struct ubifs_info *c = wbuf->c;
502 int err, dirt, sync_len;
503
504 cancel_wbuf_timer_nolock(wbuf);
505 if (!wbuf->used || wbuf->lnum == -1)
506 /* Write-buffer is empty or not seeked */
507 return 0;
508
509 dbg_io("LEB %d:%d, %d bytes, jhead %s",
510 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
511 ubifs_assert(!(wbuf->avail & 7));
512 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
513 ubifs_assert(wbuf->size >= c->min_io_size);
514 ubifs_assert(wbuf->size <= c->max_write_size);
515 ubifs_assert(wbuf->size % c->min_io_size == 0);
516 ubifs_assert(!c->ro_media && !c->ro_mount);
517 if (c->leb_size - wbuf->offs >= c->max_write_size)
518 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
519
520 if (c->ro_error)
521 return -EROFS;
522
523 /*
524 * Do not write whole write buffer but write only the minimum necessary
525 * amount of min. I/O units.
526 */
527 sync_len = ALIGN(wbuf->used, c->min_io_size);
528 dirt = sync_len - wbuf->used;
529 if (dirt)
530 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
531 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
532 if (err)
533 return err;
534
535 spin_lock(&wbuf->lock);
536 wbuf->offs += sync_len;
537 /*
538 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
539 * But our goal is to optimize writes and make sure we write in
540 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
541 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
542 * sure that @wbuf->offs + @wbuf->size is aligned to
543 * @c->max_write_size. This way we make sure that after next
544 * write-buffer flush we are again at the optimal offset (aligned to
545 * @c->max_write_size).
546 */
547 if (c->leb_size - wbuf->offs < c->max_write_size)
548 wbuf->size = c->leb_size - wbuf->offs;
549 else if (wbuf->offs & (c->max_write_size - 1))
550 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
551 else
552 wbuf->size = c->max_write_size;
553 wbuf->avail = wbuf->size;
554 wbuf->used = 0;
555 wbuf->next_ino = 0;
556 spin_unlock(&wbuf->lock);
557
558 if (wbuf->sync_callback)
559 err = wbuf->sync_callback(c, wbuf->lnum,
560 c->leb_size - wbuf->offs, dirt);
561 return err;
562}
563
564/**
565 * ubifs_wbuf_seek_nolock - seek write-buffer.
566 * @wbuf: write-buffer
567 * @lnum: logical eraseblock number to seek to
568 * @offs: logical eraseblock offset to seek to
569 *
570 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
571 * The write-buffer has to be empty. Returns zero in case of success and a
572 * negative error code in case of failure.
573 */
574int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
575{
576 const struct ubifs_info *c = wbuf->c;
577
578 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
579 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
580 ubifs_assert(offs >= 0 && offs <= c->leb_size);
581 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
582 ubifs_assert(lnum != wbuf->lnum);
583 ubifs_assert(wbuf->used == 0);
584
585 spin_lock(&wbuf->lock);
586 wbuf->lnum = lnum;
587 wbuf->offs = offs;
588 if (c->leb_size - wbuf->offs < c->max_write_size)
589 wbuf->size = c->leb_size - wbuf->offs;
590 else if (wbuf->offs & (c->max_write_size - 1))
591 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
592 else
593 wbuf->size = c->max_write_size;
594 wbuf->avail = wbuf->size;
595 wbuf->used = 0;
596 spin_unlock(&wbuf->lock);
597
598 return 0;
599}
600
601/**
602 * ubifs_bg_wbufs_sync - synchronize write-buffers.
603 * @c: UBIFS file-system description object
604 *
605 * This function is called by background thread to synchronize write-buffers.
606 * Returns zero in case of success and a negative error code in case of
607 * failure.
608 */
609int ubifs_bg_wbufs_sync(struct ubifs_info *c)
610{
611 int err, i;
612
613 ubifs_assert(!c->ro_media && !c->ro_mount);
614 if (!c->need_wbuf_sync)
615 return 0;
616 c->need_wbuf_sync = 0;
617
618 if (c->ro_error) {
619 err = -EROFS;
620 goto out_timers;
621 }
622
623 dbg_io("synchronize");
624 for (i = 0; i < c->jhead_cnt; i++) {
625 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
626
627 cond_resched();
628
629 /*
630 * If the mutex is locked then wbuf is being changed, so
631 * synchronization is not necessary.
632 */
633 if (mutex_is_locked(&wbuf->io_mutex))
634 continue;
635
636 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
637 if (!wbuf->need_sync) {
638 mutex_unlock(&wbuf->io_mutex);
639 continue;
640 }
641
642 err = ubifs_wbuf_sync_nolock(wbuf);
643 mutex_unlock(&wbuf->io_mutex);
644 if (err) {
645 ubifs_err(c, "cannot sync write-buffer, error %d", err);
646 ubifs_ro_mode(c, err);
647 goto out_timers;
648 }
649 }
650
651 return 0;
652
653out_timers:
654 /* Cancel all timers to prevent repeated errors */
655 for (i = 0; i < c->jhead_cnt; i++) {
656 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
657
658 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
659 cancel_wbuf_timer_nolock(wbuf);
660 mutex_unlock(&wbuf->io_mutex);
661 }
662 return err;
663}
664
665/**
666 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
667 * @wbuf: write-buffer
668 * @buf: node to write
669 * @len: node length
670 *
671 * This function writes data to flash via write-buffer @wbuf. This means that
672 * the last piece of the node won't reach the flash media immediately if it
673 * does not take whole max. write unit (@c->max_write_size). Instead, the node
674 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
675 * because more data are appended to the write-buffer).
676 *
677 * This function returns zero in case of success and a negative error code in
678 * case of failure. If the node cannot be written because there is no more
679 * space in this logical eraseblock, %-ENOSPC is returned.
680 */
681int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
682{
683 struct ubifs_info *c = wbuf->c;
684 int err, written, n, aligned_len = ALIGN(len, 8);
685
686 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
687 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
688 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
689 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
690 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
691 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
692 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
693 ubifs_assert(wbuf->size >= c->min_io_size);
694 ubifs_assert(wbuf->size <= c->max_write_size);
695 ubifs_assert(wbuf->size % c->min_io_size == 0);
696 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
697 ubifs_assert(!c->ro_media && !c->ro_mount);
698 ubifs_assert(!c->space_fixup);
699 if (c->leb_size - wbuf->offs >= c->max_write_size)
700 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
701
702 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
703 err = -ENOSPC;
704 goto out;
705 }
706
707 cancel_wbuf_timer_nolock(wbuf);
708
709 if (c->ro_error)
710 return -EROFS;
711
712 if (aligned_len <= wbuf->avail) {
713 /*
714 * The node is not very large and fits entirely within
715 * write-buffer.
716 */
717 memcpy(wbuf->buf + wbuf->used, buf, len);
718
719 if (aligned_len == wbuf->avail) {
720 dbg_io("flush jhead %s wbuf to LEB %d:%d",
721 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
722 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
723 wbuf->offs, wbuf->size);
724 if (err)
725 goto out;
726
727 spin_lock(&wbuf->lock);
728 wbuf->offs += wbuf->size;
729 if (c->leb_size - wbuf->offs >= c->max_write_size)
730 wbuf->size = c->max_write_size;
731 else
732 wbuf->size = c->leb_size - wbuf->offs;
733 wbuf->avail = wbuf->size;
734 wbuf->used = 0;
735 wbuf->next_ino = 0;
736 spin_unlock(&wbuf->lock);
737 } else {
738 spin_lock(&wbuf->lock);
739 wbuf->avail -= aligned_len;
740 wbuf->used += aligned_len;
741 spin_unlock(&wbuf->lock);
742 }
743
744 goto exit;
745 }
746
747 written = 0;
748
749 if (wbuf->used) {
750 /*
751 * The node is large enough and does not fit entirely within
752 * current available space. We have to fill and flush
753 * write-buffer and switch to the next max. write unit.
754 */
755 dbg_io("flush jhead %s wbuf to LEB %d:%d",
756 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
757 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
758 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
759 wbuf->size);
760 if (err)
761 goto out;
762
763 wbuf->offs += wbuf->size;
764 len -= wbuf->avail;
765 aligned_len -= wbuf->avail;
766 written += wbuf->avail;
767 } else if (wbuf->offs & (c->max_write_size - 1)) {
768 /*
769 * The write-buffer offset is not aligned to
770 * @c->max_write_size and @wbuf->size is less than
771 * @c->max_write_size. Write @wbuf->size bytes to make sure the
772 * following writes are done in optimal @c->max_write_size
773 * chunks.
774 */
775 dbg_io("write %d bytes to LEB %d:%d",
776 wbuf->size, wbuf->lnum, wbuf->offs);
777 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
778 wbuf->size);
779 if (err)
780 goto out;
781
782 wbuf->offs += wbuf->size;
783 len -= wbuf->size;
784 aligned_len -= wbuf->size;
785 written += wbuf->size;
786 }
787
788 /*
789 * The remaining data may take more whole max. write units, so write the
790 * remains multiple to max. write unit size directly to the flash media.
791 * We align node length to 8-byte boundary because we anyway flash wbuf
792 * if the remaining space is less than 8 bytes.
793 */
794 n = aligned_len >> c->max_write_shift;
795 if (n) {
796 n <<= c->max_write_shift;
797 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
798 wbuf->offs);
799 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
800 wbuf->offs, n);
801 if (err)
802 goto out;
803 wbuf->offs += n;
804 aligned_len -= n;
805 len -= n;
806 written += n;
807 }
808
809 spin_lock(&wbuf->lock);
810 if (aligned_len)
811 /*
812 * And now we have what's left and what does not take whole
813 * max. write unit, so write it to the write-buffer and we are
814 * done.
815 */
816 memcpy(wbuf->buf, buf + written, len);
817
818 if (c->leb_size - wbuf->offs >= c->max_write_size)
819 wbuf->size = c->max_write_size;
820 else
821 wbuf->size = c->leb_size - wbuf->offs;
822 wbuf->avail = wbuf->size - aligned_len;
823 wbuf->used = aligned_len;
824 wbuf->next_ino = 0;
825 spin_unlock(&wbuf->lock);
826
827exit:
828 if (wbuf->sync_callback) {
829 int free = c->leb_size - wbuf->offs - wbuf->used;
830
831 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
832 if (err)
833 goto out;
834 }
835
836 if (wbuf->used)
837 new_wbuf_timer_nolock(wbuf);
838
839 return 0;
840
841out:
842 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
843 len, wbuf->lnum, wbuf->offs, err);
844 ubifs_dump_node(c, buf);
845 dump_stack();
846 ubifs_dump_leb(c, wbuf->lnum);
847 return err;
848}
849
850/**
851 * ubifs_write_node - write node to the media.
852 * @c: UBIFS file-system description object
853 * @buf: the node to write
854 * @len: node length
855 * @lnum: logical eraseblock number
856 * @offs: offset within the logical eraseblock
857 *
858 * This function automatically fills node magic number, assigns sequence
859 * number, and calculates node CRC checksum. The length of the @buf buffer has
860 * to be aligned to the minimal I/O unit size. This function automatically
861 * appends padding node and padding bytes if needed. Returns zero in case of
862 * success and a negative error code in case of failure.
863 */
864int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
865 int offs)
866{
867 int err, buf_len = ALIGN(len, c->min_io_size);
868
869 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
870 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
871 buf_len);
872 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
873 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
874 ubifs_assert(!c->ro_media && !c->ro_mount);
875 ubifs_assert(!c->space_fixup);
876
877 if (c->ro_error)
878 return -EROFS;
879
880 ubifs_prepare_node(c, buf, len, 1);
881 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
882 if (err)
883 ubifs_dump_node(c, buf);
884
885 return err;
886}
887
888/**
889 * ubifs_read_node_wbuf - read node from the media or write-buffer.
890 * @wbuf: wbuf to check for un-written data
891 * @buf: buffer to read to
892 * @type: node type
893 * @len: node length
894 * @lnum: logical eraseblock number
895 * @offs: offset within the logical eraseblock
896 *
897 * This function reads a node of known type and length, checks it and stores
898 * in @buf. If the node partially or fully sits in the write-buffer, this
899 * function takes data from the buffer, otherwise it reads the flash media.
900 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
901 * error code in case of failure.
902 */
903int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
904 int lnum, int offs)
905{
906 const struct ubifs_info *c = wbuf->c;
907 int err, rlen, overlap;
908 struct ubifs_ch *ch = buf;
909
910 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
911 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
912 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
913 ubifs_assert(!(offs & 7) && offs < c->leb_size);
914 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
915
916 spin_lock(&wbuf->lock);
917 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
918 if (!overlap) {
919 /* We may safely unlock the write-buffer and read the data */
920 spin_unlock(&wbuf->lock);
921 return ubifs_read_node(c, buf, type, len, lnum, offs);
922 }
923
924 /* Don't read under wbuf */
925 rlen = wbuf->offs - offs;
926 if (rlen < 0)
927 rlen = 0;
928
929 /* Copy the rest from the write-buffer */
930 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
931 spin_unlock(&wbuf->lock);
932
933 if (rlen > 0) {
934 /* Read everything that goes before write-buffer */
935 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
936 if (err && err != -EBADMSG)
937 return err;
938 }
939
940 if (type != ch->node_type) {
941 ubifs_err(c, "bad node type (%d but expected %d)",
942 ch->node_type, type);
943 goto out;
944 }
945
946 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
947 if (err) {
948 ubifs_err(c, "expected node type %d", type);
949 return err;
950 }
951
952 rlen = le32_to_cpu(ch->len);
953 if (rlen != len) {
954 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
955 goto out;
956 }
957
958 return 0;
959
960out:
961 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
962 ubifs_dump_node(c, buf);
963 dump_stack();
964 return -EINVAL;
965}
966
967/**
968 * ubifs_read_node - read node.
969 * @c: UBIFS file-system description object
970 * @buf: buffer to read to
971 * @type: node type
972 * @len: node length (not aligned)
973 * @lnum: logical eraseblock number
974 * @offs: offset within the logical eraseblock
975 *
976 * This function reads a node of known type and and length, checks it and
977 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
978 * and a negative error code in case of failure.
979 */
980int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
981 int lnum, int offs)
982{
983 int err, l;
984 struct ubifs_ch *ch = buf;
985
986 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
987 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
988 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
989 ubifs_assert(!(offs & 7) && offs < c->leb_size);
990 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
991
992 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
993 if (err && err != -EBADMSG)
994 return err;
995
996 if (type != ch->node_type) {
997 ubifs_errc(c, "bad node type (%d but expected %d)",
998 ch->node_type, type);
999 goto out;
1000 }
1001
1002 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1003 if (err) {
1004 ubifs_errc(c, "expected node type %d", type);
1005 return err;
1006 }
1007
1008 l = le32_to_cpu(ch->len);
1009 if (l != len) {
1010 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1011 goto out;
1012 }
1013
1014 return 0;
1015
1016out:
1017 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1018 offs, ubi_is_mapped(c->ubi, lnum));
1019 if (!c->probing) {
1020 ubifs_dump_node(c, buf);
1021 dump_stack();
1022 }
1023 return -EINVAL;
1024}
1025
1026/**
1027 * ubifs_wbuf_init - initialize write-buffer.
1028 * @c: UBIFS file-system description object
1029 * @wbuf: write-buffer to initialize
1030 *
1031 * This function initializes write-buffer. Returns zero in case of success
1032 * %-ENOMEM in case of failure.
1033 */
1034int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1035{
1036 size_t size;
1037
1038 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1039 if (!wbuf->buf)
1040 return -ENOMEM;
1041
1042 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1043 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1044 if (!wbuf->inodes) {
1045 kfree(wbuf->buf);
1046 wbuf->buf = NULL;
1047 return -ENOMEM;
1048 }
1049
1050 wbuf->used = 0;
1051 wbuf->lnum = wbuf->offs = -1;
1052 /*
1053 * If the LEB starts at the max. write size aligned address, then
1054 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1055 * set it to something smaller so that it ends at the closest max.
1056 * write size boundary.
1057 */
1058 size = c->max_write_size - (c->leb_start % c->max_write_size);
1059 wbuf->avail = wbuf->size = size;
1060 wbuf->sync_callback = NULL;
1061 mutex_init(&wbuf->io_mutex);
1062 spin_lock_init(&wbuf->lock);
1063 wbuf->c = c;
1064 wbuf->next_ino = 0;
1065
1066 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1067 wbuf->timer.function = wbuf_timer_callback_nolock;
1068 return 0;
1069}
1070
1071/**
1072 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1073 * @wbuf: the write-buffer where to add
1074 * @inum: the inode number
1075 *
1076 * This function adds an inode number to the inode array of the write-buffer.
1077 */
1078void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1079{
1080 if (!wbuf->buf)
1081 /* NOR flash or something similar */
1082 return;
1083
1084 spin_lock(&wbuf->lock);
1085 if (wbuf->used)
1086 wbuf->inodes[wbuf->next_ino++] = inum;
1087 spin_unlock(&wbuf->lock);
1088}
1089
1090/**
1091 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1092 * @wbuf: the write-buffer
1093 * @inum: the inode number
1094 *
1095 * This function returns with %1 if the write-buffer contains some data from the
1096 * given inode otherwise it returns with %0.
1097 */
1098static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1099{
1100 int i, ret = 0;
1101
1102 spin_lock(&wbuf->lock);
1103 for (i = 0; i < wbuf->next_ino; i++)
1104 if (inum == wbuf->inodes[i]) {
1105 ret = 1;
1106 break;
1107 }
1108 spin_unlock(&wbuf->lock);
1109
1110 return ret;
1111}
1112
1113/**
1114 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1115 * @c: UBIFS file-system description object
1116 * @inode: inode to synchronize
1117 *
1118 * This function synchronizes write-buffers which contain nodes belonging to
1119 * @inode. Returns zero in case of success and a negative error code in case of
1120 * failure.
1121 */
1122int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1123{
1124 int i, err = 0;
1125
1126 for (i = 0; i < c->jhead_cnt; i++) {
1127 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1128
1129 if (i == GCHD)
1130 /*
1131 * GC head is special, do not look at it. Even if the
1132 * head contains something related to this inode, it is
1133 * a _copy_ of corresponding on-flash node which sits
1134 * somewhere else.
1135 */
1136 continue;
1137
1138 if (!wbuf_has_ino(wbuf, inode->i_ino))
1139 continue;
1140
1141 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1142 if (wbuf_has_ino(wbuf, inode->i_ino))
1143 err = ubifs_wbuf_sync_nolock(wbuf);
1144 mutex_unlock(&wbuf->io_mutex);
1145
1146 if (err) {
1147 ubifs_ro_mode(c, err);
1148 return err;
1149 }
1150 }
1151 return 0;
1152}