Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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/config.h> |
| 15 | #include <linux/time.h> |
Randy Dunlap | 16f7e0f | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 16 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/fs.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 | |
| 24 | /* |
| 25 | * balloc.c contains the blocks allocation and deallocation routines |
| 26 | */ |
| 27 | |
| 28 | /* |
| 29 | * The free blocks are managed by bitmaps. A file system contains several |
| 30 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap |
| 31 | * block for inodes, N blocks for the inode table and data blocks. |
| 32 | * |
| 33 | * The file system contains group descriptors which are located after the |
| 34 | * super block. Each descriptor contains the number of the bitmap block and |
| 35 | * the free blocks count in the block. The descriptors are loaded in memory |
| 36 | * when a file system is mounted (see ext3_read_super). |
| 37 | */ |
| 38 | |
| 39 | |
| 40 | #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) |
| 41 | |
| 42 | struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb, |
| 43 | unsigned int block_group, |
| 44 | struct buffer_head ** bh) |
| 45 | { |
| 46 | unsigned long group_desc; |
| 47 | unsigned long offset; |
| 48 | struct ext3_group_desc * desc; |
| 49 | struct ext3_sb_info *sbi = EXT3_SB(sb); |
| 50 | |
| 51 | if (block_group >= sbi->s_groups_count) { |
| 52 | ext3_error (sb, "ext3_get_group_desc", |
| 53 | "block_group >= groups_count - " |
| 54 | "block_group = %d, groups_count = %lu", |
| 55 | block_group, sbi->s_groups_count); |
| 56 | |
| 57 | return NULL; |
| 58 | } |
| 59 | smp_rmb(); |
| 60 | |
| 61 | group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb); |
| 62 | offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1); |
| 63 | if (!sbi->s_group_desc[group_desc]) { |
| 64 | ext3_error (sb, "ext3_get_group_desc", |
| 65 | "Group descriptor not loaded - " |
| 66 | "block_group = %d, group_desc = %lu, desc = %lu", |
| 67 | block_group, group_desc, offset); |
| 68 | return NULL; |
| 69 | } |
| 70 | |
| 71 | desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data; |
| 72 | if (bh) |
| 73 | *bh = sbi->s_group_desc[group_desc]; |
| 74 | return desc + offset; |
| 75 | } |
| 76 | |
| 77 | /* |
| 78 | * Read the bitmap for a given block_group, reading into the specified |
| 79 | * slot in the superblock's bitmap cache. |
| 80 | * |
| 81 | * Return buffer_head on success or NULL in case of failure. |
| 82 | */ |
| 83 | static struct buffer_head * |
| 84 | read_block_bitmap(struct super_block *sb, unsigned int block_group) |
| 85 | { |
| 86 | struct ext3_group_desc * desc; |
| 87 | struct buffer_head * bh = NULL; |
| 88 | |
| 89 | desc = ext3_get_group_desc (sb, block_group, NULL); |
| 90 | if (!desc) |
| 91 | goto error_out; |
| 92 | bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap)); |
| 93 | if (!bh) |
| 94 | ext3_error (sb, "read_block_bitmap", |
| 95 | "Cannot read block bitmap - " |
| 96 | "block_group = %d, block_bitmap = %u", |
| 97 | block_group, le32_to_cpu(desc->bg_block_bitmap)); |
| 98 | error_out: |
| 99 | return bh; |
| 100 | } |
| 101 | /* |
| 102 | * The reservation window structure operations |
| 103 | * -------------------------------------------- |
| 104 | * Operations include: |
| 105 | * dump, find, add, remove, is_empty, find_next_reservable_window, etc. |
| 106 | * |
| 107 | * We use sorted double linked list for the per-filesystem reservation |
| 108 | * window list. (like in vm_region). |
| 109 | * |
| 110 | * Initially, we keep those small operations in the abstract functions, |
| 111 | * so later if we need a better searching tree than double linked-list, |
| 112 | * we could easily switch to that without changing too much |
| 113 | * code. |
| 114 | */ |
| 115 | #if 0 |
| 116 | static void __rsv_window_dump(struct rb_root *root, int verbose, |
| 117 | const char *fn) |
| 118 | { |
| 119 | struct rb_node *n; |
| 120 | struct ext3_reserve_window_node *rsv, *prev; |
| 121 | int bad; |
| 122 | |
| 123 | restart: |
| 124 | n = rb_first(root); |
| 125 | bad = 0; |
| 126 | prev = NULL; |
| 127 | |
| 128 | printk("Block Allocation Reservation Windows Map (%s):\n", fn); |
| 129 | while (n) { |
| 130 | rsv = list_entry(n, struct ext3_reserve_window_node, rsv_node); |
| 131 | if (verbose) |
| 132 | printk("reservation window 0x%p " |
| 133 | "start: %d, end: %d\n", |
| 134 | rsv, rsv->rsv_start, rsv->rsv_end); |
| 135 | if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) { |
| 136 | printk("Bad reservation %p (start >= end)\n", |
| 137 | rsv); |
| 138 | bad = 1; |
| 139 | } |
| 140 | if (prev && prev->rsv_end >= rsv->rsv_start) { |
| 141 | printk("Bad reservation %p (prev->end >= start)\n", |
| 142 | rsv); |
| 143 | bad = 1; |
| 144 | } |
| 145 | if (bad) { |
| 146 | if (!verbose) { |
| 147 | printk("Restarting reservation walk in verbose mode\n"); |
| 148 | verbose = 1; |
| 149 | goto restart; |
| 150 | } |
| 151 | } |
| 152 | n = rb_next(n); |
| 153 | prev = rsv; |
| 154 | } |
| 155 | printk("Window map complete.\n"); |
| 156 | if (bad) |
| 157 | BUG(); |
| 158 | } |
| 159 | #define rsv_window_dump(root, verbose) \ |
| 160 | __rsv_window_dump((root), (verbose), __FUNCTION__) |
| 161 | #else |
| 162 | #define rsv_window_dump(root, verbose) do {} while (0) |
| 163 | #endif |
| 164 | |
| 165 | static int |
| 166 | goal_in_my_reservation(struct ext3_reserve_window *rsv, int goal, |
| 167 | unsigned int group, struct super_block * sb) |
| 168 | { |
| 169 | unsigned long group_first_block, group_last_block; |
| 170 | |
| 171 | group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) + |
| 172 | group * EXT3_BLOCKS_PER_GROUP(sb); |
| 173 | group_last_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1; |
| 174 | |
| 175 | if ((rsv->_rsv_start > group_last_block) || |
| 176 | (rsv->_rsv_end < group_first_block)) |
| 177 | return 0; |
| 178 | if ((goal >= 0) && ((goal + group_first_block < rsv->_rsv_start) |
| 179 | || (goal + group_first_block > rsv->_rsv_end))) |
| 180 | return 0; |
| 181 | return 1; |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * Find the reserved window which includes the goal, or the previous one |
| 186 | * if the goal is not in any window. |
| 187 | * Returns NULL if there are no windows or if all windows start after the goal. |
| 188 | */ |
| 189 | static struct ext3_reserve_window_node * |
| 190 | search_reserve_window(struct rb_root *root, unsigned long goal) |
| 191 | { |
| 192 | struct rb_node *n = root->rb_node; |
| 193 | struct ext3_reserve_window_node *rsv; |
| 194 | |
| 195 | if (!n) |
| 196 | return NULL; |
| 197 | |
| 198 | do { |
| 199 | rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node); |
| 200 | |
| 201 | if (goal < rsv->rsv_start) |
| 202 | n = n->rb_left; |
| 203 | else if (goal > rsv->rsv_end) |
| 204 | n = n->rb_right; |
| 205 | else |
| 206 | return rsv; |
| 207 | } while (n); |
| 208 | /* |
| 209 | * We've fallen off the end of the tree: the goal wasn't inside |
| 210 | * any particular node. OK, the previous node must be to one |
| 211 | * side of the interval containing the goal. If it's the RHS, |
| 212 | * we need to back up one. |
| 213 | */ |
| 214 | if (rsv->rsv_start > goal) { |
| 215 | n = rb_prev(&rsv->rsv_node); |
| 216 | rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node); |
| 217 | } |
| 218 | return rsv; |
| 219 | } |
| 220 | |
| 221 | void ext3_rsv_window_add(struct super_block *sb, |
| 222 | struct ext3_reserve_window_node *rsv) |
| 223 | { |
| 224 | struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root; |
| 225 | struct rb_node *node = &rsv->rsv_node; |
| 226 | unsigned int start = rsv->rsv_start; |
| 227 | |
| 228 | struct rb_node ** p = &root->rb_node; |
| 229 | struct rb_node * parent = NULL; |
| 230 | struct ext3_reserve_window_node *this; |
| 231 | |
| 232 | while (*p) |
| 233 | { |
| 234 | parent = *p; |
| 235 | this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node); |
| 236 | |
| 237 | if (start < this->rsv_start) |
| 238 | p = &(*p)->rb_left; |
| 239 | else if (start > this->rsv_end) |
| 240 | p = &(*p)->rb_right; |
| 241 | else |
| 242 | BUG(); |
| 243 | } |
| 244 | |
| 245 | rb_link_node(node, parent, p); |
| 246 | rb_insert_color(node, root); |
| 247 | } |
| 248 | |
| 249 | static void rsv_window_remove(struct super_block *sb, |
| 250 | struct ext3_reserve_window_node *rsv) |
| 251 | { |
| 252 | rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED; |
| 253 | rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED; |
| 254 | rsv->rsv_alloc_hit = 0; |
| 255 | rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root); |
| 256 | } |
| 257 | |
| 258 | static inline int rsv_is_empty(struct ext3_reserve_window *rsv) |
| 259 | { |
| 260 | /* a valid reservation end block could not be 0 */ |
| 261 | return (rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED); |
| 262 | } |
| 263 | void ext3_init_block_alloc_info(struct inode *inode) |
| 264 | { |
| 265 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 266 | struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info; |
| 267 | struct super_block *sb = inode->i_sb; |
| 268 | |
| 269 | block_i = kmalloc(sizeof(*block_i), GFP_NOFS); |
| 270 | if (block_i) { |
| 271 | struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node; |
| 272 | |
| 273 | rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED; |
| 274 | rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED; |
| 275 | |
| 276 | /* |
| 277 | * if filesystem is mounted with NORESERVATION, the goal |
| 278 | * reservation window size is set to zero to indicate |
| 279 | * block reservation is off |
| 280 | */ |
| 281 | if (!test_opt(sb, RESERVATION)) |
| 282 | rsv->rsv_goal_size = 0; |
| 283 | else |
| 284 | rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS; |
| 285 | rsv->rsv_alloc_hit = 0; |
| 286 | block_i->last_alloc_logical_block = 0; |
| 287 | block_i->last_alloc_physical_block = 0; |
| 288 | } |
| 289 | ei->i_block_alloc_info = block_i; |
| 290 | } |
| 291 | |
| 292 | void ext3_discard_reservation(struct inode *inode) |
| 293 | { |
| 294 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 295 | struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info; |
| 296 | struct ext3_reserve_window_node *rsv; |
| 297 | spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock; |
| 298 | |
| 299 | if (!block_i) |
| 300 | return; |
| 301 | |
| 302 | rsv = &block_i->rsv_window_node; |
| 303 | if (!rsv_is_empty(&rsv->rsv_window)) { |
| 304 | spin_lock(rsv_lock); |
| 305 | if (!rsv_is_empty(&rsv->rsv_window)) |
| 306 | rsv_window_remove(inode->i_sb, rsv); |
| 307 | spin_unlock(rsv_lock); |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | /* Free given blocks, update quota and i_blocks field */ |
| 312 | void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb, |
| 313 | unsigned long block, unsigned long count, |
| 314 | int *pdquot_freed_blocks) |
| 315 | { |
| 316 | struct buffer_head *bitmap_bh = NULL; |
| 317 | struct buffer_head *gd_bh; |
| 318 | unsigned long block_group; |
| 319 | unsigned long bit; |
| 320 | unsigned long i; |
| 321 | unsigned long overflow; |
| 322 | struct ext3_group_desc * desc; |
| 323 | struct ext3_super_block * es; |
| 324 | struct ext3_sb_info *sbi; |
| 325 | int err = 0, ret; |
| 326 | unsigned group_freed; |
| 327 | |
| 328 | *pdquot_freed_blocks = 0; |
| 329 | sbi = EXT3_SB(sb); |
| 330 | es = sbi->s_es; |
| 331 | if (block < le32_to_cpu(es->s_first_data_block) || |
| 332 | block + count < block || |
| 333 | block + count > le32_to_cpu(es->s_blocks_count)) { |
| 334 | ext3_error (sb, "ext3_free_blocks", |
| 335 | "Freeing blocks not in datazone - " |
| 336 | "block = %lu, count = %lu", block, count); |
| 337 | goto error_return; |
| 338 | } |
| 339 | |
| 340 | ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1); |
| 341 | |
| 342 | do_more: |
| 343 | overflow = 0; |
| 344 | block_group = (block - le32_to_cpu(es->s_first_data_block)) / |
| 345 | EXT3_BLOCKS_PER_GROUP(sb); |
| 346 | bit = (block - le32_to_cpu(es->s_first_data_block)) % |
| 347 | EXT3_BLOCKS_PER_GROUP(sb); |
| 348 | /* |
| 349 | * Check to see if we are freeing blocks across a group |
| 350 | * boundary. |
| 351 | */ |
| 352 | if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) { |
| 353 | overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb); |
| 354 | count -= overflow; |
| 355 | } |
| 356 | brelse(bitmap_bh); |
| 357 | bitmap_bh = read_block_bitmap(sb, block_group); |
| 358 | if (!bitmap_bh) |
| 359 | goto error_return; |
| 360 | desc = ext3_get_group_desc (sb, block_group, &gd_bh); |
| 361 | if (!desc) |
| 362 | goto error_return; |
| 363 | |
| 364 | if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) || |
| 365 | in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) || |
| 366 | in_range (block, le32_to_cpu(desc->bg_inode_table), |
| 367 | sbi->s_itb_per_group) || |
| 368 | in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table), |
| 369 | sbi->s_itb_per_group)) |
| 370 | ext3_error (sb, "ext3_free_blocks", |
| 371 | "Freeing blocks in system zones - " |
| 372 | "Block = %lu, count = %lu", |
| 373 | block, count); |
| 374 | |
| 375 | /* |
| 376 | * We are about to start releasing blocks in the bitmap, |
| 377 | * so we need undo access. |
| 378 | */ |
| 379 | /* @@@ check errors */ |
| 380 | BUFFER_TRACE(bitmap_bh, "getting undo access"); |
| 381 | err = ext3_journal_get_undo_access(handle, bitmap_bh); |
| 382 | if (err) |
| 383 | goto error_return; |
| 384 | |
| 385 | /* |
| 386 | * We are about to modify some metadata. Call the journal APIs |
| 387 | * to unshare ->b_data if a currently-committing transaction is |
| 388 | * using it |
| 389 | */ |
| 390 | BUFFER_TRACE(gd_bh, "get_write_access"); |
| 391 | err = ext3_journal_get_write_access(handle, gd_bh); |
| 392 | if (err) |
| 393 | goto error_return; |
| 394 | |
| 395 | jbd_lock_bh_state(bitmap_bh); |
| 396 | |
| 397 | for (i = 0, group_freed = 0; i < count; i++) { |
| 398 | /* |
| 399 | * An HJ special. This is expensive... |
| 400 | */ |
| 401 | #ifdef CONFIG_JBD_DEBUG |
| 402 | jbd_unlock_bh_state(bitmap_bh); |
| 403 | { |
| 404 | struct buffer_head *debug_bh; |
| 405 | debug_bh = sb_find_get_block(sb, block + i); |
| 406 | if (debug_bh) { |
| 407 | BUFFER_TRACE(debug_bh, "Deleted!"); |
| 408 | if (!bh2jh(bitmap_bh)->b_committed_data) |
| 409 | BUFFER_TRACE(debug_bh, |
| 410 | "No commited data in bitmap"); |
| 411 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); |
| 412 | __brelse(debug_bh); |
| 413 | } |
| 414 | } |
| 415 | jbd_lock_bh_state(bitmap_bh); |
| 416 | #endif |
| 417 | if (need_resched()) { |
| 418 | jbd_unlock_bh_state(bitmap_bh); |
| 419 | cond_resched(); |
| 420 | jbd_lock_bh_state(bitmap_bh); |
| 421 | } |
| 422 | /* @@@ This prevents newly-allocated data from being |
| 423 | * freed and then reallocated within the same |
| 424 | * transaction. |
| 425 | * |
| 426 | * Ideally we would want to allow that to happen, but to |
| 427 | * do so requires making journal_forget() capable of |
| 428 | * revoking the queued write of a data block, which |
| 429 | * implies blocking on the journal lock. *forget() |
| 430 | * cannot block due to truncate races. |
| 431 | * |
| 432 | * Eventually we can fix this by making journal_forget() |
| 433 | * return a status indicating whether or not it was able |
| 434 | * to revoke the buffer. On successful revoke, it is |
| 435 | * safe not to set the allocation bit in the committed |
| 436 | * bitmap, because we know that there is no outstanding |
| 437 | * activity on the buffer any more and so it is safe to |
| 438 | * reallocate it. |
| 439 | */ |
| 440 | BUFFER_TRACE(bitmap_bh, "set in b_committed_data"); |
| 441 | J_ASSERT_BH(bitmap_bh, |
| 442 | bh2jh(bitmap_bh)->b_committed_data != NULL); |
| 443 | ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i, |
| 444 | bh2jh(bitmap_bh)->b_committed_data); |
| 445 | |
| 446 | /* |
| 447 | * We clear the bit in the bitmap after setting the committed |
| 448 | * data bit, because this is the reverse order to that which |
| 449 | * the allocator uses. |
| 450 | */ |
| 451 | BUFFER_TRACE(bitmap_bh, "clear bit"); |
| 452 | if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
| 453 | bit + i, bitmap_bh->b_data)) { |
| 454 | jbd_unlock_bh_state(bitmap_bh); |
| 455 | ext3_error(sb, __FUNCTION__, |
| 456 | "bit already cleared for block %lu", block + i); |
| 457 | jbd_lock_bh_state(bitmap_bh); |
| 458 | BUFFER_TRACE(bitmap_bh, "bit already cleared"); |
| 459 | } else { |
| 460 | group_freed++; |
| 461 | } |
| 462 | } |
| 463 | jbd_unlock_bh_state(bitmap_bh); |
| 464 | |
| 465 | spin_lock(sb_bgl_lock(sbi, block_group)); |
| 466 | desc->bg_free_blocks_count = |
| 467 | cpu_to_le16(le16_to_cpu(desc->bg_free_blocks_count) + |
| 468 | group_freed); |
| 469 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
| 470 | percpu_counter_mod(&sbi->s_freeblocks_counter, count); |
| 471 | |
| 472 | /* We dirtied the bitmap block */ |
| 473 | BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); |
| 474 | err = ext3_journal_dirty_metadata(handle, bitmap_bh); |
| 475 | |
| 476 | /* And the group descriptor block */ |
| 477 | BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); |
| 478 | ret = ext3_journal_dirty_metadata(handle, gd_bh); |
| 479 | if (!err) err = ret; |
| 480 | *pdquot_freed_blocks += group_freed; |
| 481 | |
| 482 | if (overflow && !err) { |
| 483 | block += count; |
| 484 | count = overflow; |
| 485 | goto do_more; |
| 486 | } |
| 487 | sb->s_dirt = 1; |
| 488 | error_return: |
| 489 | brelse(bitmap_bh); |
| 490 | ext3_std_error(sb, err); |
| 491 | return; |
| 492 | } |
| 493 | |
| 494 | /* Free given blocks, update quota and i_blocks field */ |
| 495 | void ext3_free_blocks(handle_t *handle, struct inode *inode, |
| 496 | unsigned long block, unsigned long count) |
| 497 | { |
| 498 | struct super_block * sb; |
| 499 | int dquot_freed_blocks; |
| 500 | |
| 501 | sb = inode->i_sb; |
| 502 | if (!sb) { |
| 503 | printk ("ext3_free_blocks: nonexistent device"); |
| 504 | return; |
| 505 | } |
| 506 | ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks); |
| 507 | if (dquot_freed_blocks) |
| 508 | DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); |
| 509 | return; |
| 510 | } |
| 511 | |
| 512 | /* |
| 513 | * For ext3 allocations, we must not reuse any blocks which are |
| 514 | * allocated in the bitmap buffer's "last committed data" copy. This |
| 515 | * prevents deletes from freeing up the page for reuse until we have |
| 516 | * committed the delete transaction. |
| 517 | * |
| 518 | * If we didn't do this, then deleting something and reallocating it as |
| 519 | * data would allow the old block to be overwritten before the |
| 520 | * transaction committed (because we force data to disk before commit). |
| 521 | * This would lead to corruption if we crashed between overwriting the |
| 522 | * data and committing the delete. |
| 523 | * |
| 524 | * @@@ We may want to make this allocation behaviour conditional on |
| 525 | * data-writes at some point, and disable it for metadata allocations or |
| 526 | * sync-data inodes. |
| 527 | */ |
| 528 | static int ext3_test_allocatable(int nr, struct buffer_head *bh) |
| 529 | { |
| 530 | int ret; |
| 531 | struct journal_head *jh = bh2jh(bh); |
| 532 | |
| 533 | if (ext3_test_bit(nr, bh->b_data)) |
| 534 | return 0; |
| 535 | |
| 536 | jbd_lock_bh_state(bh); |
| 537 | if (!jh->b_committed_data) |
| 538 | ret = 1; |
| 539 | else |
| 540 | ret = !ext3_test_bit(nr, jh->b_committed_data); |
| 541 | jbd_unlock_bh_state(bh); |
| 542 | return ret; |
| 543 | } |
| 544 | |
| 545 | static int |
| 546 | bitmap_search_next_usable_block(int start, struct buffer_head *bh, |
| 547 | int maxblocks) |
| 548 | { |
| 549 | int next; |
| 550 | struct journal_head *jh = bh2jh(bh); |
| 551 | |
| 552 | /* |
| 553 | * The bitmap search --- search forward alternately through the actual |
| 554 | * bitmap and the last-committed copy until we find a bit free in |
| 555 | * both |
| 556 | */ |
| 557 | while (start < maxblocks) { |
| 558 | next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start); |
| 559 | if (next >= maxblocks) |
| 560 | return -1; |
| 561 | if (ext3_test_allocatable(next, bh)) |
| 562 | return next; |
| 563 | jbd_lock_bh_state(bh); |
| 564 | if (jh->b_committed_data) |
| 565 | start = ext3_find_next_zero_bit(jh->b_committed_data, |
| 566 | maxblocks, next); |
| 567 | jbd_unlock_bh_state(bh); |
| 568 | } |
| 569 | return -1; |
| 570 | } |
| 571 | |
| 572 | /* |
| 573 | * Find an allocatable block in a bitmap. We honour both the bitmap and |
| 574 | * its last-committed copy (if that exists), and perform the "most |
| 575 | * appropriate allocation" algorithm of looking for a free block near |
| 576 | * the initial goal; then for a free byte somewhere in the bitmap; then |
| 577 | * for any free bit in the bitmap. |
| 578 | */ |
| 579 | static int |
| 580 | find_next_usable_block(int start, struct buffer_head *bh, int maxblocks) |
| 581 | { |
| 582 | int here, next; |
| 583 | char *p, *r; |
| 584 | |
| 585 | if (start > 0) { |
| 586 | /* |
| 587 | * The goal was occupied; search forward for a free |
| 588 | * block within the next XX blocks. |
| 589 | * |
| 590 | * end_goal is more or less random, but it has to be |
| 591 | * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the |
| 592 | * next 64-bit boundary is simple.. |
| 593 | */ |
| 594 | int end_goal = (start + 63) & ~63; |
| 595 | if (end_goal > maxblocks) |
| 596 | end_goal = maxblocks; |
| 597 | here = ext3_find_next_zero_bit(bh->b_data, end_goal, start); |
| 598 | if (here < end_goal && ext3_test_allocatable(here, bh)) |
| 599 | return here; |
| 600 | ext3_debug("Bit not found near goal\n"); |
| 601 | } |
| 602 | |
| 603 | here = start; |
| 604 | if (here < 0) |
| 605 | here = 0; |
| 606 | |
| 607 | p = ((char *)bh->b_data) + (here >> 3); |
| 608 | r = memscan(p, 0, (maxblocks - here + 7) >> 3); |
| 609 | next = (r - ((char *)bh->b_data)) << 3; |
| 610 | |
| 611 | if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh)) |
| 612 | return next; |
| 613 | |
| 614 | /* |
| 615 | * The bitmap search --- search forward alternately through the actual |
| 616 | * bitmap and the last-committed copy until we find a bit free in |
| 617 | * both |
| 618 | */ |
| 619 | here = bitmap_search_next_usable_block(here, bh, maxblocks); |
| 620 | return here; |
| 621 | } |
| 622 | |
| 623 | /* |
| 624 | * We think we can allocate this block in this bitmap. Try to set the bit. |
| 625 | * If that succeeds then check that nobody has allocated and then freed the |
| 626 | * block since we saw that is was not marked in b_committed_data. If it _was_ |
| 627 | * allocated and freed then clear the bit in the bitmap again and return |
| 628 | * zero (failure). |
| 629 | */ |
| 630 | static inline int |
| 631 | claim_block(spinlock_t *lock, int block, struct buffer_head *bh) |
| 632 | { |
| 633 | struct journal_head *jh = bh2jh(bh); |
| 634 | int ret; |
| 635 | |
| 636 | if (ext3_set_bit_atomic(lock, block, bh->b_data)) |
| 637 | return 0; |
| 638 | jbd_lock_bh_state(bh); |
| 639 | if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) { |
| 640 | ext3_clear_bit_atomic(lock, block, bh->b_data); |
| 641 | ret = 0; |
| 642 | } else { |
| 643 | ret = 1; |
| 644 | } |
| 645 | jbd_unlock_bh_state(bh); |
| 646 | return ret; |
| 647 | } |
| 648 | |
| 649 | /* |
| 650 | * If we failed to allocate the desired block then we may end up crossing to a |
| 651 | * new bitmap. In that case we must release write access to the old one via |
| 652 | * ext3_journal_release_buffer(), else we'll run out of credits. |
| 653 | */ |
| 654 | static int |
| 655 | ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group, |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 656 | struct buffer_head *bitmap_bh, int goal, |
| 657 | unsigned long *count, struct ext3_reserve_window *my_rsv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | { |
| 659 | int group_first_block, start, end; |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 660 | unsigned long num = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | |
| 662 | /* we do allocation within the reservation window if we have a window */ |
| 663 | if (my_rsv) { |
| 664 | group_first_block = |
| 665 | le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) + |
| 666 | group * EXT3_BLOCKS_PER_GROUP(sb); |
| 667 | if (my_rsv->_rsv_start >= group_first_block) |
| 668 | start = my_rsv->_rsv_start - group_first_block; |
| 669 | else |
| 670 | /* reservation window cross group boundary */ |
| 671 | start = 0; |
| 672 | end = my_rsv->_rsv_end - group_first_block + 1; |
| 673 | if (end > EXT3_BLOCKS_PER_GROUP(sb)) |
| 674 | /* reservation window crosses group boundary */ |
| 675 | end = EXT3_BLOCKS_PER_GROUP(sb); |
| 676 | if ((start <= goal) && (goal < end)) |
| 677 | start = goal; |
| 678 | else |
| 679 | goal = -1; |
| 680 | } else { |
| 681 | if (goal > 0) |
| 682 | start = goal; |
| 683 | else |
| 684 | start = 0; |
| 685 | end = EXT3_BLOCKS_PER_GROUP(sb); |
| 686 | } |
| 687 | |
| 688 | BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb)); |
| 689 | |
| 690 | repeat: |
| 691 | if (goal < 0 || !ext3_test_allocatable(goal, bitmap_bh)) { |
| 692 | goal = find_next_usable_block(start, bitmap_bh, end); |
| 693 | if (goal < 0) |
| 694 | goto fail_access; |
| 695 | if (!my_rsv) { |
| 696 | int i; |
| 697 | |
| 698 | for (i = 0; i < 7 && goal > start && |
| 699 | ext3_test_allocatable(goal - 1, |
| 700 | bitmap_bh); |
| 701 | i++, goal--) |
| 702 | ; |
| 703 | } |
| 704 | } |
| 705 | start = goal; |
| 706 | |
| 707 | if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group), goal, bitmap_bh)) { |
| 708 | /* |
| 709 | * The block was allocated by another thread, or it was |
| 710 | * allocated and then freed by another thread |
| 711 | */ |
| 712 | start++; |
| 713 | goal++; |
| 714 | if (start >= end) |
| 715 | goto fail_access; |
| 716 | goto repeat; |
| 717 | } |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 718 | num++; |
| 719 | goal++; |
| 720 | while (num < *count && goal < end |
| 721 | && ext3_test_allocatable(goal, bitmap_bh) |
| 722 | && claim_block(sb_bgl_lock(EXT3_SB(sb), group), goal, bitmap_bh)) { |
| 723 | num++; |
| 724 | goal++; |
| 725 | } |
| 726 | *count = num; |
| 727 | return goal - num; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | fail_access: |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 729 | *count = num; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 730 | return -1; |
| 731 | } |
| 732 | |
| 733 | /** |
| 734 | * find_next_reservable_window(): |
| 735 | * find a reservable space within the given range. |
| 736 | * It does not allocate the reservation window for now: |
| 737 | * alloc_new_reservation() will do the work later. |
| 738 | * |
| 739 | * @search_head: the head of the searching list; |
| 740 | * This is not necessarily the list head of the whole filesystem |
| 741 | * |
| 742 | * We have both head and start_block to assist the search |
| 743 | * for the reservable space. The list starts from head, |
| 744 | * but we will shift to the place where start_block is, |
| 745 | * then start from there, when looking for a reservable space. |
| 746 | * |
| 747 | * @size: the target new reservation window size |
| 748 | * |
| 749 | * @group_first_block: the first block we consider to start |
| 750 | * the real search from |
| 751 | * |
| 752 | * @last_block: |
| 753 | * the maximum block number that our goal reservable space |
| 754 | * could start from. This is normally the last block in this |
| 755 | * group. The search will end when we found the start of next |
| 756 | * possible reservable space is out of this boundary. |
| 757 | * This could handle the cross boundary reservation window |
| 758 | * request. |
| 759 | * |
| 760 | * basically we search from the given range, rather than the whole |
| 761 | * reservation double linked list, (start_block, last_block) |
| 762 | * to find a free region that is of my size and has not |
| 763 | * been reserved. |
| 764 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | */ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 766 | static int find_next_reservable_window( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | struct ext3_reserve_window_node *search_head, |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 768 | struct ext3_reserve_window_node *my_rsv, |
| 769 | struct super_block * sb, int start_block, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | int last_block) |
| 771 | { |
| 772 | struct rb_node *next; |
| 773 | struct ext3_reserve_window_node *rsv, *prev; |
| 774 | int cur; |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 775 | int size = my_rsv->rsv_goal_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 776 | |
| 777 | /* TODO: make the start of the reservation window byte-aligned */ |
| 778 | /* cur = *start_block & ~7;*/ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 779 | cur = start_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | rsv = search_head; |
| 781 | if (!rsv) |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 782 | return -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | |
| 784 | while (1) { |
| 785 | if (cur <= rsv->rsv_end) |
| 786 | cur = rsv->rsv_end + 1; |
| 787 | |
| 788 | /* TODO? |
| 789 | * in the case we could not find a reservable space |
| 790 | * that is what is expected, during the re-search, we could |
| 791 | * remember what's the largest reservable space we could have |
| 792 | * and return that one. |
| 793 | * |
| 794 | * For now it will fail if we could not find the reservable |
| 795 | * space with expected-size (or more)... |
| 796 | */ |
| 797 | if (cur > last_block) |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 798 | return -1; /* fail */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | |
| 800 | prev = rsv; |
| 801 | next = rb_next(&rsv->rsv_node); |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 802 | rsv = list_entry(next,struct ext3_reserve_window_node,rsv_node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 803 | |
| 804 | /* |
| 805 | * Reached the last reservation, we can just append to the |
| 806 | * previous one. |
| 807 | */ |
| 808 | if (!next) |
| 809 | break; |
| 810 | |
| 811 | if (cur + size <= rsv->rsv_start) { |
| 812 | /* |
| 813 | * Found a reserveable space big enough. We could |
| 814 | * have a reservation across the group boundary here |
| 815 | */ |
| 816 | break; |
| 817 | } |
| 818 | } |
| 819 | /* |
| 820 | * we come here either : |
| 821 | * when we reach the end of the whole list, |
| 822 | * and there is empty reservable space after last entry in the list. |
| 823 | * append it to the end of the list. |
| 824 | * |
| 825 | * or we found one reservable space in the middle of the list, |
| 826 | * return the reservation window that we could append to. |
| 827 | * succeed. |
| 828 | */ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 829 | |
| 830 | if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window))) |
| 831 | rsv_window_remove(sb, my_rsv); |
| 832 | |
| 833 | /* |
| 834 | * Let's book the whole avaliable window for now. We will check the |
| 835 | * disk bitmap later and then, if there are free blocks then we adjust |
| 836 | * the window size if it's larger than requested. |
| 837 | * Otherwise, we will remove this node from the tree next time |
| 838 | * call find_next_reservable_window. |
| 839 | */ |
| 840 | my_rsv->rsv_start = cur; |
| 841 | my_rsv->rsv_end = cur + size - 1; |
| 842 | my_rsv->rsv_alloc_hit = 0; |
| 843 | |
| 844 | if (prev != my_rsv) |
| 845 | ext3_rsv_window_add(sb, my_rsv); |
| 846 | |
| 847 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | } |
| 849 | |
| 850 | /** |
| 851 | * alloc_new_reservation()--allocate a new reservation window |
| 852 | * |
| 853 | * To make a new reservation, we search part of the filesystem |
| 854 | * reservation list (the list that inside the group). We try to |
| 855 | * allocate a new reservation window near the allocation goal, |
| 856 | * or the beginning of the group, if there is no goal. |
| 857 | * |
| 858 | * We first find a reservable space after the goal, then from |
| 859 | * there, we check the bitmap for the first free block after |
| 860 | * it. If there is no free block until the end of group, then the |
| 861 | * whole group is full, we failed. Otherwise, check if the free |
| 862 | * block is inside the expected reservable space, if so, we |
| 863 | * succeed. |
| 864 | * If the first free block is outside the reservable space, then |
| 865 | * start from the first free block, we search for next available |
| 866 | * space, and go on. |
| 867 | * |
| 868 | * on succeed, a new reservation will be found and inserted into the list |
| 869 | * It contains at least one free block, and it does not overlap with other |
| 870 | * reservation windows. |
| 871 | * |
| 872 | * failed: we failed to find a reservation window in this group |
| 873 | * |
| 874 | * @rsv: the reservation |
| 875 | * |
| 876 | * @goal: The goal (group-relative). It is where the search for a |
| 877 | * free reservable space should start from. |
| 878 | * if we have a goal(goal >0 ), then start from there, |
| 879 | * no goal(goal = -1), we start from the first block |
| 880 | * of the group. |
| 881 | * |
| 882 | * @sb: the super block |
| 883 | * @group: the group we are trying to allocate in |
| 884 | * @bitmap_bh: the block group block bitmap |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 885 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | */ |
| 887 | static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv, |
| 888 | int goal, struct super_block *sb, |
| 889 | unsigned int group, struct buffer_head *bitmap_bh) |
| 890 | { |
| 891 | struct ext3_reserve_window_node *search_head; |
| 892 | int group_first_block, group_end_block, start_block; |
| 893 | int first_free_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root; |
| 895 | unsigned long size; |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 896 | int ret; |
| 897 | spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 898 | |
| 899 | group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) + |
| 900 | group * EXT3_BLOCKS_PER_GROUP(sb); |
| 901 | group_end_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1; |
| 902 | |
| 903 | if (goal < 0) |
| 904 | start_block = group_first_block; |
| 905 | else |
| 906 | start_block = goal + group_first_block; |
| 907 | |
| 908 | size = my_rsv->rsv_goal_size; |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 909 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 910 | if (!rsv_is_empty(&my_rsv->rsv_window)) { |
| 911 | /* |
| 912 | * if the old reservation is cross group boundary |
| 913 | * and if the goal is inside the old reservation window, |
| 914 | * we will come here when we just failed to allocate from |
| 915 | * the first part of the window. We still have another part |
| 916 | * that belongs to the next group. In this case, there is no |
| 917 | * point to discard our window and try to allocate a new one |
| 918 | * in this group(which will fail). we should |
| 919 | * keep the reservation window, just simply move on. |
| 920 | * |
| 921 | * Maybe we could shift the start block of the reservation |
| 922 | * window to the first block of next group. |
| 923 | */ |
| 924 | |
| 925 | if ((my_rsv->rsv_start <= group_end_block) && |
| 926 | (my_rsv->rsv_end > group_end_block) && |
| 927 | (start_block >= my_rsv->rsv_start)) |
| 928 | return -1; |
| 929 | |
| 930 | if ((my_rsv->rsv_alloc_hit > |
| 931 | (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) { |
| 932 | /* |
| 933 | * if we previously allocation hit ration is greater than half |
| 934 | * we double the size of reservation window next time |
| 935 | * otherwise keep the same |
| 936 | */ |
| 937 | size = size * 2; |
| 938 | if (size > EXT3_MAX_RESERVE_BLOCKS) |
| 939 | size = EXT3_MAX_RESERVE_BLOCKS; |
| 940 | my_rsv->rsv_goal_size= size; |
| 941 | } |
| 942 | } |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 943 | |
| 944 | spin_lock(rsv_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 945 | /* |
| 946 | * shift the search start to the window near the goal block |
| 947 | */ |
| 948 | search_head = search_reserve_window(fs_rsv_root, start_block); |
| 949 | |
| 950 | /* |
| 951 | * find_next_reservable_window() simply finds a reservable window |
| 952 | * inside the given range(start_block, group_end_block). |
| 953 | * |
| 954 | * To make sure the reservation window has a free bit inside it, we |
| 955 | * need to check the bitmap after we found a reservable window. |
| 956 | */ |
| 957 | retry: |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 958 | ret = find_next_reservable_window(search_head, my_rsv, sb, |
| 959 | start_block, group_end_block); |
| 960 | |
| 961 | if (ret == -1) { |
| 962 | if (!rsv_is_empty(&my_rsv->rsv_window)) |
| 963 | rsv_window_remove(sb, my_rsv); |
| 964 | spin_unlock(rsv_lock); |
| 965 | return -1; |
| 966 | } |
| 967 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 968 | /* |
| 969 | * On success, find_next_reservable_window() returns the |
| 970 | * reservation window where there is a reservable space after it. |
| 971 | * Before we reserve this reservable space, we need |
| 972 | * to make sure there is at least a free block inside this region. |
| 973 | * |
| 974 | * searching the first free bit on the block bitmap and copy of |
| 975 | * last committed bitmap alternatively, until we found a allocatable |
| 976 | * block. Search start from the start block of the reservable space |
| 977 | * we just found. |
| 978 | */ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 979 | spin_unlock(rsv_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | first_free_block = bitmap_search_next_usable_block( |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 981 | my_rsv->rsv_start - group_first_block, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | bitmap_bh, group_end_block - group_first_block + 1); |
| 983 | |
| 984 | if (first_free_block < 0) { |
| 985 | /* |
| 986 | * no free block left on the bitmap, no point |
| 987 | * to reserve the space. return failed. |
| 988 | */ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 989 | spin_lock(rsv_lock); |
| 990 | if (!rsv_is_empty(&my_rsv->rsv_window)) |
| 991 | rsv_window_remove(sb, my_rsv); |
| 992 | spin_unlock(rsv_lock); |
| 993 | return -1; /* failed */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | } |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 995 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | start_block = first_free_block + group_first_block; |
| 997 | /* |
| 998 | * check if the first free block is within the |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 999 | * free space we just reserved |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | */ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 1001 | if (start_block >= my_rsv->rsv_start && start_block < my_rsv->rsv_end) |
| 1002 | return 0; /* success */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1003 | /* |
| 1004 | * if the first free bit we found is out of the reservable space |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 1005 | * continue search for next reservable space, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1006 | * start from where the free block is, |
| 1007 | * we also shift the list head to where we stopped last time |
| 1008 | */ |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 1009 | search_head = my_rsv; |
| 1010 | spin_lock(rsv_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | goto retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1012 | } |
| 1013 | |
| 1014 | /* |
| 1015 | * This is the main function used to allocate a new block and its reservation |
| 1016 | * window. |
| 1017 | * |
| 1018 | * Each time when a new block allocation is need, first try to allocate from |
| 1019 | * its own reservation. If it does not have a reservation window, instead of |
| 1020 | * looking for a free bit on bitmap first, then look up the reservation list to |
| 1021 | * see if it is inside somebody else's reservation window, we try to allocate a |
| 1022 | * reservation window for it starting from the goal first. Then do the block |
| 1023 | * allocation within the reservation window. |
| 1024 | * |
| 1025 | * This will avoid keeping on searching the reservation list again and |
Glauber de Oliveira Costa | 5b11687 | 2005-10-30 15:02:48 -0800 | [diff] [blame] | 1026 | * again when somebody is looking for a free block (without |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1027 | * reservation), and there are lots of free blocks, but they are all |
| 1028 | * being reserved. |
| 1029 | * |
| 1030 | * We use a sorted double linked list for the per-filesystem reservation list. |
| 1031 | * The insert, remove and find a free space(non-reserved) operations for the |
| 1032 | * sorted double linked list should be fast. |
| 1033 | * |
| 1034 | */ |
| 1035 | static int |
| 1036 | ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle, |
| 1037 | unsigned int group, struct buffer_head *bitmap_bh, |
| 1038 | int goal, struct ext3_reserve_window_node * my_rsv, |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1039 | unsigned long *count, int *errp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | unsigned long group_first_block; |
| 1042 | int ret = 0; |
| 1043 | int fatal; |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1044 | unsigned long num = *count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1045 | |
| 1046 | *errp = 0; |
| 1047 | |
| 1048 | /* |
| 1049 | * Make sure we use undo access for the bitmap, because it is critical |
| 1050 | * that we do the frozen_data COW on bitmap buffers in all cases even |
| 1051 | * if the buffer is in BJ_Forget state in the committing transaction. |
| 1052 | */ |
| 1053 | BUFFER_TRACE(bitmap_bh, "get undo access for new block"); |
| 1054 | fatal = ext3_journal_get_undo_access(handle, bitmap_bh); |
| 1055 | if (fatal) { |
| 1056 | *errp = fatal; |
| 1057 | return -1; |
| 1058 | } |
| 1059 | |
| 1060 | /* |
| 1061 | * we don't deal with reservation when |
| 1062 | * filesystem is mounted without reservation |
| 1063 | * or the file is not a regular file |
| 1064 | * or last attempt to allocate a block with reservation turned on failed |
| 1065 | */ |
| 1066 | if (my_rsv == NULL ) { |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1067 | ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, |
| 1068 | goal, count, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | goto out; |
| 1070 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1071 | /* |
| 1072 | * goal is a group relative block number (if there is a goal) |
| 1073 | * 0 < goal < EXT3_BLOCKS_PER_GROUP(sb) |
| 1074 | * first block is a filesystem wide block number |
| 1075 | * first block is the block number of the first block in this group |
| 1076 | */ |
| 1077 | group_first_block = le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block) + |
| 1078 | group * EXT3_BLOCKS_PER_GROUP(sb); |
| 1079 | |
| 1080 | /* |
| 1081 | * Basically we will allocate a new block from inode's reservation |
| 1082 | * window. |
| 1083 | * |
| 1084 | * We need to allocate a new reservation window, if: |
| 1085 | * a) inode does not have a reservation window; or |
| 1086 | * b) last attempt to allocate a block from existing reservation |
| 1087 | * failed; or |
| 1088 | * c) we come here with a goal and with a reservation window |
| 1089 | * |
| 1090 | * We do not need to allocate a new reservation window if we come here |
| 1091 | * at the beginning with a goal and the goal is inside the window, or |
| 1092 | * we don't have a goal but already have a reservation window. |
| 1093 | * then we could go to allocate from the reservation window directly. |
| 1094 | */ |
| 1095 | while (1) { |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 1096 | if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) || |
| 1097 | !goal_in_my_reservation(&my_rsv->rsv_window, goal, group, sb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1098 | ret = alloc_new_reservation(my_rsv, goal, sb, |
| 1099 | group, bitmap_bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | if (ret < 0) |
| 1101 | break; /* failed */ |
| 1102 | |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 1103 | if (!goal_in_my_reservation(&my_rsv->rsv_window, goal, group, sb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | goal = -1; |
| 1105 | } |
Mingming Cao | 21fe347 | 2005-06-28 20:45:16 -0700 | [diff] [blame] | 1106 | if ((my_rsv->rsv_start >= group_first_block + EXT3_BLOCKS_PER_GROUP(sb)) |
| 1107 | || (my_rsv->rsv_end < group_first_block)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | BUG(); |
| 1109 | ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, goal, |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1110 | &num, &my_rsv->rsv_window); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | if (ret >= 0) { |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1112 | my_rsv->rsv_alloc_hit += num; |
| 1113 | *count = num; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1114 | break; /* succeed */ |
| 1115 | } |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1116 | num = *count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1117 | } |
| 1118 | out: |
| 1119 | if (ret >= 0) { |
| 1120 | BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for " |
| 1121 | "bitmap block"); |
| 1122 | fatal = ext3_journal_dirty_metadata(handle, bitmap_bh); |
| 1123 | if (fatal) { |
| 1124 | *errp = fatal; |
| 1125 | return -1; |
| 1126 | } |
| 1127 | return ret; |
| 1128 | } |
| 1129 | |
| 1130 | BUFFER_TRACE(bitmap_bh, "journal_release_buffer"); |
| 1131 | ext3_journal_release_buffer(handle, bitmap_bh); |
| 1132 | return ret; |
| 1133 | } |
| 1134 | |
| 1135 | static int ext3_has_free_blocks(struct ext3_sb_info *sbi) |
| 1136 | { |
| 1137 | int free_blocks, root_blocks; |
| 1138 | |
| 1139 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); |
| 1140 | root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count); |
| 1141 | if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && |
| 1142 | sbi->s_resuid != current->fsuid && |
| 1143 | (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) { |
| 1144 | return 0; |
| 1145 | } |
| 1146 | return 1; |
| 1147 | } |
| 1148 | |
| 1149 | /* |
| 1150 | * ext3_should_retry_alloc() is called when ENOSPC is returned, and if |
| 1151 | * it is profitable to retry the operation, this function will wait |
| 1152 | * for the current or commiting transaction to complete, and then |
| 1153 | * return TRUE. |
| 1154 | */ |
| 1155 | int ext3_should_retry_alloc(struct super_block *sb, int *retries) |
| 1156 | { |
| 1157 | if (!ext3_has_free_blocks(EXT3_SB(sb)) || (*retries)++ > 3) |
| 1158 | return 0; |
| 1159 | |
| 1160 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); |
| 1161 | |
| 1162 | return journal_force_commit_nested(EXT3_SB(sb)->s_journal); |
| 1163 | } |
| 1164 | |
| 1165 | /* |
| 1166 | * ext3_new_block uses a goal block to assist allocation. If the goal is |
| 1167 | * free, or there is a free block within 32 blocks of the goal, that block |
| 1168 | * is allocated. Otherwise a forward search is made for a free block; within |
| 1169 | * each block group the search first looks for an entire free byte in the block |
| 1170 | * bitmap, and then for any free bit if that fails. |
| 1171 | * This function also updates quota and i_blocks field. |
| 1172 | */ |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1173 | int ext3_new_blocks(handle_t *handle, struct inode *inode, |
| 1174 | unsigned long goal, unsigned long *count, int *errp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1175 | { |
| 1176 | struct buffer_head *bitmap_bh = NULL; |
| 1177 | struct buffer_head *gdp_bh; |
| 1178 | int group_no; |
| 1179 | int goal_group; |
| 1180 | int ret_block; |
| 1181 | int bgi; /* blockgroup iteration index */ |
| 1182 | int target_block; |
| 1183 | int fatal = 0, err; |
| 1184 | int performed_allocation = 0; |
| 1185 | int free_blocks; |
| 1186 | struct super_block *sb; |
| 1187 | struct ext3_group_desc *gdp; |
| 1188 | struct ext3_super_block *es; |
| 1189 | struct ext3_sb_info *sbi; |
| 1190 | struct ext3_reserve_window_node *my_rsv = NULL; |
| 1191 | struct ext3_block_alloc_info *block_i; |
| 1192 | unsigned short windowsz = 0; |
| 1193 | #ifdef EXT3FS_DEBUG |
| 1194 | static int goal_hits, goal_attempts; |
| 1195 | #endif |
| 1196 | unsigned long ngroups; |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1197 | unsigned long num = *count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | |
| 1199 | *errp = -ENOSPC; |
| 1200 | sb = inode->i_sb; |
| 1201 | if (!sb) { |
| 1202 | printk("ext3_new_block: nonexistent device"); |
| 1203 | return 0; |
| 1204 | } |
| 1205 | |
| 1206 | /* |
| 1207 | * Check quota for allocation of this block. |
| 1208 | */ |
| 1209 | if (DQUOT_ALLOC_BLOCK(inode, 1)) { |
| 1210 | *errp = -EDQUOT; |
| 1211 | return 0; |
| 1212 | } |
| 1213 | |
| 1214 | sbi = EXT3_SB(sb); |
| 1215 | es = EXT3_SB(sb)->s_es; |
| 1216 | ext3_debug("goal=%lu.\n", goal); |
| 1217 | /* |
| 1218 | * Allocate a block from reservation only when |
| 1219 | * filesystem is mounted with reservation(default,-o reservation), and |
| 1220 | * it's a regular file, and |
| 1221 | * the desired window size is greater than 0 (One could use ioctl |
| 1222 | * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off |
| 1223 | * reservation on that particular file) |
| 1224 | */ |
| 1225 | block_i = EXT3_I(inode)->i_block_alloc_info; |
| 1226 | if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) |
| 1227 | my_rsv = &block_i->rsv_window_node; |
| 1228 | |
| 1229 | if (!ext3_has_free_blocks(sbi)) { |
| 1230 | *errp = -ENOSPC; |
| 1231 | goto out; |
| 1232 | } |
| 1233 | |
| 1234 | /* |
| 1235 | * First, test whether the goal block is free. |
| 1236 | */ |
| 1237 | if (goal < le32_to_cpu(es->s_first_data_block) || |
| 1238 | goal >= le32_to_cpu(es->s_blocks_count)) |
| 1239 | goal = le32_to_cpu(es->s_first_data_block); |
| 1240 | group_no = (goal - le32_to_cpu(es->s_first_data_block)) / |
| 1241 | EXT3_BLOCKS_PER_GROUP(sb); |
| 1242 | gdp = ext3_get_group_desc(sb, group_no, &gdp_bh); |
| 1243 | if (!gdp) |
| 1244 | goto io_error; |
| 1245 | |
| 1246 | goal_group = group_no; |
| 1247 | retry: |
| 1248 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
| 1249 | /* |
| 1250 | * if there is not enough free blocks to make a new resevation |
| 1251 | * turn off reservation for this allocation |
| 1252 | */ |
| 1253 | if (my_rsv && (free_blocks < windowsz) |
| 1254 | && (rsv_is_empty(&my_rsv->rsv_window))) |
| 1255 | my_rsv = NULL; |
| 1256 | |
| 1257 | if (free_blocks > 0) { |
| 1258 | ret_block = ((goal - le32_to_cpu(es->s_first_data_block)) % |
| 1259 | EXT3_BLOCKS_PER_GROUP(sb)); |
| 1260 | bitmap_bh = read_block_bitmap(sb, group_no); |
| 1261 | if (!bitmap_bh) |
| 1262 | goto io_error; |
| 1263 | ret_block = ext3_try_to_allocate_with_rsv(sb, handle, group_no, |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1264 | bitmap_bh, ret_block, my_rsv, &num, &fatal); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | if (fatal) |
| 1266 | goto out; |
| 1267 | if (ret_block >= 0) |
| 1268 | goto allocated; |
| 1269 | } |
| 1270 | |
| 1271 | ngroups = EXT3_SB(sb)->s_groups_count; |
| 1272 | smp_rmb(); |
| 1273 | |
| 1274 | /* |
| 1275 | * Now search the rest of the groups. We assume that |
| 1276 | * i and gdp correctly point to the last group visited. |
| 1277 | */ |
| 1278 | for (bgi = 0; bgi < ngroups; bgi++) { |
| 1279 | group_no++; |
| 1280 | if (group_no >= ngroups) |
| 1281 | group_no = 0; |
| 1282 | gdp = ext3_get_group_desc(sb, group_no, &gdp_bh); |
| 1283 | if (!gdp) { |
| 1284 | *errp = -EIO; |
| 1285 | goto out; |
| 1286 | } |
| 1287 | free_blocks = le16_to_cpu(gdp->bg_free_blocks_count); |
| 1288 | /* |
| 1289 | * skip this group if the number of |
| 1290 | * free blocks is less than half of the reservation |
| 1291 | * window size. |
| 1292 | */ |
| 1293 | if (free_blocks <= (windowsz/2)) |
| 1294 | continue; |
| 1295 | |
| 1296 | brelse(bitmap_bh); |
| 1297 | bitmap_bh = read_block_bitmap(sb, group_no); |
| 1298 | if (!bitmap_bh) |
| 1299 | goto io_error; |
| 1300 | ret_block = ext3_try_to_allocate_with_rsv(sb, handle, group_no, |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1301 | bitmap_bh, -1, my_rsv, &num, &fatal); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1302 | if (fatal) |
| 1303 | goto out; |
| 1304 | if (ret_block >= 0) |
| 1305 | goto allocated; |
| 1306 | } |
| 1307 | /* |
| 1308 | * We may end up a bogus ealier ENOSPC error due to |
| 1309 | * filesystem is "full" of reservations, but |
| 1310 | * there maybe indeed free blocks avaliable on disk |
| 1311 | * In this case, we just forget about the reservations |
| 1312 | * just do block allocation as without reservations. |
| 1313 | */ |
| 1314 | if (my_rsv) { |
| 1315 | my_rsv = NULL; |
| 1316 | group_no = goal_group; |
| 1317 | goto retry; |
| 1318 | } |
| 1319 | /* No space left on the device */ |
| 1320 | *errp = -ENOSPC; |
| 1321 | goto out; |
| 1322 | |
| 1323 | allocated: |
| 1324 | |
| 1325 | ext3_debug("using block group %d(%d)\n", |
| 1326 | group_no, gdp->bg_free_blocks_count); |
| 1327 | |
| 1328 | BUFFER_TRACE(gdp_bh, "get_write_access"); |
| 1329 | fatal = ext3_journal_get_write_access(handle, gdp_bh); |
| 1330 | if (fatal) |
| 1331 | goto out; |
| 1332 | |
| 1333 | target_block = ret_block + group_no * EXT3_BLOCKS_PER_GROUP(sb) |
| 1334 | + le32_to_cpu(es->s_first_data_block); |
| 1335 | |
| 1336 | if (target_block == le32_to_cpu(gdp->bg_block_bitmap) || |
| 1337 | target_block == le32_to_cpu(gdp->bg_inode_bitmap) || |
| 1338 | in_range(target_block, le32_to_cpu(gdp->bg_inode_table), |
| 1339 | EXT3_SB(sb)->s_itb_per_group)) |
| 1340 | ext3_error(sb, "ext3_new_block", |
| 1341 | "Allocating block in system zone - " |
| 1342 | "block = %u", target_block); |
| 1343 | |
| 1344 | performed_allocation = 1; |
| 1345 | |
| 1346 | #ifdef CONFIG_JBD_DEBUG |
| 1347 | { |
| 1348 | struct buffer_head *debug_bh; |
| 1349 | |
| 1350 | /* Record bitmap buffer state in the newly allocated block */ |
| 1351 | debug_bh = sb_find_get_block(sb, target_block); |
| 1352 | if (debug_bh) { |
| 1353 | BUFFER_TRACE(debug_bh, "state when allocated"); |
| 1354 | BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state"); |
| 1355 | brelse(debug_bh); |
| 1356 | } |
| 1357 | } |
| 1358 | jbd_lock_bh_state(bitmap_bh); |
| 1359 | spin_lock(sb_bgl_lock(sbi, group_no)); |
| 1360 | if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) { |
| 1361 | if (ext3_test_bit(ret_block, |
| 1362 | bh2jh(bitmap_bh)->b_committed_data)) { |
| 1363 | printk("%s: block was unexpectedly set in " |
| 1364 | "b_committed_data\n", __FUNCTION__); |
| 1365 | } |
| 1366 | } |
| 1367 | ext3_debug("found bit %d\n", ret_block); |
| 1368 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
| 1369 | jbd_unlock_bh_state(bitmap_bh); |
| 1370 | #endif |
| 1371 | |
| 1372 | /* ret_block was blockgroup-relative. Now it becomes fs-relative */ |
| 1373 | ret_block = target_block; |
| 1374 | |
| 1375 | if (ret_block >= le32_to_cpu(es->s_blocks_count)) { |
| 1376 | ext3_error(sb, "ext3_new_block", |
| 1377 | "block(%d) >= blocks count(%d) - " |
| 1378 | "block_group = %d, es == %p ", ret_block, |
| 1379 | le32_to_cpu(es->s_blocks_count), group_no, es); |
| 1380 | goto out; |
| 1381 | } |
| 1382 | |
| 1383 | /* |
| 1384 | * It is up to the caller to add the new buffer to a journal |
| 1385 | * list of some description. We don't know in advance whether |
| 1386 | * the caller wants to use it as metadata or data. |
| 1387 | */ |
| 1388 | ext3_debug("allocating block %d. Goal hits %d of %d.\n", |
| 1389 | ret_block, goal_hits, goal_attempts); |
| 1390 | |
| 1391 | spin_lock(sb_bgl_lock(sbi, group_no)); |
| 1392 | gdp->bg_free_blocks_count = |
| 1393 | cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - 1); |
| 1394 | spin_unlock(sb_bgl_lock(sbi, group_no)); |
| 1395 | percpu_counter_mod(&sbi->s_freeblocks_counter, -1); |
| 1396 | |
| 1397 | BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); |
| 1398 | err = ext3_journal_dirty_metadata(handle, gdp_bh); |
| 1399 | if (!fatal) |
| 1400 | fatal = err; |
| 1401 | |
| 1402 | sb->s_dirt = 1; |
| 1403 | if (fatal) |
| 1404 | goto out; |
| 1405 | |
| 1406 | *errp = 0; |
| 1407 | brelse(bitmap_bh); |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1408 | *count = num; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | return ret_block; |
| 1410 | |
| 1411 | io_error: |
| 1412 | *errp = -EIO; |
| 1413 | out: |
| 1414 | if (fatal) { |
| 1415 | *errp = fatal; |
| 1416 | ext3_std_error(sb, fatal); |
| 1417 | } |
| 1418 | /* |
| 1419 | * Undo the block allocation |
| 1420 | */ |
| 1421 | if (!performed_allocation) |
| 1422 | DQUOT_FREE_BLOCK(inode, 1); |
| 1423 | brelse(bitmap_bh); |
| 1424 | return 0; |
| 1425 | } |
| 1426 | |
Mingming Cao | b54e41e | 2006-03-26 01:37:57 -0800 | [diff] [blame^] | 1427 | int ext3_new_block(handle_t *handle, struct inode *inode, |
| 1428 | unsigned long goal, int *errp) |
| 1429 | { |
| 1430 | unsigned long count = 1; |
| 1431 | |
| 1432 | return ext3_new_blocks(handle, inode, goal, &count, errp); |
| 1433 | } |
| 1434 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1435 | unsigned long ext3_count_free_blocks(struct super_block *sb) |
| 1436 | { |
| 1437 | unsigned long desc_count; |
| 1438 | struct ext3_group_desc *gdp; |
| 1439 | int i; |
Glauber de Oliveira Costa | 8bdac5d | 2005-09-22 21:44:26 -0700 | [diff] [blame] | 1440 | unsigned long ngroups = EXT3_SB(sb)->s_groups_count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1441 | #ifdef EXT3FS_DEBUG |
| 1442 | struct ext3_super_block *es; |
| 1443 | unsigned long bitmap_count, x; |
| 1444 | struct buffer_head *bitmap_bh = NULL; |
| 1445 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | es = EXT3_SB(sb)->s_es; |
| 1447 | desc_count = 0; |
| 1448 | bitmap_count = 0; |
| 1449 | gdp = NULL; |
Glauber de Oliveira Costa | 8bdac5d | 2005-09-22 21:44:26 -0700 | [diff] [blame] | 1450 | |
Glauber de Oliveira Costa | 5b11687 | 2005-10-30 15:02:48 -0800 | [diff] [blame] | 1451 | smp_rmb(); |
Glauber de Oliveira Costa | 8bdac5d | 2005-09-22 21:44:26 -0700 | [diff] [blame] | 1452 | for (i = 0; i < ngroups; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | gdp = ext3_get_group_desc(sb, i, NULL); |
| 1454 | if (!gdp) |
| 1455 | continue; |
| 1456 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); |
| 1457 | brelse(bitmap_bh); |
| 1458 | bitmap_bh = read_block_bitmap(sb, i); |
| 1459 | if (bitmap_bh == NULL) |
| 1460 | continue; |
| 1461 | |
| 1462 | x = ext3_count_free(bitmap_bh, sb->s_blocksize); |
| 1463 | printk("group %d: stored = %d, counted = %lu\n", |
| 1464 | i, le16_to_cpu(gdp->bg_free_blocks_count), x); |
| 1465 | bitmap_count += x; |
| 1466 | } |
| 1467 | brelse(bitmap_bh); |
| 1468 | printk("ext3_count_free_blocks: stored = %u, computed = %lu, %lu\n", |
| 1469 | le32_to_cpu(es->s_free_blocks_count), desc_count, bitmap_count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | return bitmap_count; |
| 1471 | #else |
| 1472 | desc_count = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | smp_rmb(); |
| 1474 | for (i = 0; i < ngroups; i++) { |
| 1475 | gdp = ext3_get_group_desc(sb, i, NULL); |
| 1476 | if (!gdp) |
| 1477 | continue; |
| 1478 | desc_count += le16_to_cpu(gdp->bg_free_blocks_count); |
| 1479 | } |
| 1480 | |
| 1481 | return desc_count; |
| 1482 | #endif |
| 1483 | } |
| 1484 | |
| 1485 | static inline int |
| 1486 | block_in_use(unsigned long block, struct super_block *sb, unsigned char *map) |
| 1487 | { |
| 1488 | return ext3_test_bit ((block - |
| 1489 | le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) % |
| 1490 | EXT3_BLOCKS_PER_GROUP(sb), map); |
| 1491 | } |
| 1492 | |
| 1493 | static inline int test_root(int a, int b) |
| 1494 | { |
| 1495 | int num = b; |
| 1496 | |
| 1497 | while (a > num) |
| 1498 | num *= b; |
| 1499 | return num == a; |
| 1500 | } |
| 1501 | |
| 1502 | static int ext3_group_sparse(int group) |
| 1503 | { |
| 1504 | if (group <= 1) |
| 1505 | return 1; |
| 1506 | if (!(group & 1)) |
| 1507 | return 0; |
| 1508 | return (test_root(group, 7) || test_root(group, 5) || |
| 1509 | test_root(group, 3)); |
| 1510 | } |
| 1511 | |
| 1512 | /** |
| 1513 | * ext3_bg_has_super - number of blocks used by the superblock in group |
| 1514 | * @sb: superblock for filesystem |
| 1515 | * @group: group number to check |
| 1516 | * |
| 1517 | * Return the number of blocks used by the superblock (primary or backup) |
| 1518 | * in this group. Currently this will be only 0 or 1. |
| 1519 | */ |
| 1520 | int ext3_bg_has_super(struct super_block *sb, int group) |
| 1521 | { |
Glauber de Oliveira Costa | b5a7c4f | 2006-03-24 03:18:37 -0800 | [diff] [blame] | 1522 | if (EXT3_HAS_RO_COMPAT_FEATURE(sb, |
| 1523 | EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) && |
| 1524 | !ext3_group_sparse(group)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | return 0; |
| 1526 | return 1; |
| 1527 | } |
| 1528 | |
Glauber de Oliveira Costa | b5a7c4f | 2006-03-24 03:18:37 -0800 | [diff] [blame] | 1529 | static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group) |
| 1530 | { |
| 1531 | unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb); |
| 1532 | unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb); |
| 1533 | unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1; |
| 1534 | |
| 1535 | if (group == first || group == first + 1 || group == last) |
| 1536 | return 1; |
| 1537 | return 0; |
| 1538 | } |
| 1539 | |
| 1540 | static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group) |
| 1541 | { |
| 1542 | if (EXT3_HAS_RO_COMPAT_FEATURE(sb, |
| 1543 | EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) && |
| 1544 | !ext3_group_sparse(group)) |
| 1545 | return 0; |
| 1546 | return EXT3_SB(sb)->s_gdb_count; |
| 1547 | } |
| 1548 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1549 | /** |
| 1550 | * ext3_bg_num_gdb - number of blocks used by the group table in group |
| 1551 | * @sb: superblock for filesystem |
| 1552 | * @group: group number to check |
| 1553 | * |
| 1554 | * Return the number of blocks used by the group descriptor table |
| 1555 | * (primary or backup) in this group. In the future there may be a |
| 1556 | * different number of descriptor blocks in each group. |
| 1557 | */ |
| 1558 | unsigned long ext3_bg_num_gdb(struct super_block *sb, int group) |
| 1559 | { |
Glauber de Oliveira Costa | b5a7c4f | 2006-03-24 03:18:37 -0800 | [diff] [blame] | 1560 | unsigned long first_meta_bg = |
| 1561 | le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg); |
| 1562 | unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1563 | |
Glauber de Oliveira Costa | b5a7c4f | 2006-03-24 03:18:37 -0800 | [diff] [blame] | 1564 | if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) || |
| 1565 | metagroup < first_meta_bg) |
| 1566 | return ext3_bg_num_gdb_nometa(sb,group); |
| 1567 | |
| 1568 | return ext3_bg_num_gdb_meta(sb,group); |
| 1569 | |
| 1570 | } |