fs/ubifs/recovery.c

   1 /*
   2  * This file is part of UBIFS.
   3  *
   4  * Copyright (C) 2006-2008 Nokia Corporation
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 as published by
   8  * the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc., 51
  17  * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  18  *
  19  * Authors: Adrian Hunter
  20  *          Artem Bityutskiy (Битюцкий Артём)
  21  */
  22
  23 /*
  24  * This file implements functions needed to recover from unclean un-mounts.
  25  * When UBIFS is mounted, it checks a flag on the master node to determine if
  26  * an un-mount was completed sucessfully. If not, the process of mounting
  27  * incorparates additional checking and fixing of on-flash data structures.
  28  * UBIFS always cleans away all remnants of an unclean un-mount, so that
  29  * errors do not accumulate. However UBIFS defers recovery if it is mounted
  30  * read-only, and the flash is not modified in that case.
  31  */
  32
  33 #include <linux/crc32.h>
  34 #include "ubifs.h"
  35
  36 /**
  37  * is_empty - determine whether a buffer is empty (contains all 0xff).
  38  * @buf: buffer to clean
  39  * @len: length of buffer
  40  *
  41  * This function returns %1 if the buffer is empty (contains all 0xff) otherwise
  42  * %0 is returned.
  43  */
  44 static int is_empty(void *buf, int len)
  45 {
  46         uint8_t *p = buf;
  47         int i;
  48
  49         for (i = 0; i < len; i++)
  50                 if (*p++ != 0xff)
  51                         return 0;
  52         return 1;
  53 }
  54
  55 /**
  56  * get_master_node - get the last valid master node allowing for corruption.
  57  * @c: UBIFS file-system description object
  58  * @lnum: LEB number
  59  * @pbuf: buffer containing the LEB read, is returned here
  60  * @mst: master node, if found, is returned here
  61  * @cor: corruption, if found, is returned here
  62  *
  63  * This function allocates a buffer, reads the LEB into it, and finds and
  64  * returns the last valid master node allowing for one area of corruption.
  65  * The corrupt area, if there is one, must be consistent with the assumption
  66  * that it is the result of an unclean unmount while the master node was being
  67  * written. Under those circumstances, it is valid to use the previously written
  68  * master node.
  69  *
  70  * This function returns %0 on success and a negative error code on failure.
  71  */
  72 static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
  73                            struct ubifs_mst_node **mst, void **cor)
  74 {
  75         const int sz = c->mst_node_alsz;
  76         int err, offs, len;
  77         void *sbuf, *buf;
  78
  79         sbuf = vmalloc(c->leb_size);
  80         if (!sbuf)
  81                 return -ENOMEM;
  82
  83         err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size);
  84         if (err && err != -EBADMSG)
  85                 goto out_free;
  86
  87         /* Find the first position that is definitely not a node */
  88         offs = 0;
  89         buf = sbuf;
  90         len = c->leb_size;
  91         while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
  92                 struct ubifs_ch *ch = buf;
  93
  94                 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
  95                         break;
  96                 offs += sz;
  97                 buf  += sz;
  98                 len  -= sz;
  99         }
 100         /* See if there was a valid master node before that */
 101         if (offs) {
 102                 int ret;
 103
 104                 offs -= sz;
 105                 buf  -= sz;
 106                 len  += sz;
 107                 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
 108                 if (ret != SCANNED_A_NODE && offs) {
 109                         /* Could have been corruption so check one place back */
 110                         offs -= sz;
 111                         buf  -= sz;
 112                         len  += sz;
 113                         ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
 114                         if (ret != SCANNED_A_NODE)
 115                                 /*
 116                                  * We accept only one area of corruption because
 117                                  * we are assuming that it was caused while
 118                                  * trying to write a master node.
 119                                  */
 120                                 goto out_err;
 121                 }
 122                 if (ret == SCANNED_A_NODE) {
 123                         struct ubifs_ch *ch = buf;
 124
 125                         if (ch->node_type != UBIFS_MST_NODE)
 126                                 goto out_err;
 127                         dbg_rcvry("found a master node at %d:%d", lnum, offs);
 128                         *mst = buf;
 129                         offs += sz;
 130                         buf  += sz;
 131                         len  -= sz;
 132                 }
 133         }
 134         /* Check for corruption */
 135         if (offs < c->leb_size) {
 136                 if (!is_empty(buf, min_t(int, len, sz))) {
 137                         *cor = buf;
 138                         dbg_rcvry("found corruption at %d:%d", lnum, offs);
 139                 }
 140                 offs += sz;
 141                 buf  += sz;
 142                 len  -= sz;
 143         }
 144         /* Check remaining empty space */
 145         if (offs < c->leb_size)
 146                 if (!is_empty(buf, len))
 147                         goto out_err;
 148         *pbuf = sbuf;
 149         return 0;
 150
 151 out_err:
 152         err = -EINVAL;
 153 out_free:
 154         vfree(sbuf);
 155         *mst = NULL;
 156         *cor = NULL;
 157         return err;
 158 }
 159
 160 /**
 161  * write_rcvrd_mst_node - write recovered master node.
 162  * @c: UBIFS file-system description object
 163  * @mst: master node
 164  *
 165  * This function returns %0 on success and a negative error code on failure.
 166  */
 167 static int write_rcvrd_mst_node(struct ubifs_info *c,
 168                                 struct ubifs_mst_node *mst)
 169 {
 170         int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
 171         __le32 save_flags;
 172
 173         dbg_rcvry("recovery");
 174
 175         save_flags = mst->flags;
 176         mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY);
 177
 178         ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
 179         err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM);
 180         if (err)
 181                 goto out;
 182         err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM);
 183         if (err)
 184                 goto out;
 185 out:
 186         mst->flags = save_flags;
 187         return err;
 188 }
 189
 190 /**
 191  * ubifs_recover_master_node - recover the master node.
 192  * @c: UBIFS file-system description object
 193  *
 194  * This function recovers the master node from corruption that may occur due to
 195  * an unclean unmount.
 196  *
 197  * This function returns %0 on success and a negative error code on failure.
 198  */
 199 int ubifs_recover_master_node(struct ubifs_info *c)
 200 {
 201         void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
 202         struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
 203         const int sz = c->mst_node_alsz;
 204         int err, offs1, offs2;
 205
 206         dbg_rcvry("recovery");
 207
 208         err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
 209         if (err)
 210                 goto out_free;
 211
 212         err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
 213         if (err)
 214                 goto out_free;
 215
 216         if (mst1) {
 217                 offs1 = (void *)mst1 - buf1;
 218                 if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
 219                     (offs1 == 0 && !cor1)) {
 220                         /*
 221                          * mst1 was written by recovery at offset 0 with no
 222                          * corruption.
 223                          */
 224                         dbg_rcvry("recovery recovery");
 225                         mst = mst1;
 226                 } else if (mst2) {
 227                         offs2 = (void *)mst2 - buf2;
 228                         if (offs1 == offs2) {
 229                                 /* Same offset, so must be the same */
 230                                 if (memcmp((void *)mst1 + UBIFS_CH_SZ,
 231                                            (void *)mst2 + UBIFS_CH_SZ,
 232                                            UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
 233                                         goto out_err;
 234                                 mst = mst1;
 235                         } else if (offs2 + sz == offs1) {
 236                                 /* 1st LEB was written, 2nd was not */
 237                                 if (cor1)
 238                                         goto out_err;
 239                                 mst = mst1;
 240                         } else if (offs1 == 0 && offs2 + sz >= c->leb_size) {
 241                                 /* 1st LEB was unmapped and written, 2nd not */
 242                                 if (cor1)
 243                                         goto out_err;
 244                                 mst = mst1;
 245                         } else
 246                                 goto out_err;
 247                 } else {
 248                         /*
 249                          * 2nd LEB was unmapped and about to be written, so
 250                          * there must be only one master node in the first LEB
 251                          * and no corruption.
 252                          */
 253                         if (offs1 != 0 || cor1)
 254                                 goto out_err;
 255                         mst = mst1;
 256                 }
 257         } else {
 258                 if (!mst2)
 259                         goto out_err;
 260                 /*
 261                  * 1st LEB was unmapped and about to be written, so there must
 262                  * be no room left in 2nd LEB.
 263                  */
 264                 offs2 = (void *)mst2 - buf2;
 265                 if (offs2 + sz + sz <= c->leb_size)
 266                         goto out_err;
 267                 mst = mst2;
 268         }
 269
 270         dbg_rcvry("recovered master node from LEB %d",
 271                   (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
 272
 273         memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
 274
 275         if ((c->vfs_sb->s_flags & MS_RDONLY)) {
 276                 /* Read-only mode. Keep a copy for switching to rw mode */
 277                 c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
 278                 if (!c->rcvrd_mst_node) {
 279                         err = -ENOMEM;
 280                         goto out_free;
 281                 }
 282                 memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
 283         } else {
 284                 /* Write the recovered master node */
 285                 c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
 286                 err = write_rcvrd_mst_node(c, c->mst_node);
 287                 if (err)
 288                         goto out_free;
 289         }
 290
 291         vfree(buf2);
 292         vfree(buf1);
 293
 294         return 0;
 295
 296 out_err:
 297         err = -EINVAL;
 298 out_free:
 299         ubifs_err("failed to recover master node");
 300         if (mst1) {
 301                 dbg_err("dumping first master node");
 302                 dbg_dump_node(c, mst1);
 303         }
 304         if (mst2) {
 305                 dbg_err("dumping second master node");
 306                 dbg_dump_node(c, mst2);
 307         }
 308         vfree(buf2);
 309         vfree(buf1);
 310         return err;
 311 }
 312
 313 /**
 314  * ubifs_write_rcvrd_mst_node - write the recovered master node.
 315  * @c: UBIFS file-system description object
 316  *
 317  * This function writes the master node that was recovered during mounting in
 318  * read-only mode and must now be written because we are remounting rw.
 319  *
 320  * This function returns %0 on success and a negative error code on failure.
 321  */
 322 int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
 323 {
 324         int err;
 325
 326         if (!c->rcvrd_mst_node)
 327                 return 0;
 328         c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
 329         c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
 330         err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
 331         if (err)
 332                 return err;
 333         kfree(c->rcvrd_mst_node);
 334         c->rcvrd_mst_node = NULL;
 335         return 0;
 336 }
 337
 338 /**
 339  * is_last_write - determine if an offset was in the last write to a LEB.
 340  * @c: UBIFS file-system description object
 341  * @buf: buffer to check
 342  * @offs: offset to check
 343  *
 344  * This function returns %1 if @offs was in the last write to the LEB whose data
 345  * is in @buf, otherwise %0 is returned.  The determination is made by checking
 346  * for subsequent empty space starting from the next min_io_size boundary (or a
 347  * bit less than the common header size if min_io_size is one).
 348  */
 349 static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
 350 {
 351         int empty_offs;
 352         int check_len;
 353         uint8_t *p;
 354
 355         if (c->min_io_size == 1) {
 356                 check_len = c->leb_size - offs;
 357                 p = buf + check_len;
 358                 for (; check_len > 0; check_len--)
 359                         if (*--p != 0xff)
 360                                 break;
 361                 /*
 362                  * 'check_len' is the size of the corruption which cannot be
 363                  * more than the size of 1 node if it was caused by an unclean
 364                  * unmount.
 365                  */
 366                 if (check_len > UBIFS_MAX_NODE_SZ)
 367                         return 0;
 368                 return 1;
 369         }
 370
 371         /*
 372          * Round up to the next c->min_io_size boundary i.e. 'offs' is in the
 373          * last wbuf written. After that should be empty space.
 374          */
 375         empty_offs = ALIGN(offs + 1, c->min_io_size);
 376         check_len = c->leb_size - empty_offs;
 377         p = buf + empty_offs - offs;
 378
 379         for (; check_len > 0; check_len--)
 380                 if (*p++ != 0xff)
 381                         return 0;
 382         return 1;
 383 }
 384
 385 /**
 386  * clean_buf - clean the data from an LEB sitting in a buffer.
 387  * @c: UBIFS file-system description object
 388  * @buf: buffer to clean
 389  * @lnum: LEB number to clean
 390  * @offs: offset from which to clean
 391  * @len: length of buffer
 392  *
 393  * This function pads up to the next min_io_size boundary (if there is one) and
 394  * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
 395  * min_io_size boundary (if there is one).
 396  */
 397 static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
 398                       int *offs, int *len)
 399 {
 400         int empty_offs, pad_len;
 401
 402         lnum = lnum;
 403         dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
 404
 405         if (c->min_io_size == 1) {
 406                 memset(*buf, 0xff, c->leb_size - *offs);
 407                 return;
 408         }
 409
 410         ubifs_assert(!(*offs & 7));
 411         empty_offs = ALIGN(*offs, c->min_io_size);
 412         pad_len = empty_offs - *offs;
 413         ubifs_pad(c, *buf, pad_len);
 414         *offs += pad_len;
 415         *buf += pad_len;
 416         *len -= pad_len;
 417         memset(*buf, 0xff, c->leb_size - empty_offs);
 418 }
 419
 420 /**
 421  * no_more_nodes - determine if there are no more nodes in a buffer.
 422  * @c: UBIFS file-system description object
 423  * @buf: buffer to check
 424  * @len: length of buffer
 425  * @lnum: LEB number of the LEB from which @buf was read
 426  * @offs: offset from which @buf was read
 427  *
 428  * This function ensures that the corrupted node at @offs is the last thing
 429  * written to a LEB. This function returns %1 if more data is not found and
 430  * %0 if more data is found.
 431  */
 432 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
 433                         int lnum, int offs)
 434 {
 435         struct ubifs_ch *ch = buf;
 436         int skip, dlen = le32_to_cpu(ch->len);
 437
 438         /* Check for empty space after the corrupt node's common header */
 439         skip = ALIGN(offs + UBIFS_CH_SZ, c->min_io_size) - offs;
 440         if (is_empty(buf + skip, len - skip))
 441                 return 1;
 442         /*
 443          * The area after the common header size is not empty, so the common
 444          * header must be intact. Check it.
 445          */
 446         if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
 447                 dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
 448                 return 0;
 449         }
 450         /* Now we know the corrupt node's length we can skip over it */
 451         skip = ALIGN(offs + dlen, c->min_io_size) - offs;
 452         /* After which there should be empty space */
 453         if (is_empty(buf + skip, len - skip))
 454                 return 1;
 455         dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
 456         return 0;
 457 }
 458
 459 /**
 460  * fix_unclean_leb - fix an unclean LEB.
 461  * @c: UBIFS file-system description object
 462  * @sleb: scanned LEB information
 463  * @start: offset where scan started
 464  */
 465 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
 466                            int start)
 467 {
 468         int lnum = sleb->lnum, endpt = start;
 469
 470         /* Get the end offset of the last node we are keeping */
 471         if (!list_empty(&sleb->nodes)) {
 472                 struct ubifs_scan_node *snod;
 473
 474                 snod = list_entry(sleb->nodes.prev,
 475                                   struct ubifs_scan_node, list);
 476                 endpt = snod->offs + snod->len;
 477         }
 478
 479         if ((c->vfs_sb->s_flags & MS_RDONLY) && !c->remounting_rw) {
 480                 /* Add to recovery list */
 481                 struct ubifs_unclean_leb *ucleb;
 482
 483                 dbg_rcvry("need to fix LEB %d start %d endpt %d",
 484                           lnum, start, sleb->endpt);
 485                 ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
 486                 if (!ucleb)
 487                         return -ENOMEM;
 488                 ucleb->lnum = lnum;
 489                 ucleb->endpt = endpt;
 490                 list_add_tail(&ucleb->list, &c->unclean_leb_list);
 491         } else {
 492                 /* Write the fixed LEB back to flash */
 493                 int err;
 494
 495                 dbg_rcvry("fixing LEB %d start %d endpt %d",
 496                           lnum, start, sleb->endpt);
 497                 if (endpt == 0) {
 498                         err = ubifs_leb_unmap(c, lnum);
 499                         if (err)
 500                                 return err;
 501                 } else {
 502                         int len = ALIGN(endpt, c->min_io_size);
 503
 504                         if (start) {
 505                                 err = ubi_read(c->ubi, lnum, sleb->buf, 0,
 506                                                start);
 507                                 if (err)
 508                                         return err;
 509                         }
 510                         /* Pad to min_io_size */
 511                         if (len > endpt) {
 512                                 int pad_len = len - ALIGN(endpt, 8);
 513
 514                                 if (pad_len > 0) {
 515                                         void *buf = sleb->buf + len - pad_len;
 516
 517                                         ubifs_pad(c, buf, pad_len);
 518                                 }
 519                         }
 520                         err = ubi_leb_change(c->ubi, lnum, sleb->buf, len,
 521                                              UBI_UNKNOWN);
 522                         if (err)
 523                                 return err;
 524                 }
 525         }
 526         return 0;
 527 }
 528
 529 /**
 530  * drop_incomplete_group - drop nodes from an incomplete group.
 531  * @sleb: scanned LEB information
 532  * @offs: offset of dropped nodes is returned here
 533  *
 534  * This function returns %1 if nodes are dropped and %0 otherwise.
 535  */
 536 static int drop_incomplete_group(struct ubifs_scan_leb *sleb, int *offs)
 537 {
 538         int dropped = 0;
 539
 540         while (!list_empty(&sleb->nodes)) {
 541                 struct ubifs_scan_node *snod;
 542                 struct ubifs_ch *ch;
 543
 544                 snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
 545                                   list);
 546                 ch = snod->node;
 547                 if (ch->group_type != UBIFS_IN_NODE_GROUP)
 548                         return dropped;
 549                 dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs);
 550                 *offs = snod->offs;
 551                 list_del(&snod->list);
 552                 kfree(snod);
 553                 sleb->nodes_cnt -= 1;
 554                 dropped = 1;
 555         }
 556         return dropped;
 557 }
 558
 559 /**
 560  * ubifs_recover_leb - scan and recover a LEB.
 561  * @c: UBIFS file-system description object
 562  * @lnum: LEB number
 563  * @offs: offset
 564  * @sbuf: LEB-sized buffer to use
 565  * @grouped: nodes may be grouped for recovery
 566  *
 567  * This function does a scan of a LEB, but caters for errors that might have
 568  * been caused by the unclean unmount from which we are attempting to recover.
 569  *
 570  * This function returns %0 on success and a negative error code on failure.
 571  */
 572 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
 573                                          int offs, void *sbuf, int grouped)
 574 {
 575         int err, len = c->leb_size - offs, need_clean = 0, quiet = 1;
 576         int empty_chkd = 0, start = offs;
 577         struct ubifs_scan_leb *sleb;
 578         void *buf = sbuf + offs;
 579
 580         dbg_rcvry("%d:%d", lnum, offs);
 581
 582         sleb = ubifs_start_scan(c, lnum, offs, sbuf);
 583         if (IS_ERR(sleb))
 584                 return sleb;
 585
 586         if (sleb->ecc)
 587                 need_clean = 1;
 588
 589         while (len >= 8) {
 590                 int ret;
 591
 592                 dbg_scan("look at LEB %d:%d (%d bytes left)",
 593                          lnum, offs, len);
 594
 595                 cond_resched();
 596
 597                 /*
 598                  * Scan quietly until there is an error from which we cannot
 599                  * recover
 600                  */
 601                 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
 602
 603                 if (ret == SCANNED_A_NODE) {
 604                         /* A valid node, and not a padding node */
 605                         struct ubifs_ch *ch = buf;
 606                         int node_len;
 607
 608                         err = ubifs_add_snod(c, sleb, buf, offs);
 609                         if (err)
 610                                 goto error;
 611                         node_len = ALIGN(le32_to_cpu(ch->len), 8);
 612                         offs += node_len;
 613                         buf += node_len;
 614                         len -= node_len;
 615                         continue;
 616                 }
 617
 618                 if (ret > 0) {
 619                         /* Padding bytes or a valid padding node */
 620                         offs += ret;
 621                         buf += ret;
 622                         len -= ret;
 623                         continue;
 624                 }
 625
 626                 if (ret == SCANNED_EMPTY_SPACE) {
 627                         if (!is_empty(buf, len)) {
 628                                 if (!is_last_write(c, buf, offs))
 629                                         break;
 630                                 clean_buf(c, &buf, lnum, &offs, &len);
 631                                 need_clean = 1;
 632                         }
 633                         empty_chkd = 1;
 634                         break;
 635                 }
 636
 637                 if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE)
 638                         if (is_last_write(c, buf, offs)) {
 639                                 clean_buf(c, &buf, lnum, &offs, &len);
 640                                 need_clean = 1;
 641                                 empty_chkd = 1;
 642                                 break;
 643                         }
 644
 645                 if (ret == SCANNED_A_CORRUPT_NODE)
 646                         if (no_more_nodes(c, buf, len, lnum, offs)) {
 647                                 clean_buf(c, &buf, lnum, &offs, &len);
 648                                 need_clean = 1;
 649                                 empty_chkd = 1;
 650                                 break;
 651                         }
 652
 653                 if (quiet) {
 654                         /* Redo the last scan but noisily */
 655                         quiet = 0;
 656                         continue;
 657                 }
 658
 659                 switch (ret) {
 660                 case SCANNED_GARBAGE:
 661                         dbg_err("garbage");
 662                         goto corrupted;
 663                 case SCANNED_A_CORRUPT_NODE:
 664                 case SCANNED_A_BAD_PAD_NODE:
 665                         dbg_err("bad node");
 666                         goto corrupted;
 667                 default:
 668                         dbg_err("unknown");
 669                         goto corrupted;
 670                 }
 671         }
 672
 673         if (!empty_chkd && !is_empty(buf, len)) {
 674                 if (is_last_write(c, buf, offs)) {
 675                         clean_buf(c, &buf, lnum, &offs, &len);
 676                         need_clean = 1;
 677                 } else {
 678                         ubifs_err("corrupt empty space at LEB %d:%d",
 679                                   lnum, offs);
 680                         goto corrupted;
 681                 }
 682         }
 683
 684         /* Drop nodes from incomplete group */
 685         if (grouped && drop_incomplete_group(sleb, &offs)) {
 686                 buf = sbuf + offs;
 687                 len = c->leb_size - offs;
 688                 clean_buf(c, &buf, lnum, &offs, &len);
 689                 need_clean = 1;
 690         }
 691
 692         if (offs % c->min_io_size) {
 693                 clean_buf(c, &buf, lnum, &offs, &len);
 694                 need_clean = 1;
 695         }
 696
 697         ubifs_end_scan(c, sleb, lnum, offs);
 698
 699         if (need_clean) {
 700                 err = fix_unclean_leb(c, sleb, start);
 701                 if (err)
 702                         goto error;
 703         }
 704
 705         return sleb;
 706
 707 corrupted:
 708         ubifs_scanned_corruption(c, lnum, offs, buf);
 709         err = -EUCLEAN;
 710 error:
 711         ubifs_err("LEB %d scanning failed", lnum);
 712         ubifs_scan_destroy(sleb);
 713         return ERR_PTR(err);
 714 }
 715
 716 /**
 717  * get_cs_sqnum - get commit start sequence number.
 718  * @c: UBIFS file-system description object
 719  * @lnum: LEB number of commit start node
 720  * @offs: offset of commit start node
 721  * @cs_sqnum: commit start sequence number is returned here
 722  *
 723  * This function returns %0 on success and a negative error code on failure.
 724  */
 725 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
 726                         unsigned long long *cs_sqnum)
 727 {
 728         struct ubifs_cs_node *cs_node = NULL;
 729         int err, ret;
 730
 731         dbg_rcvry("at %d:%d", lnum, offs);
 732         cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
 733         if (!cs_node)
 734                 return -ENOMEM;
 735         if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
 736                 goto out_err;
 737         err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ);
 738         if (err && err != -EBADMSG)
 739                 goto out_free;
 740         ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
 741         if (ret != SCANNED_A_NODE) {
 742                 dbg_err("Not a valid node");
 743                 goto out_err;
 744         }
 745         if (cs_node->ch.node_type != UBIFS_CS_NODE) {
 746                 dbg_err("Node a CS node, type is %d", cs_node->ch.node_type);
 747                 goto out_err;
 748         }
 749         if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
 750                 dbg_err("CS node cmt_no %llu != current cmt_no %llu",
 751                         (unsigned long long)le64_to_cpu(cs_node->cmt_no),
 752                         c->cmt_no);
 753                 goto out_err;
 754         }
 755         *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
 756         dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
 757         kfree(cs_node);
 758         return 0;
 759
 760 out_err:
 761         err = -EINVAL;
 762 out_free:
 763         ubifs_err("failed to get CS sqnum");
 764         kfree(cs_node);
 765         return err;
 766 }
 767
 768 /**
 769  * ubifs_recover_log_leb - scan and recover a log LEB.
 770  * @c: UBIFS file-system description object
 771  * @lnum: LEB number
 772  * @offs: offset
 773  * @sbuf: LEB-sized buffer to use
 774  *
 775  * This function does a scan of a LEB, but caters for errors that might have
 776  * been caused by the unclean unmount from which we are attempting to recover.
 777  *
 778  * This function returns %0 on success and a negative error code on failure.
 779  */
 780 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
 781                                              int offs, void *sbuf)
 782 {
 783         struct ubifs_scan_leb *sleb;
 784         int next_lnum;
 785
 786         dbg_rcvry("LEB %d", lnum);
 787         next_lnum = lnum + 1;
 788         if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
 789                 next_lnum = UBIFS_LOG_LNUM;
 790         if (next_lnum != c->ltail_lnum) {
 791                 /*
 792                  * We can only recover at the end of the log, so check that the
 793                  * next log LEB is empty or out of date.
 794                  */
 795                 sleb = ubifs_scan(c, next_lnum, 0, sbuf);
 796                 if (IS_ERR(sleb))
 797                         return sleb;
 798                 if (sleb->nodes_cnt) {
 799                         struct ubifs_scan_node *snod;
 800                         unsigned long long cs_sqnum = c->cs_sqnum;
 801
 802                         snod = list_entry(sleb->nodes.next,
 803                                           struct ubifs_scan_node, list);
 804                         if (cs_sqnum == 0) {
 805                                 int err;
 806
 807                                 err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
 808                                 if (err) {
 809                                         ubifs_scan_destroy(sleb);
 810                                         return ERR_PTR(err);
 811                                 }
 812                         }
 813                         if (snod->sqnum > cs_sqnum) {
 814                                 ubifs_err("unrecoverable log corruption "
 815                                           "in LEB %d", lnum);
 816                                 ubifs_scan_destroy(sleb);
 817                                 return ERR_PTR(-EUCLEAN);
 818                         }
 819                 }
 820                 ubifs_scan_destroy(sleb);
 821         }
 822         return ubifs_recover_leb(c, lnum, offs, sbuf, 0);
 823 }
 824
 825 /**
 826  * recover_head - recover a head.
 827  * @c: UBIFS file-system description object
 828  * @lnum: LEB number of head to recover
 829  * @offs: offset of head to recover
 830  * @sbuf: LEB-sized buffer to use
 831  *
 832  * This function ensures that there is no data on the flash at a head location.
 833  *
 834  * This function returns %0 on success and a negative error code on failure.
 835  */
 836 static int recover_head(const struct ubifs_info *c, int lnum, int offs,
 837                         void *sbuf)
 838 {
 839         int len, err, need_clean = 0;
 840
 841         if (c->min_io_size > 1)
 842                 len = c->min_io_size;
 843         else
 844                 len = 512;
 845         if (offs + len > c->leb_size)
 846                 len = c->leb_size - offs;
 847
 848         if (!len)
 849                 return 0;
 850
 851         /* Read at the head location and check it is empty flash */
 852         err = ubi_read(c->ubi, lnum, sbuf, offs, len);
 853         if (err)
 854                 need_clean = 1;
 855         else {
 856                 uint8_t *p = sbuf;
 857
 858                 while (len--)
 859                         if (*p++ != 0xff) {
 860                                 need_clean = 1;
 861                                 break;
 862                         }
 863         }
 864
 865         if (need_clean) {
 866                 dbg_rcvry("cleaning head at %d:%d", lnum, offs);
 867                 if (offs == 0)
 868                         return ubifs_leb_unmap(c, lnum);
 869                 err = ubi_read(c->ubi, lnum, sbuf, 0, offs);
 870                 if (err)
 871                         return err;
 872                 return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN);
 873         }
 874
 875         return 0;
 876 }
 877
 878 /**
 879  * ubifs_recover_inl_heads - recover index and LPT heads.
 880  * @c: UBIFS file-system description object
 881  * @sbuf: LEB-sized buffer to use
 882  *
 883  * This function ensures that there is no data on the flash at the index and
 884  * LPT head locations.
 885  *
 886  * This deals with the recovery of a half-completed journal commit. UBIFS is
 887  * careful never to overwrite the last version of the index or the LPT. Because
 888  * the index and LPT are wandering trees, data from a half-completed commit will
 889  * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
 890  * assumed to be empty and will be unmapped anyway before use, or in the index
 891  * and LPT heads.
 892  *
 893  * This function returns %0 on success and a negative error code on failure.
 894  */
 895 int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf)
 896 {
 897         int err;
 898
 899         ubifs_assert(!(c->vfs_sb->s_flags & MS_RDONLY) || c->remounting_rw);
 900
 901         dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
 902         err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
 903         if (err)
 904                 return err;
 905
 906         dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
 907         err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
 908         if (err)
 909                 return err;
 910
 911         return 0;
 912 }
 913
 914 /**
 915  *  clean_an_unclean_leb - read and write a LEB to remove corruption.
 916  * @c: UBIFS file-system description object
 917  * @ucleb: unclean LEB information
 918  * @sbuf: LEB-sized buffer to use
 919  *
 920  * This function reads a LEB up to a point pre-determined by the mount recovery,
 921  * checks the nodes, and writes the result back to the flash, thereby cleaning
 922  * off any following corruption, or non-fatal ECC errors.
 923  *
 924  * This function returns %0 on success and a negative error code on failure.
 925  */
 926 static int clean_an_unclean_leb(const struct ubifs_info *c,
 927                                 struct ubifs_unclean_leb *ucleb, void *sbuf)
 928 {
 929         int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
 930         void *buf = sbuf;
 931
 932         dbg_rcvry("LEB %d len %d", lnum, len);
 933
 934         if (len == 0) {
 935                 /* Nothing to read, just unmap it */
 936                 err = ubifs_leb_unmap(c, lnum);
 937                 if (err)
 938                         return err;
 939                 return 0;
 940         }
 941
 942         err = ubi_read(c->ubi, lnum, buf, offs, len);
 943         if (err && err != -EBADMSG)
 944                 return err;
 945
 946         while (len >= 8) {
 947                 int ret;
 948
 949                 cond_resched();
 950
 951                 /* Scan quietly until there is an error */
 952                 ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
 953
 954                 if (ret == SCANNED_A_NODE) {
 955                         /* A valid node, and not a padding node */
 956                         struct ubifs_ch *ch = buf;
 957                         int node_len;
 958
 959                         node_len = ALIGN(le32_to_cpu(ch->len), 8);
 960                         offs += node_len;
 961                         buf += node_len;
 962                         len -= node_len;
 963                         continue;
 964                 }
 965
 966                 if (ret > 0) {
 967                         /* Padding bytes or a valid padding node */
 968                         offs += ret;
 969                         buf += ret;
 970                         len -= ret;
 971                         continue;
 972                 }
 973
 974                 if (ret == SCANNED_EMPTY_SPACE) {
 975                         ubifs_err("unexpected empty space at %d:%d",
 976                                   lnum, offs);
 977                         return -EUCLEAN;
 978                 }
 979
 980                 if (quiet) {
 981                         /* Redo the last scan but noisily */
 982                         quiet = 0;
 983                         continue;
 984                 }
 985
 986                 ubifs_scanned_corruption(c, lnum, offs, buf);
 987                 return -EUCLEAN;
 988         }
 989
 990         /* Pad to min_io_size */
 991         len = ALIGN(ucleb->endpt, c->min_io_size);
 992         if (len > ucleb->endpt) {
 993                 int pad_len = len - ALIGN(ucleb->endpt, 8);
 994
 995                 if (pad_len > 0) {
 996                         buf = c->sbuf + len - pad_len;
 997                         ubifs_pad(c, buf, pad_len);
 998                 }
 999         }
1000
1001         /* Write back the LEB atomically */
1002         err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN);
1003         if (err)
1004                 return err;
1005
1006         dbg_rcvry("cleaned LEB %d", lnum);
1007
1008         return 0;
1009 }
1010
1011 /**
1012  * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
1013  * @c: UBIFS file-system description object
1014  * @sbuf: LEB-sized buffer to use
1015  *
1016  * This function cleans a LEB identified during recovery that needs to be
1017  * written but was not because UBIFS was mounted read-only. This happens when
1018  * remounting to read-write mode.
1019  *
1020  * This function returns %0 on success and a negative error code on failure.
1021  */
1022 int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf)
1023 {
1024         dbg_rcvry("recovery");
1025         while (!list_empty(&c->unclean_leb_list)) {
1026                 struct ubifs_unclean_leb *ucleb;
1027                 int err;
1028
1029                 ucleb = list_entry(c->unclean_leb_list.next,
1030                                    struct ubifs_unclean_leb, list);
1031                 err = clean_an_unclean_leb(c, ucleb, sbuf);
1032                 if (err)
1033                         return err;
1034                 list_del(&ucleb->list);
1035                 kfree(ucleb);
1036         }
1037         return 0;
1038 }
1039
1040 /**
1041  * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1042  * @c: UBIFS file-system description object
1043  *
1044  * Out-of-place garbage collection requires always one empty LEB with which to
1045  * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1046  * written to the master node on unmounting. In the case of an unclean unmount
1047  * the value of gc_lnum recorded in the master node is out of date and cannot
1048  * be used. Instead, recovery must allocate an empty LEB for this purpose.
1049  * However, there may not be enough empty space, in which case it must be
1050  * possible to GC the dirtiest LEB into the GC head LEB.
1051  *
1052  * This function also runs the commit which causes the TNC updates from
1053  * size-recovery and orphans to be written to the flash. That is important to
1054  * ensure correct replay order for subsequent mounts.
1055  *
1056  * This function returns %0 on success and a negative error code on failure.
1057  */
1058 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1059 {
1060         struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1061         struct ubifs_lprops lp;
1062         int lnum, err;
1063
1064         c->gc_lnum = -1;
1065         if (wbuf->lnum == -1) {
1066                 dbg_rcvry("no GC head LEB");
1067                 goto find_free;
1068         }
1069         /*
1070          * See whether the used space in the dirtiest LEB fits in the GC head
1071          * LEB.
1072          */
1073         if (wbuf->offs == c->leb_size) {
1074                 dbg_rcvry("no room in GC head LEB");
1075                 goto find_free;
1076         }
1077         err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1078         if (err) {
1079                 if (err == -ENOSPC)
1080                         dbg_err("could not find a dirty LEB");
1081                 return err;
1082         }
1083         ubifs_assert(!(lp.flags & LPROPS_INDEX));
1084         lnum = lp.lnum;
1085         if (lp.free + lp.dirty == c->leb_size) {
1086                 /* An empty LEB was returned */
1087                 if (lp.free != c->leb_size) {
1088                         err = ubifs_change_one_lp(c, lnum, c->leb_size,
1089                                                   0, 0, 0, 0);
1090                         if (err)
1091                                 return err;
1092                 }
1093                 err = ubifs_leb_unmap(c, lnum);
1094                 if (err)
1095                         return err;
1096                 c->gc_lnum = lnum;
1097                 dbg_rcvry("allocated LEB %d for GC", lnum);
1098                 /* Run the commit */
1099                 dbg_rcvry("committing");
1100                 return ubifs_run_commit(c);
1101         }
1102         /*
1103          * There was no empty LEB so the used space in the dirtiest LEB must fit
1104          * in the GC head LEB.
1105          */
1106         if (lp.free + lp.dirty < wbuf->offs) {
1107                 dbg_rcvry("LEB %d doesn't fit in GC head LEB %d:%d",
1108                           lnum, wbuf->lnum, wbuf->offs);
1109                 err = ubifs_return_leb(c, lnum);
1110                 if (err)
1111                         return err;
1112                 goto find_free;
1113         }
1114         /*
1115          * We run the commit before garbage collection otherwise subsequent
1116          * mounts will see the GC and orphan deletion in a different order.
1117          */
1118         dbg_rcvry("committing");
1119         err = ubifs_run_commit(c);
1120         if (err)
1121                 return err;
1122         /*
1123          * The data in the dirtiest LEB fits in the GC head LEB, so do the GC
1124          * - use locking to keep 'ubifs_assert()' happy.
1125          */
1126         dbg_rcvry("GC'ing LEB %d", lnum);
1127         mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1128         err = ubifs_garbage_collect_leb(c, &lp);
1129         if (err >= 0) {
1130                 int err2 = ubifs_wbuf_sync_nolock(wbuf);
1131
1132                 if (err2)
1133                         err = err2;
1134         }
1135         mutex_unlock(&wbuf->io_mutex);
1136         if (err < 0) {
1137                 dbg_err("GC failed, error %d", err);
1138                 if (err == -EAGAIN)
1139                         err = -EINVAL;
1140                 return err;
1141         }
1142         if (err != LEB_RETAINED) {
1143                 dbg_err("GC returned %d", err);
1144                 return -EINVAL;
1145         }
1146         err = ubifs_leb_unmap(c, c->gc_lnum);
1147         if (err)
1148                 return err;
1149         dbg_rcvry("allocated LEB %d for GC", lnum);
1150         return 0;
1151
1152 find_free:
1153         /*
1154          * There is no GC head LEB or the free space in the GC head LEB is too
1155          * small. Allocate gc_lnum by calling 'ubifs_find_free_leb_for_idx()' so
1156          * GC is not run.
1157          */
1158         lnum = ubifs_find_free_leb_for_idx(c);
1159         if (lnum < 0) {
1160                 dbg_err("could not find an empty LEB");
1161                 return lnum;
1162         }
1163         /* And reset the index flag */
1164         err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1165                                   LPROPS_INDEX, 0);
1166         if (err)
1167                 return err;
1168         c->gc_lnum = lnum;
1169         dbg_rcvry("allocated LEB %d for GC", lnum);
1170         /* Run the commit */
1171         dbg_rcvry("committing");
1172         return ubifs_run_commit(c);
1173 }
1174
1175 /**
1176  * struct size_entry - inode size information for recovery.
1177  * @rb: link in the RB-tree of sizes
1178  * @inum: inode number
1179  * @i_size: size on inode
1180  * @d_size: maximum size based on data nodes
1181  * @exists: indicates whether the inode exists
1182  * @inode: inode if pinned in memory awaiting rw mode to fix it
1183  */
1184 struct size_entry {
1185         struct rb_node rb;
1186         ino_t inum;
1187         loff_t i_size;
1188         loff_t d_size;
1189         int exists;
1190         struct inode *inode;
1191 };
1192
1193 /**
1194  * add_ino - add an entry to the size tree.
1195  * @c: UBIFS file-system description object
1196  * @inum: inode number
1197  * @i_size: size on inode
1198  * @d_size: maximum size based on data nodes
1199  * @exists: indicates whether the inode exists
1200  */
1201 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1202                    loff_t d_size, int exists)
1203 {
1204         struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1205         struct size_entry *e;
1206
1207         while (*p) {
1208                 parent = *p;
1209                 e = rb_entry(parent, struct size_entry, rb);
1210                 if (inum < e->inum)
1211                         p = &(*p)->rb_left;
1212                 else
1213                         p = &(*p)->rb_right;
1214         }
1215
1216         e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1217         if (!e)
1218                 return -ENOMEM;
1219
1220         e->inum = inum;
1221         e->i_size = i_size;
1222         e->d_size = d_size;
1223         e->exists = exists;
1224
1225         rb_link_node(&e->rb, parent, p);
1226         rb_insert_color(&e->rb, &c->size_tree);
1227
1228         return 0;
1229 }
1230
1231 /**
1232  * find_ino - find an entry on the size tree.
1233  * @c: UBIFS file-system description object
1234  * @inum: inode number
1235  */
1236 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1237 {
1238         struct rb_node *p = c->size_tree.rb_node;
1239         struct size_entry *e;
1240
1241         while (p) {
1242                 e = rb_entry(p, struct size_entry, rb);
1243                 if (inum < e->inum)
1244                         p = p->rb_left;
1245                 else if (inum > e->inum)
1246                         p = p->rb_right;
1247                 else
1248                         return e;
1249         }
1250         return NULL;
1251 }
1252
1253 /**
1254  * remove_ino - remove an entry from the size tree.
1255  * @c: UBIFS file-system description object
1256  * @inum: inode number
1257  */
1258 static void remove_ino(struct ubifs_info *c, ino_t inum)
1259 {
1260         struct size_entry *e = find_ino(c, inum);
1261
1262         if (!e)
1263                 return;
1264         rb_erase(&e->rb, &c->size_tree);
1265         kfree(e);
1266 }
1267
1268 /**
1269  * ubifs_destroy_size_tree - free resources related to the size tree.
1270  * @c: UBIFS file-system description object
1271  */
1272 void ubifs_destroy_size_tree(struct ubifs_info *c)
1273 {
1274         struct rb_node *this = c->size_tree.rb_node;
1275         struct size_entry *e;
1276
1277         while (this) {
1278                 if (this->rb_left) {
1279                         this = this->rb_left;
1280                         continue;
1281                 } else if (this->rb_right) {
1282                         this = this->rb_right;
1283                         continue;
1284                 }
1285                 e = rb_entry(this, struct size_entry, rb);
1286                 if (e->inode)
1287                         iput(e->inode);
1288                 this = rb_parent(this);
1289                 if (this) {
1290                         if (this->rb_left == &e->rb)
1291                                 this->rb_left = NULL;
1292                         else
1293                                 this->rb_right = NULL;
1294                 }
1295                 kfree(e);
1296         }
1297         c->size_tree = RB_ROOT;
1298 }
1299
1300 /**
1301  * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1302  * @c: UBIFS file-system description object
1303  * @key: node key
1304  * @deletion: node is for a deletion
1305  * @new_size: inode size
1306  *
1307  * This function has two purposes:
1308  *     1) to ensure there are no data nodes that fall outside the inode size
1309  *     2) to ensure there are no data nodes for inodes that do not exist
1310  * To accomplish those purposes, a rb-tree is constructed containing an entry
1311  * for each inode number in the journal that has not been deleted, and recording
1312  * the size from the inode node, the maximum size of any data node (also altered
1313  * by truncations) and a flag indicating a inode number for which no inode node
1314  * was present in the journal.
1315  *
1316  * Note that there is still the possibility that there are data nodes that have
1317  * been committed that are beyond the inode size, however the only way to find
1318  * them would be to scan the entire index. Alternatively, some provision could
1319  * be made to record the size of inodes at the start of commit, which would seem
1320  * very cumbersome for a scenario that is quite unlikely and the only negative
1321  * consequence of which is wasted space.
1322  *
1323  * This functions returns %0 on success and a negative error code on failure.
1324  */
1325 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1326                              int deletion, loff_t new_size)
1327 {
1328         ino_t inum = key_inum(c, key);
1329         struct size_entry *e;
1330         int err;
1331
1332         switch (key_type(c, key)) {
1333         case UBIFS_INO_KEY:
1334                 if (deletion)
1335                         remove_ino(c, inum);
1336                 else {
1337                         e = find_ino(c, inum);
1338                         if (e) {
1339                                 e->i_size = new_size;
1340                                 e->exists = 1;
1341                         } else {
1342                                 err = add_ino(c, inum, new_size, 0, 1);
1343                                 if (err)
1344                                         return err;
1345                         }
1346                 }
1347                 break;
1348         case UBIFS_DATA_KEY:
1349                 e = find_ino(c, inum);
1350                 if (e) {
1351                         if (new_size > e->d_size)
1352                                 e->d_size = new_size;
1353                 } else {
1354                         err = add_ino(c, inum, 0, new_size, 0);
1355                         if (err)
1356                                 return err;
1357                 }
1358                 break;
1359         case UBIFS_TRUN_KEY:
1360                 e = find_ino(c, inum);
1361                 if (e)
1362                         e->d_size = new_size;
1363                 break;
1364         }
1365         return 0;
1366 }
1367
1368 /**
1369  * fix_size_in_place - fix inode size in place on flash.
1370  * @c: UBIFS file-system description object
1371  * @e: inode size information for recovery
1372  */
1373 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1374 {
1375         struct ubifs_ino_node *ino = c->sbuf;
1376         unsigned char *p;
1377         union ubifs_key key;
1378         int err, lnum, offs, len;
1379         loff_t i_size;
1380         uint32_t crc;
1381
1382         /* Locate the inode node LEB number and offset */
1383         ino_key_init(c, &key, e->inum);
1384         err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1385         if (err)
1386                 goto out;
1387         /*
1388          * If the size recorded on the inode node is greater than the size that
1389          * was calculated from nodes in the journal then don't change the inode.
1390          */
1391         i_size = le64_to_cpu(ino->size);
1392         if (i_size >= e->d_size)
1393                 return 0;
1394         /* Read the LEB */
1395         err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size);
1396         if (err)
1397                 goto out;
1398         /* Change the size field and recalculate the CRC */
1399         ino = c->sbuf + offs;
1400         ino->size = cpu_to_le64(e->d_size);
1401         len = le32_to_cpu(ino->ch.len);
1402         crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1403         ino->ch.crc = cpu_to_le32(crc);
1404         /* Work out where data in the LEB ends and free space begins */
1405         p = c->sbuf;
1406         len = c->leb_size - 1;
1407         while (p[len] == 0xff)
1408                 len -= 1;
1409         len = ALIGN(len + 1, c->min_io_size);
1410         /* Atomically write the fixed LEB back again */
1411         err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN);
1412         if (err)
1413                 goto out;
1414         dbg_rcvry("inode %lu at %d:%d size %lld -> %lld ",
1415                   (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
1416         return 0;
1417
1418 out:
1419         ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
1420                    (unsigned long)e->inum, e->i_size, e->d_size, err);
1421         return err;
1422 }
1423
1424 /**
1425  * ubifs_recover_size - recover inode size.
1426  * @c: UBIFS file-system description object
1427  *
1428  * This function attempts to fix inode size discrepancies identified by the
1429  * 'ubifs_recover_size_accum()' function.
1430  *
1431  * This functions returns %0 on success and a negative error code on failure.
1432  */
1433 int ubifs_recover_size(struct ubifs_info *c)
1434 {
1435         struct rb_node *this = rb_first(&c->size_tree);
1436
1437         while (this) {
1438                 struct size_entry *e;
1439                 int err;
1440
1441                 e = rb_entry(this, struct size_entry, rb);
1442                 if (!e->exists) {
1443                         union ubifs_key key;
1444
1445                         ino_key_init(c, &key, e->inum);
1446                         err = ubifs_tnc_lookup(c, &key, c->sbuf);
1447                         if (err && err != -ENOENT)
1448                                 return err;
1449                         if (err == -ENOENT) {
1450                                 /* Remove data nodes that have no inode */
1451                                 dbg_rcvry("removing ino %lu",
1452                                           (unsigned long)e->inum);
1453                                 err = ubifs_tnc_remove_ino(c, e->inum);
1454                                 if (err)
1455                                         return err;
1456                         } else {
1457                                 struct ubifs_ino_node *ino = c->sbuf;
1458
1459                                 e->exists = 1;
1460                                 e->i_size = le64_to_cpu(ino->size);
1461                         }
1462                 }
1463                 if (e->exists && e->i_size < e->d_size) {
1464                         if (!e->inode && (c->vfs_sb->s_flags & MS_RDONLY)) {
1465                                 /* Fix the inode size and pin it in memory */
1466                                 struct inode *inode;
1467
1468                                 inode = ubifs_iget(c->vfs_sb, e->inum);
1469                                 if (IS_ERR(inode))
1470                                         return PTR_ERR(inode);
1471                                 if (inode->i_size < e->d_size) {
1472                                         dbg_rcvry("ino %lu size %lld -> %lld",
1473                                                   (unsigned long)e->inum,
1474                                                   e->d_size, inode->i_size);
1475                                         inode->i_size = e->d_size;
1476                                         ubifs_inode(inode)->ui_size = e->d_size;
1477                                         e->inode = inode;
1478                                         this = rb_next(this);
1479                                         continue;
1480                                 }
1481                                 iput(inode);
1482                         } else {
1483                                 /* Fix the size in place */
1484                                 err = fix_size_in_place(c, e);
1485                                 if (err)
1486                                         return err;
1487                                 if (e->inode)
1488                                         iput(e->inode);
1489                         }
1490                 }
1491                 this = rb_next(this);
1492                 rb_erase(&e->rb, &c->size_tree);
1493                 kfree(e);
1494         }
1495         return 0;
1496 }