1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext2/balloc.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993 11 * Big-endian to little-endian byte-swapping/bitmaps by 12 * David S. Miller (davem@caip.rutgers.edu), 1995 13 */ 14 15 #include "ext2.h" 16 #include <linux/quotaops.h> 17 #include <linux/slab.h> 18 #include <linux/sched.h> 19 #include <linux/cred.h> 20 #include <linux/buffer_head.h> 21 #include <linux/capability.h> 22 23 /* 24 * balloc.c contains the blocks allocation and deallocation routines 25 */ 26 27 /* 28 * The free blocks are managed by bitmaps. A file system contains several 29 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 30 * block for inodes, N blocks for the inode table and data blocks. 31 * 32 * The file system contains group descriptors which are located after the 33 * super block. Each descriptor contains the number of the bitmap block and 34 * the free blocks count in the block. The descriptors are loaded in memory 35 * when a file system is mounted (see ext2_fill_super). 36 */ 37 38 39 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb, 40 unsigned int block_group, 41 struct buffer_head ** bh) 42 { 43 unsigned long group_desc; 44 unsigned long offset; 45 struct ext2_group_desc * desc; 46 struct ext2_sb_info *sbi = EXT2_SB(sb); 47 48 if (block_group >= sbi->s_groups_count) { 49 WARN(1, "block_group >= groups_count - " 50 "block_group = %d, groups_count = %lu", 51 block_group, sbi->s_groups_count); 52 53 return NULL; 54 } 55 56 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb); 57 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1); 58 if (!sbi->s_group_desc[group_desc]) { 59 WARN(1, "Group descriptor not loaded - " 60 "block_group = %d, group_desc = %lu, desc = %lu", 61 block_group, group_desc, offset); 62 return NULL; 63 } 64 65 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data; 66 if (bh) 67 *bh = sbi->s_group_desc[group_desc]; 68 return desc + offset; 69 } 70 71 static int ext2_valid_block_bitmap(struct super_block *sb, 72 struct ext2_group_desc *desc, 73 unsigned int block_group, 74 struct buffer_head *bh) 75 { 76 ext2_grpblk_t offset; 77 ext2_grpblk_t next_zero_bit; 78 ext2_fsblk_t bitmap_blk; 79 ext2_fsblk_t group_first_block; 80 81 group_first_block = ext2_group_first_block_no(sb, block_group); 82 83 /* check whether block bitmap block number is set */ 84 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap); 85 offset = bitmap_blk - group_first_block; 86 if (!ext2_test_bit(offset, bh->b_data)) 87 /* bad block bitmap */ 88 goto err_out; 89 90 /* check whether the inode bitmap block number is set */ 91 bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap); 92 offset = bitmap_blk - group_first_block; 93 if (!ext2_test_bit(offset, bh->b_data)) 94 /* bad block bitmap */ 95 goto err_out; 96 97 /* check whether the inode table block number is set */ 98 bitmap_blk = le32_to_cpu(desc->bg_inode_table); 99 offset = bitmap_blk - group_first_block; 100 next_zero_bit = ext2_find_next_zero_bit(bh->b_data, 101 offset + EXT2_SB(sb)->s_itb_per_group, 102 offset); 103 if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group) 104 /* good bitmap for inode tables */ 105 return 1; 106 107 err_out: 108 ext2_error(sb, __func__, 109 "Invalid block bitmap - " 110 "block_group = %d, block = %lu", 111 block_group, bitmap_blk); 112 return 0; 113 } 114 115 /* 116 * Read the bitmap for a given block_group,and validate the 117 * bits for block/inode/inode tables are set in the bitmaps 118 * 119 * Return buffer_head on success or NULL in case of failure. 120 */ 121 static struct buffer_head * 122 read_block_bitmap(struct super_block *sb, unsigned int block_group) 123 { 124 struct ext2_group_desc * desc; 125 struct buffer_head * bh = NULL; 126 ext2_fsblk_t bitmap_blk; 127 int ret; 128 129 desc = ext2_get_group_desc(sb, block_group, NULL); 130 if (!desc) 131 return NULL; 132 bitmap_blk = le32_to_cpu(desc->bg_block_bitmap); 133 bh = sb_getblk(sb, bitmap_blk); 134 if (unlikely(!bh)) { 135 ext2_error(sb, __func__, 136 "Cannot read block bitmap - " 137 "block_group = %d, block_bitmap = %u", 138 block_group, le32_to_cpu(desc->bg_block_bitmap)); 139 return NULL; 140 } 141 ret = bh_read(bh, 0); 142 if (ret > 0) 143 return bh; 144 if (ret < 0) { 145 brelse(bh); 146 ext2_error(sb, __func__, 147 "Cannot read block bitmap - " 148 "block_group = %d, block_bitmap = %u", 149 block_group, le32_to_cpu(desc->bg_block_bitmap)); 150 return NULL; 151 } 152 153 ext2_valid_block_bitmap(sb, desc, block_group, bh); 154 /* 155 * file system mounted not to panic on error, continue with corrupt 156 * bitmap 157 */ 158 return bh; 159 } 160 161 static void group_adjust_blocks(struct super_block *sb, int group_no, 162 struct ext2_group_desc *desc, struct buffer_head *bh, int count) 163 { 164 if (count) { 165 struct ext2_sb_info *sbi = EXT2_SB(sb); 166 unsigned free_blocks; 167 168 spin_lock(sb_bgl_lock(sbi, group_no)); 169 free_blocks = le16_to_cpu(desc->bg_free_blocks_count); 170 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count); 171 spin_unlock(sb_bgl_lock(sbi, group_no)); 172 mark_buffer_dirty(bh); 173 } 174 } 175 176 /* 177 * The reservation window structure operations 178 * -------------------------------------------- 179 * Operations include: 180 * dump, find, add, remove, is_empty, find_next_reservable_window, etc. 181 * 182 * We use a red-black tree to represent per-filesystem reservation 183 * windows. 184 * 185 */ 186 187 /** 188 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map 189 * @root: root of per-filesystem reservation rb tree 190 * @verbose: verbose mode 191 * @fn: function which wishes to dump the reservation map 192 * 193 * If verbose is turned on, it will print the whole block reservation 194 * windows(start, end). Otherwise, it will only print out the "bad" windows, 195 * those windows that overlap with their immediate neighbors. 196 */ 197 #if 1 198 static void __rsv_window_dump(struct rb_root *root, int verbose, 199 const char *fn) 200 { 201 struct rb_node *n; 202 struct ext2_reserve_window_node *rsv, *prev; 203 int bad; 204 205 restart: 206 n = rb_first(root); 207 bad = 0; 208 prev = NULL; 209 210 printk("Block Allocation Reservation Windows Map (%s):\n", fn); 211 while (n) { 212 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node); 213 if (verbose) 214 printk("reservation window 0x%p " 215 "start: %lu, end: %lu\n", 216 rsv, rsv->rsv_start, rsv->rsv_end); 217 if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { 218 printk("Bad reservation %p (start >= end)\n", 219 rsv); 220 bad = 1; 221 } 222 if (prev && prev->rsv_end >= rsv->rsv_start) { 223 printk("Bad reservation %p (prev->end >= start)\n", 224 rsv); 225 bad = 1; 226 } 227 if (bad) { 228 if (!verbose) { 229 printk("Restarting reservation walk in verbose mode\n"); 230 verbose = 1; 231 goto restart; 232 } 233 } 234 n = rb_next(n); 235 prev = rsv; 236 } 237 printk("Window map complete.\n"); 238 BUG_ON(bad); 239 } 240 #define rsv_window_dump(root, verbose) \ 241 __rsv_window_dump((root), (verbose), __func__) 242 #else 243 #define rsv_window_dump(root, verbose) do {} while (0) 244 #endif 245 246 /** 247 * goal_in_my_reservation() 248 * @rsv: inode's reservation window 249 * @grp_goal: given goal block relative to the allocation block group 250 * @group: the current allocation block group 251 * @sb: filesystem super block 252 * 253 * Test if the given goal block (group relative) is within the file's 254 * own block reservation window range. 255 * 256 * If the reservation window is outside the goal allocation group, return 0; 257 * grp_goal (given goal block) could be -1, which means no specific 258 * goal block. In this case, always return 1. 259 * If the goal block is within the reservation window, return 1; 260 * otherwise, return 0; 261 */ 262 static int 263 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal, 264 unsigned int group, struct super_block * sb) 265 { 266 ext2_fsblk_t group_first_block, group_last_block; 267 268 group_first_block = ext2_group_first_block_no(sb, group); 269 group_last_block = ext2_group_last_block_no(sb, group); 270 271 if ((rsv->_rsv_start > group_last_block) || 272 (rsv->_rsv_end < group_first_block)) 273 return 0; 274 if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start) 275 || (grp_goal + group_first_block > rsv->_rsv_end))) 276 return 0; 277 return 1; 278 } 279 280 /** 281 * search_reserve_window() 282 * @root: root of reservation tree 283 * @goal: target allocation block 284 * 285 * Find the reserved window which includes the goal, or the previous one 286 * if the goal is not in any window. 287 * Returns NULL if there are no windows or if all windows start after the goal. 288 */ 289 static struct ext2_reserve_window_node * 290 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal) 291 { 292 struct rb_node *n = root->rb_node; 293 struct ext2_reserve_window_node *rsv; 294 295 if (!n) 296 return NULL; 297 298 do { 299 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node); 300 301 if (goal < rsv->rsv_start) 302 n = n->rb_left; 303 else if (goal > rsv->rsv_end) 304 n = n->rb_right; 305 else 306 return rsv; 307 } while (n); 308 /* 309 * We've fallen off the end of the tree: the goal wasn't inside 310 * any particular node. OK, the previous node must be to one 311 * side of the interval containing the goal. If it's the RHS, 312 * we need to back up one. 313 */ 314 if (rsv->rsv_start > goal) { 315 n = rb_prev(&rsv->rsv_node); 316 rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node); 317 } 318 return rsv; 319 } 320 321 /* 322 * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree. 323 * @sb: super block 324 * @rsv: reservation window to add 325 * 326 * Must be called with rsv_lock held. 327 */ 328 void ext2_rsv_window_add(struct super_block *sb, 329 struct ext2_reserve_window_node *rsv) 330 { 331 struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root; 332 struct rb_node *node = &rsv->rsv_node; 333 ext2_fsblk_t start = rsv->rsv_start; 334 335 struct rb_node ** p = &root->rb_node; 336 struct rb_node * parent = NULL; 337 struct ext2_reserve_window_node *this; 338 339 while (*p) 340 { 341 parent = *p; 342 this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node); 343 344 if (start < this->rsv_start) 345 p = &(*p)->rb_left; 346 else if (start > this->rsv_end) 347 p = &(*p)->rb_right; 348 else { 349 rsv_window_dump(root, 1); 350 BUG(); 351 } 352 } 353 354 rb_link_node(node, parent, p); 355 rb_insert_color(node, root); 356 } 357 358 /** 359 * rsv_window_remove() -- unlink a window from the reservation rb tree 360 * @sb: super block 361 * @rsv: reservation window to remove 362 * 363 * Mark the block reservation window as not allocated, and unlink it 364 * from the filesystem reservation window rb tree. Must be called with 365 * rsv_lock held. 366 */ 367 static void rsv_window_remove(struct super_block *sb, 368 struct ext2_reserve_window_node *rsv) 369 { 370 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED; 371 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED; 372 rsv->rsv_alloc_hit = 0; 373 rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root); 374 } 375 376 /* 377 * rsv_is_empty() -- Check if the reservation window is allocated. 378 * @rsv: given reservation window to check 379 * 380 * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED. 381 */ 382 static inline int rsv_is_empty(struct ext2_reserve_window *rsv) 383 { 384 /* a valid reservation end block could not be 0 */ 385 return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED); 386 } 387 388 /** 389 * ext2_init_block_alloc_info() 390 * @inode: file inode structure 391 * 392 * Allocate and initialize the reservation window structure, and 393 * link the window to the ext2 inode structure at last 394 * 395 * The reservation window structure is only dynamically allocated 396 * and linked to ext2 inode the first time the open file 397 * needs a new block. So, before every ext2_new_block(s) call, for 398 * regular files, we should check whether the reservation window 399 * structure exists or not. In the latter case, this function is called. 400 * Fail to do so will result in block reservation being turned off for that 401 * open file. 402 * 403 * This function is called from ext2_get_blocks_handle(), also called 404 * when setting the reservation window size through ioctl before the file 405 * is open for write (needs block allocation). 406 * 407 * Needs truncate_mutex protection prior to calling this function. 408 */ 409 void ext2_init_block_alloc_info(struct inode *inode) 410 { 411 struct ext2_inode_info *ei = EXT2_I(inode); 412 struct ext2_block_alloc_info *block_i; 413 struct super_block *sb = inode->i_sb; 414 415 block_i = kmalloc(sizeof(*block_i), GFP_NOFS); 416 if (block_i) { 417 struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node; 418 419 rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED; 420 rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED; 421 422 /* 423 * if filesystem is mounted with NORESERVATION, the goal 424 * reservation window size is set to zero to indicate 425 * block reservation is off 426 */ 427 if (!test_opt(sb, RESERVATION)) 428 rsv->rsv_goal_size = 0; 429 else 430 rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS; 431 rsv->rsv_alloc_hit = 0; 432 block_i->last_alloc_logical_block = 0; 433 block_i->last_alloc_physical_block = 0; 434 } 435 ei->i_block_alloc_info = block_i; 436 } 437 438 /** 439 * ext2_discard_reservation() 440 * @inode: inode 441 * 442 * Discard(free) block reservation window on last file close, or truncate 443 * or at last iput(). 444 * 445 * It is being called in three cases: 446 * ext2_release_file(): last writer closes the file 447 * ext2_clear_inode(): last iput(), when nobody links to this file. 448 * ext2_truncate(): when the block indirect map is about to change. 449 */ 450 void ext2_discard_reservation(struct inode *inode) 451 { 452 struct ext2_inode_info *ei = EXT2_I(inode); 453 struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info; 454 struct ext2_reserve_window_node *rsv; 455 spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock; 456 457 if (!block_i) 458 return; 459 460 rsv = &block_i->rsv_window_node; 461 if (!rsv_is_empty(&rsv->rsv_window)) { 462 spin_lock(rsv_lock); 463 if (!rsv_is_empty(&rsv->rsv_window)) 464 rsv_window_remove(inode->i_sb, rsv); 465 spin_unlock(rsv_lock); 466 } 467 } 468 469 /** 470 * ext2_free_blocks() -- Free given blocks and update quota and i_blocks 471 * @inode: inode 472 * @block: start physical block to free 473 * @count: number of blocks to free 474 */ 475 void ext2_free_blocks (struct inode * inode, unsigned long block, 476 unsigned long count) 477 { 478 struct buffer_head *bitmap_bh = NULL; 479 struct buffer_head * bh2; 480 unsigned long block_group; 481 unsigned long bit; 482 unsigned long i; 483 unsigned long overflow; 484 struct super_block * sb = inode->i_sb; 485 struct ext2_sb_info * sbi = EXT2_SB(sb); 486 struct ext2_group_desc * desc; 487 struct ext2_super_block * es = sbi->s_es; 488 unsigned freed = 0, group_freed; 489 490 if (!ext2_data_block_valid(sbi, block, count)) { 491 ext2_error (sb, "ext2_free_blocks", 492 "Freeing blocks not in datazone - " 493 "block = %lu, count = %lu", block, count); 494 goto error_return; 495 } 496 497 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1); 498 499 do_more: 500 overflow = 0; 501 block_group = (block - le32_to_cpu(es->s_first_data_block)) / 502 EXT2_BLOCKS_PER_GROUP(sb); 503 bit = (block - le32_to_cpu(es->s_first_data_block)) % 504 EXT2_BLOCKS_PER_GROUP(sb); 505 /* 506 * Check to see if we are freeing blocks across a group 507 * boundary. 508 */ 509 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) { 510 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb); 511 count -= overflow; 512 } 513 brelse(bitmap_bh); 514 bitmap_bh = read_block_bitmap(sb, block_group); 515 if (!bitmap_bh) 516 goto error_return; 517 518 desc = ext2_get_group_desc (sb, block_group, &bh2); 519 if (!desc) 520 goto error_return; 521 522 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) || 523 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) || 524 in_range (block, le32_to_cpu(desc->bg_inode_table), 525 sbi->s_itb_per_group) || 526 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table), 527 sbi->s_itb_per_group)) { 528 ext2_error (sb, "ext2_free_blocks", 529 "Freeing blocks in system zones - " 530 "Block = %lu, count = %lu", 531 block, count); 532 goto error_return; 533 } 534 535 for (i = 0, group_freed = 0; i < count; i++) { 536 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group), 537 bit + i, bitmap_bh->b_data)) { 538 ext2_error(sb, __func__, 539 "bit already cleared for block %lu", block + i); 540 } else { 541 group_freed++; 542 } 543 } 544 545 mark_buffer_dirty(bitmap_bh); 546 if (sb->s_flags & SB_SYNCHRONOUS) 547 sync_dirty_buffer(bitmap_bh); 548 549 group_adjust_blocks(sb, block_group, desc, bh2, group_freed); 550 freed += group_freed; 551 552 if (overflow) { 553 block += count; 554 count = overflow; 555 goto do_more; 556 } 557 error_return: 558 brelse(bitmap_bh); 559 if (freed) { 560 percpu_counter_add(&sbi->s_freeblocks_counter, freed); 561 dquot_free_block_nodirty(inode, freed); 562 mark_inode_dirty(inode); 563 } 564 } 565 566 /** 567 * bitmap_search_next_usable_block() 568 * @start: the starting block (group relative) of the search 569 * @bh: bufferhead contains the block group bitmap 570 * @maxblocks: the ending block (group relative) of the reservation 571 * 572 * The bitmap search --- search forward through the actual bitmap on disk until 573 * we find a bit free. 574 */ 575 static ext2_grpblk_t 576 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh, 577 ext2_grpblk_t maxblocks) 578 { 579 ext2_grpblk_t next; 580 581 next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start); 582 if (next >= maxblocks) 583 return -1; 584 return next; 585 } 586 587 /** 588 * find_next_usable_block() 589 * @start: the starting block (group relative) to find next 590 * allocatable block in bitmap. 591 * @bh: bufferhead contains the block group bitmap 592 * @maxblocks: the ending block (group relative) for the search 593 * 594 * Find an allocatable block in a bitmap. We perform the "most 595 * appropriate allocation" algorithm of looking for a free block near 596 * the initial goal; then for a free byte somewhere in the bitmap; 597 * then for any free bit in the bitmap. 598 */ 599 static ext2_grpblk_t 600 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks) 601 { 602 ext2_grpblk_t here, next; 603 char *p, *r; 604 605 if (start > 0) { 606 /* 607 * The goal was occupied; search forward for a free 608 * block within the next XX blocks. 609 * 610 * end_goal is more or less random, but it has to be 611 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the 612 * next 64-bit boundary is simple.. 613 */ 614 ext2_grpblk_t end_goal = (start + 63) & ~63; 615 if (end_goal > maxblocks) 616 end_goal = maxblocks; 617 here = ext2_find_next_zero_bit(bh->b_data, end_goal, start); 618 if (here < end_goal) 619 return here; 620 ext2_debug("Bit not found near goal\n"); 621 } 622 623 here = start; 624 if (here < 0) 625 here = 0; 626 627 p = ((char *)bh->b_data) + (here >> 3); 628 r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3)); 629 next = (r - ((char *)bh->b_data)) << 3; 630 631 if (next < maxblocks && next >= here) 632 return next; 633 634 here = bitmap_search_next_usable_block(here, bh, maxblocks); 635 return here; 636 } 637 638 /** 639 * ext2_try_to_allocate() 640 * @sb: superblock 641 * @group: given allocation block group 642 * @bitmap_bh: bufferhead holds the block bitmap 643 * @grp_goal: given target block within the group 644 * @count: target number of blocks to allocate 645 * @my_rsv: reservation window 646 * 647 * Attempt to allocate blocks within a give range. Set the range of allocation 648 * first, then find the first free bit(s) from the bitmap (within the range), 649 * and at last, allocate the blocks by claiming the found free bit as allocated. 650 * 651 * To set the range of this allocation: 652 * if there is a reservation window, only try to allocate block(s) 653 * from the file's own reservation window; 654 * Otherwise, the allocation range starts from the give goal block, 655 * ends at the block group's last block. 656 * 657 * If we failed to allocate the desired block then we may end up crossing to a 658 * new bitmap. 659 */ 660 static int 661 ext2_try_to_allocate(struct super_block *sb, int group, 662 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal, 663 unsigned long *count, 664 struct ext2_reserve_window *my_rsv) 665 { 666 ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group); 667 ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group); 668 ext2_grpblk_t start, end; 669 unsigned long num = 0; 670 671 start = 0; 672 end = group_last_block - group_first_block + 1; 673 /* we do allocation within the reservation window if we have a window */ 674 if (my_rsv) { 675 if (my_rsv->_rsv_start >= group_first_block) 676 start = my_rsv->_rsv_start - group_first_block; 677 if (my_rsv->_rsv_end < group_last_block) 678 end = my_rsv->_rsv_end - group_first_block + 1; 679 if (grp_goal < start || grp_goal >= end) 680 grp_goal = -1; 681 } 682 BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb)); 683 684 if (grp_goal < 0) { 685 grp_goal = find_next_usable_block(start, bitmap_bh, end); 686 if (grp_goal < 0) 687 goto fail_access; 688 if (!my_rsv) { 689 int i; 690 691 for (i = 0; i < 7 && grp_goal > start && 692 !ext2_test_bit(grp_goal - 1, 693 bitmap_bh->b_data); 694 i++, grp_goal--) 695 ; 696 } 697 } 698 699 for (; num < *count && grp_goal < end; grp_goal++) { 700 if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), 701 grp_goal, bitmap_bh->b_data)) { 702 if (num == 0) 703 continue; 704 break; 705 } 706 num++; 707 } 708 709 if (num == 0) 710 goto fail_access; 711 712 *count = num; 713 return grp_goal - num; 714 fail_access: 715 return -1; 716 } 717 718 /** 719 * find_next_reservable_window(): 720 * find a reservable space within the given range. 721 * It does not allocate the reservation window for now: 722 * alloc_new_reservation() will do the work later. 723 * 724 * @search_head: the head of the searching list; 725 * This is not necessarily the list head of the whole filesystem 726 * 727 * We have both head and start_block to assist the search 728 * for the reservable space. The list starts from head, 729 * but we will shift to the place where start_block is, 730 * then start from there, when looking for a reservable space. 731 * 732 * @sb: the super block. 733 * 734 * @start_block: the first block we consider to start the real search from 735 * 736 * @last_block: 737 * the maximum block number that our goal reservable space 738 * could start from. This is normally the last block in this 739 * group. The search will end when we found the start of next 740 * possible reservable space is out of this boundary. 741 * This could handle the cross boundary reservation window 742 * request. 743 * 744 * basically we search from the given range, rather than the whole 745 * reservation double linked list, (start_block, last_block) 746 * to find a free region that is of my size and has not 747 * been reserved. 748 * 749 */ 750 static int find_next_reservable_window( 751 struct ext2_reserve_window_node *search_head, 752 struct ext2_reserve_window_node *my_rsv, 753 struct super_block * sb, 754 ext2_fsblk_t start_block, 755 ext2_fsblk_t last_block) 756 { 757 struct rb_node *next; 758 struct ext2_reserve_window_node *rsv, *prev; 759 ext2_fsblk_t cur; 760 int size = my_rsv->rsv_goal_size; 761 762 /* TODO: make the start of the reservation window byte-aligned */ 763 /* cur = *start_block & ~7;*/ 764 cur = start_block; 765 rsv = search_head; 766 if (!rsv) 767 return -1; 768 769 while (1) { 770 if (cur <= rsv->rsv_end) 771 cur = rsv->rsv_end + 1; 772 773 /* TODO? 774 * in the case we could not find a reservable space 775 * that is what is expected, during the re-search, we could 776 * remember what's the largest reservable space we could have 777 * and return that one. 778 * 779 * For now it will fail if we could not find the reservable 780 * space with expected-size (or more)... 781 */ 782 if (cur > last_block) 783 return -1; /* fail */ 784 785 prev = rsv; 786 next = rb_next(&rsv->rsv_node); 787 rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node); 788 789 /* 790 * Reached the last reservation, we can just append to the 791 * previous one. 792 */ 793 if (!next) 794 break; 795 796 if (cur + size <= rsv->rsv_start) { 797 /* 798 * Found a reserveable space big enough. We could 799 * have a reservation across the group boundary here 800 */ 801 break; 802 } 803 } 804 /* 805 * we come here either : 806 * when we reach the end of the whole list, 807 * and there is empty reservable space after last entry in the list. 808 * append it to the end of the list. 809 * 810 * or we found one reservable space in the middle of the list, 811 * return the reservation window that we could append to. 812 * succeed. 813 */ 814 815 if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) 816 rsv_window_remove(sb, my_rsv); 817 818 /* 819 * Let's book the whole available window for now. We will check the 820 * disk bitmap later and then, if there are free blocks then we adjust 821 * the window size if it's larger than requested. 822 * Otherwise, we will remove this node from the tree next time 823 * call find_next_reservable_window. 824 */ 825 my_rsv->rsv_start = cur; 826 my_rsv->rsv_end = cur + size - 1; 827 my_rsv->rsv_alloc_hit = 0; 828 829 if (prev != my_rsv) 830 ext2_rsv_window_add(sb, my_rsv); 831 832 return 0; 833 } 834 835 /** 836 * alloc_new_reservation()--allocate a new reservation window 837 * 838 * To make a new reservation, we search part of the filesystem 839 * reservation list (the list that inside the group). We try to 840 * allocate a new reservation window near the allocation goal, 841 * or the beginning of the group, if there is no goal. 842 * 843 * We first find a reservable space after the goal, then from 844 * there, we check the bitmap for the first free block after 845 * it. If there is no free block until the end of group, then the 846 * whole group is full, we failed. Otherwise, check if the free 847 * block is inside the expected reservable space, if so, we 848 * succeed. 849 * If the first free block is outside the reservable space, then 850 * start from the first free block, we search for next available 851 * space, and go on. 852 * 853 * on succeed, a new reservation will be found and inserted into the list 854 * It contains at least one free block, and it does not overlap with other 855 * reservation windows. 856 * 857 * failed: we failed to find a reservation window in this group 858 * 859 * @my_rsv: the reservation 860 * 861 * @grp_goal: The goal (group-relative). It is where the search for a 862 * free reservable space should start from. 863 * if we have a goal(goal >0 ), then start from there, 864 * no goal(goal = -1), we start from the first block 865 * of the group. 866 * 867 * @sb: the super block 868 * @group: the group we are trying to allocate in 869 * @bitmap_bh: the block group block bitmap 870 * 871 */ 872 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv, 873 ext2_grpblk_t grp_goal, struct super_block *sb, 874 unsigned int group, struct buffer_head *bitmap_bh) 875 { 876 struct ext2_reserve_window_node *search_head; 877 ext2_fsblk_t group_first_block, group_end_block, start_block; 878 ext2_grpblk_t first_free_block; 879 struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root; 880 unsigned long size; 881 int ret; 882 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock; 883 884 group_first_block = ext2_group_first_block_no(sb, group); 885 group_end_block = ext2_group_last_block_no(sb, group); 886 887 if (grp_goal < 0) 888 start_block = group_first_block; 889 else 890 start_block = grp_goal + group_first_block; 891 892 size = my_rsv->rsv_goal_size; 893 894 if (!rsv_is_empty(&my_rsv->rsv_window)) { 895 /* 896 * if the old reservation is cross group boundary 897 * and if the goal is inside the old reservation window, 898 * we will come here when we just failed to allocate from 899 * the first part of the window. We still have another part 900 * that belongs to the next group. In this case, there is no 901 * point to discard our window and try to allocate a new one 902 * in this group(which will fail). we should 903 * keep the reservation window, just simply move on. 904 * 905 * Maybe we could shift the start block of the reservation 906 * window to the first block of next group. 907 */ 908 909 if ((my_rsv->rsv_start <= group_end_block) && 910 (my_rsv->rsv_end > group_end_block) && 911 (start_block >= my_rsv->rsv_start)) 912 return -1; 913 914 if ((my_rsv->rsv_alloc_hit > 915 (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { 916 /* 917 * if the previously allocation hit ratio is 918 * greater than 1/2, then we double the size of 919 * the reservation window the next time, 920 * otherwise we keep the same size window 921 */ 922 size = size * 2; 923 if (size > EXT2_MAX_RESERVE_BLOCKS) 924 size = EXT2_MAX_RESERVE_BLOCKS; 925 my_rsv->rsv_goal_size= size; 926 } 927 } 928 929 spin_lock(rsv_lock); 930 /* 931 * shift the search start to the window near the goal block 932 */ 933 search_head = search_reserve_window(fs_rsv_root, start_block); 934 935 /* 936 * find_next_reservable_window() simply finds a reservable window 937 * inside the given range(start_block, group_end_block). 938 * 939 * To make sure the reservation window has a free bit inside it, we 940 * need to check the bitmap after we found a reservable window. 941 */ 942 retry: 943 ret = find_next_reservable_window(search_head, my_rsv, sb, 944 start_block, group_end_block); 945 946 if (ret == -1) { 947 if (!rsv_is_empty(&my_rsv->rsv_window)) 948 rsv_window_remove(sb, my_rsv); 949 spin_unlock(rsv_lock); 950 return -1; 951 } 952 953 /* 954 * On success, find_next_reservable_window() returns the 955 * reservation window where there is a reservable space after it. 956 * Before we reserve this reservable space, we need 957 * to make sure there is at least a free block inside this region. 958 * 959 * Search the first free bit on the block bitmap. Search starts from 960 * the start block of the reservable space we just found. 961 */ 962 spin_unlock(rsv_lock); 963 first_free_block = bitmap_search_next_usable_block( 964 my_rsv->rsv_start - group_first_block, 965 bitmap_bh, group_end_block - group_first_block + 1); 966 967 if (first_free_block < 0) { 968 /* 969 * no free block left on the bitmap, no point 970 * to reserve the space. return failed. 971 */ 972 spin_lock(rsv_lock); 973 if (!rsv_is_empty(&my_rsv->rsv_window)) 974 rsv_window_remove(sb, my_rsv); 975 spin_unlock(rsv_lock); 976 return -1; /* failed */ 977 } 978 979 start_block = first_free_block + group_first_block; 980 /* 981 * check if the first free block is within the 982 * free space we just reserved 983 */ 984 if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end) 985 return 0; /* success */ 986 /* 987 * if the first free bit we found is out of the reservable space 988 * continue search for next reservable space, 989 * start from where the free block is, 990 * we also shift the list head to where we stopped last time 991 */ 992 search_head = my_rsv; 993 spin_lock(rsv_lock); 994 goto retry; 995 } 996 997 /** 998 * try_to_extend_reservation() 999 * @my_rsv: given reservation window 1000 * @sb: super block 1001 * @size: the delta to extend 1002 * 1003 * Attempt to expand the reservation window large enough to have 1004 * required number of free blocks 1005 * 1006 * Since ext2_try_to_allocate() will always allocate blocks within 1007 * the reservation window range, if the window size is too small, 1008 * multiple blocks allocation has to stop at the end of the reservation 1009 * window. To make this more efficient, given the total number of 1010 * blocks needed and the current size of the window, we try to 1011 * expand the reservation window size if necessary on a best-effort 1012 * basis before ext2_new_blocks() tries to allocate blocks. 1013 */ 1014 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv, 1015 struct super_block *sb, int size) 1016 { 1017 struct ext2_reserve_window_node *next_rsv; 1018 struct rb_node *next; 1019 spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock; 1020 1021 if (!spin_trylock(rsv_lock)) 1022 return; 1023 1024 next = rb_next(&my_rsv->rsv_node); 1025 1026 if (!next) 1027 my_rsv->rsv_end += size; 1028 else { 1029 next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node); 1030 1031 if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size) 1032 my_rsv->rsv_end += size; 1033 else 1034 my_rsv->rsv_end = next_rsv->rsv_start - 1; 1035 } 1036 spin_unlock(rsv_lock); 1037 } 1038 1039 /** 1040 * ext2_try_to_allocate_with_rsv() 1041 * @sb: superblock 1042 * @group: given allocation block group 1043 * @bitmap_bh: bufferhead holds the block bitmap 1044 * @grp_goal: given target block within the group 1045 * @count: target number of blocks to allocate 1046 * @my_rsv: reservation window 1047 * 1048 * This is the main function used to allocate a new block and its reservation 1049 * window. 1050 * 1051 * Each time when a new block allocation is need, first try to allocate from 1052 * its own reservation. If it does not have a reservation window, instead of 1053 * looking for a free bit on bitmap first, then look up the reservation list to 1054 * see if it is inside somebody else's reservation window, we try to allocate a 1055 * reservation window for it starting from the goal first. Then do the block 1056 * allocation within the reservation window. 1057 * 1058 * This will avoid keeping on searching the reservation list again and 1059 * again when somebody is looking for a free block (without 1060 * reservation), and there are lots of free blocks, but they are all 1061 * being reserved. 1062 * 1063 * We use a red-black tree for the per-filesystem reservation list. 1064 */ 1065 static ext2_grpblk_t 1066 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group, 1067 struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal, 1068 struct ext2_reserve_window_node * my_rsv, 1069 unsigned long *count) 1070 { 1071 ext2_fsblk_t group_first_block, group_last_block; 1072 ext2_grpblk_t ret = 0; 1073 unsigned long num = *count; 1074 1075 /* 1076 * we don't deal with reservation when 1077 * filesystem is mounted without reservation 1078 * or the file is not a regular file 1079 * or last attempt to allocate a block with reservation turned on failed 1080 */ 1081 if (my_rsv == NULL) { 1082 return ext2_try_to_allocate(sb, group, bitmap_bh, 1083 grp_goal, count, NULL); 1084 } 1085 /* 1086 * grp_goal is a group relative block number (if there is a goal) 1087 * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb) 1088 * first block is a filesystem wide block number 1089 * first block is the block number of the first block in this group 1090 */ 1091 group_first_block = ext2_group_first_block_no(sb, group); 1092 group_last_block = ext2_group_last_block_no(sb, group); 1093 1094 /* 1095 * Basically we will allocate a new block from inode's reservation 1096 * window. 1097 * 1098 * We need to allocate a new reservation window, if: 1099 * a) inode does not have a reservation window; or 1100 * b) last attempt to allocate a block from existing reservation 1101 * failed; or 1102 * c) we come here with a goal and with a reservation window 1103 * 1104 * We do not need to allocate a new reservation window if we come here 1105 * at the beginning with a goal and the goal is inside the window, or 1106 * we don't have a goal but already have a reservation window. 1107 * then we could go to allocate from the reservation window directly. 1108 */ 1109 while (1) { 1110 if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || 1111 !goal_in_my_reservation(&my_rsv->rsv_window, 1112 grp_goal, group, sb)) { 1113 if (my_rsv->rsv_goal_size < *count) 1114 my_rsv->rsv_goal_size = *count; 1115 ret = alloc_new_reservation(my_rsv, grp_goal, sb, 1116 group, bitmap_bh); 1117 if (ret < 0) 1118 break; /* failed */ 1119 1120 if (!goal_in_my_reservation(&my_rsv->rsv_window, 1121 grp_goal, group, sb)) 1122 grp_goal = -1; 1123 } else if (grp_goal >= 0) { 1124 int curr = my_rsv->rsv_end - 1125 (grp_goal + group_first_block) + 1; 1126 1127 if (curr < *count) 1128 try_to_extend_reservation(my_rsv, sb, 1129 *count - curr); 1130 } 1131 1132 if ((my_rsv->rsv_start > group_last_block) || 1133 (my_rsv->rsv_end < group_first_block)) { 1134 rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1); 1135 BUG(); 1136 } 1137 ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal, 1138 &num, &my_rsv->rsv_window); 1139 if (ret >= 0) { 1140 my_rsv->rsv_alloc_hit += num; 1141 *count = num; 1142 break; /* succeed */ 1143 } 1144 num = *count; 1145 } 1146 return ret; 1147 } 1148 1149 /** 1150 * ext2_has_free_blocks() 1151 * @sbi: in-core super block structure. 1152 * 1153 * Check if filesystem has at least 1 free block available for allocation. 1154 */ 1155 static int ext2_has_free_blocks(struct ext2_sb_info *sbi) 1156 { 1157 ext2_fsblk_t free_blocks, root_blocks; 1158 1159 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); 1160 root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count); 1161 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && 1162 !uid_eq(sbi->s_resuid, current_fsuid()) && 1163 (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) || 1164 !in_group_p (sbi->s_resgid))) { 1165 return 0; 1166 } 1167 return 1; 1168 } 1169 1170 /* 1171 * Returns 1 if the passed-in block region is valid; 0 if some part overlaps 1172 * with filesystem metadata blocks. 1173 */ 1174 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk, 1175 unsigned int count) 1176 { 1177 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 1178 (start_blk + count - 1 < start_blk) || 1179 (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count))) 1180 return 0; 1181 1182 /* Ensure we do not step over superblock */ 1183 if ((start_blk <= sbi->s_sb_block) && 1184 (start_blk + count - 1 >= sbi->s_sb_block)) 1185 return 0; 1186 1187 return 1; 1188 } 1189 1190 /* 1191 * ext2_new_blocks() -- core block(s) allocation function 1192 * @inode: file inode 1193 * @goal: given target block(filesystem wide) 1194 * @count: target number of blocks to allocate 1195 * @errp: error code 1196 * 1197 * ext2_new_blocks uses a goal block to assist allocation. If the goal is 1198 * free, or there is a free block within 32 blocks of the goal, that block 1199 * is allocated. Otherwise a forward search is made for a free block; within 1200 * each block group the search first looks for an entire free byte in the block 1201 * bitmap, and then for any free bit if that fails. 1202 * This function also updates quota and i_blocks field. 1203 */ 1204 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal, 1205 unsigned long *count, int *errp) 1206 { 1207 struct buffer_head *bitmap_bh = NULL; 1208 struct buffer_head *gdp_bh; 1209 int group_no; 1210 int goal_group; 1211 ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */ 1212 ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/ 1213 ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */ 1214 int bgi; /* blockgroup iteration index */ 1215 int performed_allocation = 0; 1216 ext2_grpblk_t free_blocks; /* number of free blocks in a group */ 1217 struct super_block *sb; 1218 struct ext2_group_desc *gdp; 1219 struct ext2_super_block *es; 1220 struct ext2_sb_info *sbi; 1221 struct ext2_reserve_window_node *my_rsv = NULL; 1222 struct ext2_block_alloc_info *block_i; 1223 unsigned short windowsz = 0; 1224 unsigned long ngroups; 1225 unsigned long num = *count; 1226 int ret; 1227 1228 *errp = -ENOSPC; 1229 sb = inode->i_sb; 1230 1231 /* 1232 * Check quota for allocation of this block. 1233 */ 1234 ret = dquot_alloc_block(inode, num); 1235 if (ret) { 1236 *errp = ret; 1237 return 0; 1238 } 1239 1240 sbi = EXT2_SB(sb); 1241 es = EXT2_SB(sb)->s_es; 1242 ext2_debug("goal=%lu.\n", goal); 1243 /* 1244 * Allocate a block from reservation only when 1245 * filesystem is mounted with reservation(default,-o reservation), and 1246 * it's a regular file, and 1247 * the desired window size is greater than 0 (One could use ioctl 1248 * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off 1249 * reservation on that particular file) 1250 */ 1251 block_i = EXT2_I(inode)->i_block_alloc_info; 1252 if (block_i) { 1253 windowsz = block_i->rsv_window_node.rsv_goal_size; 1254 if (windowsz > 0) 1255 my_rsv = &block_i->rsv_window_node; 1256 } 1257 1258 if (!ext2_has_free_blocks(sbi)) { 1259 *errp = -ENOSPC; 1260 goto out; 1261 } 1262 1263 /* 1264 * First, test whether the goal block is free. 1265 */ 1266 if (goal < le32_to_cpu(es->s_first_data_block) || 1267 goal >= le32_to_cpu(es->s_blocks_count)) 1268 goal = le32_to_cpu(es->s_first_data_block); 1269 group_no = (goal - le32_to_cpu(es->s_first_data_block)) / 1270 EXT2_BLOCKS_PER_GROUP(sb); 1271 goal_group = group_no; 1272 retry_alloc: 1273 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh); 1274 if (!gdp) 1275 goto io_error; 1276 1277 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); 1278 /* 1279 * if there is not enough free blocks to make a new resevation 1280 * turn off reservation for this allocation 1281 */ 1282 if (my_rsv && (free_blocks < windowsz) 1283 && (free_blocks > 0) 1284 && (rsv_is_empty(&my_rsv->rsv_window))) 1285 my_rsv = NULL; 1286 1287 if (free_blocks > 0) { 1288 grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) % 1289 EXT2_BLOCKS_PER_GROUP(sb)); 1290 /* 1291 * In case we retry allocation (due to fs reservation not 1292 * working out or fs corruption), the bitmap_bh is non-null 1293 * pointer and we have to release it before calling 1294 * read_block_bitmap(). 1295 */ 1296 brelse(bitmap_bh); 1297 bitmap_bh = read_block_bitmap(sb, group_no); 1298 if (!bitmap_bh) 1299 goto io_error; 1300 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no, 1301 bitmap_bh, grp_target_blk, 1302 my_rsv, &num); 1303 if (grp_alloc_blk >= 0) 1304 goto allocated; 1305 } 1306 1307 ngroups = EXT2_SB(sb)->s_groups_count; 1308 smp_rmb(); 1309 1310 /* 1311 * Now search the rest of the groups. We assume that 1312 * group_no and gdp correctly point to the last group visited. 1313 */ 1314 for (bgi = 0; bgi < ngroups; bgi++) { 1315 group_no++; 1316 if (group_no >= ngroups) 1317 group_no = 0; 1318 gdp = ext2_get_group_desc(sb, group_no, &gdp_bh); 1319 if (!gdp) 1320 goto io_error; 1321 1322 free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); 1323 /* 1324 * skip this group (and avoid loading bitmap) if there 1325 * are no free blocks 1326 */ 1327 if (!free_blocks) 1328 continue; 1329 /* 1330 * skip this group if the number of 1331 * free blocks is less than half of the reservation 1332 * window size. 1333 */ 1334 if (my_rsv && (free_blocks <= (windowsz/2))) 1335 continue; 1336 1337 brelse(bitmap_bh); 1338 bitmap_bh = read_block_bitmap(sb, group_no); 1339 if (!bitmap_bh) 1340 goto io_error; 1341 /* 1342 * try to allocate block(s) from this group, without a goal(-1). 1343 */ 1344 grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no, 1345 bitmap_bh, -1, my_rsv, &num); 1346 if (grp_alloc_blk >= 0) 1347 goto allocated; 1348 } 1349 /* 1350 * We may end up a bogus earlier ENOSPC error due to 1351 * filesystem is "full" of reservations, but 1352 * there maybe indeed free blocks available on disk 1353 * In this case, we just forget about the reservations 1354 * just do block allocation as without reservations. 1355 */ 1356 if (my_rsv) { 1357 my_rsv = NULL; 1358 windowsz = 0; 1359 group_no = goal_group; 1360 goto retry_alloc; 1361 } 1362 /* No space left on the device */ 1363 *errp = -ENOSPC; 1364 goto out; 1365 1366 allocated: 1367 1368 ext2_debug("using block group %d(%d)\n", 1369 group_no, gdp->bg_free_blocks_count); 1370 1371 ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no); 1372 1373 if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) || 1374 in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) || 1375 in_range(ret_block, le32_to_cpu(gdp->bg_inode_table), 1376 EXT2_SB(sb)->s_itb_per_group) || 1377 in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table), 1378 EXT2_SB(sb)->s_itb_per_group)) { 1379 ext2_error(sb, "ext2_new_blocks", 1380 "Allocating block in system zone - " 1381 "blocks from "E2FSBLK", length %lu", 1382 ret_block, num); 1383 /* 1384 * ext2_try_to_allocate marked the blocks we allocated as in 1385 * use. So we may want to selectively mark some of the blocks 1386 * as free 1387 */ 1388 num = *count; 1389 goto retry_alloc; 1390 } 1391 1392 performed_allocation = 1; 1393 1394 if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) { 1395 ext2_error(sb, "ext2_new_blocks", 1396 "block("E2FSBLK") >= blocks count(%d) - " 1397 "block_group = %d, es == %p ", ret_block, 1398 le32_to_cpu(es->s_blocks_count), group_no, es); 1399 goto out; 1400 } 1401 1402 group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num); 1403 percpu_counter_sub(&sbi->s_freeblocks_counter, num); 1404 1405 mark_buffer_dirty(bitmap_bh); 1406 if (sb->s_flags & SB_SYNCHRONOUS) 1407 sync_dirty_buffer(bitmap_bh); 1408 1409 *errp = 0; 1410 brelse(bitmap_bh); 1411 if (num < *count) { 1412 dquot_free_block_nodirty(inode, *count-num); 1413 mark_inode_dirty(inode); 1414 *count = num; 1415 } 1416 return ret_block; 1417 1418 io_error: 1419 *errp = -EIO; 1420 out: 1421 /* 1422 * Undo the block allocation 1423 */ 1424 if (!performed_allocation) { 1425 dquot_free_block_nodirty(inode, *count); 1426 mark_inode_dirty(inode); 1427 } 1428 brelse(bitmap_bh); 1429 return 0; 1430 } 1431 1432 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp) 1433 { 1434 unsigned long count = 1; 1435 1436 return ext2_new_blocks(inode, goal, &count, errp); 1437 } 1438 1439 #ifdef EXT2FS_DEBUG 1440 1441 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars) 1442 { 1443 return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars); 1444 } 1445 1446 #endif /* EXT2FS_DEBUG */ 1447 1448 unsigned long ext2_count_free_blocks (struct super_block * sb) 1449 { 1450 struct ext2_group_desc * desc; 1451 unsigned long desc_count = 0; 1452 int i; 1453 #ifdef EXT2FS_DEBUG 1454 unsigned long bitmap_count, x; 1455 struct ext2_super_block *es; 1456 1457 es = EXT2_SB(sb)->s_es; 1458 desc_count = 0; 1459 bitmap_count = 0; 1460 desc = NULL; 1461 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 1462 struct buffer_head *bitmap_bh; 1463 desc = ext2_get_group_desc (sb, i, NULL); 1464 if (!desc) 1465 continue; 1466 desc_count += le16_to_cpu(desc->bg_free_blocks_count); 1467 bitmap_bh = read_block_bitmap(sb, i); 1468 if (!bitmap_bh) 1469 continue; 1470 1471 x = ext2_count_free(bitmap_bh, sb->s_blocksize); 1472 printk ("group %d: stored = %d, counted = %lu\n", 1473 i, le16_to_cpu(desc->bg_free_blocks_count), x); 1474 bitmap_count += x; 1475 brelse(bitmap_bh); 1476 } 1477 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n", 1478 (long)le32_to_cpu(es->s_free_blocks_count), 1479 desc_count, bitmap_count); 1480 return bitmap_count; 1481 #else 1482 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 1483 desc = ext2_get_group_desc(sb, i, NULL); 1484 if (!desc) 1485 continue; 1486 desc_count += le16_to_cpu(desc->bg_free_blocks_count); 1487 } 1488 return desc_count; 1489 #endif 1490 } 1491 1492 static inline int test_root(int a, int b) 1493 { 1494 int num = b; 1495 1496 while (a > num) 1497 num *= b; 1498 return num == a; 1499 } 1500 1501 static int ext2_group_sparse(int group) 1502 { 1503 if (group <= 1) 1504 return 1; 1505 return (test_root(group, 3) || test_root(group, 5) || 1506 test_root(group, 7)); 1507 } 1508 1509 /** 1510 * ext2_bg_has_super - number of blocks used by the superblock in group 1511 * @sb: superblock for filesystem 1512 * @group: group number to check 1513 * 1514 * Return the number of blocks used by the superblock (primary or backup) 1515 * in this group. Currently this will be only 0 or 1. 1516 */ 1517 int ext2_bg_has_super(struct super_block *sb, int group) 1518 { 1519 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&& 1520 !ext2_group_sparse(group)) 1521 return 0; 1522 return 1; 1523 } 1524 1525 /** 1526 * ext2_bg_num_gdb - number of blocks used by the group table in group 1527 * @sb: superblock for filesystem 1528 * @group: group number to check 1529 * 1530 * Return the number of blocks used by the group descriptor table 1531 * (primary or backup) in this group. In the future there may be a 1532 * different number of descriptor blocks in each group. 1533 */ 1534 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group) 1535 { 1536 return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0; 1537 } 1538 1539