fs/ubifs/orphan.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  * Author: Adrian Hunter
  20  */
  21
  22 #include "ubifs.h"
  23
  24 /*
  25  * An orphan is an inode number whose inode node has been committed to the index
  26  * with a link count of zero. That happens when an open file is deleted
  27  * (unlinked) and then a commit is run. In the normal course of events the inode
  28  * would be deleted when the file is closed. However in the case of an unclean
  29  * unmount, orphans need to be accounted for. After an unclean unmount, the
  30  * orphans' inodes must be deleted which means either scanning the entire index
  31  * looking for them, or keeping a list on flash somewhere. This unit implements
  32  * the latter approach.
  33  *
  34  * The orphan area is a fixed number of LEBs situated between the LPT area and
  35  * the main area. The number of orphan area LEBs is specified when the file
  36  * system is created. The minimum number is 1. The size of the orphan area
  37  * should be so that it can hold the maximum number of orphans that are expected
  38  * to ever exist at one time.
  39  *
  40  * The number of orphans that can fit in a LEB is:
  41  *
  42  *         (c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64)
  43  *
  44  * For example: a 15872 byte LEB can fit 1980 orphans so 1 LEB may be enough.
  45  *
  46  * Orphans are accumulated in a rb-tree. When an inode's link count drops to
  47  * zero, the inode number is added to the rb-tree. It is removed from the tree
  48  * when the inode is deleted.  Any new orphans that are in the orphan tree when
  49  * the commit is run, are written to the orphan area in 1 or more orphan nodes.
  50  * If the orphan area is full, it is consolidated to make space.  There is
  51  * always enough space because validation prevents the user from creating more
  52  * than the maximum number of orphans allowed.
  53  */
  54
  55 static int dbg_check_orphans(struct ubifs_info *c);
  56
  57 /**
  58  * ubifs_add_orphan - add an orphan.
  59  * @c: UBIFS file-system description object
  60  * @inum: orphan inode number
  61  *
  62  * Add an orphan. This function is called when an inodes link count drops to
  63  * zero.
  64  */
  65 int ubifs_add_orphan(struct ubifs_info *c, ino_t inum)
  66 {
  67         struct ubifs_orphan *orphan, *o;
  68         struct rb_node **p, *parent = NULL;
  69
  70         orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_NOFS);
  71         if (!orphan)
  72                 return -ENOMEM;
  73         orphan->inum = inum;
  74         orphan->new = 1;
  75
  76         spin_lock(&c->orphan_lock);
  77         if (c->tot_orphans >= c->max_orphans) {
  78                 spin_unlock(&c->orphan_lock);
  79                 kfree(orphan);
  80                 return -ENFILE;
  81         }
  82         p = &c->orph_tree.rb_node;
  83         while (*p) {
  84                 parent = *p;
  85                 o = rb_entry(parent, struct ubifs_orphan, rb);
  86                 if (inum < o->inum)
  87                         p = &(*p)->rb_left;
  88                 else if (inum > o->inum)
  89                         p = &(*p)->rb_right;
  90                 else {
  91                         ubifs_err("orphaned twice");
  92                         spin_unlock(&c->orphan_lock);
  93                         kfree(orphan);
  94                         return 0;
  95                 }
  96         }
  97         c->tot_orphans += 1;
  98         c->new_orphans += 1;
  99         rb_link_node(&orphan->rb, parent, p);
 100         rb_insert_color(&orphan->rb, &c->orph_tree);
 101         list_add_tail(&orphan->list, &c->orph_list);
 102         list_add_tail(&orphan->new_list, &c->orph_new);
 103         spin_unlock(&c->orphan_lock);
 104         dbg_gen("ino %lu", (unsigned long)inum);
 105         return 0;
 106 }
 107
 108 /**
 109  * ubifs_delete_orphan - delete an orphan.
 110  * @c: UBIFS file-system description object
 111  * @inum: orphan inode number
 112  *
 113  * Delete an orphan. This function is called when an inode is deleted.
 114  */
 115 void ubifs_delete_orphan(struct ubifs_info *c, ino_t inum)
 116 {
 117         struct ubifs_orphan *o;
 118         struct rb_node *p;
 119
 120         spin_lock(&c->orphan_lock);
 121         p = c->orph_tree.rb_node;
 122         while (p) {
 123                 o = rb_entry(p, struct ubifs_orphan, rb);
 124                 if (inum < o->inum)
 125                         p = p->rb_left;
 126                 else if (inum > o->inum)
 127                         p = p->rb_right;
 128                 else {
 129                         if (o->del) {
 130                                 spin_unlock(&c->orphan_lock);
 131                                 dbg_gen("deleted twice ino %lu",
 132                                         (unsigned long)inum);
 133                                 return;
 134                         }
 135                         if (o->cmt) {
 136                                 o->del = 1;
 137                                 o->dnext = c->orph_dnext;
 138                                 c->orph_dnext = o;
 139                                 spin_unlock(&c->orphan_lock);
 140                                 dbg_gen("delete later ino %lu",
 141                                         (unsigned long)inum);
 142                                 return;
 143                         }
 144                         rb_erase(p, &c->orph_tree);
 145                         list_del(&o->list);
 146                         c->tot_orphans -= 1;
 147                         if (o->new) {
 148                                 list_del(&o->new_list);
 149                                 c->new_orphans -= 1;
 150                         }
 151                         spin_unlock(&c->orphan_lock);
 152                         kfree(o);
 153                         dbg_gen("inum %lu", (unsigned long)inum);
 154                         return;
 155                 }
 156         }
 157         spin_unlock(&c->orphan_lock);
 158         ubifs_err("missing orphan ino %lu", (unsigned long)inum);
 159         dump_stack();
 160 }
 161
 162 /**
 163  * ubifs_orphan_start_commit - start commit of orphans.
 164  * @c: UBIFS file-system description object
 165  *
 166  * Start commit of orphans.
 167  */
 168 int ubifs_orphan_start_commit(struct ubifs_info *c)
 169 {
 170         struct ubifs_orphan *orphan, **last;
 171
 172         spin_lock(&c->orphan_lock);
 173         last = &c->orph_cnext;
 174         list_for_each_entry(orphan, &c->orph_new, new_list) {
 175                 ubifs_assert(orphan->new);
 176                 ubifs_assert(!orphan->cmt);
 177                 orphan->new = 0;
 178                 orphan->cmt = 1;
 179                 *last = orphan;
 180                 last = &orphan->cnext;
 181         }
 182         *last = NULL;
 183         c->cmt_orphans = c->new_orphans;
 184         c->new_orphans = 0;
 185         dbg_cmt("%d orphans to commit", c->cmt_orphans);
 186         INIT_LIST_HEAD(&c->orph_new);
 187         if (c->tot_orphans == 0)
 188                 c->no_orphs = 1;
 189         else
 190                 c->no_orphs = 0;
 191         spin_unlock(&c->orphan_lock);
 192         return 0;
 193 }
 194
 195 /**
 196  * avail_orphs - calculate available space.
 197  * @c: UBIFS file-system description object
 198  *
 199  * This function returns the number of orphans that can be written in the
 200  * available space.
 201  */
 202 static int avail_orphs(struct ubifs_info *c)
 203 {
 204         int avail_lebs, avail, gap;
 205
 206         avail_lebs = c->orph_lebs - (c->ohead_lnum - c->orph_first) - 1;
 207         avail = avail_lebs *
 208                ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
 209         gap = c->leb_size - c->ohead_offs;
 210         if (gap >= UBIFS_ORPH_NODE_SZ + sizeof(__le64))
 211                 avail += (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
 212         return avail;
 213 }
 214
 215 /**
 216  * tot_avail_orphs - calculate total space.
 217  * @c: UBIFS file-system description object
 218  *
 219  * This function returns the number of orphans that can be written in half
 220  * the total space. That leaves half the space for adding new orphans.
 221  */
 222 static int tot_avail_orphs(struct ubifs_info *c)
 223 {
 224         int avail_lebs, avail;
 225
 226         avail_lebs = c->orph_lebs;
 227         avail = avail_lebs *
 228                ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
 229         return avail / 2;
 230 }
 231
 232 /**
 233  * do_write_orph_node - write a node to the orphan head.
 234  * @c: UBIFS file-system description object
 235  * @len: length of node
 236  * @atomic: write atomically
 237  *
 238  * This function writes a node to the orphan head from the orphan buffer. If
 239  * %atomic is not zero, then the write is done atomically. On success, %0 is
 240  * returned, otherwise a negative error code is returned.
 241  */
 242 static int do_write_orph_node(struct ubifs_info *c, int len, int atomic)
 243 {
 244         int err = 0;
 245
 246         if (atomic) {
 247                 ubifs_assert(c->ohead_offs == 0);
 248                 ubifs_prepare_node(c, c->orph_buf, len, 1);
 249                 len = ALIGN(len, c->min_io_size);
 250                 err = ubifs_leb_change(c, c->ohead_lnum, c->orph_buf, len);
 251         } else {
 252                 if (c->ohead_offs == 0) {
 253                         /* Ensure LEB has been unmapped */
 254                         err = ubifs_leb_unmap(c, c->ohead_lnum);
 255                         if (err)
 256                                 return err;
 257                 }
 258                 err = ubifs_write_node(c, c->orph_buf, len, c->ohead_lnum,
 259                                        c->ohead_offs);
 260         }
 261         return err;
 262 }
 263
 264 /**
 265  * write_orph_node - write an orphan node.
 266  * @c: UBIFS file-system description object
 267  * @atomic: write atomically
 268  *
 269  * This function builds an orphan node from the cnext list and writes it to the
 270  * orphan head. On success, %0 is returned, otherwise a negative error code
 271  * is returned.
 272  */
 273 static int write_orph_node(struct ubifs_info *c, int atomic)
 274 {
 275         struct ubifs_orphan *orphan, *cnext;
 276         struct ubifs_orph_node *orph;
 277         int gap, err, len, cnt, i;
 278
 279         ubifs_assert(c->cmt_orphans > 0);
 280         gap = c->leb_size - c->ohead_offs;
 281         if (gap < UBIFS_ORPH_NODE_SZ + sizeof(__le64)) {
 282                 c->ohead_lnum += 1;
 283                 c->ohead_offs = 0;
 284                 gap = c->leb_size;
 285                 if (c->ohead_lnum > c->orph_last) {
 286                         /*
 287                          * We limit the number of orphans so that this should
 288                          * never happen.
 289                          */
 290                         ubifs_err("out of space in orphan area");
 291                         return -EINVAL;
 292                 }
 293         }
 294         cnt = (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
 295         if (cnt > c->cmt_orphans)
 296                 cnt = c->cmt_orphans;
 297         len = UBIFS_ORPH_NODE_SZ + cnt * sizeof(__le64);
 298         ubifs_assert(c->orph_buf);
 299         orph = c->orph_buf;
 300         orph->ch.node_type = UBIFS_ORPH_NODE;
 301         spin_lock(&c->orphan_lock);
 302         cnext = c->orph_cnext;
 303         for (i = 0; i < cnt; i++) {
 304                 orphan = cnext;
 305                 ubifs_assert(orphan->cmt);
 306                 orph->inos[i] = cpu_to_le64(orphan->inum);
 307                 orphan->cmt = 0;
 308                 cnext = orphan->cnext;
 309                 orphan->cnext = NULL;
 310         }
 311         c->orph_cnext = cnext;
 312         c->cmt_orphans -= cnt;
 313         spin_unlock(&c->orphan_lock);
 314         if (c->cmt_orphans)
 315                 orph->cmt_no = cpu_to_le64(c->cmt_no);
 316         else
 317                 /* Mark the last node of the commit */
 318                 orph->cmt_no = cpu_to_le64((c->cmt_no) | (1ULL << 63));
 319         ubifs_assert(c->ohead_offs + len <= c->leb_size);
 320         ubifs_assert(c->ohead_lnum >= c->orph_first);
 321         ubifs_assert(c->ohead_lnum <= c->orph_last);
 322         err = do_write_orph_node(c, len, atomic);
 323         c->ohead_offs += ALIGN(len, c->min_io_size);
 324         c->ohead_offs = ALIGN(c->ohead_offs, 8);
 325         return err;
 326 }
 327
 328 /**
 329  * write_orph_nodes - write orphan nodes until there are no more to commit.
 330  * @c: UBIFS file-system description object
 331  * @atomic: write atomically
 332  *
 333  * This function writes orphan nodes for all the orphans to commit. On success,
 334  * %0 is returned, otherwise a negative error code is returned.
 335  */
 336 static int write_orph_nodes(struct ubifs_info *c, int atomic)
 337 {
 338         int err;
 339
 340         while (c->cmt_orphans > 0) {
 341                 err = write_orph_node(c, atomic);
 342                 if (err)
 343                         return err;
 344         }
 345         if (atomic) {
 346                 int lnum;
 347
 348                 /* Unmap any unused LEBs after consolidation */
 349                 lnum = c->ohead_lnum + 1;
 350                 for (lnum = c->ohead_lnum + 1; lnum <= c->orph_last; lnum++) {
 351                         err = ubifs_leb_unmap(c, lnum);
 352                         if (err)
 353                                 return err;
 354                 }
 355         }
 356         return 0;
 357 }
 358
 359 /**
 360  * consolidate - consolidate the orphan area.
 361  * @c: UBIFS file-system description object
 362  *
 363  * This function enables consolidation by putting all the orphans into the list
 364  * to commit. The list is in the order that the orphans were added, and the
 365  * LEBs are written atomically in order, so at no time can orphans be lost by
 366  * an unclean unmount.
 367  *
 368  * This function returns %0 on success and a negative error code on failure.
 369  */
 370 static int consolidate(struct ubifs_info *c)
 371 {
 372         int tot_avail = tot_avail_orphs(c), err = 0;
 373
 374         spin_lock(&c->orphan_lock);
 375         dbg_cmt("there is space for %d orphans and there are %d",
 376                 tot_avail, c->tot_orphans);
 377         if (c->tot_orphans - c->new_orphans <= tot_avail) {
 378                 struct ubifs_orphan *orphan, **last;
 379                 int cnt = 0;
 380
 381                 /* Change the cnext list to include all non-new orphans */
 382                 last = &c->orph_cnext;
 383                 list_for_each_entry(orphan, &c->orph_list, list) {
 384                         if (orphan->new)
 385                                 continue;
 386                         orphan->cmt = 1;
 387                         *last = orphan;
 388                         last = &orphan->cnext;
 389                         cnt += 1;
 390                 }
 391                 *last = NULL;
 392                 ubifs_assert(cnt == c->tot_orphans - c->new_orphans);
 393                 c->cmt_orphans = cnt;
 394                 c->ohead_lnum = c->orph_first;
 395                 c->ohead_offs = 0;
 396         } else {
 397                 /*
 398                  * We limit the number of orphans so that this should
 399                  * never happen.
 400                  */
 401                 ubifs_err("out of space in orphan area");
 402                 err = -EINVAL;
 403         }
 404         spin_unlock(&c->orphan_lock);
 405         return err;
 406 }
 407
 408 /**
 409  * commit_orphans - commit orphans.
 410  * @c: UBIFS file-system description object
 411  *
 412  * This function commits orphans to flash. On success, %0 is returned,
 413  * otherwise a negative error code is returned.
 414  */
 415 static int commit_orphans(struct ubifs_info *c)
 416 {
 417         int avail, atomic = 0, err;
 418
 419         ubifs_assert(c->cmt_orphans > 0);
 420         avail = avail_orphs(c);
 421         if (avail < c->cmt_orphans) {
 422                 /* Not enough space to write new orphans, so consolidate */
 423                 err = consolidate(c);
 424                 if (err)
 425                         return err;
 426                 atomic = 1;
 427         }
 428         err = write_orph_nodes(c, atomic);
 429         return err;
 430 }
 431
 432 /**
 433  * erase_deleted - erase the orphans marked for deletion.
 434  * @c: UBIFS file-system description object
 435  *
 436  * During commit, the orphans being committed cannot be deleted, so they are
 437  * marked for deletion and deleted by this function. Also, the recovery
 438  * adds killed orphans to the deletion list, and therefore they are deleted
 439  * here too.
 440  */
 441 static void erase_deleted(struct ubifs_info *c)
 442 {
 443         struct ubifs_orphan *orphan, *dnext;
 444
 445         spin_lock(&c->orphan_lock);
 446         dnext = c->orph_dnext;
 447         while (dnext) {
 448                 orphan = dnext;
 449                 dnext = orphan->dnext;
 450                 ubifs_assert(!orphan->new);
 451                 ubifs_assert(orphan->del);
 452                 rb_erase(&orphan->rb, &c->orph_tree);
 453                 list_del(&orphan->list);
 454                 c->tot_orphans -= 1;
 455                 dbg_gen("deleting orphan ino %lu", (unsigned long)orphan->inum);
 456                 kfree(orphan);
 457         }
 458         c->orph_dnext = NULL;
 459         spin_unlock(&c->orphan_lock);
 460 }
 461
 462 /**
 463  * ubifs_orphan_end_commit - end commit of orphans.
 464  * @c: UBIFS file-system description object
 465  *
 466  * End commit of orphans.
 467  */
 468 int ubifs_orphan_end_commit(struct ubifs_info *c)
 469 {
 470         int err;
 471
 472         if (c->cmt_orphans != 0) {
 473                 err = commit_orphans(c);
 474                 if (err)
 475                         return err;
 476         }
 477         erase_deleted(c);
 478         err = dbg_check_orphans(c);
 479         return err;
 480 }
 481
 482 /**
 483  * ubifs_clear_orphans - erase all LEBs used for orphans.
 484  * @c: UBIFS file-system description object
 485  *
 486  * If recovery is not required, then the orphans from the previous session
 487  * are not needed. This function locates the LEBs used to record
 488  * orphans, and un-maps them.
 489  */
 490 int ubifs_clear_orphans(struct ubifs_info *c)
 491 {
 492         int lnum, err;
 493
 494         for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
 495                 err = ubifs_leb_unmap(c, lnum);
 496                 if (err)
 497                         return err;
 498         }
 499         c->ohead_lnum = c->orph_first;
 500         c->ohead_offs = 0;
 501         return 0;
 502 }
 503
 504 /**
 505  * insert_dead_orphan - insert an orphan.
 506  * @c: UBIFS file-system description object
 507  * @inum: orphan inode number
 508  *
 509  * This function is a helper to the 'do_kill_orphans()' function. The orphan
 510  * must be kept until the next commit, so it is added to the rb-tree and the
 511  * deletion list.
 512  */
 513 static int insert_dead_orphan(struct ubifs_info *c, ino_t inum)
 514 {
 515         struct ubifs_orphan *orphan, *o;
 516         struct rb_node **p, *parent = NULL;
 517
 518         orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_KERNEL);
 519         if (!orphan)
 520                 return -ENOMEM;
 521         orphan->inum = inum;
 522
 523         p = &c->orph_tree.rb_node;
 524         while (*p) {
 525                 parent = *p;
 526                 o = rb_entry(parent, struct ubifs_orphan, rb);
 527                 if (inum < o->inum)
 528                         p = &(*p)->rb_left;
 529                 else if (inum > o->inum)
 530                         p = &(*p)->rb_right;
 531                 else {
 532                         /* Already added - no problem */
 533                         kfree(orphan);
 534                         return 0;
 535                 }
 536         }
 537         c->tot_orphans += 1;
 538         rb_link_node(&orphan->rb, parent, p);
 539         rb_insert_color(&orphan->rb, &c->orph_tree);
 540         list_add_tail(&orphan->list, &c->orph_list);
 541         orphan->del = 1;
 542         orphan->dnext = c->orph_dnext;
 543         c->orph_dnext = orphan;
 544         dbg_mnt("ino %lu, new %d, tot %d", (unsigned long)inum,
 545                 c->new_orphans, c->tot_orphans);
 546         return 0;
 547 }
 548
 549 /**
 550  * do_kill_orphans - remove orphan inodes from the index.
 551  * @c: UBIFS file-system description object
 552  * @sleb: scanned LEB
 553  * @last_cmt_no: cmt_no of last orphan node read is passed and returned here
 554  * @outofdate: whether the LEB is out of date is returned here
 555  * @last_flagged: whether the end orphan node is encountered
 556  *
 557  * This function is a helper to the 'kill_orphans()' function. It goes through
 558  * every orphan node in a LEB and for every inode number recorded, removes
 559  * all keys for that inode from the TNC.
 560  */
 561 static int do_kill_orphans(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
 562                            unsigned long long *last_cmt_no, int *outofdate,
 563                            int *last_flagged)
 564 {
 565         struct ubifs_scan_node *snod;
 566         struct ubifs_orph_node *orph;
 567         unsigned long long cmt_no;
 568         ino_t inum;
 569         int i, n, err, first = 1;
 570
 571         list_for_each_entry(snod, &sleb->nodes, list) {
 572                 if (snod->type != UBIFS_ORPH_NODE) {
 573                         ubifs_err("invalid node type %d in orphan area at %d:%d",
 574                                   snod->type, sleb->lnum, snod->offs);
 575                         ubifs_dump_node(c, snod->node);
 576                         return -EINVAL;
 577                 }
 578
 579                 orph = snod->node;
 580
 581                 /* Check commit number */
 582                 cmt_no = le64_to_cpu(orph->cmt_no) & LLONG_MAX;
 583                 /*
 584                  * The commit number on the master node may be less, because
 585                  * of a failed commit. If there are several failed commits in a
 586                  * row, the commit number written on orphan nodes will continue
 587                  * to increase (because the commit number is adjusted here) even
 588                  * though the commit number on the master node stays the same
 589                  * because the master node has not been re-written.
 590                  */
 591                 if (cmt_no > c->cmt_no)
 592                         c->cmt_no = cmt_no;
 593                 if (cmt_no < *last_cmt_no && *last_flagged) {
 594                         /*
 595                          * The last orphan node had a higher commit number and
 596                          * was flagged as the last written for that commit
 597                          * number. That makes this orphan node, out of date.
 598                          */
 599                         if (!first) {
 600                                 ubifs_err("out of order commit number %llu in orphan node at %d:%d",
 601                                           cmt_no, sleb->lnum, snod->offs);
 602                                 ubifs_dump_node(c, snod->node);
 603                                 return -EINVAL;
 604                         }
 605                         dbg_rcvry("out of date LEB %d", sleb->lnum);
 606                         *outofdate = 1;
 607                         return 0;
 608                 }
 609
 610                 if (first)
 611                         first = 0;
 612
 613                 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
 614                 for (i = 0; i < n; i++) {
 615                         inum = le64_to_cpu(orph->inos[i]);
 616                         dbg_rcvry("deleting orphaned inode %lu",
 617                                   (unsigned long)inum);
 618                         err = ubifs_tnc_remove_ino(c, inum);
 619                         if (err)
 620                                 return err;
 621                         err = insert_dead_orphan(c, inum);
 622                         if (err)
 623                                 return err;
 624                 }
 625
 626                 *last_cmt_no = cmt_no;
 627                 if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) {
 628                         dbg_rcvry("last orph node for commit %llu at %d:%d",
 629                                   cmt_no, sleb->lnum, snod->offs);
 630                         *last_flagged = 1;
 631                 } else
 632                         *last_flagged = 0;
 633         }
 634
 635         return 0;
 636 }
 637
 638 /**
 639  * kill_orphans - remove all orphan inodes from the index.
 640  * @c: UBIFS file-system description object
 641  *
 642  * If recovery is required, then orphan inodes recorded during the previous
 643  * session (which ended with an unclean unmount) must be deleted from the index.
 644  * This is done by updating the TNC, but since the index is not updated until
 645  * the next commit, the LEBs where the orphan information is recorded are not
 646  * erased until the next commit.
 647  */
 648 static int kill_orphans(struct ubifs_info *c)
 649 {
 650         unsigned long long last_cmt_no = 0;
 651         int lnum, err = 0, outofdate = 0, last_flagged = 0;
 652
 653         c->ohead_lnum = c->orph_first;
 654         c->ohead_offs = 0;
 655         /* Check no-orphans flag and skip this if no orphans */
 656         if (c->no_orphs) {
 657                 dbg_rcvry("no orphans");
 658                 return 0;
 659         }
 660         /*
 661          * Orph nodes always start at c->orph_first and are written to each
 662          * successive LEB in turn. Generally unused LEBs will have been unmapped
 663          * but may contain out of date orphan nodes if the unmap didn't go
 664          * through. In addition, the last orphan node written for each commit is
 665          * marked (top bit of orph->cmt_no is set to 1). It is possible that
 666          * there are orphan nodes from the next commit (i.e. the commit did not
 667          * complete successfully). In that case, no orphans will have been lost
 668          * due to the way that orphans are written, and any orphans added will
 669          * be valid orphans anyway and so can be deleted.
 670          */
 671         for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
 672                 struct ubifs_scan_leb *sleb;
 673
 674                 dbg_rcvry("LEB %d", lnum);
 675                 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 1);
 676                 if (IS_ERR(sleb)) {
 677                         if (PTR_ERR(sleb) == -EUCLEAN)
 678                                 sleb = ubifs_recover_leb(c, lnum, 0,
 679                                                          c->sbuf, -1);
 680                         if (IS_ERR(sleb)) {
 681                                 err = PTR_ERR(sleb);
 682                                 break;
 683                         }
 684                 }
 685                 err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate,
 686                                       &last_flagged);
 687                 if (err || outofdate) {
 688                         ubifs_scan_destroy(sleb);
 689                         break;
 690                 }
 691                 if (sleb->endpt) {
 692                         c->ohead_lnum = lnum;
 693                         c->ohead_offs = sleb->endpt;
 694                 }
 695                 ubifs_scan_destroy(sleb);
 696         }
 697         return err;
 698 }
 699
 700 /**
 701  * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them.
 702  * @c: UBIFS file-system description object
 703  * @unclean: indicates recovery from unclean unmount
 704  * @read_only: indicates read only mount
 705  *
 706  * This function is called when mounting to erase orphans from the previous
 707  * session. If UBIFS was not unmounted cleanly, then the inodes recorded as
 708  * orphans are deleted.
 709  */
 710 int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only)
 711 {
 712         int err = 0;
 713
 714         c->max_orphans = tot_avail_orphs(c);
 715
 716         if (!read_only) {
 717                 c->orph_buf = vmalloc(c->leb_size);
 718                 if (!c->orph_buf)
 719                         return -ENOMEM;
 720         }
 721
 722         if (unclean)
 723                 err = kill_orphans(c);
 724         else if (!read_only)
 725                 err = ubifs_clear_orphans(c);
 726
 727         return err;
 728 }
 729
 730 /*
 731  * Everything below is related to debugging.
 732  */
 733
 734 struct check_orphan {
 735         struct rb_node rb;
 736         ino_t inum;
 737 };
 738
 739 struct check_info {
 740         unsigned long last_ino;
 741         unsigned long tot_inos;
 742         unsigned long missing;
 743         unsigned long long leaf_cnt;
 744         struct ubifs_ino_node *node;
 745         struct rb_root root;
 746 };
 747
 748 static int dbg_find_orphan(struct ubifs_info *c, ino_t inum)
 749 {
 750         struct ubifs_orphan *o;
 751         struct rb_node *p;
 752
 753         spin_lock(&c->orphan_lock);
 754         p = c->orph_tree.rb_node;
 755         while (p) {
 756                 o = rb_entry(p, struct ubifs_orphan, rb);
 757                 if (inum < o->inum)
 758                         p = p->rb_left;
 759                 else if (inum > o->inum)
 760                         p = p->rb_right;
 761                 else {
 762                         spin_unlock(&c->orphan_lock);
 763                         return 1;
 764                 }
 765         }
 766         spin_unlock(&c->orphan_lock);
 767         return 0;
 768 }
 769
 770 static int dbg_ins_check_orphan(struct rb_root *root, ino_t inum)
 771 {
 772         struct check_orphan *orphan, *o;
 773         struct rb_node **p, *parent = NULL;
 774
 775         orphan = kzalloc(sizeof(struct check_orphan), GFP_NOFS);
 776         if (!orphan)
 777                 return -ENOMEM;
 778         orphan->inum = inum;
 779
 780         p = &root->rb_node;
 781         while (*p) {
 782                 parent = *p;
 783                 o = rb_entry(parent, struct check_orphan, rb);
 784                 if (inum < o->inum)
 785                         p = &(*p)->rb_left;
 786                 else if (inum > o->inum)
 787                         p = &(*p)->rb_right;
 788                 else {
 789                         kfree(orphan);
 790                         return 0;
 791                 }
 792         }
 793         rb_link_node(&orphan->rb, parent, p);
 794         rb_insert_color(&orphan->rb, root);
 795         return 0;
 796 }
 797
 798 static int dbg_find_check_orphan(struct rb_root *root, ino_t inum)
 799 {
 800         struct check_orphan *o;
 801         struct rb_node *p;
 802
 803         p = root->rb_node;
 804         while (p) {
 805                 o = rb_entry(p, struct check_orphan, rb);
 806                 if (inum < o->inum)
 807                         p = p->rb_left;
 808                 else if (inum > o->inum)
 809                         p = p->rb_right;
 810                 else
 811                         return 1;
 812         }
 813         return 0;
 814 }
 815
 816 static void dbg_free_check_tree(struct rb_root *root)
 817 {
 818         struct check_orphan *o, *n;
 819
 820         rbtree_postorder_for_each_entry_safe(o, n, root, rb)
 821                 kfree(o);
 822 }
 823
 824 static int dbg_orphan_check(struct ubifs_info *c, struct ubifs_zbranch *zbr,
 825                             void *priv)
 826 {
 827         struct check_info *ci = priv;
 828         ino_t inum;
 829         int err;
 830
 831         inum = key_inum(c, &zbr->key);
 832         if (inum != ci->last_ino) {
 833                 /* Lowest node type is the inode node, so it comes first */
 834                 if (key_type(c, &zbr->key) != UBIFS_INO_KEY)
 835                         ubifs_err("found orphan node ino %lu, type %d",
 836                                   (unsigned long)inum, key_type(c, &zbr->key));
 837                 ci->last_ino = inum;
 838                 ci->tot_inos += 1;
 839                 err = ubifs_tnc_read_node(c, zbr, ci->node);
 840                 if (err) {
 841                         ubifs_err("node read failed, error %d", err);
 842                         return err;
 843                 }
 844                 if (ci->node->nlink == 0)
 845                         /* Must be recorded as an orphan */
 846                         if (!dbg_find_check_orphan(&ci->root, inum) &&
 847                             !dbg_find_orphan(c, inum)) {
 848                                 ubifs_err("missing orphan, ino %lu",
 849                                           (unsigned long)inum);
 850                                 ci->missing += 1;
 851                         }
 852         }
 853         ci->leaf_cnt += 1;
 854         return 0;
 855 }
 856
 857 static int dbg_read_orphans(struct check_info *ci, struct ubifs_scan_leb *sleb)
 858 {
 859         struct ubifs_scan_node *snod;
 860         struct ubifs_orph_node *orph;
 861         ino_t inum;
 862         int i, n, err;
 863
 864         list_for_each_entry(snod, &sleb->nodes, list) {
 865                 cond_resched();
 866                 if (snod->type != UBIFS_ORPH_NODE)
 867                         continue;
 868                 orph = snod->node;
 869                 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
 870                 for (i = 0; i < n; i++) {
 871                         inum = le64_to_cpu(orph->inos[i]);
 872                         err = dbg_ins_check_orphan(&ci->root, inum);
 873                         if (err)
 874                                 return err;
 875                 }
 876         }
 877         return 0;
 878 }
 879
 880 static int dbg_scan_orphans(struct ubifs_info *c, struct check_info *ci)
 881 {
 882         int lnum, err = 0;
 883         void *buf;
 884
 885         /* Check no-orphans flag and skip this if no orphans */
 886         if (c->no_orphs)
 887                 return 0;
 888
 889         buf = __vmalloc(c->leb_size, GFP_NOFS, PAGE_KERNEL);
 890         if (!buf) {
 891                 ubifs_err("cannot allocate memory to check orphans");
 892                 return 0;
 893         }
 894
 895         for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
 896                 struct ubifs_scan_leb *sleb;
 897
 898                 sleb = ubifs_scan(c, lnum, 0, buf, 0);
 899                 if (IS_ERR(sleb)) {
 900                         err = PTR_ERR(sleb);
 901                         break;
 902                 }
 903
 904                 err = dbg_read_orphans(ci, sleb);
 905                 ubifs_scan_destroy(sleb);
 906                 if (err)
 907                         break;
 908         }
 909
 910         vfree(buf);
 911         return err;
 912 }
 913
 914 static int dbg_check_orphans(struct ubifs_info *c)
 915 {
 916         struct check_info ci;
 917         int err;
 918
 919         if (!dbg_is_chk_orph(c))
 920                 return 0;
 921
 922         ci.last_ino = 0;
 923         ci.tot_inos = 0;
 924         ci.missing  = 0;
 925         ci.leaf_cnt = 0;
 926         ci.root = RB_ROOT;
 927         ci.node = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS);
 928         if (!ci.node) {
 929                 ubifs_err("out of memory");
 930                 return -ENOMEM;
 931         }
 932
 933         err = dbg_scan_orphans(c, &ci);
 934         if (err)
 935                 goto out;
 936
 937         err = dbg_walk_index(c, &dbg_orphan_check, NULL, &ci);
 938         if (err) {
 939                 ubifs_err("cannot scan TNC, error %d", err);
 940                 goto out;
 941         }
 942
 943         if (ci.missing) {
 944                 ubifs_err("%lu missing orphan(s)", ci.missing);
 945                 err = -EINVAL;
 946                 goto out;
 947         }
 948
 949         dbg_cmt("last inode number is %lu", ci.last_ino);
 950         dbg_cmt("total number of inodes is %lu", ci.tot_inos);
 951         dbg_cmt("total number of leaf nodes is %llu", ci.leaf_cnt);
 952
 953 out:
 954         dbg_free_check_tree(&ci.root);
 955         kfree(ci.node);
 956         return err;
 957 }