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;
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);
281out:
282 if (!quiet) {
283 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
284 ubifs_dump_node(c, buf);
285 dump_stack();
286 }
287 return err;
288}
289
290/**
291 * ubifs_pad - pad flash space.
292 * @c: UBIFS file-system description object
293 * @buf: buffer to put padding to
294 * @pad: how many bytes to pad
295 *
296 * The flash media obliges us to write only in chunks of %c->min_io_size and
297 * when we have to write less data we add padding node to the write-buffer and
298 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
299 * media is being scanned. If the amount of wasted space is not enough to fit a
300 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
301 * pattern (%UBIFS_PADDING_BYTE).
302 *
303 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
304 * used.
305 */
306void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
307{
308 uint32_t crc;
309
310 ubifs_assert(c, pad >= 0 && !(pad & 7));
311
312 if (pad >= UBIFS_PAD_NODE_SZ) {
313 struct ubifs_ch *ch = buf;
314 struct ubifs_pad_node *pad_node = buf;
315
316 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
317 ch->node_type = UBIFS_PAD_NODE;
318 ch->group_type = UBIFS_NO_NODE_GROUP;
319 ch->padding[0] = ch->padding[1] = 0;
320 ch->sqnum = 0;
321 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
322 pad -= UBIFS_PAD_NODE_SZ;
323 pad_node->pad_len = cpu_to_le32(pad);
324 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
325 ch->crc = cpu_to_le32(crc);
326 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
327 } else if (pad > 0)
328 /* Too little space, padding node won't fit */
329 memset(buf, UBIFS_PADDING_BYTE, pad);
330}
331
332/**
333 * next_sqnum - get next sequence number.
334 * @c: UBIFS file-system description object
335 */
336static unsigned long long next_sqnum(struct ubifs_info *c)
337{
338 unsigned long long sqnum;
339
340 spin_lock(&c->cnt_lock);
341 sqnum = ++c->max_sqnum;
342 spin_unlock(&c->cnt_lock);
343
344 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
345 if (sqnum >= SQNUM_WATERMARK) {
346 ubifs_err(c, "sequence number overflow %llu, end of life",
347 sqnum);
348 ubifs_ro_mode(c, -EINVAL);
349 }
350 ubifs_warn(c, "running out of sequence numbers, end of life soon");
351 }
352
353 return sqnum;
354}
355
356void ubifs_init_node(struct ubifs_info *c, void *node, int len, int pad)
357{
358 struct ubifs_ch *ch = node;
359 unsigned long long sqnum = next_sqnum(c);
360
361 ubifs_assert(c, len >= UBIFS_CH_SZ);
362
363 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
364 ch->len = cpu_to_le32(len);
365 ch->group_type = UBIFS_NO_NODE_GROUP;
366 ch->sqnum = cpu_to_le64(sqnum);
367 ch->padding[0] = ch->padding[1] = 0;
368
369 if (pad) {
370 len = ALIGN(len, 8);
371 pad = ALIGN(len, c->min_io_size) - len;
372 ubifs_pad(c, node + len, pad);
373 }
374}
375
376void ubifs_crc_node(struct ubifs_info *c, void *node, int len)
377{
378 struct ubifs_ch *ch = node;
379 uint32_t crc;
380
381 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
382 ch->crc = cpu_to_le32(crc);
383}
384
385/**
386 * ubifs_prepare_node_hmac - prepare node to be written to flash.
387 * @c: UBIFS file-system description object
388 * @node: the node to pad
389 * @len: node length
390 * @hmac_offs: offset of the HMAC in the node
391 * @pad: if the buffer has to be padded
392 *
393 * This function prepares node at @node to be written to the media - it
394 * calculates node CRC, fills the common header, and adds proper padding up to
395 * the next minimum I/O unit if @pad is not zero. if @hmac_offs is positive then
396 * a HMAC is inserted into the node at the given offset.
397 *
398 * This function returns 0 for success or a negative error code otherwise.
399 */
400int ubifs_prepare_node_hmac(struct ubifs_info *c, void *node, int len,
401 int hmac_offs, int pad)
402{
403 int err;
404
405 ubifs_init_node(c, node, len, pad);
406
407 if (hmac_offs > 0) {
408 err = ubifs_node_insert_hmac(c, node, len, hmac_offs);
409 if (err)
410 return err;
411 }
412
413 ubifs_crc_node(c, node, len);
414
415 return 0;
416}
417
418/**
419 * ubifs_prepare_node - prepare node to be written to flash.
420 * @c: UBIFS file-system description object
421 * @node: the node to pad
422 * @len: node length
423 * @pad: if the buffer has to be padded
424 *
425 * This function prepares node at @node to be written to the media - it
426 * calculates node CRC, fills the common header, and adds proper padding up to
427 * the next minimum I/O unit if @pad is not zero.
428 */
429void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
430{
431 /*
432 * Deliberately ignore return value since this function can only fail
433 * when a hmac offset is given.
434 */
435 ubifs_prepare_node_hmac(c, node, len, 0, pad);
436}
437
438/**
439 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
440 * @c: UBIFS file-system description object
441 * @node: the node to pad
442 * @len: node length
443 * @last: indicates the last node of the group
444 *
445 * This function prepares node at @node to be written to the media - it
446 * calculates node CRC and fills the common header.
447 */
448void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
449{
450 uint32_t crc;
451 struct ubifs_ch *ch = node;
452 unsigned long long sqnum = next_sqnum(c);
453
454 ubifs_assert(c, len >= UBIFS_CH_SZ);
455
456 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
457 ch->len = cpu_to_le32(len);
458 if (last)
459 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
460 else
461 ch->group_type = UBIFS_IN_NODE_GROUP;
462 ch->sqnum = cpu_to_le64(sqnum);
463 ch->padding[0] = ch->padding[1] = 0;
464 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
465 ch->crc = cpu_to_le32(crc);
466}
467
468/**
469 * wbuf_timer_callback - write-buffer timer callback function.
470 * @timer: timer data (write-buffer descriptor)
471 *
472 * This function is called when the write-buffer timer expires.
473 */
474static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
475{
476 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
477
478 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
479 wbuf->need_sync = 1;
480 wbuf->c->need_wbuf_sync = 1;
481 ubifs_wake_up_bgt(wbuf->c);
482 return HRTIMER_NORESTART;
483}
484
485/**
486 * new_wbuf_timer - start new write-buffer timer.
487 * @c: UBIFS file-system description object
488 * @wbuf: write-buffer descriptor
489 */
490static void new_wbuf_timer_nolock(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
491{
492 ktime_t softlimit = ms_to_ktime(dirty_writeback_interval * 10);
493 unsigned long long delta = dirty_writeback_interval;
494
495 /* centi to milli, milli to nano, then 10% */
496 delta *= 10ULL * NSEC_PER_MSEC / 10ULL;
497
498 ubifs_assert(c, !hrtimer_active(&wbuf->timer));
499 ubifs_assert(c, delta <= ULONG_MAX);
500
501 if (wbuf->no_timer)
502 return;
503 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
504 dbg_jhead(wbuf->jhead),
505 div_u64(ktime_to_ns(softlimit), USEC_PER_SEC),
506 div_u64(ktime_to_ns(softlimit) + delta, USEC_PER_SEC));
507 hrtimer_start_range_ns(&wbuf->timer, softlimit, delta,
508 HRTIMER_MODE_REL);
509}
510
511/**
512 * cancel_wbuf_timer - cancel write-buffer timer.
513 * @wbuf: write-buffer descriptor
514 */
515static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
516{
517 if (wbuf->no_timer)
518 return;
519 wbuf->need_sync = 0;
520 hrtimer_cancel(&wbuf->timer);
521}
522
523/**
524 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
525 * @wbuf: write-buffer to synchronize
526 *
527 * This function synchronizes write-buffer @buf and returns zero in case of
528 * success or a negative error code in case of failure.
529 *
530 * Note, although write-buffers are of @c->max_write_size, this function does
531 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
532 * if the write-buffer is only partially filled with data, only the used part
533 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
534 * This way we waste less space.
535 */
536int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
537{
538 struct ubifs_info *c = wbuf->c;
539 int err, dirt, sync_len;
540
541 cancel_wbuf_timer_nolock(wbuf);
542 if (!wbuf->used || wbuf->lnum == -1)
543 /* Write-buffer is empty or not seeked */
544 return 0;
545
546 dbg_io("LEB %d:%d, %d bytes, jhead %s",
547 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
548 ubifs_assert(c, !(wbuf->avail & 7));
549 ubifs_assert(c, wbuf->offs + wbuf->size <= c->leb_size);
550 ubifs_assert(c, wbuf->size >= c->min_io_size);
551 ubifs_assert(c, wbuf->size <= c->max_write_size);
552 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
553 ubifs_assert(c, !c->ro_media && !c->ro_mount);
554 if (c->leb_size - wbuf->offs >= c->max_write_size)
555 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
556
557 if (c->ro_error)
558 return -EROFS;
559
560 /*
561 * Do not write whole write buffer but write only the minimum necessary
562 * amount of min. I/O units.
563 */
564 sync_len = ALIGN(wbuf->used, c->min_io_size);
565 dirt = sync_len - wbuf->used;
566 if (dirt)
567 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
568 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
569 if (err)
570 return err;
571
572 spin_lock(&wbuf->lock);
573 wbuf->offs += sync_len;
574 /*
575 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
576 * But our goal is to optimize writes and make sure we write in
577 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
578 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
579 * sure that @wbuf->offs + @wbuf->size is aligned to
580 * @c->max_write_size. This way we make sure that after next
581 * write-buffer flush we are again at the optimal offset (aligned to
582 * @c->max_write_size).
583 */
584 if (c->leb_size - wbuf->offs < c->max_write_size)
585 wbuf->size = c->leb_size - wbuf->offs;
586 else if (wbuf->offs & (c->max_write_size - 1))
587 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
588 else
589 wbuf->size = c->max_write_size;
590 wbuf->avail = wbuf->size;
591 wbuf->used = 0;
592 wbuf->next_ino = 0;
593 spin_unlock(&wbuf->lock);
594
595 if (wbuf->sync_callback)
596 err = wbuf->sync_callback(c, wbuf->lnum,
597 c->leb_size - wbuf->offs, dirt);
598 return err;
599}
600
601/**
602 * ubifs_wbuf_seek_nolock - seek write-buffer.
603 * @wbuf: write-buffer
604 * @lnum: logical eraseblock number to seek to
605 * @offs: logical eraseblock offset to seek to
606 *
607 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
608 * The write-buffer has to be empty. Returns zero in case of success and a
609 * negative error code in case of failure.
610 */
611int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
612{
613 const struct ubifs_info *c = wbuf->c;
614
615 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
616 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt);
617 ubifs_assert(c, offs >= 0 && offs <= c->leb_size);
618 ubifs_assert(c, offs % c->min_io_size == 0 && !(offs & 7));
619 ubifs_assert(c, lnum != wbuf->lnum);
620 ubifs_assert(c, wbuf->used == 0);
621
622 spin_lock(&wbuf->lock);
623 wbuf->lnum = lnum;
624 wbuf->offs = offs;
625 if (c->leb_size - wbuf->offs < c->max_write_size)
626 wbuf->size = c->leb_size - wbuf->offs;
627 else if (wbuf->offs & (c->max_write_size - 1))
628 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
629 else
630 wbuf->size = c->max_write_size;
631 wbuf->avail = wbuf->size;
632 wbuf->used = 0;
633 spin_unlock(&wbuf->lock);
634
635 return 0;
636}
637
638/**
639 * ubifs_bg_wbufs_sync - synchronize write-buffers.
640 * @c: UBIFS file-system description object
641 *
642 * This function is called by background thread to synchronize write-buffers.
643 * Returns zero in case of success and a negative error code in case of
644 * failure.
645 */
646int ubifs_bg_wbufs_sync(struct ubifs_info *c)
647{
648 int err, i;
649
650 ubifs_assert(c, !c->ro_media && !c->ro_mount);
651 if (!c->need_wbuf_sync)
652 return 0;
653 c->need_wbuf_sync = 0;
654
655 if (c->ro_error) {
656 err = -EROFS;
657 goto out_timers;
658 }
659
660 dbg_io("synchronize");
661 for (i = 0; i < c->jhead_cnt; i++) {
662 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
663
664 cond_resched();
665
666 /*
667 * If the mutex is locked then wbuf is being changed, so
668 * synchronization is not necessary.
669 */
670 if (mutex_is_locked(&wbuf->io_mutex))
671 continue;
672
673 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
674 if (!wbuf->need_sync) {
675 mutex_unlock(&wbuf->io_mutex);
676 continue;
677 }
678
679 err = ubifs_wbuf_sync_nolock(wbuf);
680 mutex_unlock(&wbuf->io_mutex);
681 if (err) {
682 ubifs_err(c, "cannot sync write-buffer, error %d", err);
683 ubifs_ro_mode(c, err);
684 goto out_timers;
685 }
686 }
687
688 return 0;
689
690out_timers:
691 /* Cancel all timers to prevent repeated errors */
692 for (i = 0; i < c->jhead_cnt; i++) {
693 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
694
695 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
696 cancel_wbuf_timer_nolock(wbuf);
697 mutex_unlock(&wbuf->io_mutex);
698 }
699 return err;
700}
701
702/**
703 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
704 * @wbuf: write-buffer
705 * @buf: node to write
706 * @len: node length
707 *
708 * This function writes data to flash via write-buffer @wbuf. This means that
709 * the last piece of the node won't reach the flash media immediately if it
710 * does not take whole max. write unit (@c->max_write_size). Instead, the node
711 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
712 * because more data are appended to the write-buffer).
713 *
714 * This function returns zero in case of success and a negative error code in
715 * case of failure. If the node cannot be written because there is no more
716 * space in this logical eraseblock, %-ENOSPC is returned.
717 */
718int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
719{
720 struct ubifs_info *c = wbuf->c;
721 int err, written, n, aligned_len = ALIGN(len, 8);
722
723 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
724 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
725 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
726 ubifs_assert(c, len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
727 ubifs_assert(c, wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
728 ubifs_assert(c, !(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
729 ubifs_assert(c, wbuf->avail > 0 && wbuf->avail <= wbuf->size);
730 ubifs_assert(c, wbuf->size >= c->min_io_size);
731 ubifs_assert(c, wbuf->size <= c->max_write_size);
732 ubifs_assert(c, wbuf->size % c->min_io_size == 0);
733 ubifs_assert(c, mutex_is_locked(&wbuf->io_mutex));
734 ubifs_assert(c, !c->ro_media && !c->ro_mount);
735 ubifs_assert(c, !c->space_fixup);
736 if (c->leb_size - wbuf->offs >= c->max_write_size)
737 ubifs_assert(c, !((wbuf->offs + wbuf->size) % c->max_write_size));
738
739 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
740 err = -ENOSPC;
741 goto out;
742 }
743
744 cancel_wbuf_timer_nolock(wbuf);
745
746 if (c->ro_error)
747 return -EROFS;
748
749 if (aligned_len <= wbuf->avail) {
750 /*
751 * The node is not very large and fits entirely within
752 * write-buffer.
753 */
754 memcpy(wbuf->buf + wbuf->used, buf, len);
755
756 if (aligned_len == wbuf->avail) {
757 dbg_io("flush jhead %s wbuf to LEB %d:%d",
758 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
759 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
760 wbuf->offs, wbuf->size);
761 if (err)
762 goto out;
763
764 spin_lock(&wbuf->lock);
765 wbuf->offs += wbuf->size;
766 if (c->leb_size - wbuf->offs >= c->max_write_size)
767 wbuf->size = c->max_write_size;
768 else
769 wbuf->size = c->leb_size - wbuf->offs;
770 wbuf->avail = wbuf->size;
771 wbuf->used = 0;
772 wbuf->next_ino = 0;
773 spin_unlock(&wbuf->lock);
774 } else {
775 spin_lock(&wbuf->lock);
776 wbuf->avail -= aligned_len;
777 wbuf->used += aligned_len;
778 spin_unlock(&wbuf->lock);
779 }
780
781 goto exit;
782 }
783
784 written = 0;
785
786 if (wbuf->used) {
787 /*
788 * The node is large enough and does not fit entirely within
789 * current available space. We have to fill and flush
790 * write-buffer and switch to the next max. write unit.
791 */
792 dbg_io("flush jhead %s wbuf to LEB %d:%d",
793 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
794 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
795 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
796 wbuf->size);
797 if (err)
798 goto out;
799
800 wbuf->offs += wbuf->size;
801 len -= wbuf->avail;
802 aligned_len -= wbuf->avail;
803 written += wbuf->avail;
804 } else if (wbuf->offs & (c->max_write_size - 1)) {
805 /*
806 * The write-buffer offset is not aligned to
807 * @c->max_write_size and @wbuf->size is less than
808 * @c->max_write_size. Write @wbuf->size bytes to make sure the
809 * following writes are done in optimal @c->max_write_size
810 * chunks.
811 */
812 dbg_io("write %d bytes to LEB %d:%d",
813 wbuf->size, wbuf->lnum, wbuf->offs);
814 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
815 wbuf->size);
816 if (err)
817 goto out;
818
819 wbuf->offs += wbuf->size;
820 len -= wbuf->size;
821 aligned_len -= wbuf->size;
822 written += wbuf->size;
823 }
824
825 /*
826 * The remaining data may take more whole max. write units, so write the
827 * remains multiple to max. write unit size directly to the flash media.
828 * We align node length to 8-byte boundary because we anyway flash wbuf
829 * if the remaining space is less than 8 bytes.
830 */
831 n = aligned_len >> c->max_write_shift;
832 if (n) {
833 n <<= c->max_write_shift;
834 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
835 wbuf->offs);
836 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
837 wbuf->offs, n);
838 if (err)
839 goto out;
840 wbuf->offs += n;
841 aligned_len -= n;
842 len -= n;
843 written += n;
844 }
845
846 spin_lock(&wbuf->lock);
847 if (aligned_len)
848 /*
849 * And now we have what's left and what does not take whole
850 * max. write unit, so write it to the write-buffer and we are
851 * done.
852 */
853 memcpy(wbuf->buf, buf + written, len);
854
855 if (c->leb_size - wbuf->offs >= c->max_write_size)
856 wbuf->size = c->max_write_size;
857 else
858 wbuf->size = c->leb_size - wbuf->offs;
859 wbuf->avail = wbuf->size - aligned_len;
860 wbuf->used = aligned_len;
861 wbuf->next_ino = 0;
862 spin_unlock(&wbuf->lock);
863
864exit:
865 if (wbuf->sync_callback) {
866 int free = c->leb_size - wbuf->offs - wbuf->used;
867
868 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
869 if (err)
870 goto out;
871 }
872
873 if (wbuf->used)
874 new_wbuf_timer_nolock(c, wbuf);
875
876 return 0;
877
878out:
879 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
880 len, wbuf->lnum, wbuf->offs, err);
881 ubifs_dump_node(c, buf);
882 dump_stack();
883 ubifs_dump_leb(c, wbuf->lnum);
884 return err;
885}
886
887/**
888 * ubifs_write_node_hmac - write node to the media.
889 * @c: UBIFS file-system description object
890 * @buf: the node to write
891 * @len: node length
892 * @lnum: logical eraseblock number
893 * @offs: offset within the logical eraseblock
894 * @hmac_offs: offset of the HMAC within the node
895 *
896 * This function automatically fills node magic number, assigns sequence
897 * number, and calculates node CRC checksum. The length of the @buf buffer has
898 * to be aligned to the minimal I/O unit size. This function automatically
899 * appends padding node and padding bytes if needed. Returns zero in case of
900 * success and a negative error code in case of failure.
901 */
902int ubifs_write_node_hmac(struct ubifs_info *c, void *buf, int len, int lnum,
903 int offs, int hmac_offs)
904{
905 int err, buf_len = ALIGN(len, c->min_io_size);
906
907 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
908 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
909 buf_len);
910 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
911 ubifs_assert(c, offs % c->min_io_size == 0 && offs < c->leb_size);
912 ubifs_assert(c, !c->ro_media && !c->ro_mount);
913 ubifs_assert(c, !c->space_fixup);
914
915 if (c->ro_error)
916 return -EROFS;
917
918 err = ubifs_prepare_node_hmac(c, buf, len, hmac_offs, 1);
919 if (err)
920 return err;
921
922 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
923 if (err)
924 ubifs_dump_node(c, buf);
925
926 return err;
927}
928
929/**
930 * ubifs_write_node - write node to the media.
931 * @c: UBIFS file-system description object
932 * @buf: the node to write
933 * @len: node length
934 * @lnum: logical eraseblock number
935 * @offs: offset within the logical eraseblock
936 *
937 * This function automatically fills node magic number, assigns sequence
938 * number, and calculates node CRC checksum. The length of the @buf buffer has
939 * to be aligned to the minimal I/O unit size. This function automatically
940 * appends padding node and padding bytes if needed. Returns zero in case of
941 * success and a negative error code in case of failure.
942 */
943int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
944 int offs)
945{
946 return ubifs_write_node_hmac(c, buf, len, lnum, offs, -1);
947}
948
949/**
950 * ubifs_read_node_wbuf - read node from the media or write-buffer.
951 * @wbuf: wbuf to check for un-written data
952 * @buf: buffer to read to
953 * @type: node type
954 * @len: node length
955 * @lnum: logical eraseblock number
956 * @offs: offset within the logical eraseblock
957 *
958 * This function reads a node of known type and length, checks it and stores
959 * in @buf. If the node partially or fully sits in the write-buffer, this
960 * function takes data from the buffer, otherwise it reads the flash media.
961 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
962 * error code in case of failure.
963 */
964int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
965 int lnum, int offs)
966{
967 const struct ubifs_info *c = wbuf->c;
968 int err, rlen, overlap;
969 struct ubifs_ch *ch = buf;
970
971 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
972 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
973 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
974 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
975 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
976
977 spin_lock(&wbuf->lock);
978 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
979 if (!overlap) {
980 /* We may safely unlock the write-buffer and read the data */
981 spin_unlock(&wbuf->lock);
982 return ubifs_read_node(c, buf, type, len, lnum, offs);
983 }
984
985 /* Don't read under wbuf */
986 rlen = wbuf->offs - offs;
987 if (rlen < 0)
988 rlen = 0;
989
990 /* Copy the rest from the write-buffer */
991 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
992 spin_unlock(&wbuf->lock);
993
994 if (rlen > 0) {
995 /* Read everything that goes before write-buffer */
996 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
997 if (err && err != -EBADMSG)
998 return err;
999 }
1000
1001 if (type != ch->node_type) {
1002 ubifs_err(c, "bad node type (%d but expected %d)",
1003 ch->node_type, type);
1004 goto out;
1005 }
1006
1007 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1008 if (err) {
1009 ubifs_err(c, "expected node type %d", type);
1010 return err;
1011 }
1012
1013 rlen = le32_to_cpu(ch->len);
1014 if (rlen != len) {
1015 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
1016 goto out;
1017 }
1018
1019 return 0;
1020
1021out:
1022 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
1023 ubifs_dump_node(c, buf);
1024 dump_stack();
1025 return -EINVAL;
1026}
1027
1028/**
1029 * ubifs_read_node - read node.
1030 * @c: UBIFS file-system description object
1031 * @buf: buffer to read to
1032 * @type: node type
1033 * @len: node length (not aligned)
1034 * @lnum: logical eraseblock number
1035 * @offs: offset within the logical eraseblock
1036 *
1037 * This function reads a node of known type and and length, checks it and
1038 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
1039 * and a negative error code in case of failure.
1040 */
1041int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
1042 int lnum, int offs)
1043{
1044 int err, l;
1045 struct ubifs_ch *ch = buf;
1046
1047 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
1048 ubifs_assert(c, lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1049 ubifs_assert(c, len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
1050 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1051 ubifs_assert(c, type >= 0 && type < UBIFS_NODE_TYPES_CNT);
1052
1053 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1054 if (err && err != -EBADMSG)
1055 return err;
1056
1057 if (type != ch->node_type) {
1058 ubifs_errc(c, "bad node type (%d but expected %d)",
1059 ch->node_type, type);
1060 goto out;
1061 }
1062
1063 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1064 if (err) {
1065 ubifs_errc(c, "expected node type %d", type);
1066 return err;
1067 }
1068
1069 l = le32_to_cpu(ch->len);
1070 if (l != len) {
1071 ubifs_errc(c, "bad node length %d, expected %d", l, len);
1072 goto out;
1073 }
1074
1075 return 0;
1076
1077out:
1078 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1079 offs, ubi_is_mapped(c->ubi, lnum));
1080 if (!c->probing) {
1081 ubifs_dump_node(c, buf);
1082 dump_stack();
1083 }
1084 return -EINVAL;
1085}
1086
1087/**
1088 * ubifs_wbuf_init - initialize write-buffer.
1089 * @c: UBIFS file-system description object
1090 * @wbuf: write-buffer to initialize
1091 *
1092 * This function initializes write-buffer. Returns zero in case of success
1093 * %-ENOMEM in case of failure.
1094 */
1095int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1096{
1097 size_t size;
1098
1099 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1100 if (!wbuf->buf)
1101 return -ENOMEM;
1102
1103 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1104 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1105 if (!wbuf->inodes) {
1106 kfree(wbuf->buf);
1107 wbuf->buf = NULL;
1108 return -ENOMEM;
1109 }
1110
1111 wbuf->used = 0;
1112 wbuf->lnum = wbuf->offs = -1;
1113 /*
1114 * If the LEB starts at the max. write size aligned address, then
1115 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1116 * set it to something smaller so that it ends at the closest max.
1117 * write size boundary.
1118 */
1119 size = c->max_write_size - (c->leb_start % c->max_write_size);
1120 wbuf->avail = wbuf->size = size;
1121 wbuf->sync_callback = NULL;
1122 mutex_init(&wbuf->io_mutex);
1123 spin_lock_init(&wbuf->lock);
1124 wbuf->c = c;
1125 wbuf->next_ino = 0;
1126
1127 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1128 wbuf->timer.function = wbuf_timer_callback_nolock;
1129 return 0;
1130}
1131
1132/**
1133 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1134 * @wbuf: the write-buffer where to add
1135 * @inum: the inode number
1136 *
1137 * This function adds an inode number to the inode array of the write-buffer.
1138 */
1139void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1140{
1141 if (!wbuf->buf)
1142 /* NOR flash or something similar */
1143 return;
1144
1145 spin_lock(&wbuf->lock);
1146 if (wbuf->used)
1147 wbuf->inodes[wbuf->next_ino++] = inum;
1148 spin_unlock(&wbuf->lock);
1149}
1150
1151/**
1152 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1153 * @wbuf: the write-buffer
1154 * @inum: the inode number
1155 *
1156 * This function returns with %1 if the write-buffer contains some data from the
1157 * given inode otherwise it returns with %0.
1158 */
1159static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1160{
1161 int i, ret = 0;
1162
1163 spin_lock(&wbuf->lock);
1164 for (i = 0; i < wbuf->next_ino; i++)
1165 if (inum == wbuf->inodes[i]) {
1166 ret = 1;
1167 break;
1168 }
1169 spin_unlock(&wbuf->lock);
1170
1171 return ret;
1172}
1173
1174/**
1175 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1176 * @c: UBIFS file-system description object
1177 * @inode: inode to synchronize
1178 *
1179 * This function synchronizes write-buffers which contain nodes belonging to
1180 * @inode. Returns zero in case of success and a negative error code in case of
1181 * failure.
1182 */
1183int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1184{
1185 int i, err = 0;
1186
1187 for (i = 0; i < c->jhead_cnt; i++) {
1188 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1189
1190 if (i == GCHD)
1191 /*
1192 * GC head is special, do not look at it. Even if the
1193 * head contains something related to this inode, it is
1194 * a _copy_ of corresponding on-flash node which sits
1195 * somewhere else.
1196 */
1197 continue;
1198
1199 if (!wbuf_has_ino(wbuf, inode->i_ino))
1200 continue;
1201
1202 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1203 if (wbuf_has_ino(wbuf, inode->i_ino))
1204 err = ubifs_wbuf_sync_nolock(wbuf);
1205 mutex_unlock(&wbuf->io_mutex);
1206
1207 if (err) {
1208 ubifs_ro_mode(c, err);
1209 return err;
1210 }
1211 }
1212 return 0;
1213}
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 |= MS_RDONLY;
88 ubifs_warn("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("reading %d bytes from LEB %d:%d failed, error %d",
111 len, lnum, offs, err);
112 dbg_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, int dtype)
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, dtype);
127 else
128 err = dbg_leb_write(c, lnum, buf, offs, len, dtype);
129 if (err) {
130 ubifs_err("writing %d bytes to LEB %d:%d failed, error %d",
131 len, lnum, offs, err);
132 ubifs_ro_mode(c, err);
133 dbg_dump_stack();
134 }
135 return err;
136}
137
138int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len,
139 int dtype)
140{
141 int err;
142
143 ubifs_assert(!c->ro_media && !c->ro_mount);
144 if (c->ro_error)
145 return -EROFS;
146 if (!dbg_is_tst_rcvry(c))
147 err = ubi_leb_change(c->ubi, lnum, buf, len, dtype);
148 else
149 err = dbg_leb_change(c, lnum, buf, len, dtype);
150 if (err) {
151 ubifs_err("changing %d bytes in LEB %d failed, error %d",
152 len, lnum, err);
153 ubifs_ro_mode(c, err);
154 dbg_dump_stack();
155 }
156 return err;
157}
158
159int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
160{
161 int err;
162
163 ubifs_assert(!c->ro_media && !c->ro_mount);
164 if (c->ro_error)
165 return -EROFS;
166 if (!dbg_is_tst_rcvry(c))
167 err = ubi_leb_unmap(c->ubi, lnum);
168 else
169 err = dbg_leb_unmap(c, lnum);
170 if (err) {
171 ubifs_err("unmap LEB %d failed, error %d", lnum, err);
172 ubifs_ro_mode(c, err);
173 dbg_dump_stack();
174 }
175 return err;
176}
177
178int ubifs_leb_map(struct ubifs_info *c, int lnum, int dtype)
179{
180 int err;
181
182 ubifs_assert(!c->ro_media && !c->ro_mount);
183 if (c->ro_error)
184 return -EROFS;
185 if (!dbg_is_tst_rcvry(c))
186 err = ubi_leb_map(c->ubi, lnum, dtype);
187 else
188 err = dbg_leb_map(c, lnum, dtype);
189 if (err) {
190 ubifs_err("mapping LEB %d failed, error %d", lnum, err);
191 ubifs_ro_mode(c, err);
192 dbg_dump_stack();
193 }
194 return err;
195}
196
197int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
198{
199 int err;
200
201 err = ubi_is_mapped(c->ubi, lnum);
202 if (err < 0) {
203 ubifs_err("ubi_is_mapped failed for LEB %d, error %d",
204 lnum, err);
205 dbg_dump_stack();
206 }
207 return err;
208}
209
210/**
211 * ubifs_check_node - check node.
212 * @c: UBIFS file-system description object
213 * @buf: node to check
214 * @lnum: logical eraseblock number
215 * @offs: offset within the logical eraseblock
216 * @quiet: print no messages
217 * @must_chk_crc: indicates whether to always check the CRC
218 *
219 * This function checks node magic number and CRC checksum. This function also
220 * validates node length to prevent UBIFS from becoming crazy when an attacker
221 * feeds it a file-system image with incorrect nodes. For example, too large
222 * node length in the common header could cause UBIFS to read memory outside of
223 * allocated buffer when checking the CRC checksum.
224 *
225 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
226 * true, which is controlled by corresponding UBIFS mount option. However, if
227 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
228 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
229 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
230 * is checked. This is because during mounting or re-mounting from R/O mode to
231 * R/W mode we may read journal nodes (when replying the journal or doing the
232 * recovery) and the journal nodes may potentially be corrupted, so checking is
233 * required.
234 *
235 * This function returns zero in case of success and %-EUCLEAN in case of bad
236 * CRC or magic.
237 */
238int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
239 int offs, int quiet, int must_chk_crc)
240{
241 int err = -EINVAL, type, node_len;
242 uint32_t crc, node_crc, magic;
243 const struct ubifs_ch *ch = buf;
244
245 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
246 ubifs_assert(!(offs & 7) && offs < c->leb_size);
247
248 magic = le32_to_cpu(ch->magic);
249 if (magic != UBIFS_NODE_MAGIC) {
250 if (!quiet)
251 ubifs_err("bad magic %#08x, expected %#08x",
252 magic, UBIFS_NODE_MAGIC);
253 err = -EUCLEAN;
254 goto out;
255 }
256
257 type = ch->node_type;
258 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
259 if (!quiet)
260 ubifs_err("bad node type %d", type);
261 goto out;
262 }
263
264 node_len = le32_to_cpu(ch->len);
265 if (node_len + offs > c->leb_size)
266 goto out_len;
267
268 if (c->ranges[type].max_len == 0) {
269 if (node_len != c->ranges[type].len)
270 goto out_len;
271 } else if (node_len < c->ranges[type].min_len ||
272 node_len > c->ranges[type].max_len)
273 goto out_len;
274
275 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
276 !c->remounting_rw && c->no_chk_data_crc)
277 return 0;
278
279 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
280 node_crc = le32_to_cpu(ch->crc);
281 if (crc != node_crc) {
282 if (!quiet)
283 ubifs_err("bad CRC: calculated %#08x, read %#08x",
284 crc, node_crc);
285 err = -EUCLEAN;
286 goto out;
287 }
288
289 return 0;
290
291out_len:
292 if (!quiet)
293 ubifs_err("bad node length %d", node_len);
294out:
295 if (!quiet) {
296 ubifs_err("bad node at LEB %d:%d", lnum, offs);
297 dbg_dump_node(c, buf);
298 dbg_dump_stack();
299 }
300 return err;
301}
302
303/**
304 * ubifs_pad - pad flash space.
305 * @c: UBIFS file-system description object
306 * @buf: buffer to put padding to
307 * @pad: how many bytes to pad
308 *
309 * The flash media obliges us to write only in chunks of %c->min_io_size and
310 * when we have to write less data we add padding node to the write-buffer and
311 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
312 * media is being scanned. If the amount of wasted space is not enough to fit a
313 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
314 * pattern (%UBIFS_PADDING_BYTE).
315 *
316 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
317 * used.
318 */
319void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
320{
321 uint32_t crc;
322
323 ubifs_assert(pad >= 0 && !(pad & 7));
324
325 if (pad >= UBIFS_PAD_NODE_SZ) {
326 struct ubifs_ch *ch = buf;
327 struct ubifs_pad_node *pad_node = buf;
328
329 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
330 ch->node_type = UBIFS_PAD_NODE;
331 ch->group_type = UBIFS_NO_NODE_GROUP;
332 ch->padding[0] = ch->padding[1] = 0;
333 ch->sqnum = 0;
334 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
335 pad -= UBIFS_PAD_NODE_SZ;
336 pad_node->pad_len = cpu_to_le32(pad);
337 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
338 ch->crc = cpu_to_le32(crc);
339 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
340 } else if (pad > 0)
341 /* Too little space, padding node won't fit */
342 memset(buf, UBIFS_PADDING_BYTE, pad);
343}
344
345/**
346 * next_sqnum - get next sequence number.
347 * @c: UBIFS file-system description object
348 */
349static unsigned long long next_sqnum(struct ubifs_info *c)
350{
351 unsigned long long sqnum;
352
353 spin_lock(&c->cnt_lock);
354 sqnum = ++c->max_sqnum;
355 spin_unlock(&c->cnt_lock);
356
357 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
358 if (sqnum >= SQNUM_WATERMARK) {
359 ubifs_err("sequence number overflow %llu, end of life",
360 sqnum);
361 ubifs_ro_mode(c, -EINVAL);
362 }
363 ubifs_warn("running out of sequence numbers, end of life soon");
364 }
365
366 return sqnum;
367}
368
369/**
370 * ubifs_prepare_node - prepare node to be written to flash.
371 * @c: UBIFS file-system description object
372 * @node: the node to pad
373 * @len: node length
374 * @pad: if the buffer has to be padded
375 *
376 * This function prepares node at @node to be written to the media - it
377 * calculates node CRC, fills the common header, and adds proper padding up to
378 * the next minimum I/O unit if @pad is not zero.
379 */
380void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
381{
382 uint32_t crc;
383 struct ubifs_ch *ch = node;
384 unsigned long long sqnum = next_sqnum(c);
385
386 ubifs_assert(len >= UBIFS_CH_SZ);
387
388 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
389 ch->len = cpu_to_le32(len);
390 ch->group_type = UBIFS_NO_NODE_GROUP;
391 ch->sqnum = cpu_to_le64(sqnum);
392 ch->padding[0] = ch->padding[1] = 0;
393 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
394 ch->crc = cpu_to_le32(crc);
395
396 if (pad) {
397 len = ALIGN(len, 8);
398 pad = ALIGN(len, c->min_io_size) - len;
399 ubifs_pad(c, node + len, pad);
400 }
401}
402
403/**
404 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
405 * @c: UBIFS file-system description object
406 * @node: the node to pad
407 * @len: node length
408 * @last: indicates the last node of the group
409 *
410 * This function prepares node at @node to be written to the media - it
411 * calculates node CRC and fills the common header.
412 */
413void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
414{
415 uint32_t crc;
416 struct ubifs_ch *ch = node;
417 unsigned long long sqnum = next_sqnum(c);
418
419 ubifs_assert(len >= UBIFS_CH_SZ);
420
421 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
422 ch->len = cpu_to_le32(len);
423 if (last)
424 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
425 else
426 ch->group_type = UBIFS_IN_NODE_GROUP;
427 ch->sqnum = cpu_to_le64(sqnum);
428 ch->padding[0] = ch->padding[1] = 0;
429 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
430 ch->crc = cpu_to_le32(crc);
431}
432
433/**
434 * wbuf_timer_callback - write-buffer timer callback function.
435 * @data: timer data (write-buffer descriptor)
436 *
437 * This function is called when the write-buffer timer expires.
438 */
439static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
440{
441 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
442
443 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
444 wbuf->need_sync = 1;
445 wbuf->c->need_wbuf_sync = 1;
446 ubifs_wake_up_bgt(wbuf->c);
447 return HRTIMER_NORESTART;
448}
449
450/**
451 * new_wbuf_timer - start new write-buffer timer.
452 * @wbuf: write-buffer descriptor
453 */
454static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
455{
456 ubifs_assert(!hrtimer_active(&wbuf->timer));
457
458 if (wbuf->no_timer)
459 return;
460 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
461 dbg_jhead(wbuf->jhead),
462 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
463 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
464 USEC_PER_SEC));
465 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
466 HRTIMER_MODE_REL);
467}
468
469/**
470 * cancel_wbuf_timer - cancel write-buffer timer.
471 * @wbuf: write-buffer descriptor
472 */
473static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
474{
475 if (wbuf->no_timer)
476 return;
477 wbuf->need_sync = 0;
478 hrtimer_cancel(&wbuf->timer);
479}
480
481/**
482 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
483 * @wbuf: write-buffer to synchronize
484 *
485 * This function synchronizes write-buffer @buf and returns zero in case of
486 * success or a negative error code in case of failure.
487 *
488 * Note, although write-buffers are of @c->max_write_size, this function does
489 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
490 * if the write-buffer is only partially filled with data, only the used part
491 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
492 * This way we waste less space.
493 */
494int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
495{
496 struct ubifs_info *c = wbuf->c;
497 int err, dirt, sync_len;
498
499 cancel_wbuf_timer_nolock(wbuf);
500 if (!wbuf->used || wbuf->lnum == -1)
501 /* Write-buffer is empty or not seeked */
502 return 0;
503
504 dbg_io("LEB %d:%d, %d bytes, jhead %s",
505 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
506 ubifs_assert(!(wbuf->avail & 7));
507 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
508 ubifs_assert(wbuf->size >= c->min_io_size);
509 ubifs_assert(wbuf->size <= c->max_write_size);
510 ubifs_assert(wbuf->size % c->min_io_size == 0);
511 ubifs_assert(!c->ro_media && !c->ro_mount);
512 if (c->leb_size - wbuf->offs >= c->max_write_size)
513 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
514
515 if (c->ro_error)
516 return -EROFS;
517
518 /*
519 * Do not write whole write buffer but write only the minimum necessary
520 * amount of min. I/O units.
521 */
522 sync_len = ALIGN(wbuf->used, c->min_io_size);
523 dirt = sync_len - wbuf->used;
524 if (dirt)
525 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
526 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len,
527 wbuf->dtype);
528 if (err)
529 return err;
530
531 spin_lock(&wbuf->lock);
532 wbuf->offs += sync_len;
533 /*
534 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
535 * But our goal is to optimize writes and make sure we write in
536 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
537 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
538 * sure that @wbuf->offs + @wbuf->size is aligned to
539 * @c->max_write_size. This way we make sure that after next
540 * write-buffer flush we are again at the optimal offset (aligned to
541 * @c->max_write_size).
542 */
543 if (c->leb_size - wbuf->offs < c->max_write_size)
544 wbuf->size = c->leb_size - wbuf->offs;
545 else if (wbuf->offs & (c->max_write_size - 1))
546 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
547 else
548 wbuf->size = c->max_write_size;
549 wbuf->avail = wbuf->size;
550 wbuf->used = 0;
551 wbuf->next_ino = 0;
552 spin_unlock(&wbuf->lock);
553
554 if (wbuf->sync_callback)
555 err = wbuf->sync_callback(c, wbuf->lnum,
556 c->leb_size - wbuf->offs, dirt);
557 return err;
558}
559
560/**
561 * ubifs_wbuf_seek_nolock - seek write-buffer.
562 * @wbuf: write-buffer
563 * @lnum: logical eraseblock number to seek to
564 * @offs: logical eraseblock offset to seek to
565 * @dtype: data type
566 *
567 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
568 * The write-buffer has to be empty. Returns zero in case of success and a
569 * negative error code in case of failure.
570 */
571int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs,
572 int dtype)
573{
574 const struct ubifs_info *c = wbuf->c;
575
576 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
577 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
578 ubifs_assert(offs >= 0 && offs <= c->leb_size);
579 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
580 ubifs_assert(lnum != wbuf->lnum);
581 ubifs_assert(wbuf->used == 0);
582
583 spin_lock(&wbuf->lock);
584 wbuf->lnum = lnum;
585 wbuf->offs = offs;
586 if (c->leb_size - wbuf->offs < c->max_write_size)
587 wbuf->size = c->leb_size - wbuf->offs;
588 else if (wbuf->offs & (c->max_write_size - 1))
589 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
590 else
591 wbuf->size = c->max_write_size;
592 wbuf->avail = wbuf->size;
593 wbuf->used = 0;
594 spin_unlock(&wbuf->lock);
595 wbuf->dtype = dtype;
596
597 return 0;
598}
599
600/**
601 * ubifs_bg_wbufs_sync - synchronize write-buffers.
602 * @c: UBIFS file-system description object
603 *
604 * This function is called by background thread to synchronize write-buffers.
605 * Returns zero in case of success and a negative error code in case of
606 * failure.
607 */
608int ubifs_bg_wbufs_sync(struct ubifs_info *c)
609{
610 int err, i;
611
612 ubifs_assert(!c->ro_media && !c->ro_mount);
613 if (!c->need_wbuf_sync)
614 return 0;
615 c->need_wbuf_sync = 0;
616
617 if (c->ro_error) {
618 err = -EROFS;
619 goto out_timers;
620 }
621
622 dbg_io("synchronize");
623 for (i = 0; i < c->jhead_cnt; i++) {
624 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
625
626 cond_resched();
627
628 /*
629 * If the mutex is locked then wbuf is being changed, so
630 * synchronization is not necessary.
631 */
632 if (mutex_is_locked(&wbuf->io_mutex))
633 continue;
634
635 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
636 if (!wbuf->need_sync) {
637 mutex_unlock(&wbuf->io_mutex);
638 continue;
639 }
640
641 err = ubifs_wbuf_sync_nolock(wbuf);
642 mutex_unlock(&wbuf->io_mutex);
643 if (err) {
644 ubifs_err("cannot sync write-buffer, error %d", err);
645 ubifs_ro_mode(c, err);
646 goto out_timers;
647 }
648 }
649
650 return 0;
651
652out_timers:
653 /* Cancel all timers to prevent repeated errors */
654 for (i = 0; i < c->jhead_cnt; i++) {
655 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
656
657 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
658 cancel_wbuf_timer_nolock(wbuf);
659 mutex_unlock(&wbuf->io_mutex);
660 }
661 return err;
662}
663
664/**
665 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
666 * @wbuf: write-buffer
667 * @buf: node to write
668 * @len: node length
669 *
670 * This function writes data to flash via write-buffer @wbuf. This means that
671 * the last piece of the node won't reach the flash media immediately if it
672 * does not take whole max. write unit (@c->max_write_size). Instead, the node
673 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
674 * because more data are appended to the write-buffer).
675 *
676 * This function returns zero in case of success and a negative error code in
677 * case of failure. If the node cannot be written because there is no more
678 * space in this logical eraseblock, %-ENOSPC is returned.
679 */
680int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
681{
682 struct ubifs_info *c = wbuf->c;
683 int err, written, n, aligned_len = ALIGN(len, 8);
684
685 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
686 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
687 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
688 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
689 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
690 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
691 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
692 ubifs_assert(wbuf->size >= c->min_io_size);
693 ubifs_assert(wbuf->size <= c->max_write_size);
694 ubifs_assert(wbuf->size % c->min_io_size == 0);
695 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
696 ubifs_assert(!c->ro_media && !c->ro_mount);
697 ubifs_assert(!c->space_fixup);
698 if (c->leb_size - wbuf->offs >= c->max_write_size)
699 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
700
701 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
702 err = -ENOSPC;
703 goto out;
704 }
705
706 cancel_wbuf_timer_nolock(wbuf);
707
708 if (c->ro_error)
709 return -EROFS;
710
711 if (aligned_len <= wbuf->avail) {
712 /*
713 * The node is not very large and fits entirely within
714 * write-buffer.
715 */
716 memcpy(wbuf->buf + wbuf->used, buf, len);
717
718 if (aligned_len == wbuf->avail) {
719 dbg_io("flush jhead %s wbuf to LEB %d:%d",
720 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
721 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
722 wbuf->offs, wbuf->size,
723 wbuf->dtype);
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, wbuf->dtype);
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, wbuf->dtype);
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, wbuf->dtype);
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("cannot write %d bytes to LEB %d:%d, error %d",
843 len, wbuf->lnum, wbuf->offs, err);
844 dbg_dump_node(c, buf);
845 dbg_dump_stack();
846 dbg_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 * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
858 *
859 * This function automatically fills node magic number, assigns sequence
860 * number, and calculates node CRC checksum. The length of the @buf buffer has
861 * to be aligned to the minimal I/O unit size. This function automatically
862 * appends padding node and padding bytes if needed. Returns zero in case of
863 * success and a negative error code in case of failure.
864 */
865int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
866 int offs, int dtype)
867{
868 int err, buf_len = ALIGN(len, c->min_io_size);
869
870 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
871 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
872 buf_len);
873 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
874 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
875 ubifs_assert(!c->ro_media && !c->ro_mount);
876 ubifs_assert(!c->space_fixup);
877
878 if (c->ro_error)
879 return -EROFS;
880
881 ubifs_prepare_node(c, buf, len, 1);
882 err = ubifs_leb_write(c, lnum, buf, offs, buf_len, dtype);
883 if (err)
884 dbg_dump_node(c, buf);
885
886 return err;
887}
888
889/**
890 * ubifs_read_node_wbuf - read node from the media or write-buffer.
891 * @wbuf: wbuf to check for un-written data
892 * @buf: buffer to read to
893 * @type: node type
894 * @len: node length
895 * @lnum: logical eraseblock number
896 * @offs: offset within the logical eraseblock
897 *
898 * This function reads a node of known type and length, checks it and stores
899 * in @buf. If the node partially or fully sits in the write-buffer, this
900 * function takes data from the buffer, otherwise it reads the flash media.
901 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
902 * error code in case of failure.
903 */
904int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
905 int lnum, int offs)
906{
907 const struct ubifs_info *c = wbuf->c;
908 int err, rlen, overlap;
909 struct ubifs_ch *ch = buf;
910
911 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
912 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
913 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
914 ubifs_assert(!(offs & 7) && offs < c->leb_size);
915 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
916
917 spin_lock(&wbuf->lock);
918 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
919 if (!overlap) {
920 /* We may safely unlock the write-buffer and read the data */
921 spin_unlock(&wbuf->lock);
922 return ubifs_read_node(c, buf, type, len, lnum, offs);
923 }
924
925 /* Don't read under wbuf */
926 rlen = wbuf->offs - offs;
927 if (rlen < 0)
928 rlen = 0;
929
930 /* Copy the rest from the write-buffer */
931 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
932 spin_unlock(&wbuf->lock);
933
934 if (rlen > 0) {
935 /* Read everything that goes before write-buffer */
936 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
937 if (err && err != -EBADMSG)
938 return err;
939 }
940
941 if (type != ch->node_type) {
942 ubifs_err("bad node type (%d but expected %d)",
943 ch->node_type, type);
944 goto out;
945 }
946
947 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
948 if (err) {
949 ubifs_err("expected node type %d", type);
950 return err;
951 }
952
953 rlen = le32_to_cpu(ch->len);
954 if (rlen != len) {
955 ubifs_err("bad node length %d, expected %d", rlen, len);
956 goto out;
957 }
958
959 return 0;
960
961out:
962 ubifs_err("bad node at LEB %d:%d", lnum, offs);
963 dbg_dump_node(c, buf);
964 dbg_dump_stack();
965 return -EINVAL;
966}
967
968/**
969 * ubifs_read_node - read node.
970 * @c: UBIFS file-system description object
971 * @buf: buffer to read to
972 * @type: node type
973 * @len: node length (not aligned)
974 * @lnum: logical eraseblock number
975 * @offs: offset within the logical eraseblock
976 *
977 * This function reads a node of known type and and length, checks it and
978 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
979 * and a negative error code in case of failure.
980 */
981int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
982 int lnum, int offs)
983{
984 int err, l;
985 struct ubifs_ch *ch = buf;
986
987 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
988 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
989 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
990 ubifs_assert(!(offs & 7) && offs < c->leb_size);
991 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
992
993 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
994 if (err && err != -EBADMSG)
995 return err;
996
997 if (type != ch->node_type) {
998 ubifs_err("bad node type (%d but expected %d)",
999 ch->node_type, type);
1000 goto out;
1001 }
1002
1003 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1004 if (err) {
1005 ubifs_err("expected node type %d", type);
1006 return err;
1007 }
1008
1009 l = le32_to_cpu(ch->len);
1010 if (l != len) {
1011 ubifs_err("bad node length %d, expected %d", l, len);
1012 goto out;
1013 }
1014
1015 return 0;
1016
1017out:
1018 ubifs_err("bad node at LEB %d:%d, LEB mapping status %d", lnum, offs,
1019 ubi_is_mapped(c->ubi, lnum));
1020 dbg_dump_node(c, buf);
1021 dbg_dump_stack();
1022 return -EINVAL;
1023}
1024
1025/**
1026 * ubifs_wbuf_init - initialize write-buffer.
1027 * @c: UBIFS file-system description object
1028 * @wbuf: write-buffer to initialize
1029 *
1030 * This function initializes write-buffer. Returns zero in case of success
1031 * %-ENOMEM in case of failure.
1032 */
1033int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1034{
1035 size_t size;
1036
1037 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1038 if (!wbuf->buf)
1039 return -ENOMEM;
1040
1041 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1042 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1043 if (!wbuf->inodes) {
1044 kfree(wbuf->buf);
1045 wbuf->buf = NULL;
1046 return -ENOMEM;
1047 }
1048
1049 wbuf->used = 0;
1050 wbuf->lnum = wbuf->offs = -1;
1051 /*
1052 * If the LEB starts at the max. write size aligned address, then
1053 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1054 * set it to something smaller so that it ends at the closest max.
1055 * write size boundary.
1056 */
1057 size = c->max_write_size - (c->leb_start % c->max_write_size);
1058 wbuf->avail = wbuf->size = size;
1059 wbuf->dtype = UBI_UNKNOWN;
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 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1069 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1070 wbuf->delta *= 1000000000ULL;
1071 ubifs_assert(wbuf->delta <= ULONG_MAX);
1072 return 0;
1073}
1074
1075/**
1076 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1077 * @wbuf: the write-buffer where to add
1078 * @inum: the inode number
1079 *
1080 * This function adds an inode number to the inode array of the write-buffer.
1081 */
1082void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1083{
1084 if (!wbuf->buf)
1085 /* NOR flash or something similar */
1086 return;
1087
1088 spin_lock(&wbuf->lock);
1089 if (wbuf->used)
1090 wbuf->inodes[wbuf->next_ino++] = inum;
1091 spin_unlock(&wbuf->lock);
1092}
1093
1094/**
1095 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1096 * @wbuf: the write-buffer
1097 * @inum: the inode number
1098 *
1099 * This function returns with %1 if the write-buffer contains some data from the
1100 * given inode otherwise it returns with %0.
1101 */
1102static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1103{
1104 int i, ret = 0;
1105
1106 spin_lock(&wbuf->lock);
1107 for (i = 0; i < wbuf->next_ino; i++)
1108 if (inum == wbuf->inodes[i]) {
1109 ret = 1;
1110 break;
1111 }
1112 spin_unlock(&wbuf->lock);
1113
1114 return ret;
1115}
1116
1117/**
1118 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1119 * @c: UBIFS file-system description object
1120 * @inode: inode to synchronize
1121 *
1122 * This function synchronizes write-buffers which contain nodes belonging to
1123 * @inode. Returns zero in case of success and a negative error code in case of
1124 * failure.
1125 */
1126int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1127{
1128 int i, err = 0;
1129
1130 for (i = 0; i < c->jhead_cnt; i++) {
1131 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1132
1133 if (i == GCHD)
1134 /*
1135 * GC head is special, do not look at it. Even if the
1136 * head contains something related to this inode, it is
1137 * a _copy_ of corresponding on-flash node which sits
1138 * somewhere else.
1139 */
1140 continue;
1141
1142 if (!wbuf_has_ino(wbuf, inode->i_ino))
1143 continue;
1144
1145 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1146 if (wbuf_has_ino(wbuf, inode->i_ino))
1147 err = ubifs_wbuf_sync_nolock(wbuf);
1148 mutex_unlock(&wbuf->io_mutex);
1149
1150 if (err) {
1151 ubifs_ro_mode(c, err);
1152 return err;
1153 }
1154 }
1155 return 0;
1156}