Amir Goldstein | dae1e52 | 2011-06-27 19:40:50 -0400 | [diff] [blame] | 1 | /* |
| 2 | * linux/fs/ext4/indirect.c |
| 3 | * |
| 4 | * from |
| 5 | * |
| 6 | * linux/fs/ext4/inode.c |
| 7 | * |
| 8 | * Copyright (C) 1992, 1993, 1994, 1995 |
| 9 | * Remy Card (card@masi.ibp.fr) |
| 10 | * Laboratoire MASI - Institut Blaise Pascal |
| 11 | * Universite Pierre et Marie Curie (Paris VI) |
| 12 | * |
| 13 | * from |
| 14 | * |
| 15 | * linux/fs/minix/inode.c |
| 16 | * |
| 17 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 18 | * |
| 19 | * Goal-directed block allocation by Stephen Tweedie |
| 20 | * (sct@redhat.com), 1993, 1998 |
| 21 | */ |
| 22 | |
| 23 | #include <linux/module.h> |
| 24 | #include "ext4_jbd2.h" |
| 25 | #include "truncate.h" |
| 26 | |
| 27 | #include <trace/events/ext4.h> |
| 28 | |
| 29 | typedef struct { |
| 30 | __le32 *p; |
| 31 | __le32 key; |
| 32 | struct buffer_head *bh; |
| 33 | } Indirect; |
| 34 | |
| 35 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) |
| 36 | { |
| 37 | p->key = *(p->p = v); |
| 38 | p->bh = bh; |
| 39 | } |
| 40 | |
| 41 | /** |
| 42 | * ext4_block_to_path - parse the block number into array of offsets |
| 43 | * @inode: inode in question (we are only interested in its superblock) |
| 44 | * @i_block: block number to be parsed |
| 45 | * @offsets: array to store the offsets in |
| 46 | * @boundary: set this non-zero if the referred-to block is likely to be |
| 47 | * followed (on disk) by an indirect block. |
| 48 | * |
| 49 | * To store the locations of file's data ext4 uses a data structure common |
| 50 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
| 51 | * data blocks at leaves and indirect blocks in intermediate nodes. |
| 52 | * This function translates the block number into path in that tree - |
| 53 | * return value is the path length and @offsets[n] is the offset of |
| 54 | * pointer to (n+1)th node in the nth one. If @block is out of range |
| 55 | * (negative or too large) warning is printed and zero returned. |
| 56 | * |
| 57 | * Note: function doesn't find node addresses, so no IO is needed. All |
| 58 | * we need to know is the capacity of indirect blocks (taken from the |
| 59 | * inode->i_sb). |
| 60 | */ |
| 61 | |
| 62 | /* |
| 63 | * Portability note: the last comparison (check that we fit into triple |
| 64 | * indirect block) is spelled differently, because otherwise on an |
| 65 | * architecture with 32-bit longs and 8Kb pages we might get into trouble |
| 66 | * if our filesystem had 8Kb blocks. We might use long long, but that would |
| 67 | * kill us on x86. Oh, well, at least the sign propagation does not matter - |
| 68 | * i_block would have to be negative in the very beginning, so we would not |
| 69 | * get there at all. |
| 70 | */ |
| 71 | |
| 72 | static int ext4_block_to_path(struct inode *inode, |
| 73 | ext4_lblk_t i_block, |
| 74 | ext4_lblk_t offsets[4], int *boundary) |
| 75 | { |
| 76 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
| 77 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); |
| 78 | const long direct_blocks = EXT4_NDIR_BLOCKS, |
| 79 | indirect_blocks = ptrs, |
| 80 | double_blocks = (1 << (ptrs_bits * 2)); |
| 81 | int n = 0; |
| 82 | int final = 0; |
| 83 | |
| 84 | if (i_block < direct_blocks) { |
| 85 | offsets[n++] = i_block; |
| 86 | final = direct_blocks; |
| 87 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
| 88 | offsets[n++] = EXT4_IND_BLOCK; |
| 89 | offsets[n++] = i_block; |
| 90 | final = ptrs; |
| 91 | } else if ((i_block -= indirect_blocks) < double_blocks) { |
| 92 | offsets[n++] = EXT4_DIND_BLOCK; |
| 93 | offsets[n++] = i_block >> ptrs_bits; |
| 94 | offsets[n++] = i_block & (ptrs - 1); |
| 95 | final = ptrs; |
| 96 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { |
| 97 | offsets[n++] = EXT4_TIND_BLOCK; |
| 98 | offsets[n++] = i_block >> (ptrs_bits * 2); |
| 99 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); |
| 100 | offsets[n++] = i_block & (ptrs - 1); |
| 101 | final = ptrs; |
| 102 | } else { |
| 103 | ext4_warning(inode->i_sb, "block %lu > max in inode %lu", |
| 104 | i_block + direct_blocks + |
| 105 | indirect_blocks + double_blocks, inode->i_ino); |
| 106 | } |
| 107 | if (boundary) |
| 108 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
| 109 | return n; |
| 110 | } |
| 111 | |
| 112 | /** |
| 113 | * ext4_get_branch - read the chain of indirect blocks leading to data |
| 114 | * @inode: inode in question |
| 115 | * @depth: depth of the chain (1 - direct pointer, etc.) |
| 116 | * @offsets: offsets of pointers in inode/indirect blocks |
| 117 | * @chain: place to store the result |
| 118 | * @err: here we store the error value |
| 119 | * |
| 120 | * Function fills the array of triples <key, p, bh> and returns %NULL |
| 121 | * if everything went OK or the pointer to the last filled triple |
| 122 | * (incomplete one) otherwise. Upon the return chain[i].key contains |
| 123 | * the number of (i+1)-th block in the chain (as it is stored in memory, |
| 124 | * i.e. little-endian 32-bit), chain[i].p contains the address of that |
| 125 | * number (it points into struct inode for i==0 and into the bh->b_data |
| 126 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect |
| 127 | * block for i>0 and NULL for i==0. In other words, it holds the block |
| 128 | * numbers of the chain, addresses they were taken from (and where we can |
| 129 | * verify that chain did not change) and buffer_heads hosting these |
| 130 | * numbers. |
| 131 | * |
| 132 | * Function stops when it stumbles upon zero pointer (absent block) |
| 133 | * (pointer to last triple returned, *@err == 0) |
| 134 | * or when it gets an IO error reading an indirect block |
| 135 | * (ditto, *@err == -EIO) |
| 136 | * or when it reads all @depth-1 indirect blocks successfully and finds |
| 137 | * the whole chain, all way to the data (returns %NULL, *err == 0). |
| 138 | * |
| 139 | * Need to be called with |
| 140 | * down_read(&EXT4_I(inode)->i_data_sem) |
| 141 | */ |
| 142 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
| 143 | ext4_lblk_t *offsets, |
| 144 | Indirect chain[4], int *err) |
| 145 | { |
| 146 | struct super_block *sb = inode->i_sb; |
| 147 | Indirect *p = chain; |
| 148 | struct buffer_head *bh; |
| 149 | |
| 150 | *err = 0; |
| 151 | /* i_data is not going away, no lock needed */ |
| 152 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
| 153 | if (!p->key) |
| 154 | goto no_block; |
| 155 | while (--depth) { |
| 156 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
| 157 | if (unlikely(!bh)) |
| 158 | goto failure; |
| 159 | |
| 160 | if (!bh_uptodate_or_lock(bh)) { |
| 161 | if (bh_submit_read(bh) < 0) { |
| 162 | put_bh(bh); |
| 163 | goto failure; |
| 164 | } |
| 165 | /* validate block references */ |
| 166 | if (ext4_check_indirect_blockref(inode, bh)) { |
| 167 | put_bh(bh); |
| 168 | goto failure; |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
| 173 | /* Reader: end */ |
| 174 | if (!p->key) |
| 175 | goto no_block; |
| 176 | } |
| 177 | return NULL; |
| 178 | |
| 179 | failure: |
| 180 | *err = -EIO; |
| 181 | no_block: |
| 182 | return p; |
| 183 | } |
| 184 | |
| 185 | /** |
| 186 | * ext4_find_near - find a place for allocation with sufficient locality |
| 187 | * @inode: owner |
| 188 | * @ind: descriptor of indirect block. |
| 189 | * |
| 190 | * This function returns the preferred place for block allocation. |
| 191 | * It is used when heuristic for sequential allocation fails. |
| 192 | * Rules are: |
| 193 | * + if there is a block to the left of our position - allocate near it. |
| 194 | * + if pointer will live in indirect block - allocate near that block. |
| 195 | * + if pointer will live in inode - allocate in the same |
| 196 | * cylinder group. |
| 197 | * |
| 198 | * In the latter case we colour the starting block by the callers PID to |
| 199 | * prevent it from clashing with concurrent allocations for a different inode |
| 200 | * in the same block group. The PID is used here so that functionally related |
| 201 | * files will be close-by on-disk. |
| 202 | * |
| 203 | * Caller must make sure that @ind is valid and will stay that way. |
| 204 | */ |
| 205 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
| 206 | { |
| 207 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 208 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
| 209 | __le32 *p; |
Amir Goldstein | dae1e52 | 2011-06-27 19:40:50 -0400 | [diff] [blame] | 210 | |
| 211 | /* Try to find previous block */ |
| 212 | for (p = ind->p - 1; p >= start; p--) { |
| 213 | if (*p) |
| 214 | return le32_to_cpu(*p); |
| 215 | } |
| 216 | |
| 217 | /* No such thing, so let's try location of indirect block */ |
| 218 | if (ind->bh) |
| 219 | return ind->bh->b_blocknr; |
| 220 | |
| 221 | /* |
| 222 | * It is going to be referred to from the inode itself? OK, just put it |
| 223 | * into the same cylinder group then. |
| 224 | */ |
Eric Sandeen | f86186b | 2011-06-28 10:01:31 -0400 | [diff] [blame] | 225 | return ext4_inode_to_goal_block(inode); |
Amir Goldstein | dae1e52 | 2011-06-27 19:40:50 -0400 | [diff] [blame] | 226 | } |
| 227 | |
| 228 | /** |
| 229 | * ext4_find_goal - find a preferred place for allocation. |
| 230 | * @inode: owner |
| 231 | * @block: block we want |
| 232 | * @partial: pointer to the last triple within a chain |
| 233 | * |
| 234 | * Normally this function find the preferred place for block allocation, |
| 235 | * returns it. |
| 236 | * Because this is only used for non-extent files, we limit the block nr |
| 237 | * to 32 bits. |
| 238 | */ |
| 239 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
| 240 | Indirect *partial) |
| 241 | { |
| 242 | ext4_fsblk_t goal; |
| 243 | |
| 244 | /* |
| 245 | * XXX need to get goal block from mballoc's data structures |
| 246 | */ |
| 247 | |
| 248 | goal = ext4_find_near(inode, partial); |
| 249 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; |
| 250 | return goal; |
| 251 | } |
| 252 | |
| 253 | /** |
| 254 | * ext4_blks_to_allocate - Look up the block map and count the number |
| 255 | * of direct blocks need to be allocated for the given branch. |
| 256 | * |
| 257 | * @branch: chain of indirect blocks |
| 258 | * @k: number of blocks need for indirect blocks |
| 259 | * @blks: number of data blocks to be mapped. |
| 260 | * @blocks_to_boundary: the offset in the indirect block |
| 261 | * |
| 262 | * return the total number of blocks to be allocate, including the |
| 263 | * direct and indirect blocks. |
| 264 | */ |
| 265 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
| 266 | int blocks_to_boundary) |
| 267 | { |
| 268 | unsigned int count = 0; |
| 269 | |
| 270 | /* |
| 271 | * Simple case, [t,d]Indirect block(s) has not allocated yet |
| 272 | * then it's clear blocks on that path have not allocated |
| 273 | */ |
| 274 | if (k > 0) { |
| 275 | /* right now we don't handle cross boundary allocation */ |
| 276 | if (blks < blocks_to_boundary + 1) |
| 277 | count += blks; |
| 278 | else |
| 279 | count += blocks_to_boundary + 1; |
| 280 | return count; |
| 281 | } |
| 282 | |
| 283 | count++; |
| 284 | while (count < blks && count <= blocks_to_boundary && |
| 285 | le32_to_cpu(*(branch[0].p + count)) == 0) { |
| 286 | count++; |
| 287 | } |
| 288 | return count; |
| 289 | } |
| 290 | |
| 291 | /** |
| 292 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
| 293 | * @handle: handle for this transaction |
| 294 | * @inode: inode which needs allocated blocks |
| 295 | * @iblock: the logical block to start allocated at |
| 296 | * @goal: preferred physical block of allocation |
| 297 | * @indirect_blks: the number of blocks need to allocate for indirect |
| 298 | * blocks |
| 299 | * @blks: number of desired blocks |
| 300 | * @new_blocks: on return it will store the new block numbers for |
| 301 | * the indirect blocks(if needed) and the first direct block, |
| 302 | * @err: on return it will store the error code |
| 303 | * |
| 304 | * This function will return the number of blocks allocated as |
| 305 | * requested by the passed-in parameters. |
| 306 | */ |
| 307 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
| 308 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
| 309 | int indirect_blks, int blks, |
| 310 | ext4_fsblk_t new_blocks[4], int *err) |
| 311 | { |
| 312 | struct ext4_allocation_request ar; |
| 313 | int target, i; |
| 314 | unsigned long count = 0, blk_allocated = 0; |
| 315 | int index = 0; |
| 316 | ext4_fsblk_t current_block = 0; |
| 317 | int ret = 0; |
| 318 | |
| 319 | /* |
| 320 | * Here we try to allocate the requested multiple blocks at once, |
| 321 | * on a best-effort basis. |
| 322 | * To build a branch, we should allocate blocks for |
| 323 | * the indirect blocks(if not allocated yet), and at least |
| 324 | * the first direct block of this branch. That's the |
| 325 | * minimum number of blocks need to allocate(required) |
| 326 | */ |
| 327 | /* first we try to allocate the indirect blocks */ |
| 328 | target = indirect_blks; |
| 329 | while (target > 0) { |
| 330 | count = target; |
| 331 | /* allocating blocks for indirect blocks and direct blocks */ |
| 332 | current_block = ext4_new_meta_blocks(handle, inode, goal, |
| 333 | 0, &count, err); |
| 334 | if (*err) |
| 335 | goto failed_out; |
| 336 | |
| 337 | if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) { |
| 338 | EXT4_ERROR_INODE(inode, |
| 339 | "current_block %llu + count %lu > %d!", |
| 340 | current_block, count, |
| 341 | EXT4_MAX_BLOCK_FILE_PHYS); |
| 342 | *err = -EIO; |
| 343 | goto failed_out; |
| 344 | } |
| 345 | |
| 346 | target -= count; |
| 347 | /* allocate blocks for indirect blocks */ |
| 348 | while (index < indirect_blks && count) { |
| 349 | new_blocks[index++] = current_block++; |
| 350 | count--; |
| 351 | } |
| 352 | if (count > 0) { |
| 353 | /* |
| 354 | * save the new block number |
| 355 | * for the first direct block |
| 356 | */ |
| 357 | new_blocks[index] = current_block; |
| 358 | printk(KERN_INFO "%s returned more blocks than " |
| 359 | "requested\n", __func__); |
| 360 | WARN_ON(1); |
| 361 | break; |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | target = blks - count ; |
| 366 | blk_allocated = count; |
| 367 | if (!target) |
| 368 | goto allocated; |
| 369 | /* Now allocate data blocks */ |
| 370 | memset(&ar, 0, sizeof(ar)); |
| 371 | ar.inode = inode; |
| 372 | ar.goal = goal; |
| 373 | ar.len = target; |
| 374 | ar.logical = iblock; |
| 375 | if (S_ISREG(inode->i_mode)) |
| 376 | /* enable in-core preallocation only for regular files */ |
| 377 | ar.flags = EXT4_MB_HINT_DATA; |
| 378 | |
| 379 | current_block = ext4_mb_new_blocks(handle, &ar, err); |
| 380 | if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) { |
| 381 | EXT4_ERROR_INODE(inode, |
| 382 | "current_block %llu + ar.len %d > %d!", |
| 383 | current_block, ar.len, |
| 384 | EXT4_MAX_BLOCK_FILE_PHYS); |
| 385 | *err = -EIO; |
| 386 | goto failed_out; |
| 387 | } |
| 388 | |
| 389 | if (*err && (target == blks)) { |
| 390 | /* |
| 391 | * if the allocation failed and we didn't allocate |
| 392 | * any blocks before |
| 393 | */ |
| 394 | goto failed_out; |
| 395 | } |
| 396 | if (!*err) { |
| 397 | if (target == blks) { |
| 398 | /* |
| 399 | * save the new block number |
| 400 | * for the first direct block |
| 401 | */ |
| 402 | new_blocks[index] = current_block; |
| 403 | } |
| 404 | blk_allocated += ar.len; |
| 405 | } |
| 406 | allocated: |
| 407 | /* total number of blocks allocated for direct blocks */ |
| 408 | ret = blk_allocated; |
| 409 | *err = 0; |
| 410 | return ret; |
| 411 | failed_out: |
| 412 | for (i = 0; i < index; i++) |
| 413 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); |
| 414 | return ret; |
| 415 | } |
| 416 | |
| 417 | /** |
| 418 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
| 419 | * @handle: handle for this transaction |
| 420 | * @inode: owner |
| 421 | * @indirect_blks: number of allocated indirect blocks |
| 422 | * @blks: number of allocated direct blocks |
| 423 | * @goal: preferred place for allocation |
| 424 | * @offsets: offsets (in the blocks) to store the pointers to next. |
| 425 | * @branch: place to store the chain in. |
| 426 | * |
| 427 | * This function allocates blocks, zeroes out all but the last one, |
| 428 | * links them into chain and (if we are synchronous) writes them to disk. |
| 429 | * In other words, it prepares a branch that can be spliced onto the |
| 430 | * inode. It stores the information about that chain in the branch[], in |
| 431 | * the same format as ext4_get_branch() would do. We are calling it after |
| 432 | * we had read the existing part of chain and partial points to the last |
| 433 | * triple of that (one with zero ->key). Upon the exit we have the same |
| 434 | * picture as after the successful ext4_get_block(), except that in one |
| 435 | * place chain is disconnected - *branch->p is still zero (we did not |
| 436 | * set the last link), but branch->key contains the number that should |
| 437 | * be placed into *branch->p to fill that gap. |
| 438 | * |
| 439 | * If allocation fails we free all blocks we've allocated (and forget |
| 440 | * their buffer_heads) and return the error value the from failed |
| 441 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
| 442 | * as described above and return 0. |
| 443 | */ |
| 444 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
| 445 | ext4_lblk_t iblock, int indirect_blks, |
| 446 | int *blks, ext4_fsblk_t goal, |
| 447 | ext4_lblk_t *offsets, Indirect *branch) |
| 448 | { |
| 449 | int blocksize = inode->i_sb->s_blocksize; |
| 450 | int i, n = 0; |
| 451 | int err = 0; |
| 452 | struct buffer_head *bh; |
| 453 | int num; |
| 454 | ext4_fsblk_t new_blocks[4]; |
| 455 | ext4_fsblk_t current_block; |
| 456 | |
| 457 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
| 458 | *blks, new_blocks, &err); |
| 459 | if (err) |
| 460 | return err; |
| 461 | |
| 462 | branch[0].key = cpu_to_le32(new_blocks[0]); |
| 463 | /* |
| 464 | * metadata blocks and data blocks are allocated. |
| 465 | */ |
| 466 | for (n = 1; n <= indirect_blks; n++) { |
| 467 | /* |
| 468 | * Get buffer_head for parent block, zero it out |
| 469 | * and set the pointer to new one, then send |
| 470 | * parent to disk. |
| 471 | */ |
| 472 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); |
| 473 | if (unlikely(!bh)) { |
| 474 | err = -EIO; |
| 475 | goto failed; |
| 476 | } |
| 477 | |
| 478 | branch[n].bh = bh; |
| 479 | lock_buffer(bh); |
| 480 | BUFFER_TRACE(bh, "call get_create_access"); |
| 481 | err = ext4_journal_get_create_access(handle, bh); |
| 482 | if (err) { |
| 483 | /* Don't brelse(bh) here; it's done in |
| 484 | * ext4_journal_forget() below */ |
| 485 | unlock_buffer(bh); |
| 486 | goto failed; |
| 487 | } |
| 488 | |
| 489 | memset(bh->b_data, 0, blocksize); |
| 490 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; |
| 491 | branch[n].key = cpu_to_le32(new_blocks[n]); |
| 492 | *branch[n].p = branch[n].key; |
| 493 | if (n == indirect_blks) { |
| 494 | current_block = new_blocks[n]; |
| 495 | /* |
| 496 | * End of chain, update the last new metablock of |
| 497 | * the chain to point to the new allocated |
| 498 | * data blocks numbers |
| 499 | */ |
| 500 | for (i = 1; i < num; i++) |
| 501 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
| 502 | } |
| 503 | BUFFER_TRACE(bh, "marking uptodate"); |
| 504 | set_buffer_uptodate(bh); |
| 505 | unlock_buffer(bh); |
| 506 | |
| 507 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 508 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 509 | if (err) |
| 510 | goto failed; |
| 511 | } |
| 512 | *blks = num; |
| 513 | return err; |
| 514 | failed: |
| 515 | /* Allocation failed, free what we already allocated */ |
| 516 | ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0); |
| 517 | for (i = 1; i <= n ; i++) { |
| 518 | /* |
| 519 | * branch[i].bh is newly allocated, so there is no |
| 520 | * need to revoke the block, which is why we don't |
| 521 | * need to set EXT4_FREE_BLOCKS_METADATA. |
| 522 | */ |
| 523 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, |
| 524 | EXT4_FREE_BLOCKS_FORGET); |
| 525 | } |
| 526 | for (i = n+1; i < indirect_blks; i++) |
| 527 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); |
| 528 | |
| 529 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0); |
| 530 | |
| 531 | return err; |
| 532 | } |
| 533 | |
| 534 | /** |
| 535 | * ext4_splice_branch - splice the allocated branch onto inode. |
| 536 | * @handle: handle for this transaction |
| 537 | * @inode: owner |
| 538 | * @block: (logical) number of block we are adding |
| 539 | * @chain: chain of indirect blocks (with a missing link - see |
| 540 | * ext4_alloc_branch) |
| 541 | * @where: location of missing link |
| 542 | * @num: number of indirect blocks we are adding |
| 543 | * @blks: number of direct blocks we are adding |
| 544 | * |
| 545 | * This function fills the missing link and does all housekeeping needed in |
| 546 | * inode (->i_blocks, etc.). In case of success we end up with the full |
| 547 | * chain to new block and return 0. |
| 548 | */ |
| 549 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
| 550 | ext4_lblk_t block, Indirect *where, int num, |
| 551 | int blks) |
| 552 | { |
| 553 | int i; |
| 554 | int err = 0; |
| 555 | ext4_fsblk_t current_block; |
| 556 | |
| 557 | /* |
| 558 | * If we're splicing into a [td]indirect block (as opposed to the |
| 559 | * inode) then we need to get write access to the [td]indirect block |
| 560 | * before the splice. |
| 561 | */ |
| 562 | if (where->bh) { |
| 563 | BUFFER_TRACE(where->bh, "get_write_access"); |
| 564 | err = ext4_journal_get_write_access(handle, where->bh); |
| 565 | if (err) |
| 566 | goto err_out; |
| 567 | } |
| 568 | /* That's it */ |
| 569 | |
| 570 | *where->p = where->key; |
| 571 | |
| 572 | /* |
| 573 | * Update the host buffer_head or inode to point to more just allocated |
| 574 | * direct blocks blocks |
| 575 | */ |
| 576 | if (num == 0 && blks > 1) { |
| 577 | current_block = le32_to_cpu(where->key) + 1; |
| 578 | for (i = 1; i < blks; i++) |
| 579 | *(where->p + i) = cpu_to_le32(current_block++); |
| 580 | } |
| 581 | |
| 582 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
| 583 | /* had we spliced it onto indirect block? */ |
| 584 | if (where->bh) { |
| 585 | /* |
| 586 | * If we spliced it onto an indirect block, we haven't |
| 587 | * altered the inode. Note however that if it is being spliced |
| 588 | * onto an indirect block at the very end of the file (the |
| 589 | * file is growing) then we *will* alter the inode to reflect |
| 590 | * the new i_size. But that is not done here - it is done in |
| 591 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
| 592 | */ |
| 593 | jbd_debug(5, "splicing indirect only\n"); |
| 594 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
| 595 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); |
| 596 | if (err) |
| 597 | goto err_out; |
| 598 | } else { |
| 599 | /* |
| 600 | * OK, we spliced it into the inode itself on a direct block. |
| 601 | */ |
| 602 | ext4_mark_inode_dirty(handle, inode); |
| 603 | jbd_debug(5, "splicing direct\n"); |
| 604 | } |
| 605 | return err; |
| 606 | |
| 607 | err_out: |
| 608 | for (i = 1; i <= num; i++) { |
| 609 | /* |
| 610 | * branch[i].bh is newly allocated, so there is no |
| 611 | * need to revoke the block, which is why we don't |
| 612 | * need to set EXT4_FREE_BLOCKS_METADATA. |
| 613 | */ |
| 614 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, |
| 615 | EXT4_FREE_BLOCKS_FORGET); |
| 616 | } |
| 617 | ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key), |
| 618 | blks, 0); |
| 619 | |
| 620 | return err; |
| 621 | } |
| 622 | |
| 623 | /* |
| 624 | * The ext4_ind_map_blocks() function handles non-extents inodes |
| 625 | * (i.e., using the traditional indirect/double-indirect i_blocks |
| 626 | * scheme) for ext4_map_blocks(). |
| 627 | * |
| 628 | * Allocation strategy is simple: if we have to allocate something, we will |
| 629 | * have to go the whole way to leaf. So let's do it before attaching anything |
| 630 | * to tree, set linkage between the newborn blocks, write them if sync is |
| 631 | * required, recheck the path, free and repeat if check fails, otherwise |
| 632 | * set the last missing link (that will protect us from any truncate-generated |
| 633 | * removals - all blocks on the path are immune now) and possibly force the |
| 634 | * write on the parent block. |
| 635 | * That has a nice additional property: no special recovery from the failed |
| 636 | * allocations is needed - we simply release blocks and do not touch anything |
| 637 | * reachable from inode. |
| 638 | * |
| 639 | * `handle' can be NULL if create == 0. |
| 640 | * |
| 641 | * return > 0, # of blocks mapped or allocated. |
| 642 | * return = 0, if plain lookup failed. |
| 643 | * return < 0, error case. |
| 644 | * |
| 645 | * The ext4_ind_get_blocks() function should be called with |
| 646 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem |
| 647 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or |
| 648 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system |
| 649 | * blocks. |
| 650 | */ |
| 651 | int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, |
| 652 | struct ext4_map_blocks *map, |
| 653 | int flags) |
| 654 | { |
| 655 | int err = -EIO; |
| 656 | ext4_lblk_t offsets[4]; |
| 657 | Indirect chain[4]; |
| 658 | Indirect *partial; |
| 659 | ext4_fsblk_t goal; |
| 660 | int indirect_blks; |
| 661 | int blocks_to_boundary = 0; |
| 662 | int depth; |
| 663 | int count = 0; |
| 664 | ext4_fsblk_t first_block = 0; |
| 665 | |
| 666 | trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
| 667 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); |
| 668 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
| 669 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, |
| 670 | &blocks_to_boundary); |
| 671 | |
| 672 | if (depth == 0) |
| 673 | goto out; |
| 674 | |
| 675 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
| 676 | |
| 677 | /* Simplest case - block found, no allocation needed */ |
| 678 | if (!partial) { |
| 679 | first_block = le32_to_cpu(chain[depth - 1].key); |
| 680 | count++; |
| 681 | /*map more blocks*/ |
| 682 | while (count < map->m_len && count <= blocks_to_boundary) { |
| 683 | ext4_fsblk_t blk; |
| 684 | |
| 685 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
| 686 | |
| 687 | if (blk == first_block + count) |
| 688 | count++; |
| 689 | else |
| 690 | break; |
| 691 | } |
| 692 | goto got_it; |
| 693 | } |
| 694 | |
| 695 | /* Next simple case - plain lookup or failed read of indirect block */ |
| 696 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
| 697 | goto cleanup; |
| 698 | |
| 699 | /* |
| 700 | * Okay, we need to do block allocation. |
| 701 | */ |
| 702 | goal = ext4_find_goal(inode, map->m_lblk, partial); |
| 703 | |
| 704 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
| 705 | indirect_blks = (chain + depth) - partial - 1; |
| 706 | |
| 707 | /* |
| 708 | * Next look up the indirect map to count the totoal number of |
| 709 | * direct blocks to allocate for this branch. |
| 710 | */ |
| 711 | count = ext4_blks_to_allocate(partial, indirect_blks, |
| 712 | map->m_len, blocks_to_boundary); |
| 713 | /* |
| 714 | * Block out ext4_truncate while we alter the tree |
| 715 | */ |
| 716 | err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks, |
| 717 | &count, goal, |
| 718 | offsets + (partial - chain), partial); |
| 719 | |
| 720 | /* |
| 721 | * The ext4_splice_branch call will free and forget any buffers |
| 722 | * on the new chain if there is a failure, but that risks using |
| 723 | * up transaction credits, especially for bitmaps where the |
| 724 | * credits cannot be returned. Can we handle this somehow? We |
| 725 | * may need to return -EAGAIN upwards in the worst case. --sct |
| 726 | */ |
| 727 | if (!err) |
| 728 | err = ext4_splice_branch(handle, inode, map->m_lblk, |
| 729 | partial, indirect_blks, count); |
| 730 | if (err) |
| 731 | goto cleanup; |
| 732 | |
| 733 | map->m_flags |= EXT4_MAP_NEW; |
| 734 | |
| 735 | ext4_update_inode_fsync_trans(handle, inode, 1); |
| 736 | got_it: |
| 737 | map->m_flags |= EXT4_MAP_MAPPED; |
| 738 | map->m_pblk = le32_to_cpu(chain[depth-1].key); |
| 739 | map->m_len = count; |
| 740 | if (count > blocks_to_boundary) |
| 741 | map->m_flags |= EXT4_MAP_BOUNDARY; |
| 742 | err = count; |
| 743 | /* Clean up and exit */ |
| 744 | partial = chain + depth - 1; /* the whole chain */ |
| 745 | cleanup: |
| 746 | while (partial > chain) { |
| 747 | BUFFER_TRACE(partial->bh, "call brelse"); |
| 748 | brelse(partial->bh); |
| 749 | partial--; |
| 750 | } |
| 751 | out: |
| 752 | trace_ext4_ind_map_blocks_exit(inode, map->m_lblk, |
| 753 | map->m_pblk, map->m_len, err); |
| 754 | return err; |
| 755 | } |
| 756 | |
| 757 | /* |
| 758 | * O_DIRECT for ext3 (or indirect map) based files |
| 759 | * |
| 760 | * If the O_DIRECT write will extend the file then add this inode to the |
| 761 | * orphan list. So recovery will truncate it back to the original size |
| 762 | * if the machine crashes during the write. |
| 763 | * |
| 764 | * If the O_DIRECT write is intantiating holes inside i_size and the machine |
| 765 | * crashes then stale disk data _may_ be exposed inside the file. But current |
| 766 | * VFS code falls back into buffered path in that case so we are safe. |
| 767 | */ |
| 768 | ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, |
| 769 | const struct iovec *iov, loff_t offset, |
| 770 | unsigned long nr_segs) |
| 771 | { |
| 772 | struct file *file = iocb->ki_filp; |
| 773 | struct inode *inode = file->f_mapping->host; |
| 774 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 775 | handle_t *handle; |
| 776 | ssize_t ret; |
| 777 | int orphan = 0; |
| 778 | size_t count = iov_length(iov, nr_segs); |
| 779 | int retries = 0; |
| 780 | |
| 781 | if (rw == WRITE) { |
| 782 | loff_t final_size = offset + count; |
| 783 | |
| 784 | if (final_size > inode->i_size) { |
| 785 | /* Credits for sb + inode write */ |
| 786 | handle = ext4_journal_start(inode, 2); |
| 787 | if (IS_ERR(handle)) { |
| 788 | ret = PTR_ERR(handle); |
| 789 | goto out; |
| 790 | } |
| 791 | ret = ext4_orphan_add(handle, inode); |
| 792 | if (ret) { |
| 793 | ext4_journal_stop(handle); |
| 794 | goto out; |
| 795 | } |
| 796 | orphan = 1; |
| 797 | ei->i_disksize = inode->i_size; |
| 798 | ext4_journal_stop(handle); |
| 799 | } |
| 800 | } |
| 801 | |
| 802 | retry: |
| 803 | if (rw == READ && ext4_should_dioread_nolock(inode)) |
| 804 | ret = __blockdev_direct_IO(rw, iocb, inode, |
| 805 | inode->i_sb->s_bdev, iov, |
| 806 | offset, nr_segs, |
| 807 | ext4_get_block, NULL, NULL, 0); |
| 808 | else { |
Linus Torvalds | 60ad446 | 2011-08-01 13:56:03 -1000 | [diff] [blame] | 809 | ret = blockdev_direct_IO(rw, iocb, inode, iov, |
| 810 | offset, nr_segs, ext4_get_block); |
Amir Goldstein | dae1e52 | 2011-06-27 19:40:50 -0400 | [diff] [blame] | 811 | |
| 812 | if (unlikely((rw & WRITE) && ret < 0)) { |
| 813 | loff_t isize = i_size_read(inode); |
| 814 | loff_t end = offset + iov_length(iov, nr_segs); |
| 815 | |
| 816 | if (end > isize) |
| 817 | ext4_truncate_failed_write(inode); |
| 818 | } |
| 819 | } |
| 820 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 821 | goto retry; |
| 822 | |
| 823 | if (orphan) { |
| 824 | int err; |
| 825 | |
| 826 | /* Credits for sb + inode write */ |
| 827 | handle = ext4_journal_start(inode, 2); |
| 828 | if (IS_ERR(handle)) { |
| 829 | /* This is really bad luck. We've written the data |
| 830 | * but cannot extend i_size. Bail out and pretend |
| 831 | * the write failed... */ |
| 832 | ret = PTR_ERR(handle); |
| 833 | if (inode->i_nlink) |
| 834 | ext4_orphan_del(NULL, inode); |
| 835 | |
| 836 | goto out; |
| 837 | } |
| 838 | if (inode->i_nlink) |
| 839 | ext4_orphan_del(handle, inode); |
| 840 | if (ret > 0) { |
| 841 | loff_t end = offset + ret; |
| 842 | if (end > inode->i_size) { |
| 843 | ei->i_disksize = end; |
| 844 | i_size_write(inode, end); |
| 845 | /* |
| 846 | * We're going to return a positive `ret' |
| 847 | * here due to non-zero-length I/O, so there's |
| 848 | * no way of reporting error returns from |
| 849 | * ext4_mark_inode_dirty() to userspace. So |
| 850 | * ignore it. |
| 851 | */ |
| 852 | ext4_mark_inode_dirty(handle, inode); |
| 853 | } |
| 854 | } |
| 855 | err = ext4_journal_stop(handle); |
| 856 | if (ret == 0) |
| 857 | ret = err; |
| 858 | } |
| 859 | out: |
| 860 | return ret; |
| 861 | } |
| 862 | |
| 863 | /* |
| 864 | * Calculate the number of metadata blocks need to reserve |
| 865 | * to allocate a new block at @lblocks for non extent file based file |
| 866 | */ |
| 867 | int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock) |
| 868 | { |
| 869 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 870 | sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); |
| 871 | int blk_bits; |
| 872 | |
| 873 | if (lblock < EXT4_NDIR_BLOCKS) |
| 874 | return 0; |
| 875 | |
| 876 | lblock -= EXT4_NDIR_BLOCKS; |
| 877 | |
| 878 | if (ei->i_da_metadata_calc_len && |
| 879 | (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { |
| 880 | ei->i_da_metadata_calc_len++; |
| 881 | return 0; |
| 882 | } |
| 883 | ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; |
| 884 | ei->i_da_metadata_calc_len = 1; |
| 885 | blk_bits = order_base_2(lblock); |
| 886 | return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; |
| 887 | } |
| 888 | |
| 889 | int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
| 890 | { |
| 891 | int indirects; |
| 892 | |
| 893 | /* if nrblocks are contiguous */ |
| 894 | if (chunk) { |
| 895 | /* |
| 896 | * With N contiguous data blocks, we need at most |
| 897 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks, |
| 898 | * 2 dindirect blocks, and 1 tindirect block |
| 899 | */ |
| 900 | return DIV_ROUND_UP(nrblocks, |
| 901 | EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4; |
| 902 | } |
| 903 | /* |
| 904 | * if nrblocks are not contiguous, worse case, each block touch |
| 905 | * a indirect block, and each indirect block touch a double indirect |
| 906 | * block, plus a triple indirect block |
| 907 | */ |
| 908 | indirects = nrblocks * 2 + 1; |
| 909 | return indirects; |
| 910 | } |
| 911 | |
| 912 | /* |
| 913 | * Truncate transactions can be complex and absolutely huge. So we need to |
| 914 | * be able to restart the transaction at a conventient checkpoint to make |
| 915 | * sure we don't overflow the journal. |
| 916 | * |
| 917 | * start_transaction gets us a new handle for a truncate transaction, |
| 918 | * and extend_transaction tries to extend the existing one a bit. If |
| 919 | * extend fails, we need to propagate the failure up and restart the |
| 920 | * transaction in the top-level truncate loop. --sct |
| 921 | */ |
| 922 | static handle_t *start_transaction(struct inode *inode) |
| 923 | { |
| 924 | handle_t *result; |
| 925 | |
| 926 | result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)); |
| 927 | if (!IS_ERR(result)) |
| 928 | return result; |
| 929 | |
| 930 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
| 931 | return result; |
| 932 | } |
| 933 | |
| 934 | /* |
| 935 | * Try to extend this transaction for the purposes of truncation. |
| 936 | * |
| 937 | * Returns 0 if we managed to create more room. If we can't create more |
| 938 | * room, and the transaction must be restarted we return 1. |
| 939 | */ |
| 940 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) |
| 941 | { |
| 942 | if (!ext4_handle_valid(handle)) |
| 943 | return 0; |
| 944 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) |
| 945 | return 0; |
| 946 | if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode))) |
| 947 | return 0; |
| 948 | return 1; |
| 949 | } |
| 950 | |
| 951 | /* |
| 952 | * Probably it should be a library function... search for first non-zero word |
| 953 | * or memcmp with zero_page, whatever is better for particular architecture. |
| 954 | * Linus? |
| 955 | */ |
| 956 | static inline int all_zeroes(__le32 *p, __le32 *q) |
| 957 | { |
| 958 | while (p < q) |
| 959 | if (*p++) |
| 960 | return 0; |
| 961 | return 1; |
| 962 | } |
| 963 | |
| 964 | /** |
| 965 | * ext4_find_shared - find the indirect blocks for partial truncation. |
| 966 | * @inode: inode in question |
| 967 | * @depth: depth of the affected branch |
| 968 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
| 969 | * @chain: place to store the pointers to partial indirect blocks |
| 970 | * @top: place to the (detached) top of branch |
| 971 | * |
| 972 | * This is a helper function used by ext4_truncate(). |
| 973 | * |
| 974 | * When we do truncate() we may have to clean the ends of several |
| 975 | * indirect blocks but leave the blocks themselves alive. Block is |
| 976 | * partially truncated if some data below the new i_size is referred |
| 977 | * from it (and it is on the path to the first completely truncated |
| 978 | * data block, indeed). We have to free the top of that path along |
| 979 | * with everything to the right of the path. Since no allocation |
| 980 | * past the truncation point is possible until ext4_truncate() |
| 981 | * finishes, we may safely do the latter, but top of branch may |
| 982 | * require special attention - pageout below the truncation point |
| 983 | * might try to populate it. |
| 984 | * |
| 985 | * We atomically detach the top of branch from the tree, store the |
| 986 | * block number of its root in *@top, pointers to buffer_heads of |
| 987 | * partially truncated blocks - in @chain[].bh and pointers to |
| 988 | * their last elements that should not be removed - in |
| 989 | * @chain[].p. Return value is the pointer to last filled element |
| 990 | * of @chain. |
| 991 | * |
| 992 | * The work left to caller to do the actual freeing of subtrees: |
| 993 | * a) free the subtree starting from *@top |
| 994 | * b) free the subtrees whose roots are stored in |
| 995 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) |
| 996 | * c) free the subtrees growing from the inode past the @chain[0]. |
| 997 | * (no partially truncated stuff there). */ |
| 998 | |
| 999 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
| 1000 | ext4_lblk_t offsets[4], Indirect chain[4], |
| 1001 | __le32 *top) |
| 1002 | { |
| 1003 | Indirect *partial, *p; |
| 1004 | int k, err; |
| 1005 | |
| 1006 | *top = 0; |
| 1007 | /* Make k index the deepest non-null offset + 1 */ |
| 1008 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
| 1009 | ; |
| 1010 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
| 1011 | /* Writer: pointers */ |
| 1012 | if (!partial) |
| 1013 | partial = chain + k-1; |
| 1014 | /* |
| 1015 | * If the branch acquired continuation since we've looked at it - |
| 1016 | * fine, it should all survive and (new) top doesn't belong to us. |
| 1017 | */ |
| 1018 | if (!partial->key && *partial->p) |
| 1019 | /* Writer: end */ |
| 1020 | goto no_top; |
| 1021 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
| 1022 | ; |
| 1023 | /* |
| 1024 | * OK, we've found the last block that must survive. The rest of our |
| 1025 | * branch should be detached before unlocking. However, if that rest |
| 1026 | * of branch is all ours and does not grow immediately from the inode |
| 1027 | * it's easier to cheat and just decrement partial->p. |
| 1028 | */ |
| 1029 | if (p == chain + k - 1 && p > chain) { |
| 1030 | p->p--; |
| 1031 | } else { |
| 1032 | *top = *p->p; |
| 1033 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
| 1034 | #if 0 |
| 1035 | *p->p = 0; |
| 1036 | #endif |
| 1037 | } |
| 1038 | /* Writer: end */ |
| 1039 | |
| 1040 | while (partial > p) { |
| 1041 | brelse(partial->bh); |
| 1042 | partial--; |
| 1043 | } |
| 1044 | no_top: |
| 1045 | return partial; |
| 1046 | } |
| 1047 | |
| 1048 | /* |
| 1049 | * Zero a number of block pointers in either an inode or an indirect block. |
| 1050 | * If we restart the transaction we must again get write access to the |
| 1051 | * indirect block for further modification. |
| 1052 | * |
| 1053 | * We release `count' blocks on disk, but (last - first) may be greater |
| 1054 | * than `count' because there can be holes in there. |
| 1055 | * |
| 1056 | * Return 0 on success, 1 on invalid block range |
| 1057 | * and < 0 on fatal error. |
| 1058 | */ |
| 1059 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, |
| 1060 | struct buffer_head *bh, |
| 1061 | ext4_fsblk_t block_to_free, |
| 1062 | unsigned long count, __le32 *first, |
| 1063 | __le32 *last) |
| 1064 | { |
| 1065 | __le32 *p; |
| 1066 | int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED; |
| 1067 | int err; |
| 1068 | |
| 1069 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
| 1070 | flags |= EXT4_FREE_BLOCKS_METADATA; |
| 1071 | |
| 1072 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, |
| 1073 | count)) { |
| 1074 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " |
| 1075 | "blocks %llu len %lu", |
| 1076 | (unsigned long long) block_to_free, count); |
| 1077 | return 1; |
| 1078 | } |
| 1079 | |
| 1080 | if (try_to_extend_transaction(handle, inode)) { |
| 1081 | if (bh) { |
| 1082 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 1083 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 1084 | if (unlikely(err)) |
| 1085 | goto out_err; |
| 1086 | } |
| 1087 | err = ext4_mark_inode_dirty(handle, inode); |
| 1088 | if (unlikely(err)) |
| 1089 | goto out_err; |
| 1090 | err = ext4_truncate_restart_trans(handle, inode, |
| 1091 | ext4_blocks_for_truncate(inode)); |
| 1092 | if (unlikely(err)) |
| 1093 | goto out_err; |
| 1094 | if (bh) { |
| 1095 | BUFFER_TRACE(bh, "retaking write access"); |
| 1096 | err = ext4_journal_get_write_access(handle, bh); |
| 1097 | if (unlikely(err)) |
| 1098 | goto out_err; |
| 1099 | } |
| 1100 | } |
| 1101 | |
| 1102 | for (p = first; p < last; p++) |
| 1103 | *p = 0; |
| 1104 | |
| 1105 | ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); |
| 1106 | return 0; |
| 1107 | out_err: |
| 1108 | ext4_std_error(inode->i_sb, err); |
| 1109 | return err; |
| 1110 | } |
| 1111 | |
| 1112 | /** |
| 1113 | * ext4_free_data - free a list of data blocks |
| 1114 | * @handle: handle for this transaction |
| 1115 | * @inode: inode we are dealing with |
| 1116 | * @this_bh: indirect buffer_head which contains *@first and *@last |
| 1117 | * @first: array of block numbers |
| 1118 | * @last: points immediately past the end of array |
| 1119 | * |
| 1120 | * We are freeing all blocks referred from that array (numbers are stored as |
| 1121 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. |
| 1122 | * |
| 1123 | * We accumulate contiguous runs of blocks to free. Conveniently, if these |
| 1124 | * blocks are contiguous then releasing them at one time will only affect one |
| 1125 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't |
| 1126 | * actually use a lot of journal space. |
| 1127 | * |
| 1128 | * @this_bh will be %NULL if @first and @last point into the inode's direct |
| 1129 | * block pointers. |
| 1130 | */ |
| 1131 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
| 1132 | struct buffer_head *this_bh, |
| 1133 | __le32 *first, __le32 *last) |
| 1134 | { |
| 1135 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
| 1136 | unsigned long count = 0; /* Number of blocks in the run */ |
| 1137 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
| 1138 | corresponding to |
| 1139 | block_to_free */ |
| 1140 | ext4_fsblk_t nr; /* Current block # */ |
| 1141 | __le32 *p; /* Pointer into inode/ind |
| 1142 | for current block */ |
| 1143 | int err = 0; |
| 1144 | |
| 1145 | if (this_bh) { /* For indirect block */ |
| 1146 | BUFFER_TRACE(this_bh, "get_write_access"); |
| 1147 | err = ext4_journal_get_write_access(handle, this_bh); |
| 1148 | /* Important: if we can't update the indirect pointers |
| 1149 | * to the blocks, we can't free them. */ |
| 1150 | if (err) |
| 1151 | return; |
| 1152 | } |
| 1153 | |
| 1154 | for (p = first; p < last; p++) { |
| 1155 | nr = le32_to_cpu(*p); |
| 1156 | if (nr) { |
| 1157 | /* accumulate blocks to free if they're contiguous */ |
| 1158 | if (count == 0) { |
| 1159 | block_to_free = nr; |
| 1160 | block_to_free_p = p; |
| 1161 | count = 1; |
| 1162 | } else if (nr == block_to_free + count) { |
| 1163 | count++; |
| 1164 | } else { |
| 1165 | err = ext4_clear_blocks(handle, inode, this_bh, |
| 1166 | block_to_free, count, |
| 1167 | block_to_free_p, p); |
| 1168 | if (err) |
| 1169 | break; |
| 1170 | block_to_free = nr; |
| 1171 | block_to_free_p = p; |
| 1172 | count = 1; |
| 1173 | } |
| 1174 | } |
| 1175 | } |
| 1176 | |
| 1177 | if (!err && count > 0) |
| 1178 | err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
| 1179 | count, block_to_free_p, p); |
| 1180 | if (err < 0) |
| 1181 | /* fatal error */ |
| 1182 | return; |
| 1183 | |
| 1184 | if (this_bh) { |
| 1185 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
| 1186 | |
| 1187 | /* |
| 1188 | * The buffer head should have an attached journal head at this |
| 1189 | * point. However, if the data is corrupted and an indirect |
| 1190 | * block pointed to itself, it would have been detached when |
| 1191 | * the block was cleared. Check for this instead of OOPSing. |
| 1192 | */ |
| 1193 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
| 1194 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
| 1195 | else |
| 1196 | EXT4_ERROR_INODE(inode, |
| 1197 | "circular indirect block detected at " |
| 1198 | "block %llu", |
| 1199 | (unsigned long long) this_bh->b_blocknr); |
| 1200 | } |
| 1201 | } |
| 1202 | |
| 1203 | /** |
| 1204 | * ext4_free_branches - free an array of branches |
| 1205 | * @handle: JBD handle for this transaction |
| 1206 | * @inode: inode we are dealing with |
| 1207 | * @parent_bh: the buffer_head which contains *@first and *@last |
| 1208 | * @first: array of block numbers |
| 1209 | * @last: pointer immediately past the end of array |
| 1210 | * @depth: depth of the branches to free |
| 1211 | * |
| 1212 | * We are freeing all blocks referred from these branches (numbers are |
| 1213 | * stored as little-endian 32-bit) and updating @inode->i_blocks |
| 1214 | * appropriately. |
| 1215 | */ |
| 1216 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
| 1217 | struct buffer_head *parent_bh, |
| 1218 | __le32 *first, __le32 *last, int depth) |
| 1219 | { |
| 1220 | ext4_fsblk_t nr; |
| 1221 | __le32 *p; |
| 1222 | |
| 1223 | if (ext4_handle_is_aborted(handle)) |
| 1224 | return; |
| 1225 | |
| 1226 | if (depth--) { |
| 1227 | struct buffer_head *bh; |
| 1228 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
| 1229 | p = last; |
| 1230 | while (--p >= first) { |
| 1231 | nr = le32_to_cpu(*p); |
| 1232 | if (!nr) |
| 1233 | continue; /* A hole */ |
| 1234 | |
| 1235 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), |
| 1236 | nr, 1)) { |
| 1237 | EXT4_ERROR_INODE(inode, |
| 1238 | "invalid indirect mapped " |
| 1239 | "block %lu (level %d)", |
| 1240 | (unsigned long) nr, depth); |
| 1241 | break; |
| 1242 | } |
| 1243 | |
| 1244 | /* Go read the buffer for the next level down */ |
| 1245 | bh = sb_bread(inode->i_sb, nr); |
| 1246 | |
| 1247 | /* |
| 1248 | * A read failure? Report error and clear slot |
| 1249 | * (should be rare). |
| 1250 | */ |
| 1251 | if (!bh) { |
| 1252 | EXT4_ERROR_INODE_BLOCK(inode, nr, |
| 1253 | "Read failure"); |
| 1254 | continue; |
| 1255 | } |
| 1256 | |
| 1257 | /* This zaps the entire block. Bottom up. */ |
| 1258 | BUFFER_TRACE(bh, "free child branches"); |
| 1259 | ext4_free_branches(handle, inode, bh, |
| 1260 | (__le32 *) bh->b_data, |
| 1261 | (__le32 *) bh->b_data + addr_per_block, |
| 1262 | depth); |
| 1263 | brelse(bh); |
| 1264 | |
| 1265 | /* |
| 1266 | * Everything below this this pointer has been |
| 1267 | * released. Now let this top-of-subtree go. |
| 1268 | * |
| 1269 | * We want the freeing of this indirect block to be |
| 1270 | * atomic in the journal with the updating of the |
| 1271 | * bitmap block which owns it. So make some room in |
| 1272 | * the journal. |
| 1273 | * |
| 1274 | * We zero the parent pointer *after* freeing its |
| 1275 | * pointee in the bitmaps, so if extend_transaction() |
| 1276 | * for some reason fails to put the bitmap changes and |
| 1277 | * the release into the same transaction, recovery |
| 1278 | * will merely complain about releasing a free block, |
| 1279 | * rather than leaking blocks. |
| 1280 | */ |
| 1281 | if (ext4_handle_is_aborted(handle)) |
| 1282 | return; |
| 1283 | if (try_to_extend_transaction(handle, inode)) { |
| 1284 | ext4_mark_inode_dirty(handle, inode); |
| 1285 | ext4_truncate_restart_trans(handle, inode, |
| 1286 | ext4_blocks_for_truncate(inode)); |
| 1287 | } |
| 1288 | |
| 1289 | /* |
| 1290 | * The forget flag here is critical because if |
| 1291 | * we are journaling (and not doing data |
| 1292 | * journaling), we have to make sure a revoke |
| 1293 | * record is written to prevent the journal |
| 1294 | * replay from overwriting the (former) |
| 1295 | * indirect block if it gets reallocated as a |
| 1296 | * data block. This must happen in the same |
| 1297 | * transaction where the data blocks are |
| 1298 | * actually freed. |
| 1299 | */ |
| 1300 | ext4_free_blocks(handle, inode, NULL, nr, 1, |
| 1301 | EXT4_FREE_BLOCKS_METADATA| |
| 1302 | EXT4_FREE_BLOCKS_FORGET); |
| 1303 | |
| 1304 | if (parent_bh) { |
| 1305 | /* |
| 1306 | * The block which we have just freed is |
| 1307 | * pointed to by an indirect block: journal it |
| 1308 | */ |
| 1309 | BUFFER_TRACE(parent_bh, "get_write_access"); |
| 1310 | if (!ext4_journal_get_write_access(handle, |
| 1311 | parent_bh)){ |
| 1312 | *p = 0; |
| 1313 | BUFFER_TRACE(parent_bh, |
| 1314 | "call ext4_handle_dirty_metadata"); |
| 1315 | ext4_handle_dirty_metadata(handle, |
| 1316 | inode, |
| 1317 | parent_bh); |
| 1318 | } |
| 1319 | } |
| 1320 | } |
| 1321 | } else { |
| 1322 | /* We have reached the bottom of the tree. */ |
| 1323 | BUFFER_TRACE(parent_bh, "free data blocks"); |
| 1324 | ext4_free_data(handle, inode, parent_bh, first, last); |
| 1325 | } |
| 1326 | } |
| 1327 | |
| 1328 | void ext4_ind_truncate(struct inode *inode) |
| 1329 | { |
| 1330 | handle_t *handle; |
| 1331 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1332 | __le32 *i_data = ei->i_data; |
| 1333 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
| 1334 | struct address_space *mapping = inode->i_mapping; |
| 1335 | ext4_lblk_t offsets[4]; |
| 1336 | Indirect chain[4]; |
| 1337 | Indirect *partial; |
| 1338 | __le32 nr = 0; |
| 1339 | int n = 0; |
| 1340 | ext4_lblk_t last_block, max_block; |
| 1341 | unsigned blocksize = inode->i_sb->s_blocksize; |
| 1342 | |
| 1343 | handle = start_transaction(inode); |
| 1344 | if (IS_ERR(handle)) |
| 1345 | return; /* AKPM: return what? */ |
| 1346 | |
| 1347 | last_block = (inode->i_size + blocksize-1) |
| 1348 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
| 1349 | max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) |
| 1350 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
| 1351 | |
| 1352 | if (inode->i_size & (blocksize - 1)) |
| 1353 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) |
| 1354 | goto out_stop; |
| 1355 | |
| 1356 | if (last_block != max_block) { |
| 1357 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
| 1358 | if (n == 0) |
| 1359 | goto out_stop; /* error */ |
| 1360 | } |
| 1361 | |
| 1362 | /* |
| 1363 | * OK. This truncate is going to happen. We add the inode to the |
| 1364 | * orphan list, so that if this truncate spans multiple transactions, |
| 1365 | * and we crash, we will resume the truncate when the filesystem |
| 1366 | * recovers. It also marks the inode dirty, to catch the new size. |
| 1367 | * |
| 1368 | * Implication: the file must always be in a sane, consistent |
| 1369 | * truncatable state while each transaction commits. |
| 1370 | */ |
| 1371 | if (ext4_orphan_add(handle, inode)) |
| 1372 | goto out_stop; |
| 1373 | |
| 1374 | /* |
| 1375 | * From here we block out all ext4_get_block() callers who want to |
| 1376 | * modify the block allocation tree. |
| 1377 | */ |
| 1378 | down_write(&ei->i_data_sem); |
| 1379 | |
| 1380 | ext4_discard_preallocations(inode); |
| 1381 | |
| 1382 | /* |
| 1383 | * The orphan list entry will now protect us from any crash which |
| 1384 | * occurs before the truncate completes, so it is now safe to propagate |
| 1385 | * the new, shorter inode size (held for now in i_size) into the |
| 1386 | * on-disk inode. We do this via i_disksize, which is the value which |
| 1387 | * ext4 *really* writes onto the disk inode. |
| 1388 | */ |
| 1389 | ei->i_disksize = inode->i_size; |
| 1390 | |
| 1391 | if (last_block == max_block) { |
| 1392 | /* |
| 1393 | * It is unnecessary to free any data blocks if last_block is |
| 1394 | * equal to the indirect block limit. |
| 1395 | */ |
| 1396 | goto out_unlock; |
| 1397 | } else if (n == 1) { /* direct blocks */ |
| 1398 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
| 1399 | i_data + EXT4_NDIR_BLOCKS); |
| 1400 | goto do_indirects; |
| 1401 | } |
| 1402 | |
| 1403 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
| 1404 | /* Kill the top of shared branch (not detached) */ |
| 1405 | if (nr) { |
| 1406 | if (partial == chain) { |
| 1407 | /* Shared branch grows from the inode */ |
| 1408 | ext4_free_branches(handle, inode, NULL, |
| 1409 | &nr, &nr+1, (chain+n-1) - partial); |
| 1410 | *partial->p = 0; |
| 1411 | /* |
| 1412 | * We mark the inode dirty prior to restart, |
| 1413 | * and prior to stop. No need for it here. |
| 1414 | */ |
| 1415 | } else { |
| 1416 | /* Shared branch grows from an indirect block */ |
| 1417 | BUFFER_TRACE(partial->bh, "get_write_access"); |
| 1418 | ext4_free_branches(handle, inode, partial->bh, |
| 1419 | partial->p, |
| 1420 | partial->p+1, (chain+n-1) - partial); |
| 1421 | } |
| 1422 | } |
| 1423 | /* Clear the ends of indirect blocks on the shared branch */ |
| 1424 | while (partial > chain) { |
| 1425 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
| 1426 | (__le32*)partial->bh->b_data+addr_per_block, |
| 1427 | (chain+n-1) - partial); |
| 1428 | BUFFER_TRACE(partial->bh, "call brelse"); |
| 1429 | brelse(partial->bh); |
| 1430 | partial--; |
| 1431 | } |
| 1432 | do_indirects: |
| 1433 | /* Kill the remaining (whole) subtrees */ |
| 1434 | switch (offsets[0]) { |
| 1435 | default: |
| 1436 | nr = i_data[EXT4_IND_BLOCK]; |
| 1437 | if (nr) { |
| 1438 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
| 1439 | i_data[EXT4_IND_BLOCK] = 0; |
| 1440 | } |
| 1441 | case EXT4_IND_BLOCK: |
| 1442 | nr = i_data[EXT4_DIND_BLOCK]; |
| 1443 | if (nr) { |
| 1444 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
| 1445 | i_data[EXT4_DIND_BLOCK] = 0; |
| 1446 | } |
| 1447 | case EXT4_DIND_BLOCK: |
| 1448 | nr = i_data[EXT4_TIND_BLOCK]; |
| 1449 | if (nr) { |
| 1450 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
| 1451 | i_data[EXT4_TIND_BLOCK] = 0; |
| 1452 | } |
| 1453 | case EXT4_TIND_BLOCK: |
| 1454 | ; |
| 1455 | } |
| 1456 | |
| 1457 | out_unlock: |
| 1458 | up_write(&ei->i_data_sem); |
| 1459 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
| 1460 | ext4_mark_inode_dirty(handle, inode); |
| 1461 | |
| 1462 | /* |
| 1463 | * In a multi-transaction truncate, we only make the final transaction |
| 1464 | * synchronous |
| 1465 | */ |
| 1466 | if (IS_SYNC(inode)) |
| 1467 | ext4_handle_sync(handle); |
| 1468 | out_stop: |
| 1469 | /* |
| 1470 | * If this was a simple ftruncate(), and the file will remain alive |
| 1471 | * then we need to clear up the orphan record which we created above. |
| 1472 | * However, if this was a real unlink then we were called by |
| 1473 | * ext4_delete_inode(), and we allow that function to clean up the |
| 1474 | * orphan info for us. |
| 1475 | */ |
| 1476 | if (inode->i_nlink) |
| 1477 | ext4_orphan_del(handle, inode); |
| 1478 | |
| 1479 | ext4_journal_stop(handle); |
| 1480 | trace_ext4_truncate_exit(inode); |
| 1481 | } |
| 1482 | |