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