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		dump_stack();
 113	}
 114	return err;
 115}
 116
 117int ubifs_leb_write(struct ubifs_info *c, int lnum, const void *buf, int offs,
 118		    int len)
 119{
 120	int err;
 121
 122	ubifs_assert(!c->ro_media && !c->ro_mount);
 123	if (c->ro_error)
 124		return -EROFS;
 125	if (!dbg_is_tst_rcvry(c))
 126		err = ubi_leb_write(c->ubi, lnum, buf, offs, len);
 127	else
 128		err = dbg_leb_write(c, lnum, buf, offs, len);
 129	if (err) {
 130		ubifs_err("writing %d bytes to LEB %d:%d failed, error %d",
 131			  len, lnum, offs, err);
 132		ubifs_ro_mode(c, err);
 133		dump_stack();
 134	}
 135	return err;
 136}
 137
 138int ubifs_leb_change(struct ubifs_info *c, int lnum, const void *buf, int len)
 139{
 140	int err;
 141
 142	ubifs_assert(!c->ro_media && !c->ro_mount);
 143	if (c->ro_error)
 144		return -EROFS;
 145	if (!dbg_is_tst_rcvry(c))
 146		err = ubi_leb_change(c->ubi, lnum, buf, len);
 147	else
 148		err = dbg_leb_change(c, lnum, buf, len);
 149	if (err) {
 150		ubifs_err("changing %d bytes in LEB %d failed, error %d",
 151			  len, lnum, err);
 152		ubifs_ro_mode(c, err);
 153		dump_stack();
 154	}
 155	return err;
 156}
 157
 158int ubifs_leb_unmap(struct ubifs_info *c, int lnum)
 159{
 160	int err;
 161
 162	ubifs_assert(!c->ro_media && !c->ro_mount);
 163	if (c->ro_error)
 164		return -EROFS;
 165	if (!dbg_is_tst_rcvry(c))
 166		err = ubi_leb_unmap(c->ubi, lnum);
 167	else
 168		err = dbg_leb_unmap(c, lnum);
 169	if (err) {
 170		ubifs_err("unmap LEB %d failed, error %d", lnum, err);
 171		ubifs_ro_mode(c, err);
 172		dump_stack();
 173	}
 174	return err;
 175}
 176
 177int ubifs_leb_map(struct ubifs_info *c, int lnum)
 178{
 179	int err;
 180
 181	ubifs_assert(!c->ro_media && !c->ro_mount);
 182	if (c->ro_error)
 183		return -EROFS;
 184	if (!dbg_is_tst_rcvry(c))
 185		err = ubi_leb_map(c->ubi, lnum);
 186	else
 187		err = dbg_leb_map(c, lnum);
 188	if (err) {
 189		ubifs_err("mapping LEB %d failed, error %d", lnum, err);
 190		ubifs_ro_mode(c, err);
 191		dump_stack();
 192	}
 193	return err;
 194}
 195
 196int ubifs_is_mapped(const struct ubifs_info *c, int lnum)
 197{
 198	int err;
 199
 200	err = ubi_is_mapped(c->ubi, lnum);
 201	if (err < 0) {
 202		ubifs_err("ubi_is_mapped failed for LEB %d, error %d",
 203			  lnum, err);
 204		dump_stack();
 205	}
 206	return err;
 207}
 208
 209/**
 210 * ubifs_check_node - check node.
 211 * @c: UBIFS file-system description object
 212 * @buf: node to check
 213 * @lnum: logical eraseblock number
 214 * @offs: offset within the logical eraseblock
 215 * @quiet: print no messages
 216 * @must_chk_crc: indicates whether to always check the CRC
 217 *
 218 * This function checks node magic number and CRC checksum. This function also
 219 * validates node length to prevent UBIFS from becoming crazy when an attacker
 220 * feeds it a file-system image with incorrect nodes. For example, too large
 221 * node length in the common header could cause UBIFS to read memory outside of
 222 * allocated buffer when checking the CRC checksum.
 223 *
 224 * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
 225 * true, which is controlled by corresponding UBIFS mount option. However, if
 226 * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
 227 * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
 228 * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
 229 * is checked. This is because during mounting or re-mounting from R/O mode to
 230 * R/W mode we may read journal nodes (when replying the journal or doing the
 231 * recovery) and the journal nodes may potentially be corrupted, so checking is
 232 * required.
 233 *
 234 * This function returns zero in case of success and %-EUCLEAN in case of bad
 235 * CRC or magic.
 236 */
 237int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
 238		     int offs, int quiet, int must_chk_crc)
 239{
 240	int err = -EINVAL, type, node_len;
 241	uint32_t crc, node_crc, magic;
 242	const struct ubifs_ch *ch = buf;
 243
 244	ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 245	ubifs_assert(!(offs & 7) && offs < c->leb_size);
 246
 247	magic = le32_to_cpu(ch->magic);
 248	if (magic != UBIFS_NODE_MAGIC) {
 249		if (!quiet)
 250			ubifs_err("bad magic %#08x, expected %#08x",
 251				  magic, UBIFS_NODE_MAGIC);
 252		err = -EUCLEAN;
 253		goto out;
 254	}
 255
 256	type = ch->node_type;
 257	if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
 258		if (!quiet)
 259			ubifs_err("bad node type %d", type);
 260		goto out;
 261	}
 262
 263	node_len = le32_to_cpu(ch->len);
 264	if (node_len + offs > c->leb_size)
 265		goto out_len;
 266
 267	if (c->ranges[type].max_len == 0) {
 268		if (node_len != c->ranges[type].len)
 269			goto out_len;
 270	} else if (node_len < c->ranges[type].min_len ||
 271		   node_len > c->ranges[type].max_len)
 272		goto out_len;
 273
 274	if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
 275	    !c->remounting_rw && c->no_chk_data_crc)
 276		return 0;
 277
 278	crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
 279	node_crc = le32_to_cpu(ch->crc);
 280	if (crc != node_crc) {
 281		if (!quiet)
 282			ubifs_err("bad CRC: calculated %#08x, read %#08x",
 283				  crc, node_crc);
 284		err = -EUCLEAN;
 285		goto out;
 286	}
 287
 288	return 0;
 289
 290out_len:
 291	if (!quiet)
 292		ubifs_err("bad node length %d", node_len);
 293out:
 294	if (!quiet) {
 295		ubifs_err("bad node at LEB %d:%d", lnum, offs);
 296		ubifs_dump_node(c, buf);
 297		dump_stack();
 298	}
 299	return err;
 300}
 301
 302/**
 303 * ubifs_pad - pad flash space.
 304 * @c: UBIFS file-system description object
 305 * @buf: buffer to put padding to
 306 * @pad: how many bytes to pad
 307 *
 308 * The flash media obliges us to write only in chunks of %c->min_io_size and
 309 * when we have to write less data we add padding node to the write-buffer and
 310 * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
 311 * media is being scanned. If the amount of wasted space is not enough to fit a
 312 * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
 313 * pattern (%UBIFS_PADDING_BYTE).
 314 *
 315 * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
 316 * used.
 317 */
 318void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
 319{
 320	uint32_t crc;
 321
 322	ubifs_assert(pad >= 0 && !(pad & 7));
 323
 324	if (pad >= UBIFS_PAD_NODE_SZ) {
 325		struct ubifs_ch *ch = buf;
 326		struct ubifs_pad_node *pad_node = buf;
 327
 328		ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 329		ch->node_type = UBIFS_PAD_NODE;
 330		ch->group_type = UBIFS_NO_NODE_GROUP;
 331		ch->padding[0] = ch->padding[1] = 0;
 332		ch->sqnum = 0;
 333		ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
 334		pad -= UBIFS_PAD_NODE_SZ;
 335		pad_node->pad_len = cpu_to_le32(pad);
 336		crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
 337		ch->crc = cpu_to_le32(crc);
 338		memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
 339	} else if (pad > 0)
 340		/* Too little space, padding node won't fit */
 341		memset(buf, UBIFS_PADDING_BYTE, pad);
 342}
 343
 344/**
 345 * next_sqnum - get next sequence number.
 346 * @c: UBIFS file-system description object
 347 */
 348static unsigned long long next_sqnum(struct ubifs_info *c)
 349{
 350	unsigned long long sqnum;
 351
 352	spin_lock(&c->cnt_lock);
 353	sqnum = ++c->max_sqnum;
 354	spin_unlock(&c->cnt_lock);
 355
 356	if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
 357		if (sqnum >= SQNUM_WATERMARK) {
 358			ubifs_err("sequence number overflow %llu, end of life",
 359				  sqnum);
 360			ubifs_ro_mode(c, -EINVAL);
 361		}
 362		ubifs_warn("running out of sequence numbers, end of life soon");
 363	}
 364
 365	return sqnum;
 366}
 367
 368/**
 369 * ubifs_prepare_node - prepare node to be written to flash.
 370 * @c: UBIFS file-system description object
 371 * @node: the node to pad
 372 * @len: node length
 373 * @pad: if the buffer has to be padded
 374 *
 375 * This function prepares node at @node to be written to the media - it
 376 * calculates node CRC, fills the common header, and adds proper padding up to
 377 * the next minimum I/O unit if @pad is not zero.
 378 */
 379void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
 380{
 381	uint32_t crc;
 382	struct ubifs_ch *ch = node;
 383	unsigned long long sqnum = next_sqnum(c);
 384
 385	ubifs_assert(len >= UBIFS_CH_SZ);
 386
 387	ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 388	ch->len = cpu_to_le32(len);
 389	ch->group_type = UBIFS_NO_NODE_GROUP;
 390	ch->sqnum = cpu_to_le64(sqnum);
 391	ch->padding[0] = ch->padding[1] = 0;
 392	crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
 393	ch->crc = cpu_to_le32(crc);
 394
 395	if (pad) {
 396		len = ALIGN(len, 8);
 397		pad = ALIGN(len, c->min_io_size) - len;
 398		ubifs_pad(c, node + len, pad);
 399	}
 400}
 401
 402/**
 403 * ubifs_prep_grp_node - prepare node of a group to be written to flash.
 404 * @c: UBIFS file-system description object
 405 * @node: the node to pad
 406 * @len: node length
 407 * @last: indicates the last node of the group
 408 *
 409 * This function prepares node at @node to be written to the media - it
 410 * calculates node CRC and fills the common header.
 411 */
 412void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
 413{
 414	uint32_t crc;
 415	struct ubifs_ch *ch = node;
 416	unsigned long long sqnum = next_sqnum(c);
 417
 418	ubifs_assert(len >= UBIFS_CH_SZ);
 419
 420	ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 421	ch->len = cpu_to_le32(len);
 422	if (last)
 423		ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
 424	else
 425		ch->group_type = UBIFS_IN_NODE_GROUP;
 426	ch->sqnum = cpu_to_le64(sqnum);
 427	ch->padding[0] = ch->padding[1] = 0;
 428	crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
 429	ch->crc = cpu_to_le32(crc);
 430}
 431
 432/**
 433 * wbuf_timer_callback - write-buffer timer callback function.
 434 * @data: timer data (write-buffer descriptor)
 435 *
 436 * This function is called when the write-buffer timer expires.
 437 */
 438static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
 439{
 440	struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
 441
 442	dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
 443	wbuf->need_sync = 1;
 444	wbuf->c->need_wbuf_sync = 1;
 445	ubifs_wake_up_bgt(wbuf->c);
 446	return HRTIMER_NORESTART;
 447}
 448
 449/**
 450 * new_wbuf_timer - start new write-buffer timer.
 451 * @wbuf: write-buffer descriptor
 452 */
 453static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
 454{
 455	ubifs_assert(!hrtimer_active(&wbuf->timer));
 456
 457	if (wbuf->no_timer)
 458		return;
 459	dbg_io("set timer for jhead %s, %llu-%llu millisecs",
 460	       dbg_jhead(wbuf->jhead),
 461	       div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
 462	       div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
 463		       USEC_PER_SEC));
 464	hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
 465			       HRTIMER_MODE_REL);
 466}
 467
 468/**
 469 * cancel_wbuf_timer - cancel write-buffer timer.
 470 * @wbuf: write-buffer descriptor
 471 */
 472static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
 473{
 474	if (wbuf->no_timer)
 475		return;
 476	wbuf->need_sync = 0;
 477	hrtimer_cancel(&wbuf->timer);
 478}
 479
 480/**
 481 * ubifs_wbuf_sync_nolock - synchronize write-buffer.
 482 * @wbuf: write-buffer to synchronize
 483 *
 484 * This function synchronizes write-buffer @buf and returns zero in case of
 485 * success or a negative error code in case of failure.
 486 *
 487 * Note, although write-buffers are of @c->max_write_size, this function does
 488 * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
 489 * if the write-buffer is only partially filled with data, only the used part
 490 * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
 491 * This way we waste less space.
 492 */
 493int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
 494{
 495	struct ubifs_info *c = wbuf->c;
 496	int err, dirt, sync_len;
 497
 498	cancel_wbuf_timer_nolock(wbuf);
 499	if (!wbuf->used || wbuf->lnum == -1)
 500		/* Write-buffer is empty or not seeked */
 501		return 0;
 502
 503	dbg_io("LEB %d:%d, %d bytes, jhead %s",
 504	       wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
 505	ubifs_assert(!(wbuf->avail & 7));
 506	ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
 507	ubifs_assert(wbuf->size >= c->min_io_size);
 508	ubifs_assert(wbuf->size <= c->max_write_size);
 509	ubifs_assert(wbuf->size % c->min_io_size == 0);
 510	ubifs_assert(!c->ro_media && !c->ro_mount);
 511	if (c->leb_size - wbuf->offs >= c->max_write_size)
 512		ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
 513
 514	if (c->ro_error)
 515		return -EROFS;
 516
 517	/*
 518	 * Do not write whole write buffer but write only the minimum necessary
 519	 * amount of min. I/O units.
 520	 */
 521	sync_len = ALIGN(wbuf->used, c->min_io_size);
 522	dirt = sync_len - wbuf->used;
 523	if (dirt)
 524		ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
 525	err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs, sync_len);
 526	if (err)
 527		return err;
 528
 529	spin_lock(&wbuf->lock);
 530	wbuf->offs += sync_len;
 531	/*
 532	 * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
 533	 * But our goal is to optimize writes and make sure we write in
 534	 * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
 535	 * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
 536	 * sure that @wbuf->offs + @wbuf->size is aligned to
 537	 * @c->max_write_size. This way we make sure that after next
 538	 * write-buffer flush we are again at the optimal offset (aligned to
 539	 * @c->max_write_size).
 540	 */
 541	if (c->leb_size - wbuf->offs < c->max_write_size)
 542		wbuf->size = c->leb_size - wbuf->offs;
 543	else if (wbuf->offs & (c->max_write_size - 1))
 544		wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
 545	else
 546		wbuf->size = c->max_write_size;
 547	wbuf->avail = wbuf->size;
 548	wbuf->used = 0;
 549	wbuf->next_ino = 0;
 550	spin_unlock(&wbuf->lock);
 551
 552	if (wbuf->sync_callback)
 553		err = wbuf->sync_callback(c, wbuf->lnum,
 554					  c->leb_size - wbuf->offs, dirt);
 555	return err;
 556}
 557
 558/**
 559 * ubifs_wbuf_seek_nolock - seek write-buffer.
 560 * @wbuf: write-buffer
 561 * @lnum: logical eraseblock number to seek to
 562 * @offs: logical eraseblock offset to seek to
 563 *
 564 * This function targets the write-buffer to logical eraseblock @lnum:@offs.
 565 * The write-buffer has to be empty. Returns zero in case of success and a
 566 * negative error code in case of failure.
 567 */
 568int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs)
 569{
 570	const struct ubifs_info *c = wbuf->c;
 571
 572	dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
 573	ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
 574	ubifs_assert(offs >= 0 && offs <= c->leb_size);
 575	ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
 576	ubifs_assert(lnum != wbuf->lnum);
 577	ubifs_assert(wbuf->used == 0);
 578
 579	spin_lock(&wbuf->lock);
 580	wbuf->lnum = lnum;
 581	wbuf->offs = offs;
 582	if (c->leb_size - wbuf->offs < c->max_write_size)
 583		wbuf->size = c->leb_size - wbuf->offs;
 584	else if (wbuf->offs & (c->max_write_size - 1))
 585		wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
 586	else
 587		wbuf->size = c->max_write_size;
 588	wbuf->avail = wbuf->size;
 589	wbuf->used = 0;
 590	spin_unlock(&wbuf->lock);
 591
 592	return 0;
 593}
 594
 595/**
 596 * ubifs_bg_wbufs_sync - synchronize write-buffers.
 597 * @c: UBIFS file-system description object
 598 *
 599 * This function is called by background thread to synchronize write-buffers.
 600 * Returns zero in case of success and a negative error code in case of
 601 * failure.
 602 */
 603int ubifs_bg_wbufs_sync(struct ubifs_info *c)
 604{
 605	int err, i;
 606
 607	ubifs_assert(!c->ro_media && !c->ro_mount);
 608	if (!c->need_wbuf_sync)
 609		return 0;
 610	c->need_wbuf_sync = 0;
 611
 612	if (c->ro_error) {
 613		err = -EROFS;
 614		goto out_timers;
 615	}
 616
 617	dbg_io("synchronize");
 618	for (i = 0; i < c->jhead_cnt; i++) {
 619		struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 620
 621		cond_resched();
 622
 623		/*
 624		 * If the mutex is locked then wbuf is being changed, so
 625		 * synchronization is not necessary.
 626		 */
 627		if (mutex_is_locked(&wbuf->io_mutex))
 628			continue;
 629
 630		mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 631		if (!wbuf->need_sync) {
 632			mutex_unlock(&wbuf->io_mutex);
 633			continue;
 634		}
 635
 636		err = ubifs_wbuf_sync_nolock(wbuf);
 637		mutex_unlock(&wbuf->io_mutex);
 638		if (err) {
 639			ubifs_err("cannot sync write-buffer, error %d", err);
 640			ubifs_ro_mode(c, err);
 641			goto out_timers;
 642		}
 643	}
 644
 645	return 0;
 646
 647out_timers:
 648	/* Cancel all timers to prevent repeated errors */
 649	for (i = 0; i < c->jhead_cnt; i++) {
 650		struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 651
 652		mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 653		cancel_wbuf_timer_nolock(wbuf);
 654		mutex_unlock(&wbuf->io_mutex);
 655	}
 656	return err;
 657}
 658
 659/**
 660 * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
 661 * @wbuf: write-buffer
 662 * @buf: node to write
 663 * @len: node length
 664 *
 665 * This function writes data to flash via write-buffer @wbuf. This means that
 666 * the last piece of the node won't reach the flash media immediately if it
 667 * does not take whole max. write unit (@c->max_write_size). Instead, the node
 668 * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
 669 * because more data are appended to the write-buffer).
 670 *
 671 * This function returns zero in case of success and a negative error code in
 672 * case of failure. If the node cannot be written because there is no more
 673 * space in this logical eraseblock, %-ENOSPC is returned.
 674 */
 675int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
 676{
 677	struct ubifs_info *c = wbuf->c;
 678	int err, written, n, aligned_len = ALIGN(len, 8);
 679
 680	dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
 681	       dbg_ntype(((struct ubifs_ch *)buf)->node_type),
 682	       dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
 683	ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
 684	ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
 685	ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
 686	ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
 687	ubifs_assert(wbuf->size >= c->min_io_size);
 688	ubifs_assert(wbuf->size <= c->max_write_size);
 689	ubifs_assert(wbuf->size % c->min_io_size == 0);
 690	ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
 691	ubifs_assert(!c->ro_media && !c->ro_mount);
 692	ubifs_assert(!c->space_fixup);
 693	if (c->leb_size - wbuf->offs >= c->max_write_size)
 694		ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
 695
 696	if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
 697		err = -ENOSPC;
 698		goto out;
 699	}
 700
 701	cancel_wbuf_timer_nolock(wbuf);
 702
 703	if (c->ro_error)
 704		return -EROFS;
 705
 706	if (aligned_len <= wbuf->avail) {
 707		/*
 708		 * The node is not very large and fits entirely within
 709		 * write-buffer.
 710		 */
 711		memcpy(wbuf->buf + wbuf->used, buf, len);
 712
 713		if (aligned_len == wbuf->avail) {
 714			dbg_io("flush jhead %s wbuf to LEB %d:%d",
 715			       dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
 716			err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf,
 717					      wbuf->offs, wbuf->size);
 718			if (err)
 719				goto out;
 720
 721			spin_lock(&wbuf->lock);
 722			wbuf->offs += wbuf->size;
 723			if (c->leb_size - wbuf->offs >= c->max_write_size)
 724				wbuf->size = c->max_write_size;
 725			else
 726				wbuf->size = c->leb_size - wbuf->offs;
 727			wbuf->avail = wbuf->size;
 728			wbuf->used = 0;
 729			wbuf->next_ino = 0;
 730			spin_unlock(&wbuf->lock);
 731		} else {
 732			spin_lock(&wbuf->lock);
 733			wbuf->avail -= aligned_len;
 734			wbuf->used += aligned_len;
 735			spin_unlock(&wbuf->lock);
 736		}
 737
 738		goto exit;
 739	}
 740
 741	written = 0;
 742
 743	if (wbuf->used) {
 744		/*
 745		 * The node is large enough and does not fit entirely within
 746		 * current available space. We have to fill and flush
 747		 * write-buffer and switch to the next max. write unit.
 748		 */
 749		dbg_io("flush jhead %s wbuf to LEB %d:%d",
 750		       dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
 751		memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
 752		err = ubifs_leb_write(c, wbuf->lnum, wbuf->buf, wbuf->offs,
 753				      wbuf->size);
 754		if (err)
 755			goto out;
 756
 757		wbuf->offs += wbuf->size;
 758		len -= wbuf->avail;
 759		aligned_len -= wbuf->avail;
 760		written += wbuf->avail;
 761	} else if (wbuf->offs & (c->max_write_size - 1)) {
 762		/*
 763		 * The write-buffer offset is not aligned to
 764		 * @c->max_write_size and @wbuf->size is less than
 765		 * @c->max_write_size. Write @wbuf->size bytes to make sure the
 766		 * following writes are done in optimal @c->max_write_size
 767		 * chunks.
 768		 */
 769		dbg_io("write %d bytes to LEB %d:%d",
 770		       wbuf->size, wbuf->lnum, wbuf->offs);
 771		err = ubifs_leb_write(c, wbuf->lnum, buf, wbuf->offs,
 772				      wbuf->size);
 773		if (err)
 774			goto out;
 775
 776		wbuf->offs += wbuf->size;
 777		len -= wbuf->size;
 778		aligned_len -= wbuf->size;
 779		written += wbuf->size;
 780	}
 781
 782	/*
 783	 * The remaining data may take more whole max. write units, so write the
 784	 * remains multiple to max. write unit size directly to the flash media.
 785	 * We align node length to 8-byte boundary because we anyway flash wbuf
 786	 * if the remaining space is less than 8 bytes.
 787	 */
 788	n = aligned_len >> c->max_write_shift;
 789	if (n) {
 790		n <<= c->max_write_shift;
 791		dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
 792		       wbuf->offs);
 793		err = ubifs_leb_write(c, wbuf->lnum, buf + written,
 794				      wbuf->offs, n);
 795		if (err)
 796			goto out;
 797		wbuf->offs += n;
 798		aligned_len -= n;
 799		len -= n;
 800		written += n;
 801	}
 802
 803	spin_lock(&wbuf->lock);
 804	if (aligned_len)
 805		/*
 806		 * And now we have what's left and what does not take whole
 807		 * max. write unit, so write it to the write-buffer and we are
 808		 * done.
 809		 */
 810		memcpy(wbuf->buf, buf + written, len);
 811
 812	if (c->leb_size - wbuf->offs >= c->max_write_size)
 813		wbuf->size = c->max_write_size;
 814	else
 815		wbuf->size = c->leb_size - wbuf->offs;
 816	wbuf->avail = wbuf->size - aligned_len;
 817	wbuf->used = aligned_len;
 818	wbuf->next_ino = 0;
 819	spin_unlock(&wbuf->lock);
 820
 821exit:
 822	if (wbuf->sync_callback) {
 823		int free = c->leb_size - wbuf->offs - wbuf->used;
 824
 825		err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
 826		if (err)
 827			goto out;
 828	}
 829
 830	if (wbuf->used)
 831		new_wbuf_timer_nolock(wbuf);
 832
 833	return 0;
 834
 835out:
 836	ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
 837		  len, wbuf->lnum, wbuf->offs, err);
 838	ubifs_dump_node(c, buf);
 839	dump_stack();
 840	ubifs_dump_leb(c, wbuf->lnum);
 841	return err;
 842}
 843
 844/**
 845 * ubifs_write_node - write node to the media.
 846 * @c: UBIFS file-system description object
 847 * @buf: the node to write
 848 * @len: node length
 849 * @lnum: logical eraseblock number
 850 * @offs: offset within the logical eraseblock
 851 *
 852 * This function automatically fills node magic number, assigns sequence
 853 * number, and calculates node CRC checksum. The length of the @buf buffer has
 854 * to be aligned to the minimal I/O unit size. This function automatically
 855 * appends padding node and padding bytes if needed. Returns zero in case of
 856 * success and a negative error code in case of failure.
 857 */
 858int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
 859		     int offs)
 860{
 861	int err, buf_len = ALIGN(len, c->min_io_size);
 862
 863	dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
 864	       lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
 865	       buf_len);
 866	ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 867	ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
 868	ubifs_assert(!c->ro_media && !c->ro_mount);
 869	ubifs_assert(!c->space_fixup);
 870
 871	if (c->ro_error)
 872		return -EROFS;
 873
 874	ubifs_prepare_node(c, buf, len, 1);
 875	err = ubifs_leb_write(c, lnum, buf, offs, buf_len);
 876	if (err)
 877		ubifs_dump_node(c, buf);
 878
 879	return err;
 880}
 881
 882/**
 883 * ubifs_read_node_wbuf - read node from the media or write-buffer.
 884 * @wbuf: wbuf to check for un-written data
 885 * @buf: buffer to read to
 886 * @type: node type
 887 * @len: node length
 888 * @lnum: logical eraseblock number
 889 * @offs: offset within the logical eraseblock
 890 *
 891 * This function reads a node of known type and length, checks it and stores
 892 * in @buf. If the node partially or fully sits in the write-buffer, this
 893 * function takes data from the buffer, otherwise it reads the flash media.
 894 * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
 895 * error code in case of failure.
 896 */
 897int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
 898			 int lnum, int offs)
 899{
 900	const struct ubifs_info *c = wbuf->c;
 901	int err, rlen, overlap;
 902	struct ubifs_ch *ch = buf;
 903
 904	dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
 905	       dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
 906	ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 907	ubifs_assert(!(offs & 7) && offs < c->leb_size);
 908	ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
 909
 910	spin_lock(&wbuf->lock);
 911	overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
 912	if (!overlap) {
 913		/* We may safely unlock the write-buffer and read the data */
 914		spin_unlock(&wbuf->lock);
 915		return ubifs_read_node(c, buf, type, len, lnum, offs);
 916	}
 917
 918	/* Don't read under wbuf */
 919	rlen = wbuf->offs - offs;
 920	if (rlen < 0)
 921		rlen = 0;
 922
 923	/* Copy the rest from the write-buffer */
 924	memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
 925	spin_unlock(&wbuf->lock);
 926
 927	if (rlen > 0) {
 928		/* Read everything that goes before write-buffer */
 929		err = ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
 930		if (err && err != -EBADMSG)
 931			return err;
 932	}
 933
 934	if (type != ch->node_type) {
 935		ubifs_err("bad node type (%d but expected %d)",
 936			  ch->node_type, type);
 937		goto out;
 938	}
 939
 940	err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
 941	if (err) {
 942		ubifs_err("expected node type %d", type);
 943		return err;
 944	}
 945
 946	rlen = le32_to_cpu(ch->len);
 947	if (rlen != len) {
 948		ubifs_err("bad node length %d, expected %d", rlen, len);
 949		goto out;
 950	}
 951
 952	return 0;
 953
 954out:
 955	ubifs_err("bad node at LEB %d:%d", lnum, offs);
 956	ubifs_dump_node(c, buf);
 957	dump_stack();
 958	return -EINVAL;
 959}
 960
 961/**
 962 * ubifs_read_node - read node.
 963 * @c: UBIFS file-system description object
 964 * @buf: buffer to read to
 965 * @type: node type
 966 * @len: node length (not aligned)
 967 * @lnum: logical eraseblock number
 968 * @offs: offset within the logical eraseblock
 969 *
 970 * This function reads a node of known type and and length, checks it and
 971 * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
 972 * and a negative error code in case of failure.
 973 */
 974int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
 975		    int lnum, int offs)
 976{
 977	int err, l;
 978	struct ubifs_ch *ch = buf;
 979
 980	dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
 981	ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 982	ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
 983	ubifs_assert(!(offs & 7) && offs < c->leb_size);
 984	ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
 985
 986	err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
 987	if (err && err != -EBADMSG)
 988		return err;
 989
 990	if (type != ch->node_type) {
 991		ubifs_err("bad node type (%d but expected %d)",
 992			  ch->node_type, type);
 993		goto out;
 994	}
 995
 996	err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
 997	if (err) {
 998		ubifs_err("expected node type %d", type);
 999		return err;
1000	}
1001
1002	l = le32_to_cpu(ch->len);
1003	if (l != len) {
1004		ubifs_err("bad node length %d, expected %d", l, len);
1005		goto out;
1006	}
1007
1008	return 0;
1009
1010out:
1011	ubifs_err("bad node at LEB %d:%d, LEB mapping status %d", lnum, offs,
1012		  ubi_is_mapped(c->ubi, lnum));
1013	ubifs_dump_node(c, buf);
1014	dump_stack();
 
 
1015	return -EINVAL;
1016}
1017
1018/**
1019 * ubifs_wbuf_init - initialize write-buffer.
1020 * @c: UBIFS file-system description object
1021 * @wbuf: write-buffer to initialize
1022 *
1023 * This function initializes write-buffer. Returns zero in case of success
1024 * %-ENOMEM in case of failure.
1025 */
1026int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
1027{
1028	size_t size;
1029
1030	wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
1031	if (!wbuf->buf)
1032		return -ENOMEM;
1033
1034	size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
1035	wbuf->inodes = kmalloc(size, GFP_KERNEL);
1036	if (!wbuf->inodes) {
1037		kfree(wbuf->buf);
1038		wbuf->buf = NULL;
1039		return -ENOMEM;
1040	}
1041
1042	wbuf->used = 0;
1043	wbuf->lnum = wbuf->offs = -1;
1044	/*
1045	 * If the LEB starts at the max. write size aligned address, then
1046	 * write-buffer size has to be set to @c->max_write_size. Otherwise,
1047	 * set it to something smaller so that it ends at the closest max.
1048	 * write size boundary.
1049	 */
1050	size = c->max_write_size - (c->leb_start % c->max_write_size);
1051	wbuf->avail = wbuf->size = size;
1052	wbuf->sync_callback = NULL;
1053	mutex_init(&wbuf->io_mutex);
1054	spin_lock_init(&wbuf->lock);
1055	wbuf->c = c;
1056	wbuf->next_ino = 0;
1057
1058	hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1059	wbuf->timer.function = wbuf_timer_callback_nolock;
1060	wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
1061	wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
1062	wbuf->delta *= 1000000000ULL;
1063	ubifs_assert(wbuf->delta <= ULONG_MAX);
1064	return 0;
1065}
1066
1067/**
1068 * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
1069 * @wbuf: the write-buffer where to add
1070 * @inum: the inode number
1071 *
1072 * This function adds an inode number to the inode array of the write-buffer.
1073 */
1074void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
1075{
1076	if (!wbuf->buf)
1077		/* NOR flash or something similar */
1078		return;
1079
1080	spin_lock(&wbuf->lock);
1081	if (wbuf->used)
1082		wbuf->inodes[wbuf->next_ino++] = inum;
1083	spin_unlock(&wbuf->lock);
1084}
1085
1086/**
1087 * wbuf_has_ino - returns if the wbuf contains data from the inode.
1088 * @wbuf: the write-buffer
1089 * @inum: the inode number
1090 *
1091 * This function returns with %1 if the write-buffer contains some data from the
1092 * given inode otherwise it returns with %0.
1093 */
1094static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1095{
1096	int i, ret = 0;
1097
1098	spin_lock(&wbuf->lock);
1099	for (i = 0; i < wbuf->next_ino; i++)
1100		if (inum == wbuf->inodes[i]) {
1101			ret = 1;
1102			break;
1103		}
1104	spin_unlock(&wbuf->lock);
1105
1106	return ret;
1107}
1108
1109/**
1110 * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1111 * @c: UBIFS file-system description object
1112 * @inode: inode to synchronize
1113 *
1114 * This function synchronizes write-buffers which contain nodes belonging to
1115 * @inode. Returns zero in case of success and a negative error code in case of
1116 * failure.
1117 */
1118int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1119{
1120	int i, err = 0;
1121
1122	for (i = 0; i < c->jhead_cnt; i++) {
1123		struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1124
1125		if (i == GCHD)
1126			/*
1127			 * GC head is special, do not look at it. Even if the
1128			 * head contains something related to this inode, it is
1129			 * a _copy_ of corresponding on-flash node which sits
1130			 * somewhere else.
1131			 */
1132			continue;
1133
1134		if (!wbuf_has_ino(wbuf, inode->i_ino))
1135			continue;
1136
1137		mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1138		if (wbuf_has_ino(wbuf, inode->i_ino))
1139			err = ubifs_wbuf_sync_nolock(wbuf);
1140		mutex_unlock(&wbuf->io_mutex);
1141
1142		if (err) {
1143			ubifs_ro_mode(c, err);
1144			return err;
1145		}
1146	}
1147	return 0;
1148}