fs/ubifs/tnc_commit.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 /* This file implements TNC functions for committing */
  24
  25 #include <linux/random.h>
  26 #include "ubifs.h"
  27
  28 /**
  29  * make_idx_node - make an index node for fill-the-gaps method of TNC commit.
  30  * @c: UBIFS file-system description object
  31  * @idx: buffer in which to place new index node
  32  * @znode: znode from which to make new index node
  33  * @lnum: LEB number where new index node will be written
  34  * @offs: offset where new index node will be written
  35  * @len: length of new index node
  36  */
  37 static int make_idx_node(struct ubifs_info *c, struct ubifs_idx_node *idx,
  38                          struct ubifs_znode *znode, int lnum, int offs, int len)
  39 {
  40         struct ubifs_znode *zp;
  41         int i, err;
  42
  43         /* Make index node */
  44         idx->ch.node_type = UBIFS_IDX_NODE;
  45         idx->child_cnt = cpu_to_le16(znode->child_cnt);
  46         idx->level = cpu_to_le16(znode->level);
  47         for (i = 0; i < znode->child_cnt; i++) {
  48                 struct ubifs_branch *br = ubifs_idx_branch(c, idx, i);
  49                 struct ubifs_zbranch *zbr = &znode->zbranch[i];
  50
  51                 key_write_idx(c, &zbr->key, &br->key);
  52                 br->lnum = cpu_to_le32(zbr->lnum);
  53                 br->offs = cpu_to_le32(zbr->offs);
  54                 br->len = cpu_to_le32(zbr->len);
  55                 if (!zbr->lnum || !zbr->len) {
  56                         ubifs_err(c, "bad ref in znode");
  57                         ubifs_dump_znode(c, znode);
  58                         if (zbr->znode)
  59                                 ubifs_dump_znode(c, zbr->znode);
  60                 }
  61         }
  62         ubifs_prepare_node(c, idx, len, 0);
  63
  64         znode->lnum = lnum;
  65         znode->offs = offs;
  66         znode->len = len;
  67
  68         err = insert_old_idx_znode(c, znode);
  69
  70         /* Update the parent */
  71         zp = znode->parent;
  72         if (zp) {
  73                 struct ubifs_zbranch *zbr;
  74
  75                 zbr = &zp->zbranch[znode->iip];
  76                 zbr->lnum = lnum;
  77                 zbr->offs = offs;
  78                 zbr->len = len;
  79         } else {
  80                 c->zroot.lnum = lnum;
  81                 c->zroot.offs = offs;
  82                 c->zroot.len = len;
  83         }
  84         c->calc_idx_sz += ALIGN(len, 8);
  85
  86         atomic_long_dec(&c->dirty_zn_cnt);
  87
  88         ubifs_assert(ubifs_zn_dirty(znode));
  89         ubifs_assert(ubifs_zn_cow(znode));
  90
  91         /*
  92          * Note, unlike 'write_index()' we do not add memory barriers here
  93          * because this function is called with @c->tnc_mutex locked.
  94          */
  95         __clear_bit(DIRTY_ZNODE, &znode->flags);
  96         __clear_bit(COW_ZNODE, &znode->flags);
  97
  98         return err;
  99 }
 100
 101 /**
 102  * fill_gap - make index nodes in gaps in dirty index LEBs.
 103  * @c: UBIFS file-system description object
 104  * @lnum: LEB number that gap appears in
 105  * @gap_start: offset of start of gap
 106  * @gap_end: offset of end of gap
 107  * @dirt: adds dirty space to this
 108  *
 109  * This function returns the number of index nodes written into the gap.
 110  */
 111 static int fill_gap(struct ubifs_info *c, int lnum, int gap_start, int gap_end,
 112                     int *dirt)
 113 {
 114         int len, gap_remains, gap_pos, written, pad_len;
 115
 116         ubifs_assert((gap_start & 7) == 0);
 117         ubifs_assert((gap_end & 7) == 0);
 118         ubifs_assert(gap_end >= gap_start);
 119
 120         gap_remains = gap_end - gap_start;
 121         if (!gap_remains)
 122                 return 0;
 123         gap_pos = gap_start;
 124         written = 0;
 125         while (c->enext) {
 126                 len = ubifs_idx_node_sz(c, c->enext->child_cnt);
 127                 if (len < gap_remains) {
 128                         struct ubifs_znode *znode = c->enext;
 129                         const int alen = ALIGN(len, 8);
 130                         int err;
 131
 132                         ubifs_assert(alen <= gap_remains);
 133                         err = make_idx_node(c, c->ileb_buf + gap_pos, znode,
 134                                             lnum, gap_pos, len);
 135                         if (err)
 136                                 return err;
 137                         gap_remains -= alen;
 138                         gap_pos += alen;
 139                         c->enext = znode->cnext;
 140                         if (c->enext == c->cnext)
 141                                 c->enext = NULL;
 142                         written += 1;
 143                 } else
 144                         break;
 145         }
 146         if (gap_end == c->leb_size) {
 147                 c->ileb_len = ALIGN(gap_pos, c->min_io_size);
 148                 /* Pad to end of min_io_size */
 149                 pad_len = c->ileb_len - gap_pos;
 150         } else
 151                 /* Pad to end of gap */
 152                 pad_len = gap_remains;
 153         dbg_gc("LEB %d:%d to %d len %d nodes written %d wasted bytes %d",
 154                lnum, gap_start, gap_end, gap_end - gap_start, written, pad_len);
 155         ubifs_pad(c, c->ileb_buf + gap_pos, pad_len);
 156         *dirt += pad_len;
 157         return written;
 158 }
 159
 160 /**
 161  * find_old_idx - find an index node obsoleted since the last commit start.
 162  * @c: UBIFS file-system description object
 163  * @lnum: LEB number of obsoleted index node
 164  * @offs: offset of obsoleted index node
 165  *
 166  * Returns %1 if found and %0 otherwise.
 167  */
 168 static int find_old_idx(struct ubifs_info *c, int lnum, int offs)
 169 {
 170         struct ubifs_old_idx *o;
 171         struct rb_node *p;
 172
 173         p = c->old_idx.rb_node;
 174         while (p) {
 175                 o = rb_entry(p, struct ubifs_old_idx, rb);
 176                 if (lnum < o->lnum)
 177                         p = p->rb_left;
 178                 else if (lnum > o->lnum)
 179                         p = p->rb_right;
 180                 else if (offs < o->offs)
 181                         p = p->rb_left;
 182                 else if (offs > o->offs)
 183                         p = p->rb_right;
 184                 else
 185                         return 1;
 186         }
 187         return 0;
 188 }
 189
 190 /**
 191  * is_idx_node_in_use - determine if an index node can be overwritten.
 192  * @c: UBIFS file-system description object
 193  * @key: key of index node
 194  * @level: index node level
 195  * @lnum: LEB number of index node
 196  * @offs: offset of index node
 197  *
 198  * If @key / @lnum / @offs identify an index node that was not part of the old
 199  * index, then this function returns %0 (obsolete).  Else if the index node was
 200  * part of the old index but is now dirty %1 is returned, else if it is clean %2
 201  * is returned. A negative error code is returned on failure.
 202  */
 203 static int is_idx_node_in_use(struct ubifs_info *c, union ubifs_key *key,
 204                               int level, int lnum, int offs)
 205 {
 206         int ret;
 207
 208         ret = is_idx_node_in_tnc(c, key, level, lnum, offs);
 209         if (ret < 0)
 210                 return ret; /* Error code */
 211         if (ret == 0)
 212                 if (find_old_idx(c, lnum, offs))
 213                         return 1;
 214         return ret;
 215 }
 216
 217 /**
 218  * layout_leb_in_gaps - layout index nodes using in-the-gaps method.
 219  * @c: UBIFS file-system description object
 220  * @p: return LEB number here
 221  *
 222  * This function lays out new index nodes for dirty znodes using in-the-gaps
 223  * method of TNC commit.
 224  * This function merely puts the next znode into the next gap, making no attempt
 225  * to try to maximise the number of znodes that fit.
 226  * This function returns the number of index nodes written into the gaps, or a
 227  * negative error code on failure.
 228  */
 229 static int layout_leb_in_gaps(struct ubifs_info *c, int *p)
 230 {
 231         struct ubifs_scan_leb *sleb;
 232         struct ubifs_scan_node *snod;
 233         int lnum, dirt = 0, gap_start, gap_end, err, written, tot_written;
 234
 235         tot_written = 0;
 236         /* Get an index LEB with lots of obsolete index nodes */
 237         lnum = ubifs_find_dirty_idx_leb(c);
 238         if (lnum < 0)
 239                 /*
 240                  * There also may be dirt in the index head that could be
 241                  * filled, however we do not check there at present.
 242                  */
 243                 return lnum; /* Error code */
 244         *p = lnum;
 245         dbg_gc("LEB %d", lnum);
 246         /*
 247          * Scan the index LEB.  We use the generic scan for this even though
 248          * it is more comprehensive and less efficient than is needed for this
 249          * purpose.
 250          */
 251         sleb = ubifs_scan(c, lnum, 0, c->ileb_buf, 0);
 252         c->ileb_len = 0;
 253         if (IS_ERR(sleb))
 254                 return PTR_ERR(sleb);
 255         gap_start = 0;
 256         list_for_each_entry(snod, &sleb->nodes, list) {
 257                 struct ubifs_idx_node *idx;
 258                 int in_use, level;
 259
 260                 ubifs_assert(snod->type == UBIFS_IDX_NODE);
 261                 idx = snod->node;
 262                 key_read(c, ubifs_idx_key(c, idx), &snod->key);
 263                 level = le16_to_cpu(idx->level);
 264                 /* Determine if the index node is in use (not obsolete) */
 265                 in_use = is_idx_node_in_use(c, &snod->key, level, lnum,
 266                                             snod->offs);
 267                 if (in_use < 0) {
 268                         ubifs_scan_destroy(sleb);
 269                         return in_use; /* Error code */
 270                 }
 271                 if (in_use) {
 272                         if (in_use == 1)
 273                                 dirt += ALIGN(snod->len, 8);
 274                         /*
 275                          * The obsolete index nodes form gaps that can be
 276                          * overwritten.  This gap has ended because we have
 277                          * found an index node that is still in use
 278                          * i.e. not obsolete
 279                          */
 280                         gap_end = snod->offs;
 281                         /* Try to fill gap */
 282                         written = fill_gap(c, lnum, gap_start, gap_end, &dirt);
 283                         if (written < 0) {
 284                                 ubifs_scan_destroy(sleb);
 285                                 return written; /* Error code */
 286                         }
 287                         tot_written += written;
 288                         gap_start = ALIGN(snod->offs + snod->len, 8);
 289                 }
 290         }
 291         ubifs_scan_destroy(sleb);
 292         c->ileb_len = c->leb_size;
 293         gap_end = c->leb_size;
 294         /* Try to fill gap */
 295         written = fill_gap(c, lnum, gap_start, gap_end, &dirt);
 296         if (written < 0)
 297                 return written; /* Error code */
 298         tot_written += written;
 299         if (tot_written == 0) {
 300                 struct ubifs_lprops lp;
 301
 302                 dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written);
 303                 err = ubifs_read_one_lp(c, lnum, &lp);
 304                 if (err)
 305                         return err;
 306                 if (lp.free == c->leb_size) {
 307                         /*
 308                          * We must have snatched this LEB from the idx_gc list
 309                          * so we need to correct the free and dirty space.
 310                          */
 311                         err = ubifs_change_one_lp(c, lnum,
 312                                                   c->leb_size - c->ileb_len,
 313                                                   dirt, 0, 0, 0);
 314                         if (err)
 315                                 return err;
 316                 }
 317                 return 0;
 318         }
 319         err = ubifs_change_one_lp(c, lnum, c->leb_size - c->ileb_len, dirt,
 320                                   0, 0, 0);
 321         if (err)
 322                 return err;
 323         err = ubifs_leb_change(c, lnum, c->ileb_buf, c->ileb_len);
 324         if (err)
 325                 return err;
 326         dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written);
 327         return tot_written;
 328 }
 329
 330 /**
 331  * get_leb_cnt - calculate the number of empty LEBs needed to commit.
 332  * @c: UBIFS file-system description object
 333  * @cnt: number of znodes to commit
 334  *
 335  * This function returns the number of empty LEBs needed to commit @cnt znodes
 336  * to the current index head.  The number is not exact and may be more than
 337  * needed.
 338  */
 339 static int get_leb_cnt(struct ubifs_info *c, int cnt)
 340 {
 341         int d;
 342
 343         /* Assume maximum index node size (i.e. overestimate space needed) */
 344         cnt -= (c->leb_size - c->ihead_offs) / c->max_idx_node_sz;
 345         if (cnt < 0)
 346                 cnt = 0;
 347         d = c->leb_size / c->max_idx_node_sz;
 348         return DIV_ROUND_UP(cnt, d);
 349 }
 350
 351 /**
 352  * layout_in_gaps - in-the-gaps method of committing TNC.
 353  * @c: UBIFS file-system description object
 354  * @cnt: number of dirty znodes to commit.
 355  *
 356  * This function lays out new index nodes for dirty znodes using in-the-gaps
 357  * method of TNC commit.
 358  *
 359  * This function returns %0 on success and a negative error code on failure.
 360  */
 361 static int layout_in_gaps(struct ubifs_info *c, int cnt)
 362 {
 363         int err, leb_needed_cnt, written, *p;
 364
 365         dbg_gc("%d znodes to write", cnt);
 366
 367         c->gap_lebs = kmalloc(sizeof(int) * (c->lst.idx_lebs + 1), GFP_NOFS);
 368         if (!c->gap_lebs)
 369                 return -ENOMEM;
 370
 371         p = c->gap_lebs;
 372         do {
 373                 ubifs_assert(p < c->gap_lebs + sizeof(int) * c->lst.idx_lebs);
 374                 written = layout_leb_in_gaps(c, p);
 375                 if (written < 0) {
 376                         err = written;
 377                         if (err != -ENOSPC) {
 378                                 kfree(c->gap_lebs);
 379                                 c->gap_lebs = NULL;
 380                                 return err;
 381                         }
 382                         if (!dbg_is_chk_index(c)) {
 383                                 /*
 384                                  * Do not print scary warnings if the debugging
 385                                  * option which forces in-the-gaps is enabled.
 386                                  */
 387                                 ubifs_warn(c, "out of space");
 388                                 ubifs_dump_budg(c, &c->bi);
 389                                 ubifs_dump_lprops(c);
 390                         }
 391                         /* Try to commit anyway */
 392                         break;
 393                 }
 394                 p++;
 395                 cnt -= written;
 396                 leb_needed_cnt = get_leb_cnt(c, cnt);
 397                 dbg_gc("%d znodes remaining, need %d LEBs, have %d", cnt,
 398                        leb_needed_cnt, c->ileb_cnt);
 399         } while (leb_needed_cnt > c->ileb_cnt);
 400
 401         *p = -1;
 402         return 0;
 403 }
 404
 405 /**
 406  * layout_in_empty_space - layout index nodes in empty space.
 407  * @c: UBIFS file-system description object
 408  *
 409  * This function lays out new index nodes for dirty znodes using empty LEBs.
 410  *
 411  * This function returns %0 on success and a negative error code on failure.
 412  */
 413 static int layout_in_empty_space(struct ubifs_info *c)
 414 {
 415         struct ubifs_znode *znode, *cnext, *zp;
 416         int lnum, offs, len, next_len, buf_len, buf_offs, used, avail;
 417         int wlen, blen, err;
 418
 419         cnext = c->enext;
 420         if (!cnext)
 421                 return 0;
 422
 423         lnum = c->ihead_lnum;
 424         buf_offs = c->ihead_offs;
 425
 426         buf_len = ubifs_idx_node_sz(c, c->fanout);
 427         buf_len = ALIGN(buf_len, c->min_io_size);
 428         used = 0;
 429         avail = buf_len;
 430
 431         /* Ensure there is enough room for first write */
 432         next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
 433         if (buf_offs + next_len > c->leb_size)
 434                 lnum = -1;
 435
 436         while (1) {
 437                 znode = cnext;
 438
 439                 len = ubifs_idx_node_sz(c, znode->child_cnt);
 440
 441                 /* Determine the index node position */
 442                 if (lnum == -1) {
 443                         if (c->ileb_nxt >= c->ileb_cnt) {
 444                                 ubifs_err(c, "out of space");
 445                                 return -ENOSPC;
 446                         }
 447                         lnum = c->ilebs[c->ileb_nxt++];
 448                         buf_offs = 0;
 449                         used = 0;
 450                         avail = buf_len;
 451                 }
 452
 453                 offs = buf_offs + used;
 454
 455                 znode->lnum = lnum;
 456                 znode->offs = offs;
 457                 znode->len = len;
 458
 459                 /* Update the parent */
 460                 zp = znode->parent;
 461                 if (zp) {
 462                         struct ubifs_zbranch *zbr;
 463                         int i;
 464
 465                         i = znode->iip;
 466                         zbr = &zp->zbranch[i];
 467                         zbr->lnum = lnum;
 468                         zbr->offs = offs;
 469                         zbr->len = len;
 470                 } else {
 471                         c->zroot.lnum = lnum;
 472                         c->zroot.offs = offs;
 473                         c->zroot.len = len;
 474                 }
 475                 c->calc_idx_sz += ALIGN(len, 8);
 476
 477                 /*
 478                  * Once lprops is updated, we can decrease the dirty znode count
 479                  * but it is easier to just do it here.
 480                  */
 481                 atomic_long_dec(&c->dirty_zn_cnt);
 482
 483                 /*
 484                  * Calculate the next index node length to see if there is
 485                  * enough room for it
 486                  */
 487                 cnext = znode->cnext;
 488                 if (cnext == c->cnext)
 489                         next_len = 0;
 490                 else
 491                         next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
 492
 493                 /* Update buffer positions */
 494                 wlen = used + len;
 495                 used += ALIGN(len, 8);
 496                 avail -= ALIGN(len, 8);
 497
 498                 if (next_len != 0 &&
 499                     buf_offs + used + next_len <= c->leb_size &&
 500                     avail > 0)
 501                         continue;
 502
 503                 if (avail <= 0 && next_len &&
 504                     buf_offs + used + next_len <= c->leb_size)
 505                         blen = buf_len;
 506                 else
 507                         blen = ALIGN(wlen, c->min_io_size);
 508
 509                 /* The buffer is full or there are no more znodes to do */
 510                 buf_offs += blen;
 511                 if (next_len) {
 512                         if (buf_offs + next_len > c->leb_size) {
 513                                 err = ubifs_update_one_lp(c, lnum,
 514                                         c->leb_size - buf_offs, blen - used,
 515                                         0, 0);
 516                                 if (err)
 517                                         return err;
 518                                 lnum = -1;
 519                         }
 520                         used -= blen;
 521                         if (used < 0)
 522                                 used = 0;
 523                         avail = buf_len - used;
 524                         continue;
 525                 }
 526                 err = ubifs_update_one_lp(c, lnum, c->leb_size - buf_offs,
 527                                           blen - used, 0, 0);
 528                 if (err)
 529                         return err;
 530                 break;
 531         }
 532
 533         c->dbg->new_ihead_lnum = lnum;
 534         c->dbg->new_ihead_offs = buf_offs;
 535
 536         return 0;
 537 }
 538
 539 /**
 540  * layout_commit - determine positions of index nodes to commit.
 541  * @c: UBIFS file-system description object
 542  * @no_space: indicates that insufficient empty LEBs were allocated
 543  * @cnt: number of znodes to commit
 544  *
 545  * Calculate and update the positions of index nodes to commit.  If there were
 546  * an insufficient number of empty LEBs allocated, then index nodes are placed
 547  * into the gaps created by obsolete index nodes in non-empty index LEBs.  For
 548  * this purpose, an obsolete index node is one that was not in the index as at
 549  * the end of the last commit.  To write "in-the-gaps" requires that those index
 550  * LEBs are updated atomically in-place.
 551  */
 552 static int layout_commit(struct ubifs_info *c, int no_space, int cnt)
 553 {
 554         int err;
 555
 556         if (no_space) {
 557                 err = layout_in_gaps(c, cnt);
 558                 if (err)
 559                         return err;
 560         }
 561         err = layout_in_empty_space(c);
 562         return err;
 563 }
 564
 565 /**
 566  * find_first_dirty - find first dirty znode.
 567  * @znode: znode to begin searching from
 568  */
 569 static struct ubifs_znode *find_first_dirty(struct ubifs_znode *znode)
 570 {
 571         int i, cont;
 572
 573         if (!znode)
 574                 return NULL;
 575
 576         while (1) {
 577                 if (znode->level == 0) {
 578                         if (ubifs_zn_dirty(znode))
 579                                 return znode;
 580                         return NULL;
 581                 }
 582                 cont = 0;
 583                 for (i = 0; i < znode->child_cnt; i++) {
 584                         struct ubifs_zbranch *zbr = &znode->zbranch[i];
 585
 586                         if (zbr->znode && ubifs_zn_dirty(zbr->znode)) {
 587                                 znode = zbr->znode;
 588                                 cont = 1;
 589                                 break;
 590                         }
 591                 }
 592                 if (!cont) {
 593                         if (ubifs_zn_dirty(znode))
 594                                 return znode;
 595                         return NULL;
 596                 }
 597         }
 598 }
 599
 600 /**
 601  * find_next_dirty - find next dirty znode.
 602  * @znode: znode to begin searching from
 603  */
 604 static struct ubifs_znode *find_next_dirty(struct ubifs_znode *znode)
 605 {
 606         int n = znode->iip + 1;
 607
 608         znode = znode->parent;
 609         if (!znode)
 610                 return NULL;
 611         for (; n < znode->child_cnt; n++) {
 612                 struct ubifs_zbranch *zbr = &znode->zbranch[n];
 613
 614                 if (zbr->znode && ubifs_zn_dirty(zbr->znode))
 615                         return find_first_dirty(zbr->znode);
 616         }
 617         return znode;
 618 }
 619
 620 /**
 621  * get_znodes_to_commit - create list of dirty znodes to commit.
 622  * @c: UBIFS file-system description object
 623  *
 624  * This function returns the number of znodes to commit.
 625  */
 626 static int get_znodes_to_commit(struct ubifs_info *c)
 627 {
 628         struct ubifs_znode *znode, *cnext;
 629         int cnt = 0;
 630
 631         c->cnext = find_first_dirty(c->zroot.znode);
 632         znode = c->enext = c->cnext;
 633         if (!znode) {
 634                 dbg_cmt("no znodes to commit");
 635                 return 0;
 636         }
 637         cnt += 1;
 638         while (1) {
 639                 ubifs_assert(!ubifs_zn_cow(znode));
 640                 __set_bit(COW_ZNODE, &znode->flags);
 641                 znode->alt = 0;
 642                 cnext = find_next_dirty(znode);
 643                 if (!cnext) {
 644                         znode->cnext = c->cnext;
 645                         break;
 646                 }
 647                 znode->cnext = cnext;
 648                 znode = cnext;
 649                 cnt += 1;
 650         }
 651         dbg_cmt("committing %d znodes", cnt);
 652         ubifs_assert(cnt == atomic_long_read(&c->dirty_zn_cnt));
 653         return cnt;
 654 }
 655
 656 /**
 657  * alloc_idx_lebs - allocate empty LEBs to be used to commit.
 658  * @c: UBIFS file-system description object
 659  * @cnt: number of znodes to commit
 660  *
 661  * This function returns %-ENOSPC if it cannot allocate a sufficient number of
 662  * empty LEBs.  %0 is returned on success, otherwise a negative error code
 663  * is returned.
 664  */
 665 static int alloc_idx_lebs(struct ubifs_info *c, int cnt)
 666 {
 667         int i, leb_cnt, lnum;
 668
 669         c->ileb_cnt = 0;
 670         c->ileb_nxt = 0;
 671         leb_cnt = get_leb_cnt(c, cnt);
 672         dbg_cmt("need about %d empty LEBS for TNC commit", leb_cnt);
 673         if (!leb_cnt)
 674                 return 0;
 675         c->ilebs = kmalloc(leb_cnt * sizeof(int), GFP_NOFS);
 676         if (!c->ilebs)
 677                 return -ENOMEM;
 678         for (i = 0; i < leb_cnt; i++) {
 679                 lnum = ubifs_find_free_leb_for_idx(c);
 680                 if (lnum < 0)
 681                         return lnum;
 682                 c->ilebs[c->ileb_cnt++] = lnum;
 683                 dbg_cmt("LEB %d", lnum);
 684         }
 685         if (dbg_is_chk_index(c) && !(prandom_u32() & 7))
 686                 return -ENOSPC;
 687         return 0;
 688 }
 689
 690 /**
 691  * free_unused_idx_lebs - free unused LEBs that were allocated for the commit.
 692  * @c: UBIFS file-system description object
 693  *
 694  * It is possible that we allocate more empty LEBs for the commit than we need.
 695  * This functions frees the surplus.
 696  *
 697  * This function returns %0 on success and a negative error code on failure.
 698  */
 699 static int free_unused_idx_lebs(struct ubifs_info *c)
 700 {
 701         int i, err = 0, lnum, er;
 702
 703         for (i = c->ileb_nxt; i < c->ileb_cnt; i++) {
 704                 lnum = c->ilebs[i];
 705                 dbg_cmt("LEB %d", lnum);
 706                 er = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
 707                                          LPROPS_INDEX | LPROPS_TAKEN, 0);
 708                 if (!err)
 709                         err = er;
 710         }
 711         return err;
 712 }
 713
 714 /**
 715  * free_idx_lebs - free unused LEBs after commit end.
 716  * @c: UBIFS file-system description object
 717  *
 718  * This function returns %0 on success and a negative error code on failure.
 719  */
 720 static int free_idx_lebs(struct ubifs_info *c)
 721 {
 722         int err;
 723
 724         err = free_unused_idx_lebs(c);
 725         kfree(c->ilebs);
 726         c->ilebs = NULL;
 727         return err;
 728 }
 729
 730 /**
 731  * ubifs_tnc_start_commit - start TNC commit.
 732  * @c: UBIFS file-system description object
 733  * @zroot: new index root position is returned here
 734  *
 735  * This function prepares the list of indexing nodes to commit and lays out
 736  * their positions on flash. If there is not enough free space it uses the
 737  * in-gap commit method. Returns zero in case of success and a negative error
 738  * code in case of failure.
 739  */
 740 int ubifs_tnc_start_commit(struct ubifs_info *c, struct ubifs_zbranch *zroot)
 741 {
 742         int err = 0, cnt;
 743
 744         mutex_lock(&c->tnc_mutex);
 745         err = dbg_check_tnc(c, 1);
 746         if (err)
 747                 goto out;
 748         cnt = get_znodes_to_commit(c);
 749         if (cnt != 0) {
 750                 int no_space = 0;
 751
 752                 err = alloc_idx_lebs(c, cnt);
 753                 if (err == -ENOSPC)
 754                         no_space = 1;
 755                 else if (err)
 756                         goto out_free;
 757                 err = layout_commit(c, no_space, cnt);
 758                 if (err)
 759                         goto out_free;
 760                 ubifs_assert(atomic_long_read(&c->dirty_zn_cnt) == 0);
 761                 err = free_unused_idx_lebs(c);
 762                 if (err)
 763                         goto out;
 764         }
 765         destroy_old_idx(c);
 766         memcpy(zroot, &c->zroot, sizeof(struct ubifs_zbranch));
 767
 768         err = ubifs_save_dirty_idx_lnums(c);
 769         if (err)
 770                 goto out;
 771
 772         spin_lock(&c->space_lock);
 773         /*
 774          * Although we have not finished committing yet, update size of the
 775          * committed index ('c->bi.old_idx_sz') and zero out the index growth
 776          * budget. It is OK to do this now, because we've reserved all the
 777          * space which is needed to commit the index, and it is save for the
 778          * budgeting subsystem to assume the index is already committed,
 779          * even though it is not.
 780          */
 781         ubifs_assert(c->bi.min_idx_lebs == ubifs_calc_min_idx_lebs(c));
 782         c->bi.old_idx_sz = c->calc_idx_sz;
 783         c->bi.uncommitted_idx = 0;
 784         c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c);
 785         spin_unlock(&c->space_lock);
 786         mutex_unlock(&c->tnc_mutex);
 787
 788         dbg_cmt("number of index LEBs %d", c->lst.idx_lebs);
 789         dbg_cmt("size of index %llu", c->calc_idx_sz);
 790         return err;
 791
 792 out_free:
 793         free_idx_lebs(c);
 794 out:
 795         mutex_unlock(&c->tnc_mutex);
 796         return err;
 797 }
 798
 799 /**
 800  * write_index - write index nodes.
 801  * @c: UBIFS file-system description object
 802  *
 803  * This function writes the index nodes whose positions were laid out in the
 804  * layout_in_empty_space function.
 805  */
 806 static int write_index(struct ubifs_info *c)
 807 {
 808         struct ubifs_idx_node *idx;
 809         struct ubifs_znode *znode, *cnext;
 810         int i, lnum, offs, len, next_len, buf_len, buf_offs, used;
 811         int avail, wlen, err, lnum_pos = 0, blen, nxt_offs;
 812
 813         cnext = c->enext;
 814         if (!cnext)
 815                 return 0;
 816
 817         /*
 818          * Always write index nodes to the index head so that index nodes and
 819          * other types of nodes are never mixed in the same erase block.
 820          */
 821         lnum = c->ihead_lnum;
 822         buf_offs = c->ihead_offs;
 823
 824         /* Allocate commit buffer */
 825         buf_len = ALIGN(c->max_idx_node_sz, c->min_io_size);
 826         used = 0;
 827         avail = buf_len;
 828
 829         /* Ensure there is enough room for first write */
 830         next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
 831         if (buf_offs + next_len > c->leb_size) {
 832                 err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0, 0,
 833                                           LPROPS_TAKEN);
 834                 if (err)
 835                         return err;
 836                 lnum = -1;
 837         }
 838
 839         while (1) {
 840                 cond_resched();
 841
 842                 znode = cnext;
 843                 idx = c->cbuf + used;
 844
 845                 /* Make index node */
 846                 idx->ch.node_type = UBIFS_IDX_NODE;
 847                 idx->child_cnt = cpu_to_le16(znode->child_cnt);
 848                 idx->level = cpu_to_le16(znode->level);
 849                 for (i = 0; i < znode->child_cnt; i++) {
 850                         struct ubifs_branch *br = ubifs_idx_branch(c, idx, i);
 851                         struct ubifs_zbranch *zbr = &znode->zbranch[i];
 852
 853                         key_write_idx(c, &zbr->key, &br->key);
 854                         br->lnum = cpu_to_le32(zbr->lnum);
 855                         br->offs = cpu_to_le32(zbr->offs);
 856                         br->len = cpu_to_le32(zbr->len);
 857                         if (!zbr->lnum || !zbr->len) {
 858                                 ubifs_err(c, "bad ref in znode");
 859                                 ubifs_dump_znode(c, znode);
 860                                 if (zbr->znode)
 861                                         ubifs_dump_znode(c, zbr->znode);
 862                         }
 863                 }
 864                 len = ubifs_idx_node_sz(c, znode->child_cnt);
 865                 ubifs_prepare_node(c, idx, len, 0);
 866
 867                 /* Determine the index node position */
 868                 if (lnum == -1) {
 869                         lnum = c->ilebs[lnum_pos++];
 870                         buf_offs = 0;
 871                         used = 0;
 872                         avail = buf_len;
 873                 }
 874                 offs = buf_offs + used;
 875
 876                 if (lnum != znode->lnum || offs != znode->offs ||
 877                     len != znode->len) {
 878                         ubifs_err(c, "inconsistent znode posn");
 879                         return -EINVAL;
 880                 }
 881
 882                 /* Grab some stuff from znode while we still can */
 883                 cnext = znode->cnext;
 884
 885                 ubifs_assert(ubifs_zn_dirty(znode));
 886                 ubifs_assert(ubifs_zn_cow(znode));
 887
 888                 /*
 889                  * It is important that other threads should see %DIRTY_ZNODE
 890                  * flag cleared before %COW_ZNODE. Specifically, it matters in
 891                  * the 'dirty_cow_znode()' function. This is the reason for the
 892                  * first barrier. Also, we want the bit changes to be seen to
 893                  * other threads ASAP, to avoid unnecesarry copying, which is
 894                  * the reason for the second barrier.
 895                  */
 896                 clear_bit(DIRTY_ZNODE, &znode->flags);
 897                 smp_mb__before_atomic();
 898                 clear_bit(COW_ZNODE, &znode->flags);
 899                 smp_mb__after_atomic();
 900
 901                 /*
 902                  * We have marked the znode as clean but have not updated the
 903                  * @c->clean_zn_cnt counter. If this znode becomes dirty again
 904                  * before 'free_obsolete_znodes()' is called, then
 905                  * @c->clean_zn_cnt will be decremented before it gets
 906                  * incremented (resulting in 2 decrements for the same znode).
 907                  * This means that @c->clean_zn_cnt may become negative for a
 908                  * while.
 909                  *
 910                  * Q: why we cannot increment @c->clean_zn_cnt?
 911                  * A: because we do not have the @c->tnc_mutex locked, and the
 912                  *    following code would be racy and buggy:
 913                  *
 914                  *    if (!ubifs_zn_obsolete(znode)) {
 915                  *            atomic_long_inc(&c->clean_zn_cnt);
 916                  *            atomic_long_inc(&ubifs_clean_zn_cnt);
 917                  *    }
 918                  *
 919                  *    Thus, we just delay the @c->clean_zn_cnt update until we
 920                  *    have the mutex locked.
 921                  */
 922
 923                 /* Do not access znode from this point on */
 924
 925                 /* Update buffer positions */
 926                 wlen = used + len;
 927                 used += ALIGN(len, 8);
 928                 avail -= ALIGN(len, 8);
 929
 930                 /*
 931                  * Calculate the next index node length to see if there is
 932                  * enough room for it
 933                  */
 934                 if (cnext == c->cnext)
 935                         next_len = 0;
 936                 else
 937                         next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
 938
 939                 nxt_offs = buf_offs + used + next_len;
 940                 if (next_len && nxt_offs <= c->leb_size) {
 941                         if (avail > 0)
 942                                 continue;
 943                         else
 944                                 blen = buf_len;
 945                 } else {
 946                         wlen = ALIGN(wlen, 8);
 947                         blen = ALIGN(wlen, c->min_io_size);
 948                         ubifs_pad(c, c->cbuf + wlen, blen - wlen);
 949                 }
 950
 951                 /* The buffer is full or there are no more znodes to do */
 952                 err = ubifs_leb_write(c, lnum, c->cbuf, buf_offs, blen);
 953                 if (err)
 954                         return err;
 955                 buf_offs += blen;
 956                 if (next_len) {
 957                         if (nxt_offs > c->leb_size) {
 958                                 err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0,
 959                                                           0, LPROPS_TAKEN);
 960                                 if (err)
 961                                         return err;
 962                                 lnum = -1;
 963                         }
 964                         used -= blen;
 965                         if (used < 0)
 966                                 used = 0;
 967                         avail = buf_len - used;
 968                         memmove(c->cbuf, c->cbuf + blen, used);
 969                         continue;
 970                 }
 971                 break;
 972         }
 973
 974         if (lnum != c->dbg->new_ihead_lnum ||
 975             buf_offs != c->dbg->new_ihead_offs) {
 976                 ubifs_err(c, "inconsistent ihead");
 977                 return -EINVAL;
 978         }
 979
 980         c->ihead_lnum = lnum;
 981         c->ihead_offs = buf_offs;
 982
 983         return 0;
 984 }
 985
 986 /**
 987  * free_obsolete_znodes - free obsolete znodes.
 988  * @c: UBIFS file-system description object
 989  *
 990  * At the end of commit end, obsolete znodes are freed.
 991  */
 992 static void free_obsolete_znodes(struct ubifs_info *c)
 993 {
 994         struct ubifs_znode *znode, *cnext;
 995
 996         cnext = c->cnext;
 997         do {
 998                 znode = cnext;
 999                 cnext = znode->cnext;
1000                 if (ubifs_zn_obsolete(znode))
1001                         kfree(znode);
1002                 else {
1003                         znode->cnext = NULL;
1004                         atomic_long_inc(&c->clean_zn_cnt);
1005                         atomic_long_inc(&ubifs_clean_zn_cnt);
1006                 }
1007         } while (cnext != c->cnext);
1008 }
1009
1010 /**
1011  * return_gap_lebs - return LEBs used by the in-gap commit method.
1012  * @c: UBIFS file-system description object
1013  *
1014  * This function clears the "taken" flag for the LEBs which were used by the
1015  * "commit in-the-gaps" method.
1016  */
1017 static int return_gap_lebs(struct ubifs_info *c)
1018 {
1019         int *p, err;
1020
1021         if (!c->gap_lebs)
1022                 return 0;
1023
1024         dbg_cmt("");
1025         for (p = c->gap_lebs; *p != -1; p++) {
1026                 err = ubifs_change_one_lp(c, *p, LPROPS_NC, LPROPS_NC, 0,
1027                                           LPROPS_TAKEN, 0);
1028                 if (err)
1029                         return err;
1030         }
1031
1032         kfree(c->gap_lebs);
1033         c->gap_lebs = NULL;
1034         return 0;
1035 }
1036
1037 /**
1038  * ubifs_tnc_end_commit - update the TNC for commit end.
1039  * @c: UBIFS file-system description object
1040  *
1041  * Write the dirty znodes.
1042  */
1043 int ubifs_tnc_end_commit(struct ubifs_info *c)
1044 {
1045         int err;
1046
1047         if (!c->cnext)
1048                 return 0;
1049
1050         err = return_gap_lebs(c);
1051         if (err)
1052                 return err;
1053
1054         err = write_index(c);
1055         if (err)
1056                 return err;
1057
1058         mutex_lock(&c->tnc_mutex);
1059
1060         dbg_cmt("TNC height is %d", c->zroot.znode->level + 1);
1061
1062         free_obsolete_znodes(c);
1063
1064         c->cnext = NULL;
1065         kfree(c->ilebs);
1066         c->ilebs = NULL;
1067
1068         mutex_unlock(&c->tnc_mutex);
1069
1070         return 0;
1071 }