common: Drop init.h from common header
[J-u-boot.git] / fs / ubifs / io.c
CommitLineData
83d290c5 1// SPDX-License-Identifier: GPL-2.0+
9eefe2a2
SR
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 *
9eefe2a2
SR
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 *
ff94bc40
HS
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 *
9eefe2a2
SR
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 *
ff94bc40
HS
57 * All UBIFS nodes are protected by CRC checksums and UBIFS checks CRC when
58 * they are read from the flash media.
9eefe2a2
SR
59 */
60
ff94bc40 61#ifndef __UBOOT__
691d719d 62#include <init.h>
61b29b82 63#include <dm/devres.h>
ff94bc40
HS
64#include <linux/crc32.h>
65#include <linux/slab.h>
3db71108 66#include <u-boot/crc.h>
ff94bc40
HS
67#else
68#include <linux/compat.h>
69#include <linux/err.h>
70#endif
9eefe2a2
SR
71#include "ubifs.h"
72
73/**
74 * ubifs_ro_mode - switch UBIFS to read read-only mode.
75 * @c: UBIFS file-system description object
76 * @err: error code which is the reason of switching to R/O mode
77 */
78void ubifs_ro_mode(struct ubifs_info *c, int err)
79{
ff94bc40
HS
80 if (!c->ro_error) {
81 c->ro_error = 1;
9eefe2a2 82 c->no_chk_data_crc = 0;
ff94bc40 83 c->vfs_sb->s_flags |= MS_RDONLY;
0195a7bb 84 ubifs_warn(c, "switched to read-only mode, error %d", err);
ff94bc40
HS
85 dump_stack();
86 }
87}
88
89/*
90 * Below are simple wrappers over UBI I/O functions which include some
91 * additional checks and UBIFS debugging stuff. See corresponding UBI function
92 * for more information.
93 */
94
95int ubifs_leb_read(const struct ubifs_info *c, int lnum, void *buf, int offs,
96 int len, int even_ebadmsg)
97{
98 int err;
99
100 err = ubi_read(c->ubi, lnum, buf, offs, len);
101 /*
102 * In case of %-EBADMSG print the error message only if the
103 * @even_ebadmsg is true.
104 */
105 if (err && (err != -EBADMSG || even_ebadmsg)) {
0195a7bb 106 ubifs_err(c, "reading %d bytes from LEB %d:%d failed, error %d",
ff94bc40
HS
107 len, lnum, offs, err);
108 dump_stack();
109 }
110 return err;
111}
112
113int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
114 int len)
115{
116 int err;
117
118 ubifs_assert(!c->ro_media && !c->ro_mount);
119 if (c->ro_error)
120 return -EROFS;
121 if (!dbg_is_tst_rcvry(c))
122 err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
0195a7bb 123#ifndef __UBOOT__
ff94bc40
HS
124 else
125 err = dbg_leb_write(c, lnum, buf, offs, len);
0195a7bb 126#endif
ff94bc40 127 if (err) {
0195a7bb 128 ubifs_err(c, "writing %d bytes to LEB %d:%d failed, error %d",
ff94bc40
HS
129 len, lnum, offs, err);
130 ubifs_ro_mode(c, err);
131 dump_stack();
132 }
133 return err;
134}
135
136int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
137{
138 int err;
139
140 ubifs_assert(!c->ro_media && !c->ro_mount);
141 if (c->ro_error)
142 return -EROFS;
143 if (!dbg_is_tst_rcvry(c))
144 err = ubi_leb_change(c->ubi, lnum, buf, len);
0195a7bb 145#ifndef __UBOOT__
ff94bc40
HS
146 else
147 err = dbg_leb_change(c, lnum, buf, len);
0195a7bb 148#endif
ff94bc40 149 if (err) {
0195a7bb 150 ubifs_err(c, "changing %d bytes in LEB %d failed, error %d",
ff94bc40
HS
151 len, lnum, err);
152 ubifs_ro_mode(c, err);
153 dump_stack();
9eefe2a2 154 }
ff94bc40
HS
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);
0195a7bb 167#ifndef __UBOOT__
ff94bc40
HS
168 else
169 err = dbg_leb_unmap(c, lnum);
0195a7bb 170#endif
ff94bc40 171 if (err) {
0195a7bb 172 ubifs_err(c, "unmap LEB %d failed, error %d", lnum, err);
ff94bc40
HS
173 ubifs_ro_mode(c, err);
174 dump_stack();
175 }
176 return err;
177}
178
179int ubifs_leb_map(struct ubifs_info *c, int lnum)
180{
181 int err;
182
183 ubifs_assert(!c->ro_media && !c->ro_mount);
184 if (c->ro_error)
185 return -EROFS;
186 if (!dbg_is_tst_rcvry(c))
187 err = ubi_leb_map(c->ubi, lnum);
0195a7bb 188#ifndef __UBOOT__
ff94bc40
HS
189 else
190 err = dbg_leb_map(c, lnum);
0195a7bb 191#endif
ff94bc40 192 if (err) {
0195a7bb 193 ubifs_err(c, "mapping LEB %d failed, error %d", lnum, err);
ff94bc40
HS
194 ubifs_ro_mode(c, err);
195 dump_stack();
196 }
197 return err;
198}
199
200int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
201{
202 int err;
203
204 err = ubi_is_mapped(c->ubi, lnum);
205 if (err < 0) {
0195a7bb 206 ubifs_err(c, "ubi_is_mapped failed for LEB %d, error %d",
ff94bc40
HS
207 lnum, err);
208 dump_stack();
209 }
210 return err;
9eefe2a2
SR
211}
212
213/**
214 * ubifs_check_node - check node.
215 * @c: UBIFS file-system description object
216 * @buf: node to check
217 * @lnum: logical eraseblock number
218 * @offs: offset within the logical eraseblock
219 * @quiet: print no messages
220 * @must_chk_crc: indicates whether to always check the CRC
221 *
222 * This function checks node magic number and CRC checksum. This function also
223 * validates node length to prevent UBIFS from becoming crazy when an attacker
224 * feeds it a file-system image with incorrect nodes. For example, too large
225 * node length in the common header could cause UBIFS to read memory outside of
226 * allocated buffer when checking the CRC checksum.
227 *
228 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
229 * true, which is controlled by corresponding UBIFS mount option. However, if
230 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
ff94bc40
HS
231 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
232 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
233 * is checked. This is because during mounting or re-mounting from R/O mode to
234 * R/W mode we may read journal nodes (when replying the journal or doing the
235 * recovery) and the journal nodes may potentially be corrupted, so checking is
236 * required.
9eefe2a2
SR
237 *
238 * This function returns zero in case of success and %-EUCLEAN in case of bad
239 * CRC or magic.
240 */
241int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
242 int offs, int quiet, int must_chk_crc)
243{
244 int err = -EINVAL, type, node_len;
245 uint32_t crc, node_crc, magic;
246 const struct ubifs_ch *ch = buf;
247
248 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
249 ubifs_assert(!(offs & 7) && offs < c->leb_size);
250
251 magic = le32_to_cpu(ch->magic);
252 if (magic != UBIFS_NODE_MAGIC) {
253 if (!quiet)
0195a7bb 254 ubifs_err(c, "bad magic %#08x, expected %#08x",
9eefe2a2
SR
255 magic, UBIFS_NODE_MAGIC);
256 err = -EUCLEAN;
257 goto out;
258 }
259
260 type = ch->node_type;
261 if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
262 if (!quiet)
0195a7bb 263 ubifs_err(c, "bad node type %d", type);
9eefe2a2
SR
264 goto out;
265 }
266
267 node_len = le32_to_cpu(ch->len);
268 if (node_len + offs > c->leb_size)
269 goto out_len;
270
271 if (c->ranges[type].max_len == 0) {
272 if (node_len != c->ranges[type].len)
273 goto out_len;
274 } else if (node_len < c->ranges[type].min_len ||
275 node_len > c->ranges[type].max_len)
276 goto out_len;
277
ff94bc40
HS
278 if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
279 !c->remounting_rw && c->no_chk_data_crc)
9eefe2a2
SR
280 return 0;
281
282 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
283 node_crc = le32_to_cpu(ch->crc);
284 if (crc != node_crc) {
285 if (!quiet)
0195a7bb 286 ubifs_err(c, "bad CRC: calculated %#08x, read %#08x",
9eefe2a2
SR
287 crc, node_crc);
288 err = -EUCLEAN;
289 goto out;
290 }
291
292 return 0;
293
294out_len:
295 if (!quiet)
0195a7bb 296 ubifs_err(c, "bad node length %d", node_len);
9eefe2a2
SR
297out:
298 if (!quiet) {
0195a7bb 299 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
ff94bc40
HS
300 ubifs_dump_node(c, buf);
301 dump_stack();
9eefe2a2
SR
302 }
303 return err;
304}
305
306/**
307 * ubifs_pad - pad flash space.
308 * @c: UBIFS file-system description object
309 * @buf: buffer to put padding to
310 * @pad: how many bytes to pad
311 *
312 * The flash media obliges us to write only in chunks of %c->min_io_size and
313 * when we have to write less data we add padding node to the write-buffer and
314 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
315 * media is being scanned. If the amount of wasted space is not enough to fit a
316 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
317 * pattern (%UBIFS_PADDING_BYTE).
318 *
319 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
320 * used.
321 */
322void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
323{
324 uint32_t crc;
325
326 ubifs_assert(pad >= 0 && !(pad & 7));
327
328 if (pad >= UBIFS_PAD_NODE_SZ) {
329 struct ubifs_ch *ch = buf;
330 struct ubifs_pad_node *pad_node = buf;
331
332 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
333 ch->node_type = UBIFS_PAD_NODE;
334 ch->group_type = UBIFS_NO_NODE_GROUP;
335 ch->padding[0] = ch->padding[1] = 0;
336 ch->sqnum = 0;
337 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
338 pad -= UBIFS_PAD_NODE_SZ;
339 pad_node->pad_len = cpu_to_le32(pad);
340 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
341 ch->crc = cpu_to_le32(crc);
342 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
343 } else if (pad > 0)
344 /* Too little space, padding node won't fit */
345 memset(buf, UBIFS_PADDING_BYTE, pad);
346}
347
348/**
349 * next_sqnum - get next sequence number.
350 * @c: UBIFS file-system description object
351 */
352static unsigned long long next_sqnum(struct ubifs_info *c)
353{
354 unsigned long long sqnum;
355
356 spin_lock(&c->cnt_lock);
357 sqnum = ++c->max_sqnum;
358 spin_unlock(&c->cnt_lock);
359
360 if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
361 if (sqnum >= SQNUM_WATERMARK) {
0195a7bb 362 ubifs_err(c, "sequence number overflow %llu, end of life",
9eefe2a2
SR
363 sqnum);
364 ubifs_ro_mode(c, -EINVAL);
365 }
0195a7bb 366 ubifs_warn(c, "running out of sequence numbers, end of life soon");
9eefe2a2
SR
367 }
368
369 return sqnum;
370}
371
372/**
373 * ubifs_prepare_node - prepare node to be written to flash.
374 * @c: UBIFS file-system description object
375 * @node: the node to pad
376 * @len: node length
377 * @pad: if the buffer has to be padded
378 *
379 * This function prepares node at @node to be written to the media - it
380 * calculates node CRC, fills the common header, and adds proper padding up to
381 * the next minimum I/O unit if @pad is not zero.
382 */
383void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
384{
385 uint32_t crc;
386 struct ubifs_ch *ch = node;
387 unsigned long long sqnum = next_sqnum(c);
388
389 ubifs_assert(len >= UBIFS_CH_SZ);
390
391 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
392 ch->len = cpu_to_le32(len);
393 ch->group_type = UBIFS_NO_NODE_GROUP;
394 ch->sqnum = cpu_to_le64(sqnum);
395 ch->padding[0] = ch->padding[1] = 0;
396 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
397 ch->crc = cpu_to_le32(crc);
398
399 if (pad) {
400 len = ALIGN(len, 8);
401 pad = ALIGN(len, c->min_io_size) - len;
402 ubifs_pad(c, node + len, pad);
403 }
404}
405
ff94bc40
HS
406/**
407 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
408 * @c: UBIFS file-system description object
409 * @node: the node to pad
410 * @len: node length
411 * @last: indicates the last node of the group
412 *
413 * This function prepares node at @node to be written to the media - it
414 * calculates node CRC and fills the common header.
415 */
416void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
417{
418 uint32_t crc;
419 struct ubifs_ch *ch = node;
420 unsigned long long sqnum = next_sqnum(c);
421
422 ubifs_assert(len >= UBIFS_CH_SZ);
423
424 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
425 ch->len = cpu_to_le32(len);
426 if (last)
427 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
428 else
429 ch->group_type = UBIFS_IN_NODE_GROUP;
430 ch->sqnum = cpu_to_le64(sqnum);
431 ch->padding[0] = ch->padding[1] = 0;
432 crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
433 ch->crc = cpu_to_le32(crc);
434}
435
436#ifndef __UBOOT__
437/**
438 * wbuf_timer_callback - write-buffer timer callback function.
0195a7bb 439 * @timer: timer data (write-buffer descriptor)
ff94bc40
HS
440 *
441 * This function is called when the write-buffer timer expires.
442 */
443static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
444{
445 struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
446
447 dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
448 wbuf->need_sync = 1;
449 wbuf->c->need_wbuf_sync = 1;
450 ubifs_wake_up_bgt(wbuf->c);
451 return HRTIMER_NORESTART;
452}
453
454/**
455 * new_wbuf_timer - start new write-buffer timer.
456 * @wbuf: write-buffer descriptor
457 */
458static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
459{
460 ubifs_assert(!hrtimer_active(&wbuf->timer));
461
462 if (wbuf->no_timer)
463 return;
464 dbg_io("set timer for jhead %s, %llu-%llu millisecs",
465 dbg_jhead(wbuf->jhead),
466 div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
467 div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
468 USEC_PER_SEC));
469 hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
470 HRTIMER_MODE_REL);
471}
472#endif
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#ifndef __UBOOT__
484 hrtimer_cancel(&wbuf->timer);
485#endif
486}
487
488/**
489 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
490 * @wbuf: write-buffer to synchronize
491 *
492 * This function synchronizes write-buffer @buf and returns zero in case of
493 * success or a negative error code in case of failure.
494 *
495 * Note, although write-buffers are of @c->max_write_size, this function does
496 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
497 * if the write-buffer is only partially filled with data, only the used part
498 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
499 * This way we waste less space.
500 */
501int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
502{
503 struct ubifs_info *c = wbuf->c;
504 int err, dirt, sync_len;
505
506 cancel_wbuf_timer_nolock(wbuf);
507 if (!wbuf->used || wbuf->lnum == -1)
508 /* Write-buffer is empty or not seeked */
509 return 0;
510
511 dbg_io("LEB %d:%d, %d bytes, jhead %s",
512 wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
513 ubifs_assert(!(wbuf->avail & 7));
514 ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
515 ubifs_assert(wbuf->size >= c->min_io_size);
516 ubifs_assert(wbuf->size <= c->max_write_size);
517 ubifs_assert(wbuf->size % c->min_io_size == 0);
518 ubifs_assert(!c->ro_media && !c->ro_mount);
519 if (c->leb_size - wbuf->offs >= c->max_write_size)
520 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
521
522 if (c->ro_error)
523 return -EROFS;
524
525 /*
526 * Do not write whole write buffer but write only the minimum necessary
527 * amount of min. I/O units.
528 */
529 sync_len = ALIGN(wbuf->used, c->min_io_size);
530 dirt = sync_len - wbuf->used;
531 if (dirt)
532 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
533 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
534 if (err)
535 return err;
536
537 spin_lock(&wbuf->lock);
538 wbuf->offs += sync_len;
539 /*
540 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
541 * But our goal is to optimize writes and make sure we write in
542 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
543 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
544 * sure that @wbuf->offs + @wbuf->size is aligned to
545 * @c->max_write_size. This way we make sure that after next
546 * write-buffer flush we are again at the optimal offset (aligned to
547 * @c->max_write_size).
548 */
549 if (c->leb_size - wbuf->offs < c->max_write_size)
550 wbuf->size = c->leb_size - wbuf->offs;
551 else if (wbuf->offs & (c->max_write_size - 1))
552 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
553 else
554 wbuf->size = c->max_write_size;
555 wbuf->avail = wbuf->size;
556 wbuf->used = 0;
557 wbuf->next_ino = 0;
558 spin_unlock(&wbuf->lock);
559
560 if (wbuf->sync_callback)
561 err = wbuf->sync_callback(c, wbuf->lnum,
562 c->leb_size - wbuf->offs, dirt);
563 return err;
564}
565
566/**
567 * ubifs_wbuf_seek_nolock - seek write-buffer.
568 * @wbuf: write-buffer
569 * @lnum: logical eraseblock number to seek to
570 * @offs: logical eraseblock offset to seek to
571 *
572 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
573 * The write-buffer has to be empty. Returns zero in case of success and a
574 * negative error code in case of failure.
575 */
576int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
577{
578 const struct ubifs_info *c = wbuf->c;
579
580 dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
581 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
582 ubifs_assert(offs >= 0 && offs <= c->leb_size);
583 ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
584 ubifs_assert(lnum != wbuf->lnum);
585 ubifs_assert(wbuf->used == 0);
586
587 spin_lock(&wbuf->lock);
588 wbuf->lnum = lnum;
589 wbuf->offs = offs;
590 if (c->leb_size - wbuf->offs < c->max_write_size)
591 wbuf->size = c->leb_size - wbuf->offs;
592 else if (wbuf->offs & (c->max_write_size - 1))
593 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
594 else
595 wbuf->size = c->max_write_size;
596 wbuf->avail = wbuf->size;
597 wbuf->used = 0;
598 spin_unlock(&wbuf->lock);
599
600 return 0;
601}
602
603#ifndef __UBOOT__
604/**
605 * ubifs_bg_wbufs_sync - synchronize write-buffers.
606 * @c: UBIFS file-system description object
607 *
608 * This function is called by background thread to synchronize write-buffers.
609 * Returns zero in case of success and a negative error code in case of
610 * failure.
611 */
612int ubifs_bg_wbufs_sync(struct ubifs_info *c)
613{
614 int err, i;
615
616 ubifs_assert(!c->ro_media && !c->ro_mount);
617 if (!c->need_wbuf_sync)
618 return 0;
619 c->need_wbuf_sync = 0;
620
621 if (c->ro_error) {
622 err = -EROFS;
623 goto out_timers;
624 }
625
626 dbg_io("synchronize");
627 for (i = 0; i < c->jhead_cnt; i++) {
628 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
629
630 cond_resched();
631
632 /*
633 * If the mutex is locked then wbuf is being changed, so
634 * synchronization is not necessary.
635 */
636 if (mutex_is_locked(&wbuf->io_mutex))
637 continue;
638
639 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
640 if (!wbuf->need_sync) {
641 mutex_unlock(&wbuf->io_mutex);
642 continue;
643 }
644
645 err = ubifs_wbuf_sync_nolock(wbuf);
646 mutex_unlock(&wbuf->io_mutex);
647 if (err) {
0195a7bb 648 ubifs_err(c, "cannot sync write-buffer, error %d", err);
ff94bc40
HS
649 ubifs_ro_mode(c, err);
650 goto out_timers;
651 }
652 }
653
654 return 0;
655
656out_timers:
657 /* Cancel all timers to prevent repeated errors */
658 for (i = 0; i < c->jhead_cnt; i++) {
659 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
660
661 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
662 cancel_wbuf_timer_nolock(wbuf);
663 mutex_unlock(&wbuf->io_mutex);
664 }
665 return err;
666}
667
668/**
669 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
670 * @wbuf: write-buffer
671 * @buf: node to write
672 * @len: node length
673 *
674 * This function writes data to flash via write-buffer @wbuf. This means that
675 * the last piece of the node won't reach the flash media immediately if it
676 * does not take whole max. write unit (@c->max_write_size). Instead, the node
677 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
678 * because more data are appended to the write-buffer).
679 *
680 * This function returns zero in case of success and a negative error code in
681 * case of failure. If the node cannot be written because there is no more
682 * space in this logical eraseblock, %-ENOSPC is returned.
683 */
684int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
685{
686 struct ubifs_info *c = wbuf->c;
687 int err, written, n, aligned_len = ALIGN(len, 8);
688
689 dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
690 dbg_ntype(((struct ubifs_ch *)buf)->node_type),
691 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
692 ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
693 ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
694 ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
695 ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
696 ubifs_assert(wbuf->size >= c->min_io_size);
697 ubifs_assert(wbuf->size <= c->max_write_size);
698 ubifs_assert(wbuf->size % c->min_io_size == 0);
699 ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
700 ubifs_assert(!c->ro_media && !c->ro_mount);
701 ubifs_assert(!c->space_fixup);
702 if (c->leb_size - wbuf->offs >= c->max_write_size)
703 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
704
705 if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
706 err = -ENOSPC;
707 goto out;
708 }
709
710 cancel_wbuf_timer_nolock(wbuf);
711
712 if (c->ro_error)
713 return -EROFS;
714
715 if (aligned_len <= wbuf->avail) {
716 /*
717 * The node is not very large and fits entirely within
718 * write-buffer.
719 */
720 memcpy(wbuf->buf + wbuf->used, buf, len);
721
722 if (aligned_len == wbuf->avail) {
723 dbg_io("flush jhead %s wbuf to LEB %d:%d",
724 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
725 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
726 wbuf->offs, wbuf->size);
727 if (err)
728 goto out;
729
730 spin_lock(&wbuf->lock);
731 wbuf->offs += wbuf->size;
732 if (c->leb_size - wbuf->offs >= c->max_write_size)
733 wbuf->size = c->max_write_size;
734 else
735 wbuf->size = c->leb_size - wbuf->offs;
736 wbuf->avail = wbuf->size;
737 wbuf->used = 0;
738 wbuf->next_ino = 0;
739 spin_unlock(&wbuf->lock);
740 } else {
741 spin_lock(&wbuf->lock);
742 wbuf->avail -= aligned_len;
743 wbuf->used += aligned_len;
744 spin_unlock(&wbuf->lock);
745 }
746
747 goto exit;
748 }
749
750 written = 0;
751
752 if (wbuf->used) {
753 /*
754 * The node is large enough and does not fit entirely within
755 * current available space. We have to fill and flush
756 * write-buffer and switch to the next max. write unit.
757 */
758 dbg_io("flush jhead %s wbuf to LEB %d:%d",
759 dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
760 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
761 err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
762 wbuf->size);
763 if (err)
764 goto out;
765
766 wbuf->offs += wbuf->size;
767 len -= wbuf->avail;
768 aligned_len -= wbuf->avail;
769 written += wbuf->avail;
770 } else if (wbuf->offs & (c->max_write_size - 1)) {
771 /*
772 * The write-buffer offset is not aligned to
773 * @c->max_write_size and @wbuf->size is less than
774 * @c->max_write_size. Write @wbuf->size bytes to make sure the
775 * following writes are done in optimal @c->max_write_size
776 * chunks.
777 */
778 dbg_io("write %d bytes to LEB %d:%d",
779 wbuf->size, wbuf->lnum, wbuf->offs);
780 err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
781 wbuf->size);
782 if (err)
783 goto out;
784
785 wbuf->offs += wbuf->size;
786 len -= wbuf->size;
787 aligned_len -= wbuf->size;
788 written += wbuf->size;
789 }
790
791 /*
792 * The remaining data may take more whole max. write units, so write the
793 * remains multiple to max. write unit size directly to the flash media.
794 * We align node length to 8-byte boundary because we anyway flash wbuf
795 * if the remaining space is less than 8 bytes.
796 */
797 n = aligned_len >> c->max_write_shift;
798 if (n) {
799 n <<= c->max_write_shift;
800 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
801 wbuf->offs);
802 err = ubifs_leb_write(c, wbuf->lnum, buf + written,
803 wbuf->offs, n);
804 if (err)
805 goto out;
806 wbuf->offs += n;
807 aligned_len -= n;
808 len -= n;
809 written += n;
810 }
811
812 spin_lock(&wbuf->lock);
813 if (aligned_len)
814 /*
815 * And now we have what's left and what does not take whole
816 * max. write unit, so write it to the write-buffer and we are
817 * done.
818 */
819 memcpy(wbuf->buf, buf + written, len);
820
821 if (c->leb_size - wbuf->offs >= c->max_write_size)
822 wbuf->size = c->max_write_size;
823 else
824 wbuf->size = c->leb_size - wbuf->offs;
825 wbuf->avail = wbuf->size - aligned_len;
826 wbuf->used = aligned_len;
827 wbuf->next_ino = 0;
828 spin_unlock(&wbuf->lock);
829
830exit:
831 if (wbuf->sync_callback) {
832 int free = c->leb_size - wbuf->offs - wbuf->used;
833
834 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
835 if (err)
836 goto out;
837 }
838
839 if (wbuf->used)
840 new_wbuf_timer_nolock(wbuf);
841
842 return 0;
843
844out:
0195a7bb 845 ubifs_err(c, "cannot write %d bytes to LEB %d:%d, error %d",
ff94bc40
HS
846 len, wbuf->lnum, wbuf->offs, err);
847 ubifs_dump_node(c, buf);
848 dump_stack();
849 ubifs_dump_leb(c, wbuf->lnum);
850 return err;
851}
852
853/**
854 * ubifs_write_node - write node to the media.
855 * @c: UBIFS file-system description object
856 * @buf: the node to write
857 * @len: node length
858 * @lnum: logical eraseblock number
859 * @offs: offset within the logical eraseblock
860 *
861 * This function automatically fills node magic number, assigns sequence
862 * number, and calculates node CRC checksum. The length of the @buf buffer has
863 * to be aligned to the minimal I/O unit size. This function automatically
864 * appends padding node and padding bytes if needed. Returns zero in case of
865 * success and a negative error code in case of failure.
866 */
867int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
868 int offs)
869{
870 int err, buf_len = ALIGN(len, c->min_io_size);
871
872 dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
873 lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
874 buf_len);
875 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
876 ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
877 ubifs_assert(!c->ro_media && !c->ro_mount);
878 ubifs_assert(!c->space_fixup);
879
880 if (c->ro_error)
881 return -EROFS;
882
883 ubifs_prepare_node(c, buf, len, 1);
884 err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
885 if (err)
886 ubifs_dump_node(c, buf);
887
888 return err;
889}
890#endif
891
892/**
893 * ubifs_read_node_wbuf - read node from the media or write-buffer.
894 * @wbuf: wbuf to check for un-written data
895 * @buf: buffer to read to
896 * @type: node type
897 * @len: node length
898 * @lnum: logical eraseblock number
899 * @offs: offset within the logical eraseblock
900 *
901 * This function reads a node of known type and length, checks it and stores
902 * in @buf. If the node partially or fully sits in the write-buffer, this
903 * function takes data from the buffer, otherwise it reads the flash media.
904 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
905 * error code in case of failure.
906 */
907int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
908 int lnum, int offs)
909{
910 const struct ubifs_info *c = wbuf->c;
911 int err, rlen, overlap;
912 struct ubifs_ch *ch = buf;
913
914 dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
915 dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
916 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
917 ubifs_assert(!(offs & 7) && offs < c->leb_size);
918 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
919
920 spin_lock(&wbuf->lock);
921 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
922 if (!overlap) {
923 /* We may safely unlock the write-buffer and read the data */
924 spin_unlock(&wbuf->lock);
925 return ubifs_read_node(c, buf, type, len, lnum, offs);
926 }
927
928 /* Don't read under wbuf */
929 rlen = wbuf->offs - offs;
930 if (rlen < 0)
931 rlen = 0;
932
933 /* Copy the rest from the write-buffer */
934 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
935 spin_unlock(&wbuf->lock);
936
937 if (rlen > 0) {
938 /* Read everything that goes before write-buffer */
939 err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
940 if (err && err != -EBADMSG)
941 return err;
942 }
943
944 if (type != ch->node_type) {
0195a7bb 945 ubifs_err(c, "bad node type (%d but expected %d)",
ff94bc40
HS
946 ch->node_type, type);
947 goto out;
948 }
949
950 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
951 if (err) {
0195a7bb 952 ubifs_err(c, "expected node type %d", type);
ff94bc40
HS
953 return err;
954 }
955
956 rlen = le32_to_cpu(ch->len);
957 if (rlen != len) {
0195a7bb 958 ubifs_err(c, "bad node length %d, expected %d", rlen, len);
ff94bc40
HS
959 goto out;
960 }
961
962 return 0;
963
964out:
0195a7bb 965 ubifs_err(c, "bad node at LEB %d:%d", lnum, offs);
ff94bc40
HS
966 ubifs_dump_node(c, buf);
967 dump_stack();
968 return -EINVAL;
969}
970
9eefe2a2
SR
971/**
972 * ubifs_read_node - read node.
973 * @c: UBIFS file-system description object
974 * @buf: buffer to read to
975 * @type: node type
976 * @len: node length (not aligned)
977 * @lnum: logical eraseblock number
978 * @offs: offset within the logical eraseblock
979 *
980 * This function reads a node of known type and and length, checks it and
981 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
982 * and a negative error code in case of failure.
983 */
984int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
985 int lnum, int offs)
986{
987 int err, l;
988 struct ubifs_ch *ch = buf;
989
990 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
991 ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
992 ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
993 ubifs_assert(!(offs & 7) && offs < c->leb_size);
994 ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
995
ff94bc40
HS
996 err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
997 if (err && err != -EBADMSG)
9eefe2a2 998 return err;
9eefe2a2
SR
999
1000 if (type != ch->node_type) {
0195a7bb
HS
1001 ubifs_errc(c, "bad node type (%d but expected %d)",
1002 ch->node_type, type);
9eefe2a2
SR
1003 goto out;
1004 }
1005
1006 err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
1007 if (err) {
0195a7bb 1008 ubifs_errc(c, "expected node type %d", type);
9eefe2a2
SR
1009 return err;
1010 }
1011
1012 l = le32_to_cpu(ch->len);
1013 if (l != len) {
0195a7bb 1014 ubifs_errc(c, "bad node length %d, expected %d", l, len);
9eefe2a2
SR
1015 goto out;
1016 }
1017
1018 return 0;
1019
1020out:
0195a7bb
HS
1021 ubifs_errc(c, "bad node at LEB %d:%d, LEB mapping status %d", lnum,
1022 offs, ubi_is_mapped(c->ubi, lnum));
1023 if (!c->probing) {
1024 ubifs_dump_node(c, buf);
1025 dump_stack();
1026 }
9eefe2a2
SR
1027 return -EINVAL;
1028}
ff94bc40
HS
1029
1030/**
1031 * ubifs_wbuf_init - initialize write-buffer.
1032 * @c: UBIFS file-system description object
1033 * @wbuf: write-buffer to initialize
1034 *
1035 * This function initializes write-buffer. Returns zero in case of success
1036 * %-ENOMEM in case of failure.
1037 */
1038int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1039{
1040 size_t size;
1041
1042 wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1043 if (!wbuf->buf)
1044 return -ENOMEM;
1045
1046 size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1047 wbuf->inodes = kmalloc(size, GFP_KERNEL);
1048 if (!wbuf->inodes) {
1049 kfree(wbuf->buf);
1050 wbuf->buf = NULL;
1051 return -ENOMEM;
1052 }
1053
1054 wbuf->used = 0;
1055 wbuf->lnum = wbuf->offs = -1;
1056 /*
1057 * If the LEB starts at the max. write size aligned address, then
1058 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1059 * set it to something smaller so that it ends at the closest max.
1060 * write size boundary.
1061 */
1062 size = c->max_write_size - (c->leb_start % c->max_write_size);
1063 wbuf->avail = wbuf->size = size;
1064 wbuf->sync_callback = NULL;
1065 mutex_init(&wbuf->io_mutex);
1066 spin_lock_init(&wbuf->lock);
1067 wbuf->c = c;
1068 wbuf->next_ino = 0;
1069
1070#ifndef __UBOOT__
1071 hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1072 wbuf->timer.function = wbuf_timer_callback_nolock;
1073 wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1074 wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1075 wbuf->delta *= 1000000000ULL;
1076 ubifs_assert(wbuf->delta <= ULONG_MAX);
1077#endif
1078 return 0;
1079}
1080
1081/**
1082 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1083 * @wbuf: the write-buffer where to add
1084 * @inum: the inode number
1085 *
1086 * This function adds an inode number to the inode array of the write-buffer.
1087 */
1088void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1089{
1090 if (!wbuf->buf)
1091 /* NOR flash or something similar */
1092 return;
1093
1094 spin_lock(&wbuf->lock);
1095 if (wbuf->used)
1096 wbuf->inodes[wbuf->next_ino++] = inum;
1097 spin_unlock(&wbuf->lock);
1098}
1099
1100/**
1101 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1102 * @wbuf: the write-buffer
1103 * @inum: the inode number
1104 *
1105 * This function returns with %1 if the write-buffer contains some data from the
1106 * given inode otherwise it returns with %0.
1107 */
1108static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1109{
1110 int i, ret = 0;
1111
1112 spin_lock(&wbuf->lock);
1113 for (i = 0; i < wbuf->next_ino; i++)
1114 if (inum == wbuf->inodes[i]) {
1115 ret = 1;
1116 break;
1117 }
1118 spin_unlock(&wbuf->lock);
1119
1120 return ret;
1121}
1122
1123/**
1124 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1125 * @c: UBIFS file-system description object
1126 * @inode: inode to synchronize
1127 *
1128 * This function synchronizes write-buffers which contain nodes belonging to
1129 * @inode. Returns zero in case of success and a negative error code in case of
1130 * failure.
1131 */
1132int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1133{
1134 int i, err = 0;
1135
1136 for (i = 0; i < c->jhead_cnt; i++) {
1137 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1138
1139 if (i == GCHD)
1140 /*
1141 * GC head is special, do not look at it. Even if the
1142 * head contains something related to this inode, it is
1143 * a _copy_ of corresponding on-flash node which sits
1144 * somewhere else.
1145 */
1146 continue;
1147
1148 if (!wbuf_has_ino(wbuf, inode->i_ino))
1149 continue;
1150
1151 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1152 if (wbuf_has_ino(wbuf, inode->i_ino))
1153 err = ubifs_wbuf_sync_nolock(wbuf);
1154 mutex_unlock(&wbuf->io_mutex);
1155
1156 if (err) {
1157 ubifs_ro_mode(c, err);
1158 return err;
1159 }
1160 }
1161 return 0;
1162}
This page took 0.513358 seconds and 4 git commands to generate.