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  * Write-buffers are defined by 'struct ubifs_wbuf' objects and protected by
  35  * mutexes defined inside these objects. Since sometimes upper-level code
  36  * has to lock the write-buffer (e.g. journal space reservation code), many
  37  * functions related to write-buffers have "nolock" suffix which means that the
  38  * caller has to lock the write-buffer before calling this function.
  39  *
  40  * UBIFS stores nodes at 64 bit-aligned addresses. If the node length is not
  41  * aligned, UBIFS starts the next node from the aligned address, and the padded
  42  * bytes may contain any rubbish. In other words, UBIFS does not put padding
  43  * bytes in those small gaps. Common headers of nodes store real node lengths,
  44  * not aligned lengths. Indexing nodes also store real lengths in branches.
  45  *
  46  * UBIFS uses padding when it pads to the next min. I/O unit. In this case it
  47  * uses padding nodes or padding bytes, if the padding node does not fit.
  48  *
  49  * All UBIFS nodes are protected by CRC checksums and UBIFS checks all nodes
  50  * every time they are read from the flash media.
  51  */
  52
  53 #include <linux/crc32.h>
  54 #include <linux/slab.h>
  55 #include "ubifs.h"
  56
  57 /**
  58  * ubifs_ro_mode - switch UBIFS to read read-only mode.
  59  * @c: UBIFS file-system description object
  60  * @err: error code which is the reason of switching to R/O mode
  61  */
  62 void ubifs_ro_mode(struct ubifs_info *c, int err)
  63 {
  64         if (!c->ro_media) {
  65                 c->ro_media = 1;
  66                 c->no_chk_data_crc = 0;
  67                 c->vfs_sb->s_flags |= MS_RDONLY;
  68                 ubifs_warn("switched to read-only mode, error %d", err);
  69                 dbg_dump_stack();
  70         }
  71 }
  72
  73 /**
  74  * ubifs_check_node - check node.
  75  * @c: UBIFS file-system description object
  76  * @buf: node to check
  77  * @lnum: logical eraseblock number
  78  * @offs: offset within the logical eraseblock
  79  * @quiet: print no messages
  80  * @must_chk_crc: indicates whether to always check the CRC
  81  *
  82  * This function checks node magic number and CRC checksum. This function also
  83  * validates node length to prevent UBIFS from becoming crazy when an attacker
  84  * feeds it a file-system image with incorrect nodes. For example, too large
  85  * node length in the common header could cause UBIFS to read memory outside of
  86  * allocated buffer when checking the CRC checksum.
  87  *
  88  * This function may skip data nodes CRC checking if @c->no_chk_data_crc is
  89  * true, which is controlled by corresponding UBIFS mount option. However, if
  90  * @must_chk_crc is true, then @c->no_chk_data_crc is ignored and CRC is
  91  * checked. Similarly, if @c->always_chk_crc is true, @c->no_chk_data_crc is
  92  * ignored and CRC is checked.
  93  *
  94  * This function returns zero in case of success and %-EUCLEAN in case of bad
  95  * CRC or magic.
  96  */
  97 int ubifs_check_node(const struct ubifs_info *c, const void *buf, int lnum,
  98                      int offs, int quiet, int must_chk_crc)
  99 {
 100         int err = -EINVAL, type, node_len;
 101         uint32_t crc, node_crc, magic;
 102         const struct ubifs_ch *ch = buf;
 103
 104         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 105         ubifs_assert(!(offs & 7) && offs < c->leb_size);
 106
 107         magic = le32_to_cpu(ch->magic);
 108         if (magic != UBIFS_NODE_MAGIC) {
 109                 if (!quiet)
 110                         ubifs_err("bad magic %#08x, expected %#08x",
 111                                   magic, UBIFS_NODE_MAGIC);
 112                 err = -EUCLEAN;
 113                 goto out;
 114         }
 115
 116         type = ch->node_type;
 117         if (type < 0 || type >= UBIFS_NODE_TYPES_CNT) {
 118                 if (!quiet)
 119                         ubifs_err("bad node type %d", type);
 120                 goto out;
 121         }
 122
 123         node_len = le32_to_cpu(ch->len);
 124         if (node_len + offs > c->leb_size)
 125                 goto out_len;
 126
 127         if (c->ranges[type].max_len == 0) {
 128                 if (node_len != c->ranges[type].len)
 129                         goto out_len;
 130         } else if (node_len < c->ranges[type].min_len ||
 131                    node_len > c->ranges[type].max_len)
 132                 goto out_len;
 133
 134         if (!must_chk_crc && type == UBIFS_DATA_NODE && !c->always_chk_crc &&
 135              c->no_chk_data_crc)
 136                 return 0;
 137
 138         crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
 139         node_crc = le32_to_cpu(ch->crc);
 140         if (crc != node_crc) {
 141                 if (!quiet)
 142                         ubifs_err("bad CRC: calculated %#08x, read %#08x",
 143                                   crc, node_crc);
 144                 err = -EUCLEAN;
 145                 goto out;
 146         }
 147
 148         return 0;
 149
 150 out_len:
 151         if (!quiet)
 152                 ubifs_err("bad node length %d", node_len);
 153 out:
 154         if (!quiet) {
 155                 ubifs_err("bad node at LEB %d:%d", lnum, offs);
 156                 dbg_dump_node(c, buf);
 157                 dbg_dump_stack();
 158         }
 159         return err;
 160 }
 161
 162 /**
 163  * ubifs_pad - pad flash space.
 164  * @c: UBIFS file-system description object
 165  * @buf: buffer to put padding to
 166  * @pad: how many bytes to pad
 167  *
 168  * The flash media obliges us to write only in chunks of %c->min_io_size and
 169  * when we have to write less data we add padding node to the write-buffer and
 170  * pad it to the next minimal I/O unit's boundary. Padding nodes help when the
 171  * media is being scanned. If the amount of wasted space is not enough to fit a
 172  * padding node which takes %UBIFS_PAD_NODE_SZ bytes, we write padding bytes
 173  * pattern (%UBIFS_PADDING_BYTE).
 174  *
 175  * Padding nodes are also used to fill gaps when the "commit-in-gaps" method is
 176  * used.
 177  */
 178 void ubifs_pad(const struct ubifs_info *c, void *buf, int pad)
 179 {
 180         uint32_t crc;
 181
 182         ubifs_assert(pad >= 0 && !(pad & 7));
 183
 184         if (pad >= UBIFS_PAD_NODE_SZ) {
 185                 struct ubifs_ch *ch = buf;
 186                 struct ubifs_pad_node *pad_node = buf;
 187
 188                 ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 189                 ch->node_type = UBIFS_PAD_NODE;
 190                 ch->group_type = UBIFS_NO_NODE_GROUP;
 191                 ch->padding[0] = ch->padding[1] = 0;
 192                 ch->sqnum = 0;
 193                 ch->len = cpu_to_le32(UBIFS_PAD_NODE_SZ);
 194                 pad -= UBIFS_PAD_NODE_SZ;
 195                 pad_node->pad_len = cpu_to_le32(pad);
 196                 crc = crc32(UBIFS_CRC32_INIT, buf + 8, UBIFS_PAD_NODE_SZ - 8);
 197                 ch->crc = cpu_to_le32(crc);
 198                 memset(buf + UBIFS_PAD_NODE_SZ, 0, pad);
 199         } else if (pad > 0)
 200                 /* Too little space, padding node won't fit */
 201                 memset(buf, UBIFS_PADDING_BYTE, pad);
 202 }
 203
 204 /**
 205  * next_sqnum - get next sequence number.
 206  * @c: UBIFS file-system description object
 207  */
 208 static unsigned long long next_sqnum(struct ubifs_info *c)
 209 {
 210         unsigned long long sqnum;
 211
 212         spin_lock(&c->cnt_lock);
 213         sqnum = ++c->max_sqnum;
 214         spin_unlock(&c->cnt_lock);
 215
 216         if (unlikely(sqnum >= SQNUM_WARN_WATERMARK)) {
 217                 if (sqnum >= SQNUM_WATERMARK) {
 218                         ubifs_err("sequence number overflow %llu, end of life",
 219                                   sqnum);
 220                         ubifs_ro_mode(c, -EINVAL);
 221                 }
 222                 ubifs_warn("running out of sequence numbers, end of life soon");
 223         }
 224
 225         return sqnum;
 226 }
 227
 228 /**
 229  * ubifs_prepare_node - prepare node to be written to flash.
 230  * @c: UBIFS file-system description object
 231  * @node: the node to pad
 232  * @len: node length
 233  * @pad: if the buffer has to be padded
 234  *
 235  * This function prepares node at @node to be written to the media - it
 236  * calculates node CRC, fills the common header, and adds proper padding up to
 237  * the next minimum I/O unit if @pad is not zero.
 238  */
 239 void ubifs_prepare_node(struct ubifs_info *c, void *node, int len, int pad)
 240 {
 241         uint32_t crc;
 242         struct ubifs_ch *ch = node;
 243         unsigned long long sqnum = next_sqnum(c);
 244
 245         ubifs_assert(len >= UBIFS_CH_SZ);
 246
 247         ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 248         ch->len = cpu_to_le32(len);
 249         ch->group_type = UBIFS_NO_NODE_GROUP;
 250         ch->sqnum = cpu_to_le64(sqnum);
 251         ch->padding[0] = ch->padding[1] = 0;
 252         crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
 253         ch->crc = cpu_to_le32(crc);
 254
 255         if (pad) {
 256                 len = ALIGN(len, 8);
 257                 pad = ALIGN(len, c->min_io_size) - len;
 258                 ubifs_pad(c, node + len, pad);
 259         }
 260 }
 261
 262 /**
 263  * ubifs_prep_grp_node - prepare node of a group to be written to flash.
 264  * @c: UBIFS file-system description object
 265  * @node: the node to pad
 266  * @len: node length
 267  * @last: indicates the last node of the group
 268  *
 269  * This function prepares node at @node to be written to the media - it
 270  * calculates node CRC and fills the common header.
 271  */
 272 void ubifs_prep_grp_node(struct ubifs_info *c, void *node, int len, int last)
 273 {
 274         uint32_t crc;
 275         struct ubifs_ch *ch = node;
 276         unsigned long long sqnum = next_sqnum(c);
 277
 278         ubifs_assert(len >= UBIFS_CH_SZ);
 279
 280         ch->magic = cpu_to_le32(UBIFS_NODE_MAGIC);
 281         ch->len = cpu_to_le32(len);
 282         if (last)
 283                 ch->group_type = UBIFS_LAST_OF_NODE_GROUP;
 284         else
 285                 ch->group_type = UBIFS_IN_NODE_GROUP;
 286         ch->sqnum = cpu_to_le64(sqnum);
 287         ch->padding[0] = ch->padding[1] = 0;
 288         crc = crc32(UBIFS_CRC32_INIT, node + 8, len - 8);
 289         ch->crc = cpu_to_le32(crc);
 290 }
 291
 292 /**
 293  * wbuf_timer_callback - write-buffer timer callback function.
 294  * @data: timer data (write-buffer descriptor)
 295  *
 296  * This function is called when the write-buffer timer expires.
 297  */
 298 static enum hrtimer_restart wbuf_timer_callback_nolock(struct hrtimer *timer)
 299 {
 300         struct ubifs_wbuf *wbuf = container_of(timer, struct ubifs_wbuf, timer);
 301
 302         dbg_io("jhead %s", dbg_jhead(wbuf->jhead));
 303         wbuf->need_sync = 1;
 304         wbuf->c->need_wbuf_sync = 1;
 305         ubifs_wake_up_bgt(wbuf->c);
 306         return HRTIMER_NORESTART;
 307 }
 308
 309 /**
 310  * new_wbuf_timer - start new write-buffer timer.
 311  * @wbuf: write-buffer descriptor
 312  */
 313 static void new_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
 314 {
 315         ubifs_assert(!hrtimer_active(&wbuf->timer));
 316
 317         if (wbuf->no_timer)
 318                 return;
 319         dbg_io("set timer for jhead %s, %llu-%llu millisecs",
 320                dbg_jhead(wbuf->jhead),
 321                div_u64(ktime_to_ns(wbuf->softlimit), USEC_PER_SEC),
 322                div_u64(ktime_to_ns(wbuf->softlimit) + wbuf->delta,
 323                        USEC_PER_SEC));
 324         hrtimer_start_range_ns(&wbuf->timer, wbuf->softlimit, wbuf->delta,
 325                                HRTIMER_MODE_REL);
 326 }
 327
 328 /**
 329  * cancel_wbuf_timer - cancel write-buffer timer.
 330  * @wbuf: write-buffer descriptor
 331  */
 332 static void cancel_wbuf_timer_nolock(struct ubifs_wbuf *wbuf)
 333 {
 334         if (wbuf->no_timer)
 335                 return;
 336         wbuf->need_sync = 0;
 337         hrtimer_cancel(&wbuf->timer);
 338 }
 339
 340 /**
 341  * ubifs_wbuf_sync_nolock - synchronize write-buffer.
 342  * @wbuf: write-buffer to synchronize
 343  *
 344  * This function synchronizes write-buffer @buf and returns zero in case of
 345  * success or a negative error code in case of failure.
 346  */
 347 int ubifs_wbuf_sync_nolock(struct ubifs_wbuf *wbuf)
 348 {
 349         struct ubifs_info *c = wbuf->c;
 350         int err, dirt;
 351
 352         cancel_wbuf_timer_nolock(wbuf);
 353         if (!wbuf->used || wbuf->lnum == -1)
 354                 /* Write-buffer is empty or not seeked */
 355                 return 0;
 356
 357         dbg_io("LEB %d:%d, %d bytes, jhead %s",
 358                wbuf->lnum, wbuf->offs, wbuf->used, dbg_jhead(wbuf->jhead));
 359         ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY));
 360         ubifs_assert(!(wbuf->avail & 7));
 361         ubifs_assert(wbuf->offs + c->min_io_size <= c->leb_size);
 362
 363         if (c->ro_media)
 364                 return -EROFS;
 365
 366         ubifs_pad(c, wbuf->buf + wbuf->used, wbuf->avail);
 367         err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
 368                             c->min_io_size, wbuf->dtype);
 369         if (err) {
 370                 ubifs_err("cannot write %d bytes to LEB %d:%d",
 371                           c->min_io_size, wbuf->lnum, wbuf->offs);
 372                 dbg_dump_stack();
 373                 return err;
 374         }
 375
 376         dirt = wbuf->avail;
 377
 378         spin_lock(&wbuf->lock);
 379         wbuf->offs += c->min_io_size;
 380         wbuf->avail = c->min_io_size;
 381         wbuf->used = 0;
 382         wbuf->next_ino = 0;
 383         spin_unlock(&wbuf->lock);
 384
 385         if (wbuf->sync_callback)
 386                 err = wbuf->sync_callback(c, wbuf->lnum,
 387                                           c->leb_size - wbuf->offs, dirt);
 388         return err;
 389 }
 390
 391 /**
 392  * ubifs_wbuf_seek_nolock - seek write-buffer.
 393  * @wbuf: write-buffer
 394  * @lnum: logical eraseblock number to seek to
 395  * @offs: logical eraseblock offset to seek to
 396  * @dtype: data type
 397  *
 398  * This function targets the write-buffer to logical eraseblock @lnum:@offs.
 399  * The write-buffer is synchronized if it is not empty. Returns zero in case of
 400  * success and a negative error code in case of failure.
 401  */
 402 int ubifs_wbuf_seek_nolock(struct ubifs_wbuf *wbuf, int lnum, int offs,
 403                            int dtype)
 404 {
 405         const struct ubifs_info *c = wbuf->c;
 406
 407         dbg_io("LEB %d:%d, jhead %s", lnum, offs, dbg_jhead(wbuf->jhead));
 408         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt);
 409         ubifs_assert(offs >= 0 && offs <= c->leb_size);
 410         ubifs_assert(offs % c->min_io_size == 0 && !(offs & 7));
 411         ubifs_assert(lnum != wbuf->lnum);
 412
 413         if (wbuf->used > 0) {
 414                 int err = ubifs_wbuf_sync_nolock(wbuf);
 415
 416                 if (err)
 417                         return err;
 418         }
 419
 420         spin_lock(&wbuf->lock);
 421         wbuf->lnum = lnum;
 422         wbuf->offs = offs;
 423         wbuf->avail = c->min_io_size;
 424         wbuf->used = 0;
 425         spin_unlock(&wbuf->lock);
 426         wbuf->dtype = dtype;
 427
 428         return 0;
 429 }
 430
 431 /**
 432  * ubifs_bg_wbufs_sync - synchronize write-buffers.
 433  * @c: UBIFS file-system description object
 434  *
 435  * This function is called by background thread to synchronize write-buffers.
 436  * Returns zero in case of success and a negative error code in case of
 437  * failure.
 438  */
 439 int ubifs_bg_wbufs_sync(struct ubifs_info *c)
 440 {
 441         int err, i;
 442
 443         if (!c->need_wbuf_sync)
 444                 return 0;
 445         c->need_wbuf_sync = 0;
 446
 447         if (c->ro_media) {
 448                 err = -EROFS;
 449                 goto out_timers;
 450         }
 451
 452         dbg_io("synchronize");
 453         for (i = 0; i < c->jhead_cnt; i++) {
 454                 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 455
 456                 cond_resched();
 457
 458                 /*
 459                  * If the mutex is locked then wbuf is being changed, so
 460                  * synchronization is not necessary.
 461                  */
 462                 if (mutex_is_locked(&wbuf->io_mutex))
 463                         continue;
 464
 465                 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 466                 if (!wbuf->need_sync) {
 467                         mutex_unlock(&wbuf->io_mutex);
 468                         continue;
 469                 }
 470
 471                 err = ubifs_wbuf_sync_nolock(wbuf);
 472                 mutex_unlock(&wbuf->io_mutex);
 473                 if (err) {
 474                         ubifs_err("cannot sync write-buffer, error %d", err);
 475                         ubifs_ro_mode(c, err);
 476                         goto out_timers;
 477                 }
 478         }
 479
 480         return 0;
 481
 482 out_timers:
 483         /* Cancel all timers to prevent repeated errors */
 484         for (i = 0; i < c->jhead_cnt; i++) {
 485                 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 486
 487                 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 488                 cancel_wbuf_timer_nolock(wbuf);
 489                 mutex_unlock(&wbuf->io_mutex);
 490         }
 491         return err;
 492 }
 493
 494 /**
 495  * ubifs_wbuf_write_nolock - write data to flash via write-buffer.
 496  * @wbuf: write-buffer
 497  * @buf: node to write
 498  * @len: node length
 499  *
 500  * This function writes data to flash via write-buffer @wbuf. This means that
 501  * the last piece of the node won't reach the flash media immediately if it
 502  * does not take whole minimal I/O unit. Instead, the node will sit in RAM
 503  * until the write-buffer is synchronized (e.g., by timer).
 504  *
 505  * This function returns zero in case of success and a negative error code in
 506  * case of failure. If the node cannot be written because there is no more
 507  * space in this logical eraseblock, %-ENOSPC is returned.
 508  */
 509 int ubifs_wbuf_write_nolock(struct ubifs_wbuf *wbuf, void *buf, int len)
 510 {
 511         struct ubifs_info *c = wbuf->c;
 512         int err, written, n, aligned_len = ALIGN(len, 8), offs;
 513
 514         dbg_io("%d bytes (%s) to jhead %s wbuf at LEB %d:%d", len,
 515                dbg_ntype(((struct ubifs_ch *)buf)->node_type),
 516                dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs + wbuf->used);
 517         ubifs_assert(len > 0 && wbuf->lnum >= 0 && wbuf->lnum < c->leb_cnt);
 518         ubifs_assert(wbuf->offs >= 0 && wbuf->offs % c->min_io_size == 0);
 519         ubifs_assert(!(wbuf->offs & 7) && wbuf->offs <= c->leb_size);
 520         ubifs_assert(wbuf->avail > 0 && wbuf->avail <= c->min_io_size);
 521         ubifs_assert(mutex_is_locked(&wbuf->io_mutex));
 522
 523         if (c->leb_size - wbuf->offs - wbuf->used < aligned_len) {
 524                 err = -ENOSPC;
 525                 goto out;
 526         }
 527
 528         cancel_wbuf_timer_nolock(wbuf);
 529
 530         if (c->ro_media)
 531                 return -EROFS;
 532
 533         if (aligned_len <= wbuf->avail) {
 534                 /*
 535                  * The node is not very large and fits entirely within
 536                  * write-buffer.
 537                  */
 538                 memcpy(wbuf->buf + wbuf->used, buf, len);
 539
 540                 if (aligned_len == wbuf->avail) {
 541                         dbg_io("flush jhead %s wbuf to LEB %d:%d",
 542                                dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
 543                         err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf,
 544                                             wbuf->offs, c->min_io_size,
 545                                             wbuf->dtype);
 546                         if (err)
 547                                 goto out;
 548
 549                         spin_lock(&wbuf->lock);
 550                         wbuf->offs += c->min_io_size;
 551                         wbuf->avail = c->min_io_size;
 552                         wbuf->used = 0;
 553                         wbuf->next_ino = 0;
 554                         spin_unlock(&wbuf->lock);
 555                 } else {
 556                         spin_lock(&wbuf->lock);
 557                         wbuf->avail -= aligned_len;
 558                         wbuf->used += aligned_len;
 559                         spin_unlock(&wbuf->lock);
 560                 }
 561
 562                 goto exit;
 563         }
 564
 565         /*
 566          * The node is large enough and does not fit entirely within current
 567          * minimal I/O unit. We have to fill and flush write-buffer and switch
 568          * to the next min. I/O unit.
 569          */
 570         dbg_io("flush jhead %s wbuf to LEB %d:%d",
 571                dbg_jhead(wbuf->jhead), wbuf->lnum, wbuf->offs);
 572         memcpy(wbuf->buf + wbuf->used, buf, wbuf->avail);
 573         err = ubi_leb_write(c->ubi, wbuf->lnum, wbuf->buf, wbuf->offs,
 574                             c->min_io_size, wbuf->dtype);
 575         if (err)
 576                 goto out;
 577
 578         offs = wbuf->offs + c->min_io_size;
 579         len -= wbuf->avail;
 580         aligned_len -= wbuf->avail;
 581         written = wbuf->avail;
 582
 583         /*
 584          * The remaining data may take more whole min. I/O units, so write the
 585          * remains multiple to min. I/O unit size directly to the flash media.
 586          * We align node length to 8-byte boundary because we anyway flash wbuf
 587          * if the remaining space is less than 8 bytes.
 588          */
 589         n = aligned_len >> c->min_io_shift;
 590         if (n) {
 591                 n <<= c->min_io_shift;
 592                 dbg_io("write %d bytes to LEB %d:%d", n, wbuf->lnum, offs);
 593                 err = ubi_leb_write(c->ubi, wbuf->lnum, buf + written, offs, n,
 594                                     wbuf->dtype);
 595                 if (err)
 596                         goto out;
 597                 offs += n;
 598                 aligned_len -= n;
 599                 len -= n;
 600                 written += n;
 601         }
 602
 603         spin_lock(&wbuf->lock);
 604         if (aligned_len)
 605                 /*
 606                  * And now we have what's left and what does not take whole
 607                  * min. I/O unit, so write it to the write-buffer and we are
 608                  * done.
 609                  */
 610                 memcpy(wbuf->buf, buf + written, len);
 611
 612         wbuf->offs = offs;
 613         wbuf->used = aligned_len;
 614         wbuf->avail = c->min_io_size - aligned_len;
 615         wbuf->next_ino = 0;
 616         spin_unlock(&wbuf->lock);
 617
 618 exit:
 619         if (wbuf->sync_callback) {
 620                 int free = c->leb_size - wbuf->offs - wbuf->used;
 621
 622                 err = wbuf->sync_callback(c, wbuf->lnum, free, 0);
 623                 if (err)
 624                         goto out;
 625         }
 626
 627         if (wbuf->used)
 628                 new_wbuf_timer_nolock(wbuf);
 629
 630         return 0;
 631
 632 out:
 633         ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
 634                   len, wbuf->lnum, wbuf->offs, err);
 635         dbg_dump_node(c, buf);
 636         dbg_dump_stack();
 637         dbg_dump_leb(c, wbuf->lnum);
 638         return err;
 639 }
 640
 641 /**
 642  * ubifs_write_node - write node to the media.
 643  * @c: UBIFS file-system description object
 644  * @buf: the node to write
 645  * @len: node length
 646  * @lnum: logical eraseblock number
 647  * @offs: offset within the logical eraseblock
 648  * @dtype: node life-time hint (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
 649  *
 650  * This function automatically fills node magic number, assigns sequence
 651  * number, and calculates node CRC checksum. The length of the @buf buffer has
 652  * to be aligned to the minimal I/O unit size. This function automatically
 653  * appends padding node and padding bytes if needed. Returns zero in case of
 654  * success and a negative error code in case of failure.
 655  */
 656 int ubifs_write_node(struct ubifs_info *c, void *buf, int len, int lnum,
 657                      int offs, int dtype)
 658 {
 659         int err, buf_len = ALIGN(len, c->min_io_size);
 660
 661         dbg_io("LEB %d:%d, %s, length %d (aligned %d)",
 662                lnum, offs, dbg_ntype(((struct ubifs_ch *)buf)->node_type), len,
 663                buf_len);
 664         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 665         ubifs_assert(offs % c->min_io_size == 0 && offs < c->leb_size);
 666
 667         if (c->ro_media)
 668                 return -EROFS;
 669
 670         ubifs_prepare_node(c, buf, len, 1);
 671         err = ubi_leb_write(c->ubi, lnum, buf, offs, buf_len, dtype);
 672         if (err) {
 673                 ubifs_err("cannot write %d bytes to LEB %d:%d, error %d",
 674                           buf_len, lnum, offs, err);
 675                 dbg_dump_node(c, buf);
 676                 dbg_dump_stack();
 677         }
 678
 679         return err;
 680 }
 681
 682 /**
 683  * ubifs_read_node_wbuf - read node from the media or write-buffer.
 684  * @wbuf: wbuf to check for un-written data
 685  * @buf: buffer to read to
 686  * @type: node type
 687  * @len: node length
 688  * @lnum: logical eraseblock number
 689  * @offs: offset within the logical eraseblock
 690  *
 691  * This function reads a node of known type and length, checks it and stores
 692  * in @buf. If the node partially or fully sits in the write-buffer, this
 693  * function takes data from the buffer, otherwise it reads the flash media.
 694  * Returns zero in case of success, %-EUCLEAN if CRC mismatched and a negative
 695  * error code in case of failure.
 696  */
 697 int ubifs_read_node_wbuf(struct ubifs_wbuf *wbuf, void *buf, int type, int len,
 698                          int lnum, int offs)
 699 {
 700         const struct ubifs_info *c = wbuf->c;
 701         int err, rlen, overlap;
 702         struct ubifs_ch *ch = buf;
 703
 704         dbg_io("LEB %d:%d, %s, length %d, jhead %s", lnum, offs,
 705                dbg_ntype(type), len, dbg_jhead(wbuf->jhead));
 706         ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 707         ubifs_assert(!(offs & 7) && offs < c->leb_size);
 708         ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
 709
 710         spin_lock(&wbuf->lock);
 711         overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
 712         if (!overlap) {
 713                 /* We may safely unlock the write-buffer and read the data */
 714                 spin_unlock(&wbuf->lock);
 715                 return ubifs_read_node(c, buf, type, len, lnum, offs);
 716         }
 717
 718         /* Don't read under wbuf */
 719         rlen = wbuf->offs - offs;
 720         if (rlen < 0)
 721                 rlen = 0;
 722
 723         /* Copy the rest from the write-buffer */
 724         memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
 725         spin_unlock(&wbuf->lock);
 726
 727         if (rlen > 0) {
 728                 /* Read everything that goes before write-buffer */
 729                 err = ubi_read(c->ubi, lnum, buf, offs, rlen);
 730                 if (err && err != -EBADMSG) {
 731                         ubifs_err("failed to read node %d from LEB %d:%d, "
 732                                   "error %d", type, lnum, offs, err);
 733                         dbg_dump_stack();
 734                         return err;
 735                 }
 736         }
 737
 738         if (type != ch->node_type) {
 739                 ubifs_err("bad node type (%d but expected %d)",
 740                           ch->node_type, type);
 741                 goto out;
 742         }
 743
 744         err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
 745         if (err) {
 746                 ubifs_err("expected node type %d", type);
 747                 return err;
 748         }
 749
 750         rlen = le32_to_cpu(ch->len);
 751         if (rlen != len) {
 752                 ubifs_err("bad node length %d, expected %d", rlen, len);
 753                 goto out;
 754         }
 755
 756         return 0;
 757
 758 out:
 759         ubifs_err("bad node at LEB %d:%d", lnum, offs);
 760         dbg_dump_node(c, buf);
 761         dbg_dump_stack();
 762         return -EINVAL;
 763 }
 764
 765 /**
 766  * ubifs_read_node - read node.
 767  * @c: UBIFS file-system description object
 768  * @buf: buffer to read to
 769  * @type: node type
 770  * @len: node length (not aligned)
 771  * @lnum: logical eraseblock number
 772  * @offs: offset within the logical eraseblock
 773  *
 774  * This function reads a node of known type and and length, checks it and
 775  * stores in @buf. Returns zero in case of success, %-EUCLEAN if CRC mismatched
 776  * and a negative error code in case of failure.
 777  */
 778 int ubifs_read_node(const struct ubifs_info *c, void *buf, int type, int len,
 779                     int lnum, int offs)
 780 {
 781         int err, l;
 782         struct ubifs_ch *ch = buf;
 783
 784         dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
 785         ubifs_assert(lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
 786         ubifs_assert(len >= UBIFS_CH_SZ && offs + len <= c->leb_size);
 787         ubifs_assert(!(offs & 7) && offs < c->leb_size);
 788         ubifs_assert(type >= 0 && type < UBIFS_NODE_TYPES_CNT);
 789
 790         err = ubi_read(c->ubi, lnum, buf, offs, len);
 791         if (err && err != -EBADMSG) {
 792                 ubifs_err("cannot read node %d from LEB %d:%d, error %d",
 793                           type, lnum, offs, err);
 794                 return err;
 795         }
 796
 797         if (type != ch->node_type) {
 798                 ubifs_err("bad node type (%d but expected %d)",
 799                           ch->node_type, type);
 800                 goto out;
 801         }
 802
 803         err = ubifs_check_node(c, buf, lnum, offs, 0, 0);
 804         if (err) {
 805                 ubifs_err("expected node type %d", type);
 806                 return err;
 807         }
 808
 809         l = le32_to_cpu(ch->len);
 810         if (l != len) {
 811                 ubifs_err("bad node length %d, expected %d", l, len);
 812                 goto out;
 813         }
 814
 815         return 0;
 816
 817 out:
 818         ubifs_err("bad node at LEB %d:%d", lnum, offs);
 819         dbg_dump_node(c, buf);
 820         dbg_dump_stack();
 821         return -EINVAL;
 822 }
 823
 824 /**
 825  * ubifs_wbuf_init - initialize write-buffer.
 826  * @c: UBIFS file-system description object
 827  * @wbuf: write-buffer to initialize
 828  *
 829  * This function initializes write-buffer. Returns zero in case of success
 830  * %-ENOMEM in case of failure.
 831  */
 832 int ubifs_wbuf_init(struct ubifs_info *c, struct ubifs_wbuf *wbuf)
 833 {
 834         size_t size;
 835
 836         wbuf->buf = kmalloc(c->min_io_size, GFP_KERNEL);
 837         if (!wbuf->buf)
 838                 return -ENOMEM;
 839
 840         size = (c->min_io_size / UBIFS_CH_SZ + 1) * sizeof(ino_t);
 841         wbuf->inodes = kmalloc(size, GFP_KERNEL);
 842         if (!wbuf->inodes) {
 843                 kfree(wbuf->buf);
 844                 wbuf->buf = NULL;
 845                 return -ENOMEM;
 846         }
 847
 848         wbuf->used = 0;
 849         wbuf->lnum = wbuf->offs = -1;
 850         wbuf->avail = c->min_io_size;
 851         wbuf->dtype = UBI_UNKNOWN;
 852         wbuf->sync_callback = NULL;
 853         mutex_init(&wbuf->io_mutex);
 854         spin_lock_init(&wbuf->lock);
 855         wbuf->c = c;
 856         wbuf->next_ino = 0;
 857
 858         hrtimer_init(&wbuf->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 859         wbuf->timer.function = wbuf_timer_callback_nolock;
 860         wbuf->softlimit = ktime_set(WBUF_TIMEOUT_SOFTLIMIT, 0);
 861         wbuf->delta = WBUF_TIMEOUT_HARDLIMIT - WBUF_TIMEOUT_SOFTLIMIT;
 862         wbuf->delta *= 1000000000ULL;
 863         ubifs_assert(wbuf->delta <= ULONG_MAX);
 864         return 0;
 865 }
 866
 867 /**
 868  * ubifs_wbuf_add_ino_nolock - add an inode number into the wbuf inode array.
 869  * @wbuf: the write-buffer where to add
 870  * @inum: the inode number
 871  *
 872  * This function adds an inode number to the inode array of the write-buffer.
 873  */
 874 void ubifs_wbuf_add_ino_nolock(struct ubifs_wbuf *wbuf, ino_t inum)
 875 {
 876         if (!wbuf->buf)
 877                 /* NOR flash or something similar */
 878                 return;
 879
 880         spin_lock(&wbuf->lock);
 881         if (wbuf->used)
 882                 wbuf->inodes[wbuf->next_ino++] = inum;
 883         spin_unlock(&wbuf->lock);
 884 }
 885
 886 /**
 887  * wbuf_has_ino - returns if the wbuf contains data from the inode.
 888  * @wbuf: the write-buffer
 889  * @inum: the inode number
 890  *
 891  * This function returns with %1 if the write-buffer contains some data from the
 892  * given inode otherwise it returns with %0.
 893  */
 894 static int wbuf_has_ino(struct ubifs_wbuf *wbuf, ino_t inum)
 895 {
 896         int i, ret = 0;
 897
 898         spin_lock(&wbuf->lock);
 899         for (i = 0; i < wbuf->next_ino; i++)
 900                 if (inum == wbuf->inodes[i]) {
 901                         ret = 1;
 902                         break;
 903                 }
 904         spin_unlock(&wbuf->lock);
 905
 906         return ret;
 907 }
 908
 909 /**
 910  * ubifs_sync_wbufs_by_inode - synchronize write-buffers for an inode.
 911  * @c: UBIFS file-system description object
 912  * @inode: inode to synchronize
 913  *
 914  * This function synchronizes write-buffers which contain nodes belonging to
 915  * @inode. Returns zero in case of success and a negative error code in case of
 916  * failure.
 917  */
 918 int ubifs_sync_wbufs_by_inode(struct ubifs_info *c, struct inode *inode)
 919 {
 920         int i, err = 0;
 921
 922         for (i = 0; i < c->jhead_cnt; i++) {
 923                 struct ubifs_wbuf *wbuf = &c->jheads[i].wbuf;
 924
 925                 if (i == GCHD)
 926                         /*
 927                          * GC head is special, do not look at it. Even if the
 928                          * head contains something related to this inode, it is
 929                          * a _copy_ of corresponding on-flash node which sits
 930                          * somewhere else.
 931                          */
 932                         continue;
 933
 934                 if (!wbuf_has_ino(wbuf, inode->i_ino))
 935                         continue;
 936
 937                 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 938                 if (wbuf_has_ino(wbuf, inode->i_ino))
 939                         err = ubifs_wbuf_sync_nolock(wbuf);
 940                 mutex_unlock(&wbuf->io_mutex);
 941
 942                 if (err) {
 943                         ubifs_ro_mode(c, err);
 944                         return err;
 945                 }
 946         }
 947         return 0;
 948 }