fs/ubifs/io.c

   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  */
  82 void 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                 dbg_dump_stack();
  90         }
  91 }
  92
  93 /**
  94  * ubifs_check_node - check node.
  95  * @c: UBIFS file-system description object
  96  * @buf: node to check
  97  * @lnum: logical eraseblock number
  98  * @offs: offset within the logical eraseblock
  99  * @quiet: print no messages
 100  * @must_chk_crc: indicates whether to always check the CRC
 101  *
 102  * This function checks node magic number and CRC checksum. This function also
 103  * validates node length to prevent UBIFS from becoming crazy when an attacker
 104  * feeds it a file-system image with incorrect nodes. For example, too large
 105  * node length in the common header could cause UBIFS to read memory outside of
 106  * allocated buffer when checking the CRC checksum.
 107  *
 108  * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
 109  * true, which is controlled by corresponding UBIFS mount option. However, if
 110  * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
 111  * checked. Similarly, if @c->mounting or @c->remounting_rw is true (we are
 112  * mounting or re-mounting to R/W mode), @c->no_chk_data_crc is ignored and CRC
 113  * is checked. This is because during mounting or re-mounting from R/O mode to
 114  * R/W mode we may read journal nodes (when replying the journal or doing the
 115  * recovery) and the journal nodes may potentially be corrupted, so checking is
 116  * required.
 117  *
 118  * This function returns zero in case of success and %-EUCLEAN in case of bad
 119  * CRC or magic.
 120  */
 121 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
 122                      int offs, int quiet, int must_chk_crc)
 123 {
 124         int err = -EINVAL, type, node_len;
 125         uint32_t crc, node_crc, magic;
 126         const struct ubifs_ch *ch = buf;
 127
 128         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 129         ubifs_assert(!(offs & 7) && offs < c->leb_size);
 130
 131         magic = le32_to_cpu(ch->magic);
 132         if (magic != UBIFS_NODE_MAGIC) {
 133                 if (!quiet)
 134                         ubifs_err("bad magic %#08x, expected %#08x",
 135                                   magic, UBIFS_NODE_MAGIC);
 136                 err = -EUCLEAN;
 137                 goto out;
 138         }
 139
 140         type = ch->node_type;
 141         if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
 142                 if (!quiet)
 143                         ubifs_err("bad node type %d", type);
 144                 goto out;
 145         }
 146
 147         node_len = le32_to_cpu(ch->len);
 148         if (node_len + offs > c->leb_size)
 149                 goto out_len;
 150
 151         if (c->ranges[type].max_len == 0) {
 152                 if (node_len != c->ranges[type].len)
 153                         goto out_len;
 154         } else if (node_len < c->ranges[type].min_len ||
 155                    node_len > c->ranges[type].max_len)
 156                 goto out_len;
 157
 158         if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->mounting &&
 159             !c->remounting_rw && c->no_chk_data_crc)
 160                 return 0;
 161
 162         crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
 163         node_crc = le32_to_cpu(ch->crc);
 164         if (crc != node_crc) {
 165                 if (!quiet)
 166                         ubifs_err("bad CRC: calculated %#08x, read %#08x",
 167                                   crc, node_crc);
 168                 err = -EUCLEAN;
 169                 goto out;
 170         }
 171
 172         return 0;
 173
 174 out_len:
 175         if (!quiet)
 176                 ubifs_err("bad node length %d", node_len);
 177 out:
 178         if (!quiet) {
 179                 ubifs_err("bad node at LEB %d:%d", lnum, offs);
 180                 dbg_dump_node(c, buf);
 181                 dbg_dump_stack();
 182         }
 183         return err;
 184 }
 185
 186 /**
 187  * ubifs_pad - pad flash space.
 188  * @c: UBIFS file-system description object
 189  * @buf: buffer to put padding to
 190  * @pad: how many bytes to pad
 191  *
 192  * The flash media obliges us to write only in chunks of %c->min_io_size and
 193  * when we have to write less data we add padding node to the write-buffer and
 194  * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
 195  * media is being scanned. If the amount of wasted space is not enough to fit a
 196  * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
 197  * pattern (%UBIFS_PADDING_BYTE).
 198  *
 199  * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
 200  * used.
 201  */
 202 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
 203 {
 204         uint32_t crc;
 205
 206         ubifs_assert(pad >= 0 && !(pad & 7));
 207
 208         if (pad >= UBIFS_PAD_NODE_SZ) {
 209                 struct ubifs_ch *ch = buf;
 210                 struct ubifs_pad_node *pad_node = buf;
 211
 212                 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 213                 ch->node_type = UBIFS_PAD_NODE;
 214                 ch->group_type = UBIFS_NO_NODE_GROUP;
 215                 ch->padding[0] = ch->padding[1] = 0;
 216                 ch->sqnum = 0;
 217                 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
 218                 pad -= UBIFS_PAD_NODE_SZ;
 219                 pad_node->pad_len = cpu_to_le32(pad);
 220                 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
 221                 ch->crc = cpu_to_le32(crc);
 222                 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
 223         } else if (pad > 0)
 224                 /* Too little space, padding node won't fit */
 225                 memset(buf, UBIFS_PADDING_BYTE, pad);
 226 }
 227
 228 /**
 229  * next_sqnum - get next sequence number.
 230  * @c: UBIFS file-system description object
 231  */
 232 static unsigned long long next_sqnum(struct ubifs_info *c)
 233 {
 234         unsigned long long sqnum;
 235
 236         spin_lock(&c->cnt_lock);
 237         sqnum = ++c->max_sqnum;
 238         spin_unlock(&c->cnt_lock);
 239
 240         if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
 241                 if (sqnum >= SQNUM_WATERMARK) {
 242                         ubifs_err("sequence number overflow %llu, end of life",
 243                                   sqnum);
 244                         ubifs_ro_mode(c, -EINVAL);
 245                 }
 246                 ubifs_warn("running out of sequence numbers, end of life soon");
 247         }
 248
 249         return sqnum;
 250 }
 251
 252 /**
 253  * ubifs_prepare_node - prepare node to be written to flash.
 254  * @c: UBIFS file-system description object
 255  * @node: the node to pad
 256  * @len: node length
 257  * @pad: if the buffer has to be padded
 258  *
 259  * This function prepares node at @node to be written to the media - it
 260  * calculates node CRC, fills the common header, and adds proper padding up to
 261  * the next minimum I/O unit if @pad is not zero.
 262  */
 263 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
 264 {
 265         uint32_t crc;
 266         struct ubifs_ch *ch = node;
 267         unsigned long long sqnum = next_sqnum(c);
 268
 269         ubifs_assert(len >= UBIFS_CH_SZ);
 270
 271         ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 272         ch->len = cpu_to_le32(len);
 273         ch->group_type = UBIFS_NO_NODE_GROUP;
 274         ch->sqnum = cpu_to_le64(sqnum);
 275         ch->padding[0] = ch->padding[1] = 0;
 276         crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
 277         ch->crc = cpu_to_le32(crc);
 278
 279         if (pad) {
 280                 len = ALIGN(len, 8);
 281                 pad = ALIGN(len, c->min_io_size) - len;
 282                 ubifs_pad(c, node + len, pad);
 283         }
 284 }
 285
 286 /**
 287  * ubifs_prep_grp_node - prepare node of a group to be written to flash.
 288  * @c: UBIFS file-system description object
 289  * @node: the node to pad
 290  * @len: node length
 291  * @last: indicates the last node of the group
 292  *
 293  * This function prepares node at @node to be written to the media - it
 294  * calculates node CRC and fills the common header.
 295  */
 296 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
 297 {
 298         uint32_t crc;
 299         struct ubifs_ch *ch = node;
 300         unsigned long long sqnum = next_sqnum(c);
 301
 302         ubifs_assert(len >= UBIFS_CH_SZ);
 303
 304         ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 305         ch->len = cpu_to_le32(len);
 306         if (last)
 307                 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
 308         else
 309                 ch->group_type = UBIFS_IN_NODE_GROUP;
 310         ch->sqnum = cpu_to_le64(sqnum);
 311         ch->padding[0] = ch->padding[1] = 0;
 312         crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
 313         ch->crc = cpu_to_le32(crc);
 314 }
 315
 316 /**
 317  * wbuf_timer_callback - write-buffer timer callback function.
 318  * @data: timer data (write-buffer descriptor)
 319  *
 320  * This function is called when the write-buffer timer expires.
 321  */
 322 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
 323 {
 324         struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
 325
 326         dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
 327         wbuf->need_sync = 1;
 328         wbuf->c->need_wbuf_sync = 1;
 329         ubifs_wake_up_bgt(wbuf->c);
 330         return HRTIMER_NORESTART;
 331 }
 332
 333 /**
 334  * new_wbuf_timer - start new write-buffer timer.
 335  * @wbuf: write-buffer descriptor
 336  */
 337 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
 338 {
 339         ubifs_assert(!hrtimer_active(&wbuf->timer));
 340
 341         if (wbuf->no_timer)
 342                 return;
 343         dbg_io("set timer for jhead %s, %llu-%llu millisecs",
 344                dbg_jhead(wbuf->jhead),
 345                div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
 346                div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
 347                        USEC_PER_SEC));
 348         hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
 349                                HRTIMER_MODE_REL);
 350 }
 351
 352 /**
 353  * cancel_wbuf_timer - cancel write-buffer timer.
 354  * @wbuf: write-buffer descriptor
 355  */
 356 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
 357 {
 358         if (wbuf->no_timer)
 359                 return;
 360         wbuf->need_sync = 0;
 361         hrtimer_cancel(&wbuf->timer);
 362 }
 363
 364 /**
 365  * ubifs_wbuf_sync_nolock - synchronize write-buffer.
 366  * @wbuf: write-buffer to synchronize
 367  *
 368  * This function synchronizes write-buffer @buf and returns zero in case of
 369  * success or a negative error code in case of failure.
 370  *
 371  * Note, although write-buffers are of @c->max_write_size, this function does
 372  * not necessarily writes all @c->max_write_size bytes to the flash. Instead,
 373  * if the write-buffer is only partially filled with data, only the used part
 374  * of the write-buffer (aligned on @c->min_io_size boundary) is synchronized.
 375  * This way we waste less space.
 376  */
 377 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
 378 {
 379         struct ubifs_info *c = wbuf->c;
 380         int err, dirt, sync_len;
 381
 382         cancel_wbuf_timer_nolock(wbuf);
 383         if (!wbuf->used || wbuf->lnum == -1)
 384                 /* Write-buffer is empty or not seeked */
 385                 return 0;
 386
 387         dbg_io("LEB %d:%d, %d bytes, jhead %s",
 388                wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
 389         ubifs_assert(!(wbuf->avail & 7));
 390         ubifs_assert(wbuf->offs + wbuf->size <= c->leb_size);
 391         ubifs_assert(wbuf->size >= c->min_io_size);
 392         ubifs_assert(wbuf->size <= c->max_write_size);
 393         ubifs_assert(wbuf->size % c->min_io_size == 0);
 394         ubifs_assert(!c->ro_media && !c->ro_mount);
 395         if (c->leb_size - wbuf->offs >= c->max_write_size)
 396                 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
 397
 398         if (c->ro_error)
 399                 return -EROFS;
 400
 401         /*
 402          * Do not write whole write buffer but write only the minimum necessary
 403          * amount of min. I/O units.
 404          */
 405         sync_len = ALIGN(wbuf->used, c->min_io_size);
 406         dirt = sync_len - wbuf->used;
 407         if (dirt)
 408                 ubifs_pad(c, wbuf->buf + wbuf->used, dirt);
 409         err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
 410                             sync_len, wbuf->dtype);
 411         if (err) {
 412                 ubifs_err("cannot write %d bytes to LEB %d:%d",
 413                           sync_len, wbuf->lnum, wbuf->offs);
 414                 dbg_dump_stack();
 415                 return err;
 416         }
 417
 418         spin_lock(&wbuf->lock);
 419         wbuf->offs += sync_len;
 420         /*
 421          * Now @wbuf->offs is not necessarily aligned to @c->max_write_size.
 422          * But our goal is to optimize writes and make sure we write in
 423          * @c->max_write_size chunks and to @c->max_write_size-aligned offset.
 424          * Thus, if @wbuf->offs is not aligned to @c->max_write_size now, make
 425          * sure that @wbuf->offs + @wbuf->size is aligned to
 426          * @c->max_write_size. This way we make sure that after next
 427          * write-buffer flush we are again at the optimal offset (aligned to
 428          * @c->max_write_size).
 429          */
 430         if (c->leb_size - wbuf->offs < c->max_write_size)
 431                 wbuf->size = c->leb_size - wbuf->offs;
 432         else if (wbuf->offs & (c->max_write_size - 1))
 433                 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
 434         else
 435                 wbuf->size = c->max_write_size;
 436         wbuf->avail = wbuf->size;
 437         wbuf->used = 0;
 438         wbuf->next_ino = 0;
 439         spin_unlock(&wbuf->lock);
 440
 441         if (wbuf->sync_callback)
 442                 err = wbuf->sync_callback(c, wbuf->lnum,
 443                                           c->leb_size - wbuf->offs, dirt);
 444         return err;
 445 }
 446
 447 /**
 448  * ubifs_wbuf_seek_nolock - seek write-buffer.
 449  * @wbuf: write-buffer
 450  * @lnum: logical eraseblock number to seek to
 451  * @offs: logical eraseblock offset to seek to
 452  * @dtype: data type
 453  *
 454  * This function targets the write-buffer to logical eraseblock @lnum:@offs.
 455  * The write-buffer has to be empty. Returns zero in case of success and a
 456  * negative error code in case of failure.
 457  */
 458 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs,
 459                            int dtype)
 460 {
 461         const struct ubifs_info *c = wbuf->c;
 462
 463         dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
 464         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
 465         ubifs_assert(offs >= 0 && offs <= c->leb_size);
 466         ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
 467         ubifs_assert(lnum != wbuf->lnum);
 468         ubifs_assert(wbuf->used == 0);
 469
 470         spin_lock(&wbuf->lock);
 471         wbuf->lnum = lnum;
 472         wbuf->offs = offs;
 473         if (c->leb_size - wbuf->offs < c->max_write_size)
 474                 wbuf->size = c->leb_size - wbuf->offs;
 475         else if (wbuf->offs & (c->max_write_size - 1))
 476                 wbuf->size = ALIGN(wbuf->offs, c->max_write_size) - wbuf->offs;
 477         else
 478                 wbuf->size = c->max_write_size;
 479         wbuf->avail = wbuf->size;
 480         wbuf->used = 0;
 481         spin_unlock(&wbuf->lock);
 482         wbuf->dtype = dtype;
 483
 484         return 0;
 485 }
 486
 487 /**
 488  * ubifs_bg_wbufs_sync - synchronize write-buffers.
 489  * @c: UBIFS file-system description object
 490  *
 491  * This function is called by background thread to synchronize write-buffers.
 492  * Returns zero in case of success and a negative error code in case of
 493  * failure.
 494  */
 495 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
 496 {
 497         int err, i;
 498
 499         ubifs_assert(!c->ro_media && !c->ro_mount);
 500         if (!c->need_wbuf_sync)
 501                 return 0;
 502         c->need_wbuf_sync = 0;
 503
 504         if (c->ro_error) {
 505                 err = -EROFS;
 506                 goto out_timers;
 507         }
 508
 509         dbg_io("synchronize");
 510         for (i = 0; i < c->jhead_cnt; i++) {
 511                 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 512
 513                 cond_resched();
 514
 515                 /*
 516                  * If the mutex is locked then wbuf is being changed, so
 517                  * synchronization is not necessary.
 518                  */
 519                 if (mutex_is_locked(&wbuf->io_mutex))
 520                         continue;
 521
 522                 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 523                 if (!wbuf->need_sync) {
 524                         mutex_unlock(&wbuf->io_mutex);
 525                         continue;
 526                 }
 527
 528                 err = ubifs_wbuf_sync_nolock(wbuf);
 529                 mutex_unlock(&wbuf->io_mutex);
 530                 if (err) {
 531                         ubifs_err("cannot sync write-buffer, error %d", err);
 532                         ubifs_ro_mode(c, err);
 533                         goto out_timers;
 534                 }
 535         }
 536
 537         return 0;
 538
 539 out_timers:
 540         /* Cancel all timers to prevent repeated errors */
 541         for (i = 0; i < c->jhead_cnt; i++) {
 542                 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 543
 544                 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 545                 cancel_wbuf_timer_nolock(wbuf);
 546                 mutex_unlock(&wbuf->io_mutex);
 547         }
 548         return err;
 549 }
 550
 551 /**
 552  * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
 553  * @wbuf: write-buffer
 554  * @buf: node to write
 555  * @len: node length
 556  *
 557  * This function writes data to flash via write-buffer @wbuf. This means that
 558  * the last piece of the node won't reach the flash media immediately if it
 559  * does not take whole max. write unit (@c->max_write_size). Instead, the node
 560  * will sit in RAM until the write-buffer is synchronized (e.g., by timer, or
 561  * because more data are appended to the write-buffer).
 562  *
 563  * This function returns zero in case of success and a negative error code in
 564  * case of failure. If the node cannot be written because there is no more
 565  * space in this logical eraseblock, %-ENOSPC is returned.
 566  */
 567 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
 568 {
 569         struct ubifs_info *c = wbuf->c;
 570         int err, written, n, aligned_len = ALIGN(len, 8);
 571
 572         dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
 573                dbg_ntype(((struct ubifs_ch *)buf)->node_type),
 574                dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
 575         ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
 576         ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
 577         ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
 578         ubifs_assert(wbuf->avail > 0 && wbuf->avail <= wbuf->size);
 579         ubifs_assert(wbuf->size >= c->min_io_size);
 580         ubifs_assert(wbuf->size <= c->max_write_size);
 581         ubifs_assert(wbuf->size % c->min_io_size == 0);
 582         ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
 583         ubifs_assert(!c->ro_media && !c->ro_mount);
 584         ubifs_assert(!c->space_fixup);
 585         if (c->leb_size - wbuf->offs >= c->max_write_size)
 586                 ubifs_assert(!((wbuf->offs + wbuf->size) % c->max_write_size));
 587
 588         if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
 589                 err = -ENOSPC;
 590                 goto out;
 591         }
 592
 593         cancel_wbuf_timer_nolock(wbuf);
 594
 595         if (c->ro_error)
 596                 return -EROFS;
 597
 598         if (aligned_len <= wbuf->avail) {
 599                 /*
 600                  * The node is not very large and fits entirely within
 601                  * write-buffer.
 602                  */
 603                 memcpy(wbuf->buf + wbuf->used, buf, len);
 604
 605                 if (aligned_len == wbuf->avail) {
 606                         dbg_io("flush jhead %s wbuf to LEB %d:%d",
 607                                dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
 608                         err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf,
 609                                             wbuf->offs, wbuf->size,
 610                                             wbuf->dtype);
 611                         if (err)
 612                                 goto out;
 613
 614                         spin_lock(&wbuf->lock);
 615                         wbuf->offs += wbuf->size;
 616                         if (c->leb_size - wbuf->offs >= c->max_write_size)
 617                                 wbuf->size = c->max_write_size;
 618                         else
 619                                 wbuf->size = c->leb_size - wbuf->offs;
 620                         wbuf->avail = wbuf->size;
 621                         wbuf->used = 0;
 622                         wbuf->next_ino = 0;
 623                         spin_unlock(&wbuf->lock);
 624                 } else {
 625                         spin_lock(&wbuf->lock);
 626                         wbuf->avail -= aligned_len;
 627                         wbuf->used += aligned_len;
 628                         spin_unlock(&wbuf->lock);
 629                 }
 630
 631                 goto exit;
 632         }
 633
 634         written = 0;
 635
 636         if (wbuf->used) {
 637                 /*
 638                  * The node is large enough and does not fit entirely within
 639                  * current available space. We have to fill and flush
 640                  * write-buffer and switch to the next max. write unit.
 641                  */
 642                 dbg_io("flush jhead %s wbuf to LEB %d:%d",
 643                        dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
 644                 memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
 645                 err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
 646                                     wbuf->size, wbuf->dtype);
 647                 if (err)
 648                         goto out;
 649
 650                 wbuf->offs += wbuf->size;
 651                 len -= wbuf->avail;
 652                 aligned_len -= wbuf->avail;
 653                 written += wbuf->avail;
 654         } else if (wbuf->offs & (c->max_write_size - 1)) {
 655                 /*
 656                  * The write-buffer offset is not aligned to
 657                  * @c->max_write_size and @wbuf->size is less than
 658                  * @c->max_write_size. Write @wbuf->size bytes to make sure the
 659                  * following writes are done in optimal @c->max_write_size
 660                  * chunks.
 661                  */
 662                 dbg_io("write %d bytes to LEB %d:%d",
 663                        wbuf->size, wbuf->lnum, wbuf->offs);
 664                 err = ubi_leb_write(c->ubi, wbuf->lnum, buf, wbuf->offs,
 665                                     wbuf->size, wbuf->dtype);
 666                 if (err)
 667                         goto out;
 668
 669                 wbuf->offs += wbuf->size;
 670                 len -= wbuf->size;
 671                 aligned_len -= wbuf->size;
 672                 written += wbuf->size;
 673         }
 674
 675         /*
 676          * The remaining data may take more whole max. write units, so write the
 677          * remains multiple to max. write unit size directly to the flash media.
 678          * We align node length to 8-byte boundary because we anyway flash wbuf
 679          * if the remaining space is less than 8 bytes.
 680          */
 681         n = aligned_len >> c->max_write_shift;
 682         if (n) {
 683                 n <<= c->max_write_shift;
 684                 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum,
 685                        wbuf->offs);
 686                 err = ubi_leb_write(c->ubi, wbuf->lnum, buf + written,
 687                                     wbuf->offs, n, wbuf->dtype);
 688                 if (err)
 689                         goto out;
 690                 wbuf->offs += n;
 691                 aligned_len -= n;
 692                 len -= n;
 693                 written += n;
 694         }
 695
 696         spin_lock(&wbuf->lock);
 697         if (aligned_len)
 698                 /*
 699                  * And now we have what's left and what does not take whole
 700                  * max. write unit, so write it to the write-buffer and we are
 701                  * done.
 702                  */
 703                 memcpy(wbuf->buf, buf + written, len);
 704
 705         if (c->leb_size - wbuf->offs >= c->max_write_size)
 706                 wbuf->size = c->max_write_size;
 707         else
 708                 wbuf->size = c->leb_size - wbuf->offs;
 709         wbuf->avail = wbuf->size - aligned_len;
 710         wbuf->used = aligned_len;
 711         wbuf->next_ino = 0;
 712         spin_unlock(&wbuf->lock);
 713
 714 exit:
 715         if (wbuf->sync_callback) {
 716                 int free = c->leb_size - wbuf->offs - wbuf->used;
 717
 718                 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
 719                 if (err)
 720                         goto out;
 721         }
 722
 723         if (wbuf->used)
 724                 new_wbuf_timer_nolock(wbuf);
 725
 726         return 0;
 727
 728 out:
 729         ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
 730                   len, wbuf->lnum, wbuf->offs, err);
 731         dbg_dump_node(c, buf);
 732         dbg_dump_stack();
 733         dbg_dump_leb(c, wbuf->lnum);
 734         return err;
 735 }
 736
 737 /**
 738  * ubifs_write_node - write node to the media.
 739  * @c: UBIFS file-system description object
 740  * @buf: the node to write
 741  * @len: node length
 742  * @lnum: logical eraseblock number
 743  * @offs: offset within the logical eraseblock
 744  * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
 745  *
 746  * This function automatically fills node magic number, assigns sequence
 747  * number, and calculates node CRC checksum. The length of the @buf buffer has
 748  * to be aligned to the minimal I/O unit size. This function automatically
 749  * appends padding node and padding bytes if needed. Returns zero in case of
 750  * success and a negative error code in case of failure.
 751  */
 752 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
 753                      int offs, int dtype)
 754 {
 755         int err, buf_len = ALIGN(len, c->min_io_size);
 756
 757         dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
 758                lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
 759                buf_len);
 760         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 761         ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
 762         ubifs_assert(!c->ro_media && !c->ro_mount);
 763         ubifs_assert(!c->space_fixup);
 764
 765         if (c->ro_error)
 766                 return -EROFS;
 767
 768         ubifs_prepare_node(c, buf, len, 1);
 769         err = ubi_leb_write(c->ubi, lnum, buf, offs, buf_len, dtype);
 770         if (err) {
 771                 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
 772                           buf_len, lnum, offs, err);
 773                 dbg_dump_node(c, buf);
 774                 dbg_dump_stack();
 775         }
 776
 777         return err;
 778 }
 779
 780 /**
 781  * ubifs_read_node_wbuf - read node from the media or write-buffer.
 782  * @wbuf: wbuf to check for un-written data
 783  * @buf: buffer to read to
 784  * @type: node type
 785  * @len: node length
 786  * @lnum: logical eraseblock number
 787  * @offs: offset within the logical eraseblock
 788  *
 789  * This function reads a node of known type and length, checks it and stores
 790  * in @buf. If the node partially or fully sits in the write-buffer, this
 791  * function takes data from the buffer, otherwise it reads the flash media.
 792  * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
 793  * error code in case of failure.
 794  */
 795 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
 796                          int lnum, int offs)
 797 {
 798         const struct ubifs_info *c = wbuf->c;
 799         int err, rlen, overlap;
 800         struct ubifs_ch *ch = buf;
 801
 802         dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
 803                dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
 804         ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 805         ubifs_assert(!(offs & 7) && offs < c->leb_size);
 806         ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
 807
 808         spin_lock(&wbuf->lock);
 809         overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
 810         if (!overlap) {
 811                 /* We may safely unlock the write-buffer and read the data */
 812                 spin_unlock(&wbuf->lock);
 813                 return ubifs_read_node(c, buf, type, len, lnum, offs);
 814         }
 815
 816         /* Don't read under wbuf */
 817         rlen = wbuf->offs - offs;
 818         if (rlen < 0)
 819                 rlen = 0;
 820
 821         /* Copy the rest from the write-buffer */
 822         memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
 823         spin_unlock(&wbuf->lock);
 824
 825         if (rlen > 0) {
 826                 /* Read everything that goes before write-buffer */
 827                 err = ubi_read(c->ubi, lnum, buf, offs, rlen);
 828                 if (err && err != -EBADMSG) {
 829                         ubifs_err("failed to read node %d from LEB %d:%d, "
 830                                   "error %d", type, lnum, offs, err);
 831                         dbg_dump_stack();
 832                         return err;
 833                 }
 834         }
 835
 836         if (type != ch->node_type) {
 837                 ubifs_err("bad node type (%d but expected %d)",
 838                           ch->node_type, type);
 839                 goto out;
 840         }
 841
 842         err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
 843         if (err) {
 844                 ubifs_err("expected node type %d", type);
 845                 return err;
 846         }
 847
 848         rlen = le32_to_cpu(ch->len);
 849         if (rlen != len) {
 850                 ubifs_err("bad node length %d, expected %d", rlen, len);
 851                 goto out;
 852         }
 853
 854         return 0;
 855
 856 out:
 857         ubifs_err("bad node at LEB %d:%d", lnum, offs);
 858         dbg_dump_node(c, buf);
 859         dbg_dump_stack();
 860         return -EINVAL;
 861 }
 862
 863 /**
 864  * ubifs_read_node - read node.
 865  * @c: UBIFS file-system description object
 866  * @buf: buffer to read to
 867  * @type: node type
 868  * @len: node length (not aligned)
 869  * @lnum: logical eraseblock number
 870  * @offs: offset within the logical eraseblock
 871  *
 872  * This function reads a node of known type and and length, checks it and
 873  * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
 874  * and a negative error code in case of failure.
 875  */
 876 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
 877                     int lnum, int offs)
 878 {
 879         int err, l;
 880         struct ubifs_ch *ch = buf;
 881
 882         dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
 883         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 884         ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
 885         ubifs_assert(!(offs & 7) && offs < c->leb_size);
 886         ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
 887
 888         err = ubi_read(c->ubi, lnum, buf, offs, len);
 889         if (err && err != -EBADMSG) {
 890                 ubifs_err("cannot read node %d from LEB %d:%d, error %d",
 891                           type, lnum, offs, err);
 892                 return err;
 893         }
 894
 895         if (type != ch->node_type) {
 896                 ubifs_err("bad node type (%d but expected %d)",
 897                           ch->node_type, type);
 898                 goto out;
 899         }
 900
 901         err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
 902         if (err) {
 903                 ubifs_err("expected node type %d", type);
 904                 return err;
 905         }
 906
 907         l = le32_to_cpu(ch->len);
 908         if (l != len) {
 909                 ubifs_err("bad node length %d, expected %d", l, len);
 910                 goto out;
 911         }
 912
 913         return 0;
 914
 915 out:
 916         ubifs_err("bad node at LEB %d:%d, LEB mapping status %d", lnum, offs,
 917                   ubi_is_mapped(c->ubi, lnum));
 918         dbg_dump_node(c, buf);
 919         dbg_dump_stack();
 920         return -EINVAL;
 921 }
 922
 923 /**
 924  * ubifs_wbuf_init - initialize write-buffer.
 925  * @c: UBIFS file-system description object
 926  * @wbuf: write-buffer to initialize
 927  *
 928  * This function initializes write-buffer. Returns zero in case of success
 929  * %-ENOMEM in case of failure.
 930  */
 931 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
 932 {
 933         size_t size;
 934
 935         wbuf->buf = kmalloc(c->max_write_size, GFP_KERNEL);
 936         if (!wbuf->buf)
 937                 return -ENOMEM;
 938
 939         size = (c->max_write_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
 940         wbuf->inodes = kmalloc(size, GFP_KERNEL);
 941         if (!wbuf->inodes) {
 942                 kfree(wbuf->buf);
 943                 wbuf->buf = NULL;
 944                 return -ENOMEM;
 945         }
 946
 947         wbuf->used = 0;
 948         wbuf->lnum = wbuf->offs = -1;
 949         /*
 950          * If the LEB starts at the max. write size aligned address, then
 951          * write-buffer size has to be set to @c->max_write_size. Otherwise,
 952          * set it to something smaller so that it ends at the closest max.
 953          * write size boundary.
 954          */
 955         size = c->max_write_size - (c->leb_start % c->max_write_size);
 956         wbuf->avail = wbuf->size = size;
 957         wbuf->dtype = UBI_UNKNOWN;
 958         wbuf->sync_callback = NULL;
 959         mutex_init(&wbuf->io_mutex);
 960         spin_lock_init(&wbuf->lock);
 961         wbuf->c = c;
 962         wbuf->next_ino = 0;
 963
 964         hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 965         wbuf->timer.function = wbuf_timer_callback_nolock;
 966         wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
 967         wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
 968         wbuf->delta *= 1000000000ULL;
 969         ubifs_assert(wbuf->delta <= ULONG_MAX);
 970         return 0;
 971 }
 972
 973 /**
 974  * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
 975  * @wbuf: the write-buffer where to add
 976  * @inum: the inode number
 977  *
 978  * This function adds an inode number to the inode array of the write-buffer.
 979  */
 980 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
 981 {
 982         if (!wbuf->buf)
 983                 /* NOR flash or something similar */
 984                 return;
 985
 986         spin_lock(&wbuf->lock);
 987         if (wbuf->used)
 988                 wbuf->inodes[wbuf->next_ino++] = inum;
 989         spin_unlock(&wbuf->lock);
 990 }
 991
 992 /**
 993  * wbuf_has_ino - returns if the wbuf contains data from the inode.
 994  * @wbuf: the write-buffer
 995  * @inum: the inode number
 996  *
 997  * This function returns with %1 if the write-buffer contains some data from the
 998  * given inode otherwise it returns with %0.
 999  */
1000 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
1001 {
1002         int i, ret = 0;
1003
1004         spin_lock(&wbuf->lock);
1005         for (i = 0; i < wbuf->next_ino; i++)
1006                 if (inum == wbuf->inodes[i]) {
1007                         ret = 1;
1008                         break;
1009                 }
1010         spin_unlock(&wbuf->lock);
1011
1012         return ret;
1013 }
1014
1015 /**
1016  * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
1017  * @c: UBIFS file-system description object
1018  * @inode: inode to synchronize
1019  *
1020  * This function synchronizes write-buffers which contain nodes belonging to
1021  * @inode. Returns zero in case of success and a negative error code in case of
1022  * failure.
1023  */
1024 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
1025 {
1026         int i, err = 0;
1027
1028         for (i = 0; i < c->jhead_cnt; i++) {
1029                 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
1030
1031                 if (i == GCHD)
1032                         /*
1033                          * GC head is special, do not look at it. Even if the
1034                          * head contains something related to this inode, it is
1035                          * a _copy_ of corresponding on-flash node which sits
1036                          * somewhere else.
1037                          */
1038                         continue;
1039
1040                 if (!wbuf_has_ino(wbuf, inode->i_ino))
1041                         continue;
1042
1043                 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1044                 if (wbuf_has_ino(wbuf, inode->i_ino))
1045                         err = ubifs_wbuf_sync_nolock(wbuf);
1046                 mutex_unlock(&wbuf->io_mutex);
1047
1048                 if (err) {
1049                         ubifs_ro_mode(c, err);
1050                         return err;
1051                 }
1052         }
1053         return 0;
1054 }