fs/ext3/balloc.c

   1 /*
   2  *  linux/fs/ext3/balloc.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
  10  *  Big-endian to little-endian byte-swapping/bitmaps by
  11  *        David S. Miller (davem@caip.rutgers.edu), 1995
  12  */
  13
  14 #include <linux/time.h>
  15 #include <linux/capability.h>
  16 #include <linux/fs.h>
  17 #include <linux/slab.h>
  18 #include <linux/jbd.h>
  19 #include <linux/ext3_fs.h>
  20 #include <linux/ext3_jbd.h>
  21 #include <linux/quotaops.h>
  22 #include <linux/buffer_head.h>
  23 #include <linux/blkdev.h>
  24
  25 /*
  26  * balloc.c contains the blocks allocation and deallocation routines
  27  */
  28
  29 /*
  30  * The free blocks are managed by bitmaps.  A file system contains several
  31  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  32  * block for inodes, N blocks for the inode table and data blocks.
  33  *
  34  * The file system contains group descriptors which are located after the
  35  * super block.  Each descriptor contains the number of the bitmap block and
  36  * the free blocks count in the block.  The descriptors are loaded in memory
  37  * when a file system is mounted (see ext3_fill_super).
  38  */
  39
  40
  41 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
  42
  43 /*
  44  * Calculate the block group number and offset, given a block number
  45  */
  46 static void ext3_get_group_no_and_offset(struct super_block *sb,
  47         ext3_fsblk_t blocknr, unsigned long *blockgrpp, ext3_grpblk_t *offsetp)
  48 {
  49         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
  50
  51         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
  52         if (offsetp)
  53                 *offsetp = blocknr % EXT3_BLOCKS_PER_GROUP(sb);
  54         if (blockgrpp)
  55                 *blockgrpp = blocknr / EXT3_BLOCKS_PER_GROUP(sb);
  56 }
  57
  58 /**
  59  * ext3_get_group_desc() -- load group descriptor from disk
  60  * @sb:                 super block
  61  * @block_group:        given block group
  62  * @bh:                 pointer to the buffer head to store the block
  63  *                      group descriptor
  64  */
  65 struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
  66                                              unsigned int block_group,
  67                                              struct buffer_head ** bh)
  68 {
  69         unsigned long group_desc;
  70         unsigned long offset;
  71         struct ext3_group_desc * desc;
  72         struct ext3_sb_info *sbi = EXT3_SB(sb);
  73
  74         if (block_group >= sbi->s_groups_count) {
  75                 ext3_error (sb, "ext3_get_group_desc",
  76                             "block_group >= groups_count - "
  77                             "block_group = %d, groups_count = %lu",
  78                             block_group, sbi->s_groups_count);
  79
  80                 return NULL;
  81         }
  82         smp_rmb();
  83
  84         group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
  85         offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
  86         if (!sbi->s_group_desc[group_desc]) {
  87                 ext3_error (sb, "ext3_get_group_desc",
  88                             "Group descriptor not loaded - "
  89                             "block_group = %d, group_desc = %lu, desc = %lu",
  90                              block_group, group_desc, offset);
  91                 return NULL;
  92         }
  93
  94         desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
  95         if (bh)
  96                 *bh = sbi->s_group_desc[group_desc];
  97         return desc + offset;
  98 }
  99
 100 static int ext3_valid_block_bitmap(struct super_block *sb,
 101                                         struct ext3_group_desc *desc,
 102                                         unsigned int block_group,
 103                                         struct buffer_head *bh)
 104 {
 105         ext3_grpblk_t offset;
 106         ext3_grpblk_t next_zero_bit;
 107         ext3_fsblk_t bitmap_blk;
 108         ext3_fsblk_t group_first_block;
 109
 110         group_first_block = ext3_group_first_block_no(sb, block_group);
 111
 112         /* check whether block bitmap block number is set */
 113         bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
 114         offset = bitmap_blk - group_first_block;
 115         if (!ext3_test_bit(offset, bh->b_data))
 116                 /* bad block bitmap */
 117                 goto err_out;
 118
 119         /* check whether the inode bitmap block number is set */
 120         bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
 121         offset = bitmap_blk - group_first_block;
 122         if (!ext3_test_bit(offset, bh->b_data))
 123                 /* bad block bitmap */
 124                 goto err_out;
 125
 126         /* check whether the inode table block number is set */
 127         bitmap_blk = le32_to_cpu(desc->bg_inode_table);
 128         offset = bitmap_blk - group_first_block;
 129         next_zero_bit = ext3_find_next_zero_bit(bh->b_data,
 130                                 offset + EXT3_SB(sb)->s_itb_per_group,
 131                                 offset);
 132         if (next_zero_bit >= offset + EXT3_SB(sb)->s_itb_per_group)
 133                 /* good bitmap for inode tables */
 134                 return 1;
 135
 136 err_out:
 137         ext3_error(sb, __func__,
 138                         "Invalid block bitmap - "
 139                         "block_group = %d, block = %lu",
 140                         block_group, bitmap_blk);
 141         return 0;
 142 }
 143
 144 /**
 145  * read_block_bitmap()
 146  * @sb:                 super block
 147  * @block_group:        given block group
 148  *
 149  * Read the bitmap for a given block_group,and validate the
 150  * bits for block/inode/inode tables are set in the bitmaps
 151  *
 152  * Return buffer_head on success or NULL in case of failure.
 153  */
 154 static struct buffer_head *
 155 read_block_bitmap(struct super_block *sb, unsigned int block_group)
 156 {
 157         struct ext3_group_desc * desc;
 158         struct buffer_head * bh = NULL;
 159         ext3_fsblk_t bitmap_blk;
 160
 161         desc = ext3_get_group_desc(sb, block_group, NULL);
 162         if (!desc)
 163                 return NULL;
 164         bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
 165         bh = sb_getblk(sb, bitmap_blk);
 166         if (unlikely(!bh)) {
 167                 ext3_error(sb, __func__,
 168                             "Cannot read block bitmap - "
 169                             "block_group = %d, block_bitmap = %u",
 170                             block_group, le32_to_cpu(desc->bg_block_bitmap));
 171                 return NULL;
 172         }
 173         if (likely(bh_uptodate_or_lock(bh)))
 174                 return bh;
 175
 176         if (bh_submit_read(bh) < 0) {
 177                 brelse(bh);
 178                 ext3_error(sb, __func__,
 179                             "Cannot read block bitmap - "
 180                             "block_group = %d, block_bitmap = %u",
 181                             block_group, le32_to_cpu(desc->bg_block_bitmap));
 182                 return NULL;
 183         }
 184         ext3_valid_block_bitmap(sb, desc, block_group, bh);
 185         /*
 186          * file system mounted not to panic on error, continue with corrupt
 187          * bitmap
 188          */
 189         return bh;
 190 }
 191 /*
 192  * The reservation window structure operations
 193  * --------------------------------------------
 194  * Operations include:
 195  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
 196  *
 197  * We use a red-black tree to represent per-filesystem reservation
 198  * windows.
 199  *
 200  */
 201
 202 /**
 203  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
 204  * @rb_root:            root of per-filesystem reservation rb tree
 205  * @verbose:            verbose mode
 206  * @fn:                 function which wishes to dump the reservation map
 207  *
 208  * If verbose is turned on, it will print the whole block reservation
 209  * windows(start, end). Otherwise, it will only print out the "bad" windows,
 210  * those windows that overlap with their immediate neighbors.
 211  */
 212 #if 1
 213 static void __rsv_window_dump(struct rb_root *root, int verbose,
 214                               const char *fn)
 215 {
 216         struct rb_node *n;
 217         struct ext3_reserve_window_node *rsv, *prev;
 218         int bad;
 219
 220 restart:
 221         n = rb_first(root);
 222         bad = 0;
 223         prev = NULL;
 224
 225         printk("Block Allocation Reservation Windows Map (%s):\n", fn);
 226         while (n) {
 227                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
 228                 if (verbose)
 229                         printk("reservation window 0x%p "
 230                                "start:  %lu, end:  %lu\n",
 231                                rsv, rsv->rsv_start, rsv->rsv_end);
 232                 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
 233                         printk("Bad reservation %p (start >= end)\n",
 234                                rsv);
 235                         bad = 1;
 236                 }
 237                 if (prev && prev->rsv_end >= rsv->rsv_start) {
 238                         printk("Bad reservation %p (prev->end >= start)\n",
 239                                rsv);
 240                         bad = 1;
 241                 }
 242                 if (bad) {
 243                         if (!verbose) {
 244                                 printk("Restarting reservation walk in verbose mode\n");
 245                                 verbose = 1;
 246                                 goto restart;
 247                         }
 248                 }
 249                 n = rb_next(n);
 250                 prev = rsv;
 251         }
 252         printk("Window map complete.\n");
 253         BUG_ON(bad);
 254 }
 255 #define rsv_window_dump(root, verbose) \
 256         __rsv_window_dump((root), (verbose), __func__)
 257 #else
 258 #define rsv_window_dump(root, verbose) do {} while (0)
 259 #endif
 260
 261 /**
 262  * goal_in_my_reservation()
 263  * @rsv:                inode's reservation window
 264  * @grp_goal:           given goal block relative to the allocation block group
 265  * @group:              the current allocation block group
 266  * @sb:                 filesystem super block
 267  *
 268  * Test if the given goal block (group relative) is within the file's
 269  * own block reservation window range.
 270  *
 271  * If the reservation window is outside the goal allocation group, return 0;
 272  * grp_goal (given goal block) could be -1, which means no specific
 273  * goal block. In this case, always return 1.
 274  * If the goal block is within the reservation window, return 1;
 275  * otherwise, return 0;
 276  */
 277 static int
 278 goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
 279                         unsigned int group, struct super_block * sb)
 280 {
 281         ext3_fsblk_t group_first_block, group_last_block;
 282
 283         group_first_block = ext3_group_first_block_no(sb, group);
 284         group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
 285
 286         if ((rsv->_rsv_start > group_last_block) ||
 287             (rsv->_rsv_end < group_first_block))
 288                 return 0;
 289         if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
 290                 || (grp_goal + group_first_block > rsv->_rsv_end)))
 291                 return 0;
 292         return 1;
 293 }
 294
 295 /**
 296  * search_reserve_window()
 297  * @rb_root:            root of reservation tree
 298  * @goal:               target allocation block
 299  *
 300  * Find the reserved window which includes the goal, or the previous one
 301  * if the goal is not in any window.
 302  * Returns NULL if there are no windows or if all windows start after the goal.
 303  */
 304 static struct ext3_reserve_window_node *
 305 search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
 306 {
 307         struct rb_node *n = root->rb_node;
 308         struct ext3_reserve_window_node *rsv;
 309
 310         if (!n)
 311                 return NULL;
 312
 313         do {
 314                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
 315
 316                 if (goal < rsv->rsv_start)
 317                         n = n->rb_left;
 318                 else if (goal > rsv->rsv_end)
 319                         n = n->rb_right;
 320                 else
 321                         return rsv;
 322         } while (n);
 323         /*
 324          * We've fallen off the end of the tree: the goal wasn't inside
 325          * any particular node.  OK, the previous node must be to one
 326          * side of the interval containing the goal.  If it's the RHS,
 327          * we need to back up one.
 328          */
 329         if (rsv->rsv_start > goal) {
 330                 n = rb_prev(&rsv->rsv_node);
 331                 rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
 332         }
 333         return rsv;
 334 }
 335
 336 /**
 337  * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
 338  * @sb:                 super block
 339  * @rsv:                reservation window to add
 340  *
 341  * Must be called with rsv_lock hold.
 342  */
 343 void ext3_rsv_window_add(struct super_block *sb,
 344                     struct ext3_reserve_window_node *rsv)
 345 {
 346         struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
 347         struct rb_node *node = &rsv->rsv_node;
 348         ext3_fsblk_t start = rsv->rsv_start;
 349
 350         struct rb_node ** p = &root->rb_node;
 351         struct rb_node * parent = NULL;
 352         struct ext3_reserve_window_node *this;
 353
 354         while (*p)
 355         {
 356                 parent = *p;
 357                 this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
 358
 359                 if (start < this->rsv_start)
 360                         p = &(*p)->rb_left;
 361                 else if (start > this->rsv_end)
 362                         p = &(*p)->rb_right;
 363                 else {
 364                         rsv_window_dump(root, 1);
 365                         BUG();
 366                 }
 367         }
 368
 369         rb_link_node(node, parent, p);
 370         rb_insert_color(node, root);
 371 }
 372
 373 /**
 374  * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
 375  * @sb:                 super block
 376  * @rsv:                reservation window to remove
 377  *
 378  * Mark the block reservation window as not allocated, and unlink it
 379  * from the filesystem reservation window rb tree. Must be called with
 380  * rsv_lock hold.
 381  */
 382 static void rsv_window_remove(struct super_block *sb,
 383                               struct ext3_reserve_window_node *rsv)
 384 {
 385         rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 386         rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 387         rsv->rsv_alloc_hit = 0;
 388         rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
 389 }
 390
 391 /*
 392  * rsv_is_empty() -- Check if the reservation window is allocated.
 393  * @rsv:                given reservation window to check
 394  *
 395  * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
 396  */
 397 static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
 398 {
 399         /* a valid reservation end block could not be 0 */
 400         return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 401 }
 402
 403 /**
 404  * ext3_init_block_alloc_info()
 405  * @inode:              file inode structure
 406  *
 407  * Allocate and initialize the  reservation window structure, and
 408  * link the window to the ext3 inode structure at last
 409  *
 410  * The reservation window structure is only dynamically allocated
 411  * and linked to ext3 inode the first time the open file
 412  * needs a new block. So, before every ext3_new_block(s) call, for
 413  * regular files, we should check whether the reservation window
 414  * structure exists or not. In the latter case, this function is called.
 415  * Fail to do so will result in block reservation being turned off for that
 416  * open file.
 417  *
 418  * This function is called from ext3_get_blocks_handle(), also called
 419  * when setting the reservation window size through ioctl before the file
 420  * is open for write (needs block allocation).
 421  *
 422  * Needs truncate_mutex protection prior to call this function.
 423  */
 424 void ext3_init_block_alloc_info(struct inode *inode)
 425 {
 426         struct ext3_inode_info *ei = EXT3_I(inode);
 427         struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
 428         struct super_block *sb = inode->i_sb;
 429
 430         block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
 431         if (block_i) {
 432                 struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
 433
 434                 rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 435                 rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 436
 437                 /*
 438                  * if filesystem is mounted with NORESERVATION, the goal
 439                  * reservation window size is set to zero to indicate
 440                  * block reservation is off
 441                  */
 442                 if (!test_opt(sb, RESERVATION))
 443                         rsv->rsv_goal_size = 0;
 444                 else
 445                         rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
 446                 rsv->rsv_alloc_hit = 0;
 447                 block_i->last_alloc_logical_block = 0;
 448                 block_i->last_alloc_physical_block = 0;
 449         }
 450         ei->i_block_alloc_info = block_i;
 451 }
 452
 453 /**
 454  * ext3_discard_reservation()
 455  * @inode:              inode
 456  *
 457  * Discard(free) block reservation window on last file close, or truncate
 458  * or at last iput().
 459  *
 460  * It is being called in three cases:
 461  *      ext3_release_file(): last writer close the file
 462  *      ext3_clear_inode(): last iput(), when nobody link to this file.
 463  *      ext3_truncate(): when the block indirect map is about to change.
 464  *
 465  */
 466 void ext3_discard_reservation(struct inode *inode)
 467 {
 468         struct ext3_inode_info *ei = EXT3_I(inode);
 469         struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
 470         struct ext3_reserve_window_node *rsv;
 471         spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
 472
 473         if (!block_i)
 474                 return;
 475
 476         rsv = &block_i->rsv_window_node;
 477         if (!rsv_is_empty(&rsv->rsv_window)) {
 478                 spin_lock(rsv_lock);
 479                 if (!rsv_is_empty(&rsv->rsv_window))
 480                         rsv_window_remove(inode->i_sb, rsv);
 481                 spin_unlock(rsv_lock);
 482         }
 483 }
 484
 485 /**
 486  * ext3_free_blocks_sb() -- Free given blocks and update quota
 487  * @handle:                     handle to this transaction
 488  * @sb:                         super block
 489  * @block:                      start physcial block to free
 490  * @count:                      number of blocks to free
 491  * @pdquot_freed_blocks:        pointer to quota
 492  */
 493 void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
 494                          ext3_fsblk_t block, unsigned long count,
 495                          unsigned long *pdquot_freed_blocks)
 496 {
 497         struct buffer_head *bitmap_bh = NULL;
 498         struct buffer_head *gd_bh;
 499         unsigned long block_group;
 500         ext3_grpblk_t bit;
 501         unsigned long i;
 502         unsigned long overflow;
 503         struct ext3_group_desc * desc;
 504         struct ext3_super_block * es;
 505         struct ext3_sb_info *sbi;
 506         int err = 0, ret;
 507         ext3_grpblk_t group_freed;
 508
 509         *pdquot_freed_blocks = 0;
 510         sbi = EXT3_SB(sb);
 511         es = sbi->s_es;
 512         if (block < le32_to_cpu(es->s_first_data_block) ||
 513             block + count < block ||
 514             block + count > le32_to_cpu(es->s_blocks_count)) {
 515                 ext3_error (sb, "ext3_free_blocks",
 516                             "Freeing blocks not in datazone - "
 517                             "block = "E3FSBLK", count = %lu", block, count);
 518                 goto error_return;
 519         }
 520
 521         ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
 522
 523 do_more:
 524         overflow = 0;
 525         block_group = (block - le32_to_cpu(es->s_first_data_block)) /
 526                       EXT3_BLOCKS_PER_GROUP(sb);
 527         bit = (block - le32_to_cpu(es->s_first_data_block)) %
 528                       EXT3_BLOCKS_PER_GROUP(sb);
 529         /*
 530          * Check to see if we are freeing blocks across a group
 531          * boundary.
 532          */
 533         if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
 534                 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
 535                 count -= overflow;
 536         }
 537         brelse(bitmap_bh);
 538         bitmap_bh = read_block_bitmap(sb, block_group);
 539         if (!bitmap_bh)
 540                 goto error_return;
 541         desc = ext3_get_group_desc (sb, block_group, &gd_bh);
 542         if (!desc)
 543                 goto error_return;
 544
 545         if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
 546             in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
 547             in_range (block, le32_to_cpu(desc->bg_inode_table),
 548                       sbi->s_itb_per_group) ||
 549             in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
 550                       sbi->s_itb_per_group)) {
 551                 ext3_error (sb, "ext3_free_blocks",
 552                             "Freeing blocks in system zones - "
 553                             "Block = "E3FSBLK", count = %lu",
 554                             block, count);
 555                 goto error_return;
 556         }
 557
 558         /*
 559          * We are about to start releasing blocks in the bitmap,
 560          * so we need undo access.
 561          */
 562         /* @@@ check errors */
 563         BUFFER_TRACE(bitmap_bh, "getting undo access");
 564         err = ext3_journal_get_undo_access(handle, bitmap_bh);
 565         if (err)
 566                 goto error_return;
 567
 568         /*
 569          * We are about to modify some metadata.  Call the journal APIs
 570          * to unshare ->b_data if a currently-committing transaction is
 571          * using it
 572          */
 573         BUFFER_TRACE(gd_bh, "get_write_access");
 574         err = ext3_journal_get_write_access(handle, gd_bh);
 575         if (err)
 576                 goto error_return;
 577
 578         jbd_lock_bh_state(bitmap_bh);
 579
 580         for (i = 0, group_freed = 0; i < count; i++) {
 581                 /*
 582                  * An HJ special.  This is expensive...
 583                  */
 584 #ifdef CONFIG_JBD_DEBUG
 585                 jbd_unlock_bh_state(bitmap_bh);
 586                 {
 587                         struct buffer_head *debug_bh;
 588                         debug_bh = sb_find_get_block(sb, block + i);
 589                         if (debug_bh) {
 590                                 BUFFER_TRACE(debug_bh, "Deleted!");
 591                                 if (!bh2jh(bitmap_bh)->b_committed_data)
 592                                         BUFFER_TRACE(debug_bh,
 593                                                 "No committed data in bitmap");
 594                                 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
 595                                 __brelse(debug_bh);
 596                         }
 597                 }
 598                 jbd_lock_bh_state(bitmap_bh);
 599 #endif
 600                 if (need_resched()) {
 601                         jbd_unlock_bh_state(bitmap_bh);
 602                         cond_resched();
 603                         jbd_lock_bh_state(bitmap_bh);
 604                 }
 605                 /* @@@ This prevents newly-allocated data from being
 606                  * freed and then reallocated within the same
 607                  * transaction.
 608                  *
 609                  * Ideally we would want to allow that to happen, but to
 610                  * do so requires making journal_forget() capable of
 611                  * revoking the queued write of a data block, which
 612                  * implies blocking on the journal lock.  *forget()
 613                  * cannot block due to truncate races.
 614                  *
 615                  * Eventually we can fix this by making journal_forget()
 616                  * return a status indicating whether or not it was able
 617                  * to revoke the buffer.  On successful revoke, it is
 618                  * safe not to set the allocation bit in the committed
 619                  * bitmap, because we know that there is no outstanding
 620                  * activity on the buffer any more and so it is safe to
 621                  * reallocate it.
 622                  */
 623                 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
 624                 J_ASSERT_BH(bitmap_bh,
 625                                 bh2jh(bitmap_bh)->b_committed_data != NULL);
 626                 ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
 627                                 bh2jh(bitmap_bh)->b_committed_data);
 628
 629                 /*
 630                  * We clear the bit in the bitmap after setting the committed
 631                  * data bit, because this is the reverse order to that which
 632                  * the allocator uses.
 633                  */
 634                 BUFFER_TRACE(bitmap_bh, "clear bit");
 635                 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
 636                                                 bit + i, bitmap_bh->b_data)) {
 637                         jbd_unlock_bh_state(bitmap_bh);
 638                         ext3_error(sb, __func__,
 639                                 "bit already cleared for block "E3FSBLK,
 640                                  block + i);
 641                         jbd_lock_bh_state(bitmap_bh);
 642                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
 643                 } else {
 644                         group_freed++;
 645                 }
 646         }
 647         jbd_unlock_bh_state(bitmap_bh);
 648
 649         spin_lock(sb_bgl_lock(sbi, block_group));
 650         le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
 651         spin_unlock(sb_bgl_lock(sbi, block_group));
 652         percpu_counter_add(&sbi->s_freeblocks_counter, count);
 653
 654         /* We dirtied the bitmap block */
 655         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 656         err = ext3_journal_dirty_metadata(handle, bitmap_bh);
 657
 658         /* And the group descriptor block */
 659         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 660         ret = ext3_journal_dirty_metadata(handle, gd_bh);
 661         if (!err) err = ret;
 662         *pdquot_freed_blocks += group_freed;
 663
 664         if (overflow && !err) {
 665                 block += count;
 666                 count = overflow;
 667                 goto do_more;
 668         }
 669
 670 error_return:
 671         brelse(bitmap_bh);
 672         ext3_std_error(sb, err);
 673         return;
 674 }
 675
 676 /**
 677  * ext3_free_blocks() -- Free given blocks and update quota
 678  * @handle:             handle for this transaction
 679  * @inode:              inode
 680  * @block:              start physical block to free
 681  * @count:              number of blocks to count
 682  */
 683 void ext3_free_blocks(handle_t *handle, struct inode *inode,
 684                         ext3_fsblk_t block, unsigned long count)
 685 {
 686         struct super_block * sb;
 687         unsigned long dquot_freed_blocks;
 688
 689         sb = inode->i_sb;
 690         if (!sb) {
 691                 printk ("ext3_free_blocks: nonexistent device");
 692                 return;
 693         }
 694         ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
 695         if (dquot_freed_blocks)
 696                 dquot_free_block(inode, dquot_freed_blocks);
 697         return;
 698 }
 699
 700 /**
 701  * ext3_test_allocatable()
 702  * @nr:                 given allocation block group
 703  * @bh:                 bufferhead contains the bitmap of the given block group
 704  *
 705  * For ext3 allocations, we must not reuse any blocks which are
 706  * allocated in the bitmap buffer's "last committed data" copy.  This
 707  * prevents deletes from freeing up the page for reuse until we have
 708  * committed the delete transaction.
 709  *
 710  * If we didn't do this, then deleting something and reallocating it as
 711  * data would allow the old block to be overwritten before the
 712  * transaction committed (because we force data to disk before commit).
 713  * This would lead to corruption if we crashed between overwriting the
 714  * data and committing the delete.
 715  *
 716  * @@@ We may want to make this allocation behaviour conditional on
 717  * data-writes at some point, and disable it for metadata allocations or
 718  * sync-data inodes.
 719  */
 720 static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
 721 {
 722         int ret;
 723         struct journal_head *jh = bh2jh(bh);
 724
 725         if (ext3_test_bit(nr, bh->b_data))
 726                 return 0;
 727
 728         jbd_lock_bh_state(bh);
 729         if (!jh->b_committed_data)
 730                 ret = 1;
 731         else
 732                 ret = !ext3_test_bit(nr, jh->b_committed_data);
 733         jbd_unlock_bh_state(bh);
 734         return ret;
 735 }
 736
 737 /**
 738  * bitmap_search_next_usable_block()
 739  * @start:              the starting block (group relative) of the search
 740  * @bh:                 bufferhead contains the block group bitmap
 741  * @maxblocks:          the ending block (group relative) of the reservation
 742  *
 743  * The bitmap search --- search forward alternately through the actual
 744  * bitmap on disk and the last-committed copy in journal, until we find a
 745  * bit free in both bitmaps.
 746  */
 747 static ext3_grpblk_t
 748 bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
 749                                         ext3_grpblk_t maxblocks)
 750 {
 751         ext3_grpblk_t next;
 752         struct journal_head *jh = bh2jh(bh);
 753
 754         while (start < maxblocks) {
 755                 next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
 756                 if (next >= maxblocks)
 757                         return -1;
 758                 if (ext3_test_allocatable(next, bh))
 759                         return next;
 760                 jbd_lock_bh_state(bh);
 761                 if (jh->b_committed_data)
 762                         start = ext3_find_next_zero_bit(jh->b_committed_data,
 763                                                         maxblocks, next);
 764                 jbd_unlock_bh_state(bh);
 765         }
 766         return -1;
 767 }
 768
 769 /**
 770  * find_next_usable_block()
 771  * @start:              the starting block (group relative) to find next
 772  *                      allocatable block in bitmap.
 773  * @bh:                 bufferhead contains the block group bitmap
 774  * @maxblocks:          the ending block (group relative) for the search
 775  *
 776  * Find an allocatable block in a bitmap.  We honor both the bitmap and
 777  * its last-committed copy (if that exists), and perform the "most
 778  * appropriate allocation" algorithm of looking for a free block near
 779  * the initial goal; then for a free byte somewhere in the bitmap; then
 780  * for any free bit in the bitmap.
 781  */
 782 static ext3_grpblk_t
 783 find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
 784                         ext3_grpblk_t maxblocks)
 785 {
 786         ext3_grpblk_t here, next;
 787         char *p, *r;
 788
 789         if (start > 0) {
 790                 /*
 791                  * The goal was occupied; search forward for a free
 792                  * block within the next XX blocks.
 793                  *
 794                  * end_goal is more or less random, but it has to be
 795                  * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
 796                  * next 64-bit boundary is simple..
 797                  */
 798                 ext3_grpblk_t end_goal = (start + 63) & ~63;
 799                 if (end_goal > maxblocks)
 800                         end_goal = maxblocks;
 801                 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
 802                 if (here < end_goal && ext3_test_allocatable(here, bh))
 803                         return here;
 804                 ext3_debug("Bit not found near goal\n");
 805         }
 806
 807         here = start;
 808         if (here < 0)
 809                 here = 0;
 810
 811         p = bh->b_data + (here >> 3);
 812         r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
 813         next = (r - bh->b_data) << 3;
 814
 815         if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
 816                 return next;
 817
 818         /*
 819          * The bitmap search --- search forward alternately through the actual
 820          * bitmap and the last-committed copy until we find a bit free in
 821          * both
 822          */
 823         here = bitmap_search_next_usable_block(here, bh, maxblocks);
 824         return here;
 825 }
 826
 827 /**
 828  * claim_block()
 829  * @lock:               the spin lock for this block group
 830  * @block:              the free block (group relative) to allocate
 831  * @bh:                 the buffer_head contains the block group bitmap
 832  *
 833  * We think we can allocate this block in this bitmap.  Try to set the bit.
 834  * If that succeeds then check that nobody has allocated and then freed the
 835  * block since we saw that is was not marked in b_committed_data.  If it _was_
 836  * allocated and freed then clear the bit in the bitmap again and return
 837  * zero (failure).
 838  */
 839 static inline int
 840 claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
 841 {
 842         struct journal_head *jh = bh2jh(bh);
 843         int ret;
 844
 845         if (ext3_set_bit_atomic(lock, block, bh->b_data))
 846                 return 0;
 847         jbd_lock_bh_state(bh);
 848         if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
 849                 ext3_clear_bit_atomic(lock, block, bh->b_data);
 850                 ret = 0;
 851         } else {
 852                 ret = 1;
 853         }
 854         jbd_unlock_bh_state(bh);
 855         return ret;
 856 }
 857
 858 /**
 859  * ext3_try_to_allocate()
 860  * @sb:                 superblock
 861  * @handle:             handle to this transaction
 862  * @group:              given allocation block group
 863  * @bitmap_bh:          bufferhead holds the block bitmap
 864  * @grp_goal:           given target block within the group
 865  * @count:              target number of blocks to allocate
 866  * @my_rsv:             reservation window
 867  *
 868  * Attempt to allocate blocks within a give range. Set the range of allocation
 869  * first, then find the first free bit(s) from the bitmap (within the range),
 870  * and at last, allocate the blocks by claiming the found free bit as allocated.
 871  *
 872  * To set the range of this allocation:
 873  *      if there is a reservation window, only try to allocate block(s) from the
 874  *      file's own reservation window;
 875  *      Otherwise, the allocation range starts from the give goal block, ends at
 876  *      the block group's last block.
 877  *
 878  * If we failed to allocate the desired block then we may end up crossing to a
 879  * new bitmap.  In that case we must release write access to the old one via
 880  * ext3_journal_release_buffer(), else we'll run out of credits.
 881  */
 882 static ext3_grpblk_t
 883 ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
 884                         struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
 885                         unsigned long *count, struct ext3_reserve_window *my_rsv)
 886 {
 887         ext3_fsblk_t group_first_block;
 888         ext3_grpblk_t start, end;
 889         unsigned long num = 0;
 890
 891         /* we do allocation within the reservation window if we have a window */
 892         if (my_rsv) {
 893                 group_first_block = ext3_group_first_block_no(sb, group);
 894                 if (my_rsv->_rsv_start >= group_first_block)
 895                         start = my_rsv->_rsv_start - group_first_block;
 896                 else
 897                         /* reservation window cross group boundary */
 898                         start = 0;
 899                 end = my_rsv->_rsv_end - group_first_block + 1;
 900                 if (end > EXT3_BLOCKS_PER_GROUP(sb))
 901                         /* reservation window crosses group boundary */
 902                         end = EXT3_BLOCKS_PER_GROUP(sb);
 903                 if ((start <= grp_goal) && (grp_goal < end))
 904                         start = grp_goal;
 905                 else
 906                         grp_goal = -1;
 907         } else {
 908                 if (grp_goal > 0)
 909                         start = grp_goal;
 910                 else
 911                         start = 0;
 912                 end = EXT3_BLOCKS_PER_GROUP(sb);
 913         }
 914
 915         BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
 916
 917 repeat:
 918         if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
 919                 grp_goal = find_next_usable_block(start, bitmap_bh, end);
 920                 if (grp_goal < 0)
 921                         goto fail_access;
 922                 if (!my_rsv) {
 923                         int i;
 924
 925                         for (i = 0; i < 7 && grp_goal > start &&
 926                                         ext3_test_allocatable(grp_goal - 1,
 927                                                                 bitmap_bh);
 928                                         i++, grp_goal--)
 929                                 ;
 930                 }
 931         }
 932         start = grp_goal;
 933
 934         if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
 935                 grp_goal, bitmap_bh)) {
 936                 /*
 937                  * The block was allocated by another thread, or it was
 938                  * allocated and then freed by another thread
 939                  */
 940                 start++;
 941                 grp_goal++;
 942                 if (start >= end)
 943                         goto fail_access;
 944                 goto repeat;
 945         }
 946         num++;
 947         grp_goal++;
 948         while (num < *count && grp_goal < end
 949                 && ext3_test_allocatable(grp_goal, bitmap_bh)
 950                 && claim_block(sb_bgl_lock(EXT3_SB(sb), group),
 951                                 grp_goal, bitmap_bh)) {
 952                 num++;
 953                 grp_goal++;
 954         }
 955         *count = num;
 956         return grp_goal - num;
 957 fail_access:
 958         *count = num;
 959         return -1;
 960 }
 961
 962 /**
 963  *      find_next_reservable_window():
 964  *              find a reservable space within the given range.
 965  *              It does not allocate the reservation window for now:
 966  *              alloc_new_reservation() will do the work later.
 967  *
 968  *      @search_head: the head of the searching list;
 969  *              This is not necessarily the list head of the whole filesystem
 970  *
 971  *              We have both head and start_block to assist the search
 972  *              for the reservable space. The list starts from head,
 973  *              but we will shift to the place where start_block is,
 974  *              then start from there, when looking for a reservable space.
 975  *
 976  *      @my_rsv: the reservation window
 977  *
 978  *      @sb: the super block
 979  *
 980  *      @start_block: the first block we consider to start
 981  *                      the real search from
 982  *
 983  *      @last_block:
 984  *              the maximum block number that our goal reservable space
 985  *              could start from. This is normally the last block in this
 986  *              group. The search will end when we found the start of next
 987  *              possible reservable space is out of this boundary.
 988  *              This could handle the cross boundary reservation window
 989  *              request.
 990  *
 991  *      basically we search from the given range, rather than the whole
 992  *      reservation double linked list, (start_block, last_block)
 993  *      to find a free region that is of my size and has not
 994  *      been reserved.
 995  *
 996  */
 997 static int find_next_reservable_window(
 998                                 struct ext3_reserve_window_node *search_head,
 999                                 struct ext3_reserve_window_node *my_rsv,
1000                                 struct super_block * sb,
1001                                 ext3_fsblk_t start_block,
1002                                 ext3_fsblk_t last_block)
1003 {
1004         struct rb_node *next;
1005         struct ext3_reserve_window_node *rsv, *prev;
1006         ext3_fsblk_t cur;
1007         int size = my_rsv->rsv_goal_size;
1008
1009         /* TODO: make the start of the reservation window byte-aligned */
1010         /* cur = *start_block & ~7;*/
1011         cur = start_block;
1012         rsv = search_head;
1013         if (!rsv)
1014                 return -1;
1015
1016         while (1) {
1017                 if (cur <= rsv->rsv_end)
1018                         cur = rsv->rsv_end + 1;
1019
1020                 /* TODO?
1021                  * in the case we could not find a reservable space
1022                  * that is what is expected, during the re-search, we could
1023                  * remember what's the largest reservable space we could have
1024                  * and return that one.
1025                  *
1026                  * For now it will fail if we could not find the reservable
1027                  * space with expected-size (or more)...
1028                  */
1029                 if (cur > last_block)
1030                         return -1;              /* fail */
1031
1032                 prev = rsv;
1033                 next = rb_next(&rsv->rsv_node);
1034                 rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
1035
1036                 /*
1037                  * Reached the last reservation, we can just append to the
1038                  * previous one.
1039                  */
1040                 if (!next)
1041                         break;
1042
1043                 if (cur + size <= rsv->rsv_start) {
1044                         /*
1045                          * Found a reserveable space big enough.  We could
1046                          * have a reservation across the group boundary here
1047                          */
1048                         break;
1049                 }
1050         }
1051         /*
1052          * we come here either :
1053          * when we reach the end of the whole list,
1054          * and there is empty reservable space after last entry in the list.
1055          * append it to the end of the list.
1056          *
1057          * or we found one reservable space in the middle of the list,
1058          * return the reservation window that we could append to.
1059          * succeed.
1060          */
1061
1062         if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1063                 rsv_window_remove(sb, my_rsv);
1064
1065         /*
1066          * Let's book the whole available window for now.  We will check the
1067          * disk bitmap later and then, if there are free blocks then we adjust
1068          * the window size if it's larger than requested.
1069          * Otherwise, we will remove this node from the tree next time
1070          * call find_next_reservable_window.
1071          */
1072         my_rsv->rsv_start = cur;
1073         my_rsv->rsv_end = cur + size - 1;
1074         my_rsv->rsv_alloc_hit = 0;
1075
1076         if (prev != my_rsv)
1077                 ext3_rsv_window_add(sb, my_rsv);
1078
1079         return 0;
1080 }
1081
1082 /**
1083  *      alloc_new_reservation()--allocate a new reservation window
1084  *
1085  *              To make a new reservation, we search part of the filesystem
1086  *              reservation list (the list that inside the group). We try to
1087  *              allocate a new reservation window near the allocation goal,
1088  *              or the beginning of the group, if there is no goal.
1089  *
1090  *              We first find a reservable space after the goal, then from
1091  *              there, we check the bitmap for the first free block after
1092  *              it. If there is no free block until the end of group, then the
1093  *              whole group is full, we failed. Otherwise, check if the free
1094  *              block is inside the expected reservable space, if so, we
1095  *              succeed.
1096  *              If the first free block is outside the reservable space, then
1097  *              start from the first free block, we search for next available
1098  *              space, and go on.
1099  *
1100  *      on succeed, a new reservation will be found and inserted into the list
1101  *      It contains at least one free block, and it does not overlap with other
1102  *      reservation windows.
1103  *
1104  *      failed: we failed to find a reservation window in this group
1105  *
1106  *      @my_rsv: the reservation window
1107  *
1108  *      @grp_goal: The goal (group-relative).  It is where the search for a
1109  *              free reservable space should start from.
1110  *              if we have a grp_goal(grp_goal >0 ), then start from there,
1111  *              no grp_goal(grp_goal = -1), we start from the first block
1112  *              of the group.
1113  *
1114  *      @sb: the super block
1115  *      @group: the group we are trying to allocate in
1116  *      @bitmap_bh: the block group block bitmap
1117  *
1118  */
1119 static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
1120                 ext3_grpblk_t grp_goal, struct super_block *sb,
1121                 unsigned int group, struct buffer_head *bitmap_bh)
1122 {
1123         struct ext3_reserve_window_node *search_head;
1124         ext3_fsblk_t group_first_block, group_end_block, start_block;
1125         ext3_grpblk_t first_free_block;
1126         struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1127         unsigned long size;
1128         int ret;
1129         spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1130
1131         group_first_block = ext3_group_first_block_no(sb, group);
1132         group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1133
1134         if (grp_goal < 0)
1135                 start_block = group_first_block;
1136         else
1137                 start_block = grp_goal + group_first_block;
1138
1139         size = my_rsv->rsv_goal_size;
1140
1141         if (!rsv_is_empty(&my_rsv->rsv_window)) {
1142                 /*
1143                  * if the old reservation is cross group boundary
1144                  * and if the goal is inside the old reservation window,
1145                  * we will come here when we just failed to allocate from
1146                  * the first part of the window. We still have another part
1147                  * that belongs to the next group. In this case, there is no
1148                  * point to discard our window and try to allocate a new one
1149                  * in this group(which will fail). we should
1150                  * keep the reservation window, just simply move on.
1151                  *
1152                  * Maybe we could shift the start block of the reservation
1153                  * window to the first block of next group.
1154                  */
1155
1156                 if ((my_rsv->rsv_start <= group_end_block) &&
1157                                 (my_rsv->rsv_end > group_end_block) &&
1158                                 (start_block >= my_rsv->rsv_start))
1159                         return -1;
1160
1161                 if ((my_rsv->rsv_alloc_hit >
1162                      (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1163                         /*
1164                          * if the previously allocation hit ratio is
1165                          * greater than 1/2, then we double the size of
1166                          * the reservation window the next time,
1167                          * otherwise we keep the same size window
1168                          */
1169                         size = size * 2;
1170                         if (size > EXT3_MAX_RESERVE_BLOCKS)
1171                                 size = EXT3_MAX_RESERVE_BLOCKS;
1172                         my_rsv->rsv_goal_size= size;
1173                 }
1174         }
1175
1176         spin_lock(rsv_lock);
1177         /*
1178          * shift the search start to the window near the goal block
1179          */
1180         search_head = search_reserve_window(fs_rsv_root, start_block);
1181
1182         /*
1183          * find_next_reservable_window() simply finds a reservable window
1184          * inside the given range(start_block, group_end_block).
1185          *
1186          * To make sure the reservation window has a free bit inside it, we
1187          * need to check the bitmap after we found a reservable window.
1188          */
1189 retry:
1190         ret = find_next_reservable_window(search_head, my_rsv, sb,
1191                                                 start_block, group_end_block);
1192
1193         if (ret == -1) {
1194                 if (!rsv_is_empty(&my_rsv->rsv_window))
1195                         rsv_window_remove(sb, my_rsv);
1196                 spin_unlock(rsv_lock);
1197                 return -1;
1198         }
1199
1200         /*
1201          * On success, find_next_reservable_window() returns the
1202          * reservation window where there is a reservable space after it.
1203          * Before we reserve this reservable space, we need
1204          * to make sure there is at least a free block inside this region.
1205          *
1206          * searching the first free bit on the block bitmap and copy of
1207          * last committed bitmap alternatively, until we found a allocatable
1208          * block. Search start from the start block of the reservable space
1209          * we just found.
1210          */
1211         spin_unlock(rsv_lock);
1212         first_free_block = bitmap_search_next_usable_block(
1213                         my_rsv->rsv_start - group_first_block,
1214                         bitmap_bh, group_end_block - group_first_block + 1);
1215
1216         if (first_free_block < 0) {
1217                 /*
1218                  * no free block left on the bitmap, no point
1219                  * to reserve the space. return failed.
1220                  */
1221                 spin_lock(rsv_lock);
1222                 if (!rsv_is_empty(&my_rsv->rsv_window))
1223                         rsv_window_remove(sb, my_rsv);
1224                 spin_unlock(rsv_lock);
1225                 return -1;              /* failed */
1226         }
1227
1228         start_block = first_free_block + group_first_block;
1229         /*
1230          * check if the first free block is within the
1231          * free space we just reserved
1232          */
1233         if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
1234                 return 0;               /* success */
1235         /*
1236          * if the first free bit we found is out of the reservable space
1237          * continue search for next reservable space,
1238          * start from where the free block is,
1239          * we also shift the list head to where we stopped last time
1240          */
1241         search_head = my_rsv;
1242         spin_lock(rsv_lock);
1243         goto retry;
1244 }
1245
1246 /**
1247  * try_to_extend_reservation()
1248  * @my_rsv:             given reservation window
1249  * @sb:                 super block
1250  * @size:               the delta to extend
1251  *
1252  * Attempt to expand the reservation window large enough to have
1253  * required number of free blocks
1254  *
1255  * Since ext3_try_to_allocate() will always allocate blocks within
1256  * the reservation window range, if the window size is too small,
1257  * multiple blocks allocation has to stop at the end of the reservation
1258  * window. To make this more efficient, given the total number of
1259  * blocks needed and the current size of the window, we try to
1260  * expand the reservation window size if necessary on a best-effort
1261  * basis before ext3_new_blocks() tries to allocate blocks,
1262  */
1263 static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1264                         struct super_block *sb, int size)
1265 {
1266         struct ext3_reserve_window_node *next_rsv;
1267         struct rb_node *next;
1268         spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1269
1270         if (!spin_trylock(rsv_lock))
1271                 return;
1272
1273         next = rb_next(&my_rsv->rsv_node);
1274
1275         if (!next)
1276                 my_rsv->rsv_end += size;
1277         else {
1278                 next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
1279
1280                 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1281                         my_rsv->rsv_end += size;
1282                 else
1283                         my_rsv->rsv_end = next_rsv->rsv_start - 1;
1284         }
1285         spin_unlock(rsv_lock);
1286 }
1287
1288 /**
1289  * ext3_try_to_allocate_with_rsv()
1290  * @sb:                 superblock
1291  * @handle:             handle to this transaction
1292  * @group:              given allocation block group
1293  * @bitmap_bh:          bufferhead holds the block bitmap
1294  * @grp_goal:           given target block within the group
1295  * @my_rsv:             reservation window
1296  * @count:              target number of blocks to allocate
1297  * @errp:               pointer to store the error code
1298  *
1299  * This is the main function used to allocate a new block and its reservation
1300  * window.
1301  *
1302  * Each time when a new block allocation is need, first try to allocate from
1303  * its own reservation.  If it does not have a reservation window, instead of
1304  * looking for a free bit on bitmap first, then look up the reservation list to
1305  * see if it is inside somebody else's reservation window, we try to allocate a
1306  * reservation window for it starting from the goal first. Then do the block
1307  * allocation within the reservation window.
1308  *
1309  * This will avoid keeping on searching the reservation list again and
1310  * again when somebody is looking for a free block (without
1311  * reservation), and there are lots of free blocks, but they are all
1312  * being reserved.
1313  *
1314  * We use a red-black tree for the per-filesystem reservation list.
1315  *
1316  */
1317 static ext3_grpblk_t
1318 ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1319                         unsigned int group, struct buffer_head *bitmap_bh,
1320                         ext3_grpblk_t grp_goal,
1321                         struct ext3_reserve_window_node * my_rsv,
1322                         unsigned long *count, int *errp)
1323 {
1324         ext3_fsblk_t group_first_block, group_last_block;
1325         ext3_grpblk_t ret = 0;
1326         int fatal;
1327         unsigned long num = *count;
1328
1329         *errp = 0;
1330
1331         /*
1332          * Make sure we use undo access for the bitmap, because it is critical
1333          * that we do the frozen_data COW on bitmap buffers in all cases even
1334          * if the buffer is in BJ_Forget state in the committing transaction.
1335          */
1336         BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1337         fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1338         if (fatal) {
1339                 *errp = fatal;
1340                 return -1;
1341         }
1342
1343         /*
1344          * we don't deal with reservation when
1345          * filesystem is mounted without reservation
1346          * or the file is not a regular file
1347          * or last attempt to allocate a block with reservation turned on failed
1348          */
1349         if (my_rsv == NULL ) {
1350                 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1351                                                 grp_goal, count, NULL);
1352                 goto out;
1353         }
1354         /*
1355          * grp_goal is a group relative block number (if there is a goal)
1356          * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1357          * first block is a filesystem wide block number
1358          * first block is the block number of the first block in this group
1359          */
1360         group_first_block = ext3_group_first_block_no(sb, group);
1361         group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1362
1363         /*
1364          * Basically we will allocate a new block from inode's reservation
1365          * window.
1366          *
1367          * We need to allocate a new reservation window, if:
1368          * a) inode does not have a reservation window; or
1369          * b) last attempt to allocate a block from existing reservation
1370          *    failed; or
1371          * c) we come here with a goal and with a reservation window
1372          *
1373          * We do not need to allocate a new reservation window if we come here
1374          * at the beginning with a goal and the goal is inside the window, or
1375          * we don't have a goal but already have a reservation window.
1376          * then we could go to allocate from the reservation window directly.
1377          */
1378         while (1) {
1379                 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1380                         !goal_in_my_reservation(&my_rsv->rsv_window,
1381                                                 grp_goal, group, sb)) {
1382                         if (my_rsv->rsv_goal_size < *count)
1383                                 my_rsv->rsv_goal_size = *count;
1384                         ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1385                                                         group, bitmap_bh);
1386                         if (ret < 0)
1387                                 break;                  /* failed */
1388
1389                         if (!goal_in_my_reservation(&my_rsv->rsv_window,
1390                                                         grp_goal, group, sb))
1391                                 grp_goal = -1;
1392                 } else if (grp_goal >= 0) {
1393                         int curr = my_rsv->rsv_end -
1394                                         (grp_goal + group_first_block) + 1;
1395
1396                         if (curr < *count)
1397                                 try_to_extend_reservation(my_rsv, sb,
1398                                                         *count - curr);
1399                 }
1400
1401                 if ((my_rsv->rsv_start > group_last_block) ||
1402                                 (my_rsv->rsv_end < group_first_block)) {
1403                         rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
1404                         BUG();
1405                 }
1406                 ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1407                                            grp_goal, &num, &my_rsv->rsv_window);
1408                 if (ret >= 0) {
1409                         my_rsv->rsv_alloc_hit += num;
1410                         *count = num;
1411                         break;                          /* succeed */
1412                 }
1413                 num = *count;
1414         }
1415 out:
1416         if (ret >= 0) {
1417                 BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1418                                         "bitmap block");
1419                 fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1420                 if (fatal) {
1421                         *errp = fatal;
1422                         return -1;
1423                 }
1424                 return ret;
1425         }
1426
1427         BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1428         ext3_journal_release_buffer(handle, bitmap_bh);
1429         return ret;
1430 }
1431
1432 /**
1433  * ext3_has_free_blocks()
1434  * @sbi:                in-core super block structure.
1435  *
1436  * Check if filesystem has at least 1 free block available for allocation.
1437  */
1438 static int ext3_has_free_blocks(struct ext3_sb_info *sbi)
1439 {
1440         ext3_fsblk_t free_blocks, root_blocks;
1441
1442         free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1443         root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1444         if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1445                 sbi->s_resuid != current_fsuid() &&
1446                 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
1447                 return 0;
1448         }
1449         return 1;
1450 }
1451
1452 /**
1453  * ext3_should_retry_alloc()
1454  * @sb:                 super block
1455  * @retries             number of attemps has been made
1456  *
1457  * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1458  * it is profitable to retry the operation, this function will wait
1459  * for the current or committing transaction to complete, and then
1460  * return TRUE.
1461  *
1462  * if the total number of retries exceed three times, return FALSE.
1463  */
1464 int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1465 {
1466         if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3)
1467                 return 0;
1468
1469         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1470
1471         return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1472 }
1473
1474 /**
1475  * ext3_new_blocks() -- core block(s) allocation function
1476  * @handle:             handle to this transaction
1477  * @inode:              file inode
1478  * @goal:               given target block(filesystem wide)
1479  * @count:              target number of blocks to allocate
1480  * @errp:               error code
1481  *
1482  * ext3_new_blocks uses a goal block to assist allocation.  It tries to
1483  * allocate block(s) from the block group contains the goal block first. If that
1484  * fails, it will try to allocate block(s) from other block groups without
1485  * any specific goal block.
1486  *
1487  */
1488 ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1489                         ext3_fsblk_t goal, unsigned long *count, int *errp)
1490 {
1491         struct buffer_head *bitmap_bh = NULL;
1492         struct buffer_head *gdp_bh;
1493         int group_no;
1494         int goal_group;
1495         ext3_grpblk_t grp_target_blk;   /* blockgroup relative goal block */
1496         ext3_grpblk_t grp_alloc_blk;    /* blockgroup-relative allocated block*/
1497         ext3_fsblk_t ret_block;         /* filesyetem-wide allocated block */
1498         int bgi;                        /* blockgroup iteration index */
1499         int fatal = 0, err;
1500         int performed_allocation = 0;
1501         ext3_grpblk_t free_blocks;      /* number of free blocks in a group */
1502         struct super_block *sb;
1503         struct ext3_group_desc *gdp;
1504         struct ext3_super_block *es;
1505         struct ext3_sb_info *sbi;
1506         struct ext3_reserve_window_node *my_rsv = NULL;
1507         struct ext3_block_alloc_info *block_i;
1508         unsigned short windowsz = 0;
1509 #ifdef EXT3FS_DEBUG
1510         static int goal_hits, goal_attempts;
1511 #endif
1512         unsigned long ngroups;
1513         unsigned long num = *count;
1514
1515         *errp = -ENOSPC;
1516         sb = inode->i_sb;
1517         if (!sb) {
1518                 printk("ext3_new_block: nonexistent device");
1519                 return 0;
1520         }
1521
1522         /*
1523          * Check quota for allocation of this block.
1524          */
1525         err = dquot_alloc_block(inode, num);
1526         if (err) {
1527                 *errp = err;
1528                 return 0;
1529         }
1530
1531         sbi = EXT3_SB(sb);
1532         es = EXT3_SB(sb)->s_es;
1533         ext3_debug("goal=%lu.\n", goal);
1534         /*
1535          * Allocate a block from reservation only when
1536          * filesystem is mounted with reservation(default,-o reservation), and
1537          * it's a regular file, and
1538          * the desired window size is greater than 0 (One could use ioctl
1539          * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1540          * reservation on that particular file)
1541          */
1542         block_i = EXT3_I(inode)->i_block_alloc_info;
1543         if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1544                 my_rsv = &block_i->rsv_window_node;
1545
1546         if (!ext3_has_free_blocks(sbi)) {
1547                 *errp = -ENOSPC;
1548                 goto out;
1549         }
1550
1551         /*
1552          * First, test whether the goal block is free.
1553          */
1554         if (goal < le32_to_cpu(es->s_first_data_block) ||
1555             goal >= le32_to_cpu(es->s_blocks_count))
1556                 goal = le32_to_cpu(es->s_first_data_block);
1557         group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1558                         EXT3_BLOCKS_PER_GROUP(sb);
1559         goal_group = group_no;
1560 retry_alloc:
1561         gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1562         if (!gdp)
1563                 goto io_error;
1564
1565         free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1566         /*
1567          * if there is not enough free blocks to make a new resevation
1568          * turn off reservation for this allocation
1569          */
1570         if (my_rsv && (free_blocks < windowsz)
1571                 && (free_blocks > 0)
1572                 && (rsv_is_empty(&my_rsv->rsv_window)))
1573                 my_rsv = NULL;
1574
1575         if (free_blocks > 0) {
1576                 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1577                                 EXT3_BLOCKS_PER_GROUP(sb));
1578                 bitmap_bh = read_block_bitmap(sb, group_no);
1579                 if (!bitmap_bh)
1580                         goto io_error;
1581                 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1582                                         group_no, bitmap_bh, grp_target_blk,
1583                                         my_rsv, &num, &fatal);
1584                 if (fatal)
1585                         goto out;
1586                 if (grp_alloc_blk >= 0)
1587                         goto allocated;
1588         }
1589
1590         ngroups = EXT3_SB(sb)->s_groups_count;
1591         smp_rmb();
1592
1593         /*
1594          * Now search the rest of the groups.  We assume that
1595          * group_no and gdp correctly point to the last group visited.
1596          */
1597         for (bgi = 0; bgi < ngroups; bgi++) {
1598                 group_no++;
1599                 if (group_no >= ngroups)
1600                         group_no = 0;
1601                 gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1602                 if (!gdp)
1603                         goto io_error;
1604                 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1605                 /*
1606                  * skip this group (and avoid loading bitmap) if there
1607                  * are no free blocks
1608                  */
1609                 if (!free_blocks)
1610                         continue;
1611                 /*
1612                  * skip this group if the number of
1613                  * free blocks is less than half of the reservation
1614                  * window size.
1615                  */
1616                 if (my_rsv && (free_blocks <= (windowsz/2)))
1617                         continue;
1618
1619                 brelse(bitmap_bh);
1620                 bitmap_bh = read_block_bitmap(sb, group_no);
1621                 if (!bitmap_bh)
1622                         goto io_error;
1623                 /*
1624                  * try to allocate block(s) from this group, without a goal(-1).
1625                  */
1626                 grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1627                                         group_no, bitmap_bh, -1, my_rsv,
1628                                         &num, &fatal);
1629                 if (fatal)
1630                         goto out;
1631                 if (grp_alloc_blk >= 0)
1632                         goto allocated;
1633         }
1634         /*
1635          * We may end up a bogus earlier ENOSPC error due to
1636          * filesystem is "full" of reservations, but
1637          * there maybe indeed free blocks available on disk
1638          * In this case, we just forget about the reservations
1639          * just do block allocation as without reservations.
1640          */
1641         if (my_rsv) {
1642                 my_rsv = NULL;
1643                 windowsz = 0;
1644                 group_no = goal_group;
1645                 goto retry_alloc;
1646         }
1647         /* No space left on the device */
1648         *errp = -ENOSPC;
1649         goto out;
1650
1651 allocated:
1652
1653         ext3_debug("using block group %d(%d)\n",
1654                         group_no, gdp->bg_free_blocks_count);
1655
1656         BUFFER_TRACE(gdp_bh, "get_write_access");
1657         fatal = ext3_journal_get_write_access(handle, gdp_bh);
1658         if (fatal)
1659                 goto out;
1660
1661         ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1662
1663         if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1664             in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1665             in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1666                       EXT3_SB(sb)->s_itb_per_group) ||
1667             in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1668                       EXT3_SB(sb)->s_itb_per_group)) {
1669                 ext3_error(sb, "ext3_new_block",
1670                             "Allocating block in system zone - "
1671                             "blocks from "E3FSBLK", length %lu",
1672                              ret_block, num);
1673                 /*
1674                  * claim_block() marked the blocks we allocated as in use. So we
1675                  * may want to selectively mark some of the blocks as free.
1676                  */
1677                 goto retry_alloc;
1678         }
1679
1680         performed_allocation = 1;
1681
1682 #ifdef CONFIG_JBD_DEBUG
1683         {
1684                 struct buffer_head *debug_bh;
1685
1686                 /* Record bitmap buffer state in the newly allocated block */
1687                 debug_bh = sb_find_get_block(sb, ret_block);
1688                 if (debug_bh) {
1689                         BUFFER_TRACE(debug_bh, "state when allocated");
1690                         BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1691                         brelse(debug_bh);
1692                 }
1693         }
1694         jbd_lock_bh_state(bitmap_bh);
1695         spin_lock(sb_bgl_lock(sbi, group_no));
1696         if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1697                 int i;
1698
1699                 for (i = 0; i < num; i++) {
1700                         if (ext3_test_bit(grp_alloc_blk+i,
1701                                         bh2jh(bitmap_bh)->b_committed_data)) {
1702                                 printk("%s: block was unexpectedly set in "
1703                                         "b_committed_data\n", __func__);
1704                         }
1705                 }
1706         }
1707         ext3_debug("found bit %d\n", grp_alloc_blk);
1708         spin_unlock(sb_bgl_lock(sbi, group_no));
1709         jbd_unlock_bh_state(bitmap_bh);
1710 #endif
1711
1712         if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1713                 ext3_error(sb, "ext3_new_block",
1714                             "block("E3FSBLK") >= blocks count(%d) - "
1715                             "block_group = %d, es == %p ", ret_block,
1716                         le32_to_cpu(es->s_blocks_count), group_no, es);
1717                 goto out;
1718         }
1719
1720         /*
1721          * It is up to the caller to add the new buffer to a journal
1722          * list of some description.  We don't know in advance whether
1723          * the caller wants to use it as metadata or data.
1724          */
1725         ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1726                         ret_block, goal_hits, goal_attempts);
1727
1728         spin_lock(sb_bgl_lock(sbi, group_no));
1729         le16_add_cpu(&gdp->bg_free_blocks_count, -num);
1730         spin_unlock(sb_bgl_lock(sbi, group_no));
1731         percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1732
1733         BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1734         err = ext3_journal_dirty_metadata(handle, gdp_bh);
1735         if (!fatal)
1736                 fatal = err;
1737
1738         if (fatal)
1739                 goto out;
1740
1741         *errp = 0;
1742         brelse(bitmap_bh);
1743         dquot_free_block(inode, *count-num);
1744         *count = num;
1745         return ret_block;
1746
1747 io_error:
1748         *errp = -EIO;
1749 out:
1750         if (fatal) {
1751                 *errp = fatal;
1752                 ext3_std_error(sb, fatal);
1753         }
1754         /*
1755          * Undo the block allocation
1756          */
1757         if (!performed_allocation)
1758                 dquot_free_block(inode, *count);
1759         brelse(bitmap_bh);
1760         return 0;
1761 }
1762
1763 ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1764                         ext3_fsblk_t goal, int *errp)
1765 {
1766         unsigned long count = 1;
1767
1768         return ext3_new_blocks(handle, inode, goal, &count, errp);
1769 }
1770
1771 /**
1772  * ext3_count_free_blocks() -- count filesystem free blocks
1773  * @sb:         superblock
1774  *
1775  * Adds up the number of free blocks from each block group.
1776  */
1777 ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1778 {
1779         ext3_fsblk_t desc_count;
1780         struct ext3_group_desc *gdp;
1781         int i;
1782         unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1783 #ifdef EXT3FS_DEBUG
1784         struct ext3_super_block *es;
1785         ext3_fsblk_t bitmap_count;
1786         unsigned long x;
1787         struct buffer_head *bitmap_bh = NULL;
1788
1789         es = EXT3_SB(sb)->s_es;
1790         desc_count = 0;
1791         bitmap_count = 0;
1792         gdp = NULL;
1793
1794         smp_rmb();
1795         for (i = 0; i < ngroups; i++) {
1796                 gdp = ext3_get_group_desc(sb, i, NULL);
1797                 if (!gdp)
1798                         continue;
1799                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1800                 brelse(bitmap_bh);
1801                 bitmap_bh = read_block_bitmap(sb, i);
1802                 if (bitmap_bh == NULL)
1803                         continue;
1804
1805                 x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1806                 printk("group %d: stored = %d, counted = %lu\n",
1807                         i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1808                 bitmap_count += x;
1809         }
1810         brelse(bitmap_bh);
1811         printk("ext3_count_free_blocks: stored = "E3FSBLK
1812                 ", computed = "E3FSBLK", "E3FSBLK"\n",
1813                le32_to_cpu(es->s_free_blocks_count),
1814                 desc_count, bitmap_count);
1815         return bitmap_count;
1816 #else
1817         desc_count = 0;
1818         smp_rmb();
1819         for (i = 0; i < ngroups; i++) {
1820                 gdp = ext3_get_group_desc(sb, i, NULL);
1821                 if (!gdp)
1822                         continue;
1823                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1824         }
1825
1826         return desc_count;
1827 #endif
1828 }
1829
1830 static inline int test_root(int a, int b)
1831 {
1832         int num = b;
1833
1834         while (a > num)
1835                 num *= b;
1836         return num == a;
1837 }
1838
1839 static int ext3_group_sparse(int group)
1840 {
1841         if (group <= 1)
1842                 return 1;
1843         if (!(group & 1))
1844                 return 0;
1845         return (test_root(group, 7) || test_root(group, 5) ||
1846                 test_root(group, 3));
1847 }
1848
1849 /**
1850  *      ext3_bg_has_super - number of blocks used by the superblock in group
1851  *      @sb: superblock for filesystem
1852  *      @group: group number to check
1853  *
1854  *      Return the number of blocks used by the superblock (primary or backup)
1855  *      in this group.  Currently this will be only 0 or 1.
1856  */
1857 int ext3_bg_has_super(struct super_block *sb, int group)
1858 {
1859         if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1860                                 EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1861                         !ext3_group_sparse(group))
1862                 return 0;
1863         return 1;
1864 }
1865
1866 static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1867 {
1868         unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1869         unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1870         unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1871
1872         if (group == first || group == first + 1 || group == last)
1873                 return 1;
1874         return 0;
1875 }
1876
1877 static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1878 {
1879         return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
1880 }
1881
1882 /**
1883  *      ext3_bg_num_gdb - number of blocks used by the group table in group
1884  *      @sb: superblock for filesystem
1885  *      @group: group number to check
1886  *
1887  *      Return the number of blocks used by the group descriptor table
1888  *      (primary or backup) in this group.  In the future there may be a
1889  *      different number of descriptor blocks in each group.
1890  */
1891 unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1892 {
1893         unsigned long first_meta_bg =
1894                         le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1895         unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1896
1897         if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1898                         metagroup < first_meta_bg)
1899                 return ext3_bg_num_gdb_nometa(sb,group);
1900
1901         return ext3_bg_num_gdb_meta(sb,group);
1902
1903 }
1904
1905 /**
1906  * ext3_trim_all_free -- function to trim all free space in alloc. group
1907  * @sb:                 super block for file system
1908  * @group:              allocation group to trim
1909  * @start:              first group block to examine
1910  * @max:                last group block to examine
1911  * @gdp:                allocation group description structure
1912  * @minblocks:          minimum extent block count
1913  *
1914  * ext3_trim_all_free walks through group's block bitmap searching for free
1915  * blocks. When the free block is found, it tries to allocate this block and
1916  * consequent free block to get the biggest free extent possible, until it
1917  * reaches any used block. Then issue a TRIM command on this extent and free
1918  * the extent in the block bitmap. This is done until whole group is scanned.
1919  */
1920 ext3_grpblk_t ext3_trim_all_free(struct super_block *sb, unsigned int group,
1921                                 ext3_grpblk_t start, ext3_grpblk_t max,
1922                                 ext3_grpblk_t minblocks)
1923 {
1924         handle_t *handle;
1925         ext3_grpblk_t next, free_blocks, bit, freed, count = 0;
1926         ext3_fsblk_t discard_block;
1927         struct ext3_sb_info *sbi;
1928         struct buffer_head *gdp_bh, *bitmap_bh = NULL;
1929         struct ext3_group_desc *gdp;
1930         int err = 0, ret = 0;
1931
1932         /*
1933          * We will update one block bitmap, and one group descriptor
1934          */
1935         handle = ext3_journal_start_sb(sb, 2);
1936         if (IS_ERR(handle))
1937                 return PTR_ERR(handle);
1938
1939         bitmap_bh = read_block_bitmap(sb, group);
1940         if (!bitmap_bh) {
1941                 err = -EIO;
1942                 goto err_out;
1943         }
1944
1945         BUFFER_TRACE(bitmap_bh, "getting undo access");
1946         err = ext3_journal_get_undo_access(handle, bitmap_bh);
1947         if (err)
1948                 goto err_out;
1949
1950         gdp = ext3_get_group_desc(sb, group, &gdp_bh);
1951         if (!gdp) {
1952                 err = -EIO;
1953                 goto err_out;
1954         }
1955
1956         BUFFER_TRACE(gdp_bh, "get_write_access");
1957         err = ext3_journal_get_write_access(handle, gdp_bh);
1958         if (err)
1959                 goto err_out;
1960
1961         free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1962         sbi = EXT3_SB(sb);
1963
1964          /* Walk through the whole group */
1965         while (start < max) {
1966                 start = bitmap_search_next_usable_block(start, bitmap_bh, max);
1967                 if (start < 0)
1968                         break;
1969                 next = start;
1970
1971                 /*
1972                  * Allocate contiguous free extents by setting bits in the
1973                  * block bitmap
1974                  */
1975                 while (next < max
1976                         && claim_block(sb_bgl_lock(sbi, group),
1977                                         next, bitmap_bh)) {
1978                         next++;
1979                 }
1980
1981                  /* We did not claim any blocks */
1982                 if (next == start)
1983                         continue;
1984
1985                 discard_block = (ext3_fsblk_t)start +
1986                                 ext3_group_first_block_no(sb, group);
1987
1988                 /* Update counters */
1989                 spin_lock(sb_bgl_lock(sbi, group));
1990                 le16_add_cpu(&gdp->bg_free_blocks_count, start - next);
1991                 spin_unlock(sb_bgl_lock(sbi, group));
1992                 percpu_counter_sub(&sbi->s_freeblocks_counter, next - start);
1993
1994                 free_blocks -= next - start;
1995                 /* Do not issue a TRIM on extents smaller than minblocks */
1996                 if ((next - start) < minblocks)
1997                         goto free_extent;
1998
1999                  /* Send the TRIM command down to the device */
2000                 err = sb_issue_discard(sb, discard_block, next - start,
2001                                        GFP_NOFS, 0);
2002                 count += (next - start);
2003 free_extent:
2004                 freed = 0;
2005
2006                 /*
2007                  * Clear bits in the bitmap
2008                  */
2009                 for (bit = start; bit < next; bit++) {
2010                         BUFFER_TRACE(bitmap_bh, "clear bit");
2011                         if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, group),
2012                                                 bit, bitmap_bh->b_data)) {
2013                                 ext3_error(sb, __func__,
2014                                         "bit already cleared for block "E3FSBLK,
2015                                          (unsigned long)bit);
2016                                 BUFFER_TRACE(bitmap_bh, "bit already cleared");
2017                         } else {
2018                                 freed++;
2019                         }
2020                 }
2021
2022                 /* Update couters */
2023                 spin_lock(sb_bgl_lock(sbi, group));
2024                 le16_add_cpu(&gdp->bg_free_blocks_count, freed);
2025                 spin_unlock(sb_bgl_lock(sbi, group));
2026                 percpu_counter_add(&sbi->s_freeblocks_counter, freed);
2027
2028                 start = next;
2029                 if (err < 0) {
2030                         if (err != -EOPNOTSUPP)
2031                                 ext3_warning(sb, __func__, "Discard command "
2032                                              "returned error %d\n", err);
2033                         break;
2034                 }
2035
2036                 if (fatal_signal_pending(current)) {
2037                         err = -ERESTARTSYS;
2038                         break;
2039                 }
2040
2041                 cond_resched();
2042
2043                 /* No more suitable extents */
2044                 if (free_blocks < minblocks)
2045                         break;
2046         }
2047
2048         /* We dirtied the bitmap block */
2049         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
2050         ret = ext3_journal_dirty_metadata(handle, bitmap_bh);
2051         if (!err)
2052                 err = ret;
2053
2054         /* And the group descriptor block */
2055         BUFFER_TRACE(gdp_bh, "dirtied group descriptor block");
2056         ret = ext3_journal_dirty_metadata(handle, gdp_bh);
2057         if (!err)
2058                 err = ret;
2059
2060         ext3_debug("trimmed %d blocks in the group %d\n",
2061                 count, group);
2062
2063 err_out:
2064         if (err)
2065                 count = err;
2066         ext3_journal_stop(handle);
2067         brelse(bitmap_bh);
2068
2069         return count;
2070 }
2071
2072 /**
2073  * ext3_trim_fs() -- trim ioctl handle function
2074  * @sb:                 superblock for filesystem
2075  * @start:              First Byte to trim
2076  * @len:                number of Bytes to trim from start
2077  * @minlen:             minimum extent length in Bytes
2078  *
2079  * ext3_trim_fs goes through all allocation groups containing Bytes from
2080  * start to start+len. For each such a group ext3_trim_all_free function
2081  * is invoked to trim all free space.
2082  */
2083 int ext3_trim_fs(struct super_block *sb, struct fstrim_range *range)
2084 {
2085         ext3_grpblk_t last_block, first_block, free_blocks;
2086         unsigned long first_group, last_group;
2087         unsigned long group, ngroups;
2088         struct ext3_group_desc *gdp;
2089         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2090         uint64_t start, len, minlen, trimmed;
2091         ext3_fsblk_t max_blks = le32_to_cpu(es->s_blocks_count);
2092         int ret = 0;
2093
2094         start = (range->start >> sb->s_blocksize_bits) +
2095                 le32_to_cpu(es->s_first_data_block);
2096         len = range->len >> sb->s_blocksize_bits;
2097         minlen = range->minlen >> sb->s_blocksize_bits;
2098         trimmed = 0;
2099
2100         if (unlikely(minlen > EXT3_BLOCKS_PER_GROUP(sb)))
2101                 return -EINVAL;
2102         if (start >= max_blks)
2103                 goto out;
2104         if (start + len > max_blks)
2105                 len = max_blks - start;
2106
2107         ngroups = EXT3_SB(sb)->s_groups_count;
2108         smp_rmb();
2109
2110         /* Determine first and last group to examine based on start and len */
2111         ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) start,
2112                                      &first_group, &first_block);
2113         ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) (start + len),
2114                                      &last_group, &last_block);
2115         last_group = (last_group > ngroups - 1) ? ngroups - 1 : last_group;
2116         last_block = EXT3_BLOCKS_PER_GROUP(sb);
2117
2118         if (first_group > last_group)
2119                 return -EINVAL;
2120
2121         for (group = first_group; group <= last_group; group++) {
2122                 gdp = ext3_get_group_desc(sb, group, NULL);
2123                 if (!gdp)
2124                         break;
2125
2126                 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
2127                 if (free_blocks < minlen)
2128                         continue;
2129
2130                 /*
2131                  * For all the groups except the last one, last block will
2132                  * always be EXT3_BLOCKS_PER_GROUP(sb), so we only need to
2133                  * change it for the last group in which case first_block +
2134                  * len < EXT3_BLOCKS_PER_GROUP(sb).
2135                  */
2136                 if (first_block + len < EXT3_BLOCKS_PER_GROUP(sb))
2137                         last_block = first_block + len;
2138                 len -= last_block - first_block;
2139
2140                 ret = ext3_trim_all_free(sb, group, first_block,
2141                                         last_block, minlen);
2142                 if (ret < 0)
2143                         break;
2144
2145                 trimmed += ret;
2146                 first_block = 0;
2147         }
2148
2149         if (ret >= 0)
2150                 ret = 0;
2151
2152 out:
2153         range->len = trimmed * sb->s_blocksize;
2154
2155         return ret;
2156 }