| 1 | /* |
| 2 | * Copyright (C) 2007 Oracle. All rights reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public |
| 6 | * License v2 as published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it will be useful, |
| 9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 11 | * General Public License for more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public |
| 14 | * License along with this program; if not, write to the |
| 15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 16 | * Boston, MA 021110-1307, USA. |
| 17 | */ |
| 18 | |
| 19 | #include <linux/slab.h> |
| 20 | #include <linux/blkdev.h> |
| 21 | #include <linux/writeback.h> |
| 22 | #include <linux/pagevec.h> |
| 23 | #include "ctree.h" |
| 24 | #include "transaction.h" |
| 25 | #include "btrfs_inode.h" |
| 26 | #include "extent_io.h" |
| 27 | |
| 28 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
| 29 | { |
| 30 | if (entry->file_offset + entry->len < entry->file_offset) |
| 31 | return (u64)-1; |
| 32 | return entry->file_offset + entry->len; |
| 33 | } |
| 34 | |
| 35 | /* returns NULL if the insertion worked, or it returns the node it did find |
| 36 | * in the tree |
| 37 | */ |
| 38 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
| 39 | struct rb_node *node) |
| 40 | { |
| 41 | struct rb_node **p = &root->rb_node; |
| 42 | struct rb_node *parent = NULL; |
| 43 | struct btrfs_ordered_extent *entry; |
| 44 | |
| 45 | while (*p) { |
| 46 | parent = *p; |
| 47 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
| 48 | |
| 49 | if (file_offset < entry->file_offset) |
| 50 | p = &(*p)->rb_left; |
| 51 | else if (file_offset >= entry_end(entry)) |
| 52 | p = &(*p)->rb_right; |
| 53 | else |
| 54 | return parent; |
| 55 | } |
| 56 | |
| 57 | rb_link_node(node, parent, p); |
| 58 | rb_insert_color(node, root); |
| 59 | return NULL; |
| 60 | } |
| 61 | |
| 62 | /* |
| 63 | * look for a given offset in the tree, and if it can't be found return the |
| 64 | * first lesser offset |
| 65 | */ |
| 66 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
| 67 | struct rb_node **prev_ret) |
| 68 | { |
| 69 | struct rb_node *n = root->rb_node; |
| 70 | struct rb_node *prev = NULL; |
| 71 | struct rb_node *test; |
| 72 | struct btrfs_ordered_extent *entry; |
| 73 | struct btrfs_ordered_extent *prev_entry = NULL; |
| 74 | |
| 75 | while (n) { |
| 76 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
| 77 | prev = n; |
| 78 | prev_entry = entry; |
| 79 | |
| 80 | if (file_offset < entry->file_offset) |
| 81 | n = n->rb_left; |
| 82 | else if (file_offset >= entry_end(entry)) |
| 83 | n = n->rb_right; |
| 84 | else |
| 85 | return n; |
| 86 | } |
| 87 | if (!prev_ret) |
| 88 | return NULL; |
| 89 | |
| 90 | while (prev && file_offset >= entry_end(prev_entry)) { |
| 91 | test = rb_next(prev); |
| 92 | if (!test) |
| 93 | break; |
| 94 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, |
| 95 | rb_node); |
| 96 | if (file_offset < entry_end(prev_entry)) |
| 97 | break; |
| 98 | |
| 99 | prev = test; |
| 100 | } |
| 101 | if (prev) |
| 102 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, |
| 103 | rb_node); |
| 104 | while (prev && file_offset < entry_end(prev_entry)) { |
| 105 | test = rb_prev(prev); |
| 106 | if (!test) |
| 107 | break; |
| 108 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, |
| 109 | rb_node); |
| 110 | prev = test; |
| 111 | } |
| 112 | *prev_ret = prev; |
| 113 | return NULL; |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * helper to check if a given offset is inside a given entry |
| 118 | */ |
| 119 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
| 120 | { |
| 121 | if (file_offset < entry->file_offset || |
| 122 | entry->file_offset + entry->len <= file_offset) |
| 123 | return 0; |
| 124 | return 1; |
| 125 | } |
| 126 | |
| 127 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
| 128 | u64 len) |
| 129 | { |
| 130 | if (file_offset + len <= entry->file_offset || |
| 131 | entry->file_offset + entry->len <= file_offset) |
| 132 | return 0; |
| 133 | return 1; |
| 134 | } |
| 135 | |
| 136 | /* |
| 137 | * look find the first ordered struct that has this offset, otherwise |
| 138 | * the first one less than this offset |
| 139 | */ |
| 140 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
| 141 | u64 file_offset) |
| 142 | { |
| 143 | struct rb_root *root = &tree->tree; |
| 144 | struct rb_node *prev; |
| 145 | struct rb_node *ret; |
| 146 | struct btrfs_ordered_extent *entry; |
| 147 | |
| 148 | if (tree->last) { |
| 149 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, |
| 150 | rb_node); |
| 151 | if (offset_in_entry(entry, file_offset)) |
| 152 | return tree->last; |
| 153 | } |
| 154 | ret = __tree_search(root, file_offset, &prev); |
| 155 | if (!ret) |
| 156 | ret = prev; |
| 157 | if (ret) |
| 158 | tree->last = ret; |
| 159 | return ret; |
| 160 | } |
| 161 | |
| 162 | /* allocate and add a new ordered_extent into the per-inode tree. |
| 163 | * file_offset is the logical offset in the file |
| 164 | * |
| 165 | * start is the disk block number of an extent already reserved in the |
| 166 | * extent allocation tree |
| 167 | * |
| 168 | * len is the length of the extent |
| 169 | * |
| 170 | * The tree is given a single reference on the ordered extent that was |
| 171 | * inserted. |
| 172 | */ |
| 173 | static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
| 174 | u64 start, u64 len, u64 disk_len, |
| 175 | int type, int dio) |
| 176 | { |
| 177 | struct btrfs_ordered_inode_tree *tree; |
| 178 | struct rb_node *node; |
| 179 | struct btrfs_ordered_extent *entry; |
| 180 | |
| 181 | tree = &BTRFS_I(inode)->ordered_tree; |
| 182 | entry = kzalloc(sizeof(*entry), GFP_NOFS); |
| 183 | if (!entry) |
| 184 | return -ENOMEM; |
| 185 | |
| 186 | entry->file_offset = file_offset; |
| 187 | entry->start = start; |
| 188 | entry->len = len; |
| 189 | entry->disk_len = disk_len; |
| 190 | entry->bytes_left = len; |
| 191 | entry->inode = inode; |
| 192 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
| 193 | set_bit(type, &entry->flags); |
| 194 | |
| 195 | if (dio) |
| 196 | set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); |
| 197 | |
| 198 | /* one ref for the tree */ |
| 199 | atomic_set(&entry->refs, 1); |
| 200 | init_waitqueue_head(&entry->wait); |
| 201 | INIT_LIST_HEAD(&entry->list); |
| 202 | INIT_LIST_HEAD(&entry->root_extent_list); |
| 203 | |
| 204 | spin_lock(&tree->lock); |
| 205 | node = tree_insert(&tree->tree, file_offset, |
| 206 | &entry->rb_node); |
| 207 | BUG_ON(node); |
| 208 | spin_unlock(&tree->lock); |
| 209 | |
| 210 | spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
| 211 | list_add_tail(&entry->root_extent_list, |
| 212 | &BTRFS_I(inode)->root->fs_info->ordered_extents); |
| 213 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
| 214 | |
| 215 | BUG_ON(node); |
| 216 | return 0; |
| 217 | } |
| 218 | |
| 219 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
| 220 | u64 start, u64 len, u64 disk_len, int type) |
| 221 | { |
| 222 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, |
| 223 | disk_len, type, 0); |
| 224 | } |
| 225 | |
| 226 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, |
| 227 | u64 start, u64 len, u64 disk_len, int type) |
| 228 | { |
| 229 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, |
| 230 | disk_len, type, 1); |
| 231 | } |
| 232 | |
| 233 | /* |
| 234 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted |
| 235 | * when an ordered extent is finished. If the list covers more than one |
| 236 | * ordered extent, it is split across multiples. |
| 237 | */ |
| 238 | int btrfs_add_ordered_sum(struct inode *inode, |
| 239 | struct btrfs_ordered_extent *entry, |
| 240 | struct btrfs_ordered_sum *sum) |
| 241 | { |
| 242 | struct btrfs_ordered_inode_tree *tree; |
| 243 | |
| 244 | tree = &BTRFS_I(inode)->ordered_tree; |
| 245 | spin_lock(&tree->lock); |
| 246 | list_add_tail(&sum->list, &entry->list); |
| 247 | spin_unlock(&tree->lock); |
| 248 | return 0; |
| 249 | } |
| 250 | |
| 251 | /* |
| 252 | * this is used to account for finished IO across a given range |
| 253 | * of the file. The IO should not span ordered extents. If |
| 254 | * a given ordered_extent is completely done, 1 is returned, otherwise |
| 255 | * 0. |
| 256 | * |
| 257 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used |
| 258 | * to make sure this function only returns 1 once for a given ordered extent. |
| 259 | */ |
| 260 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
| 261 | struct btrfs_ordered_extent **cached, |
| 262 | u64 file_offset, u64 io_size) |
| 263 | { |
| 264 | struct btrfs_ordered_inode_tree *tree; |
| 265 | struct rb_node *node; |
| 266 | struct btrfs_ordered_extent *entry = NULL; |
| 267 | int ret; |
| 268 | |
| 269 | tree = &BTRFS_I(inode)->ordered_tree; |
| 270 | spin_lock(&tree->lock); |
| 271 | node = tree_search(tree, file_offset); |
| 272 | if (!node) { |
| 273 | ret = 1; |
| 274 | goto out; |
| 275 | } |
| 276 | |
| 277 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 278 | if (!offset_in_entry(entry, file_offset)) { |
| 279 | ret = 1; |
| 280 | goto out; |
| 281 | } |
| 282 | |
| 283 | if (io_size > entry->bytes_left) { |
| 284 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", |
| 285 | (unsigned long long)entry->bytes_left, |
| 286 | (unsigned long long)io_size); |
| 287 | } |
| 288 | entry->bytes_left -= io_size; |
| 289 | if (entry->bytes_left == 0) |
| 290 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
| 291 | else |
| 292 | ret = 1; |
| 293 | out: |
| 294 | if (!ret && cached && entry) { |
| 295 | *cached = entry; |
| 296 | atomic_inc(&entry->refs); |
| 297 | } |
| 298 | spin_unlock(&tree->lock); |
| 299 | return ret == 0; |
| 300 | } |
| 301 | |
| 302 | /* |
| 303 | * used to drop a reference on an ordered extent. This will free |
| 304 | * the extent if the last reference is dropped |
| 305 | */ |
| 306 | int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
| 307 | { |
| 308 | struct list_head *cur; |
| 309 | struct btrfs_ordered_sum *sum; |
| 310 | |
| 311 | if (atomic_dec_and_test(&entry->refs)) { |
| 312 | while (!list_empty(&entry->list)) { |
| 313 | cur = entry->list.next; |
| 314 | sum = list_entry(cur, struct btrfs_ordered_sum, list); |
| 315 | list_del(&sum->list); |
| 316 | kfree(sum); |
| 317 | } |
| 318 | kfree(entry); |
| 319 | } |
| 320 | return 0; |
| 321 | } |
| 322 | |
| 323 | /* |
| 324 | * remove an ordered extent from the tree. No references are dropped |
| 325 | * and you must wake_up entry->wait. You must hold the tree lock |
| 326 | * while you call this function. |
| 327 | */ |
| 328 | static int __btrfs_remove_ordered_extent(struct inode *inode, |
| 329 | struct btrfs_ordered_extent *entry) |
| 330 | { |
| 331 | struct btrfs_ordered_inode_tree *tree; |
| 332 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 333 | struct rb_node *node; |
| 334 | |
| 335 | tree = &BTRFS_I(inode)->ordered_tree; |
| 336 | node = &entry->rb_node; |
| 337 | rb_erase(node, &tree->tree); |
| 338 | tree->last = NULL; |
| 339 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); |
| 340 | |
| 341 | spin_lock(&root->fs_info->ordered_extent_lock); |
| 342 | list_del_init(&entry->root_extent_list); |
| 343 | |
| 344 | /* |
| 345 | * we have no more ordered extents for this inode and |
| 346 | * no dirty pages. We can safely remove it from the |
| 347 | * list of ordered extents |
| 348 | */ |
| 349 | if (RB_EMPTY_ROOT(&tree->tree) && |
| 350 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
| 351 | list_del_init(&BTRFS_I(inode)->ordered_operations); |
| 352 | } |
| 353 | spin_unlock(&root->fs_info->ordered_extent_lock); |
| 354 | |
| 355 | return 0; |
| 356 | } |
| 357 | |
| 358 | /* |
| 359 | * remove an ordered extent from the tree. No references are dropped |
| 360 | * but any waiters are woken. |
| 361 | */ |
| 362 | int btrfs_remove_ordered_extent(struct inode *inode, |
| 363 | struct btrfs_ordered_extent *entry) |
| 364 | { |
| 365 | struct btrfs_ordered_inode_tree *tree; |
| 366 | int ret; |
| 367 | |
| 368 | tree = &BTRFS_I(inode)->ordered_tree; |
| 369 | spin_lock(&tree->lock); |
| 370 | ret = __btrfs_remove_ordered_extent(inode, entry); |
| 371 | spin_unlock(&tree->lock); |
| 372 | wake_up(&entry->wait); |
| 373 | |
| 374 | return ret; |
| 375 | } |
| 376 | |
| 377 | /* |
| 378 | * wait for all the ordered extents in a root. This is done when balancing |
| 379 | * space between drives. |
| 380 | */ |
| 381 | int btrfs_wait_ordered_extents(struct btrfs_root *root, |
| 382 | int nocow_only, int delay_iput) |
| 383 | { |
| 384 | struct list_head splice; |
| 385 | struct list_head *cur; |
| 386 | struct btrfs_ordered_extent *ordered; |
| 387 | struct inode *inode; |
| 388 | |
| 389 | INIT_LIST_HEAD(&splice); |
| 390 | |
| 391 | spin_lock(&root->fs_info->ordered_extent_lock); |
| 392 | list_splice_init(&root->fs_info->ordered_extents, &splice); |
| 393 | while (!list_empty(&splice)) { |
| 394 | cur = splice.next; |
| 395 | ordered = list_entry(cur, struct btrfs_ordered_extent, |
| 396 | root_extent_list); |
| 397 | if (nocow_only && |
| 398 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) && |
| 399 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { |
| 400 | list_move(&ordered->root_extent_list, |
| 401 | &root->fs_info->ordered_extents); |
| 402 | cond_resched_lock(&root->fs_info->ordered_extent_lock); |
| 403 | continue; |
| 404 | } |
| 405 | |
| 406 | list_del_init(&ordered->root_extent_list); |
| 407 | atomic_inc(&ordered->refs); |
| 408 | |
| 409 | /* |
| 410 | * the inode may be getting freed (in sys_unlink path). |
| 411 | */ |
| 412 | inode = igrab(ordered->inode); |
| 413 | |
| 414 | spin_unlock(&root->fs_info->ordered_extent_lock); |
| 415 | |
| 416 | if (inode) { |
| 417 | btrfs_start_ordered_extent(inode, ordered, 1); |
| 418 | btrfs_put_ordered_extent(ordered); |
| 419 | if (delay_iput) |
| 420 | btrfs_add_delayed_iput(inode); |
| 421 | else |
| 422 | iput(inode); |
| 423 | } else { |
| 424 | btrfs_put_ordered_extent(ordered); |
| 425 | } |
| 426 | |
| 427 | spin_lock(&root->fs_info->ordered_extent_lock); |
| 428 | } |
| 429 | spin_unlock(&root->fs_info->ordered_extent_lock); |
| 430 | return 0; |
| 431 | } |
| 432 | |
| 433 | /* |
| 434 | * this is used during transaction commit to write all the inodes |
| 435 | * added to the ordered operation list. These files must be fully on |
| 436 | * disk before the transaction commits. |
| 437 | * |
| 438 | * we have two modes here, one is to just start the IO via filemap_flush |
| 439 | * and the other is to wait for all the io. When we wait, we have an |
| 440 | * extra check to make sure the ordered operation list really is empty |
| 441 | * before we return |
| 442 | */ |
| 443 | int btrfs_run_ordered_operations(struct btrfs_root *root, int wait) |
| 444 | { |
| 445 | struct btrfs_inode *btrfs_inode; |
| 446 | struct inode *inode; |
| 447 | struct list_head splice; |
| 448 | |
| 449 | INIT_LIST_HEAD(&splice); |
| 450 | |
| 451 | mutex_lock(&root->fs_info->ordered_operations_mutex); |
| 452 | spin_lock(&root->fs_info->ordered_extent_lock); |
| 453 | again: |
| 454 | list_splice_init(&root->fs_info->ordered_operations, &splice); |
| 455 | |
| 456 | while (!list_empty(&splice)) { |
| 457 | btrfs_inode = list_entry(splice.next, struct btrfs_inode, |
| 458 | ordered_operations); |
| 459 | |
| 460 | inode = &btrfs_inode->vfs_inode; |
| 461 | |
| 462 | list_del_init(&btrfs_inode->ordered_operations); |
| 463 | |
| 464 | /* |
| 465 | * the inode may be getting freed (in sys_unlink path). |
| 466 | */ |
| 467 | inode = igrab(inode); |
| 468 | |
| 469 | if (!wait && inode) { |
| 470 | list_add_tail(&BTRFS_I(inode)->ordered_operations, |
| 471 | &root->fs_info->ordered_operations); |
| 472 | } |
| 473 | spin_unlock(&root->fs_info->ordered_extent_lock); |
| 474 | |
| 475 | if (inode) { |
| 476 | if (wait) |
| 477 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
| 478 | else |
| 479 | filemap_flush(inode->i_mapping); |
| 480 | btrfs_add_delayed_iput(inode); |
| 481 | } |
| 482 | |
| 483 | cond_resched(); |
| 484 | spin_lock(&root->fs_info->ordered_extent_lock); |
| 485 | } |
| 486 | if (wait && !list_empty(&root->fs_info->ordered_operations)) |
| 487 | goto again; |
| 488 | |
| 489 | spin_unlock(&root->fs_info->ordered_extent_lock); |
| 490 | mutex_unlock(&root->fs_info->ordered_operations_mutex); |
| 491 | |
| 492 | return 0; |
| 493 | } |
| 494 | |
| 495 | /* |
| 496 | * Used to start IO or wait for a given ordered extent to finish. |
| 497 | * |
| 498 | * If wait is one, this effectively waits on page writeback for all the pages |
| 499 | * in the extent, and it waits on the io completion code to insert |
| 500 | * metadata into the btree corresponding to the extent |
| 501 | */ |
| 502 | void btrfs_start_ordered_extent(struct inode *inode, |
| 503 | struct btrfs_ordered_extent *entry, |
| 504 | int wait) |
| 505 | { |
| 506 | u64 start = entry->file_offset; |
| 507 | u64 end = start + entry->len - 1; |
| 508 | |
| 509 | /* |
| 510 | * pages in the range can be dirty, clean or writeback. We |
| 511 | * start IO on any dirty ones so the wait doesn't stall waiting |
| 512 | * for pdflush to find them |
| 513 | */ |
| 514 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
| 515 | filemap_fdatawrite_range(inode->i_mapping, start, end); |
| 516 | if (wait) { |
| 517 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
| 518 | &entry->flags)); |
| 519 | } |
| 520 | } |
| 521 | |
| 522 | /* |
| 523 | * Used to wait on ordered extents across a large range of bytes. |
| 524 | */ |
| 525 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
| 526 | { |
| 527 | u64 end; |
| 528 | u64 orig_end; |
| 529 | struct btrfs_ordered_extent *ordered; |
| 530 | int found; |
| 531 | |
| 532 | if (start + len < start) { |
| 533 | orig_end = INT_LIMIT(loff_t); |
| 534 | } else { |
| 535 | orig_end = start + len - 1; |
| 536 | if (orig_end > INT_LIMIT(loff_t)) |
| 537 | orig_end = INT_LIMIT(loff_t); |
| 538 | } |
| 539 | again: |
| 540 | /* start IO across the range first to instantiate any delalloc |
| 541 | * extents |
| 542 | */ |
| 543 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
| 544 | |
| 545 | /* The compression code will leave pages locked but return from |
| 546 | * writepage without setting the page writeback. Starting again |
| 547 | * with WB_SYNC_ALL will end up waiting for the IO to actually start. |
| 548 | */ |
| 549 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
| 550 | |
| 551 | filemap_fdatawait_range(inode->i_mapping, start, orig_end); |
| 552 | |
| 553 | end = orig_end; |
| 554 | found = 0; |
| 555 | while (1) { |
| 556 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
| 557 | if (!ordered) |
| 558 | break; |
| 559 | if (ordered->file_offset > orig_end) { |
| 560 | btrfs_put_ordered_extent(ordered); |
| 561 | break; |
| 562 | } |
| 563 | if (ordered->file_offset + ordered->len < start) { |
| 564 | btrfs_put_ordered_extent(ordered); |
| 565 | break; |
| 566 | } |
| 567 | found++; |
| 568 | btrfs_start_ordered_extent(inode, ordered, 1); |
| 569 | end = ordered->file_offset; |
| 570 | btrfs_put_ordered_extent(ordered); |
| 571 | if (end == 0 || end == start) |
| 572 | break; |
| 573 | end--; |
| 574 | } |
| 575 | if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, |
| 576 | EXTENT_DELALLOC, 0, NULL)) { |
| 577 | schedule_timeout(1); |
| 578 | goto again; |
| 579 | } |
| 580 | return 0; |
| 581 | } |
| 582 | |
| 583 | /* |
| 584 | * find an ordered extent corresponding to file_offset. return NULL if |
| 585 | * nothing is found, otherwise take a reference on the extent and return it |
| 586 | */ |
| 587 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
| 588 | u64 file_offset) |
| 589 | { |
| 590 | struct btrfs_ordered_inode_tree *tree; |
| 591 | struct rb_node *node; |
| 592 | struct btrfs_ordered_extent *entry = NULL; |
| 593 | |
| 594 | tree = &BTRFS_I(inode)->ordered_tree; |
| 595 | spin_lock(&tree->lock); |
| 596 | node = tree_search(tree, file_offset); |
| 597 | if (!node) |
| 598 | goto out; |
| 599 | |
| 600 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 601 | if (!offset_in_entry(entry, file_offset)) |
| 602 | entry = NULL; |
| 603 | if (entry) |
| 604 | atomic_inc(&entry->refs); |
| 605 | out: |
| 606 | spin_unlock(&tree->lock); |
| 607 | return entry; |
| 608 | } |
| 609 | |
| 610 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
| 611 | * extents that exist in the range, rather than just the start of the range. |
| 612 | */ |
| 613 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, |
| 614 | u64 file_offset, |
| 615 | u64 len) |
| 616 | { |
| 617 | struct btrfs_ordered_inode_tree *tree; |
| 618 | struct rb_node *node; |
| 619 | struct btrfs_ordered_extent *entry = NULL; |
| 620 | |
| 621 | tree = &BTRFS_I(inode)->ordered_tree; |
| 622 | spin_lock(&tree->lock); |
| 623 | node = tree_search(tree, file_offset); |
| 624 | if (!node) { |
| 625 | node = tree_search(tree, file_offset + len); |
| 626 | if (!node) |
| 627 | goto out; |
| 628 | } |
| 629 | |
| 630 | while (1) { |
| 631 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 632 | if (range_overlaps(entry, file_offset, len)) |
| 633 | break; |
| 634 | |
| 635 | if (entry->file_offset >= file_offset + len) { |
| 636 | entry = NULL; |
| 637 | break; |
| 638 | } |
| 639 | entry = NULL; |
| 640 | node = rb_next(node); |
| 641 | if (!node) |
| 642 | break; |
| 643 | } |
| 644 | out: |
| 645 | if (entry) |
| 646 | atomic_inc(&entry->refs); |
| 647 | spin_unlock(&tree->lock); |
| 648 | return entry; |
| 649 | } |
| 650 | |
| 651 | /* |
| 652 | * lookup and return any extent before 'file_offset'. NULL is returned |
| 653 | * if none is found |
| 654 | */ |
| 655 | struct btrfs_ordered_extent * |
| 656 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
| 657 | { |
| 658 | struct btrfs_ordered_inode_tree *tree; |
| 659 | struct rb_node *node; |
| 660 | struct btrfs_ordered_extent *entry = NULL; |
| 661 | |
| 662 | tree = &BTRFS_I(inode)->ordered_tree; |
| 663 | spin_lock(&tree->lock); |
| 664 | node = tree_search(tree, file_offset); |
| 665 | if (!node) |
| 666 | goto out; |
| 667 | |
| 668 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 669 | atomic_inc(&entry->refs); |
| 670 | out: |
| 671 | spin_unlock(&tree->lock); |
| 672 | return entry; |
| 673 | } |
| 674 | |
| 675 | /* |
| 676 | * After an extent is done, call this to conditionally update the on disk |
| 677 | * i_size. i_size is updated to cover any fully written part of the file. |
| 678 | */ |
| 679 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
| 680 | struct btrfs_ordered_extent *ordered) |
| 681 | { |
| 682 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; |
| 683 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
| 684 | u64 disk_i_size; |
| 685 | u64 new_i_size; |
| 686 | u64 i_size_test; |
| 687 | u64 i_size = i_size_read(inode); |
| 688 | struct rb_node *node; |
| 689 | struct rb_node *prev = NULL; |
| 690 | struct btrfs_ordered_extent *test; |
| 691 | int ret = 1; |
| 692 | |
| 693 | if (ordered) |
| 694 | offset = entry_end(ordered); |
| 695 | else |
| 696 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); |
| 697 | |
| 698 | spin_lock(&tree->lock); |
| 699 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
| 700 | |
| 701 | /* truncate file */ |
| 702 | if (disk_i_size > i_size) { |
| 703 | BTRFS_I(inode)->disk_i_size = i_size; |
| 704 | ret = 0; |
| 705 | goto out; |
| 706 | } |
| 707 | |
| 708 | /* |
| 709 | * if the disk i_size is already at the inode->i_size, or |
| 710 | * this ordered extent is inside the disk i_size, we're done |
| 711 | */ |
| 712 | if (disk_i_size == i_size || offset <= disk_i_size) { |
| 713 | goto out; |
| 714 | } |
| 715 | |
| 716 | /* |
| 717 | * we can't update the disk_isize if there are delalloc bytes |
| 718 | * between disk_i_size and this ordered extent |
| 719 | */ |
| 720 | if (test_range_bit(io_tree, disk_i_size, offset - 1, |
| 721 | EXTENT_DELALLOC, 0, NULL)) { |
| 722 | goto out; |
| 723 | } |
| 724 | /* |
| 725 | * walk backward from this ordered extent to disk_i_size. |
| 726 | * if we find an ordered extent then we can't update disk i_size |
| 727 | * yet |
| 728 | */ |
| 729 | if (ordered) { |
| 730 | node = rb_prev(&ordered->rb_node); |
| 731 | } else { |
| 732 | prev = tree_search(tree, offset); |
| 733 | /* |
| 734 | * we insert file extents without involving ordered struct, |
| 735 | * so there should be no ordered struct cover this offset |
| 736 | */ |
| 737 | if (prev) { |
| 738 | test = rb_entry(prev, struct btrfs_ordered_extent, |
| 739 | rb_node); |
| 740 | BUG_ON(offset_in_entry(test, offset)); |
| 741 | } |
| 742 | node = prev; |
| 743 | } |
| 744 | while (node) { |
| 745 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 746 | if (test->file_offset + test->len <= disk_i_size) |
| 747 | break; |
| 748 | if (test->file_offset >= i_size) |
| 749 | break; |
| 750 | if (test->file_offset >= disk_i_size) |
| 751 | goto out; |
| 752 | node = rb_prev(node); |
| 753 | } |
| 754 | new_i_size = min_t(u64, offset, i_size); |
| 755 | |
| 756 | /* |
| 757 | * at this point, we know we can safely update i_size to at least |
| 758 | * the offset from this ordered extent. But, we need to |
| 759 | * walk forward and see if ios from higher up in the file have |
| 760 | * finished. |
| 761 | */ |
| 762 | if (ordered) { |
| 763 | node = rb_next(&ordered->rb_node); |
| 764 | } else { |
| 765 | if (prev) |
| 766 | node = rb_next(prev); |
| 767 | else |
| 768 | node = rb_first(&tree->tree); |
| 769 | } |
| 770 | i_size_test = 0; |
| 771 | if (node) { |
| 772 | /* |
| 773 | * do we have an area where IO might have finished |
| 774 | * between our ordered extent and the next one. |
| 775 | */ |
| 776 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 777 | if (test->file_offset > offset) |
| 778 | i_size_test = test->file_offset; |
| 779 | } else { |
| 780 | i_size_test = i_size; |
| 781 | } |
| 782 | |
| 783 | /* |
| 784 | * i_size_test is the end of a region after this ordered |
| 785 | * extent where there are no ordered extents. As long as there |
| 786 | * are no delalloc bytes in this area, it is safe to update |
| 787 | * disk_i_size to the end of the region. |
| 788 | */ |
| 789 | if (i_size_test > offset && |
| 790 | !test_range_bit(io_tree, offset, i_size_test - 1, |
| 791 | EXTENT_DELALLOC, 0, NULL)) { |
| 792 | new_i_size = min_t(u64, i_size_test, i_size); |
| 793 | } |
| 794 | BTRFS_I(inode)->disk_i_size = new_i_size; |
| 795 | ret = 0; |
| 796 | out: |
| 797 | /* |
| 798 | * we need to remove the ordered extent with the tree lock held |
| 799 | * so that other people calling this function don't find our fully |
| 800 | * processed ordered entry and skip updating the i_size |
| 801 | */ |
| 802 | if (ordered) |
| 803 | __btrfs_remove_ordered_extent(inode, ordered); |
| 804 | spin_unlock(&tree->lock); |
| 805 | if (ordered) |
| 806 | wake_up(&ordered->wait); |
| 807 | return ret; |
| 808 | } |
| 809 | |
| 810 | /* |
| 811 | * search the ordered extents for one corresponding to 'offset' and |
| 812 | * try to find a checksum. This is used because we allow pages to |
| 813 | * be reclaimed before their checksum is actually put into the btree |
| 814 | */ |
| 815 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
| 816 | u32 *sum) |
| 817 | { |
| 818 | struct btrfs_ordered_sum *ordered_sum; |
| 819 | struct btrfs_sector_sum *sector_sums; |
| 820 | struct btrfs_ordered_extent *ordered; |
| 821 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; |
| 822 | unsigned long num_sectors; |
| 823 | unsigned long i; |
| 824 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; |
| 825 | int ret = 1; |
| 826 | |
| 827 | ordered = btrfs_lookup_ordered_extent(inode, offset); |
| 828 | if (!ordered) |
| 829 | return 1; |
| 830 | |
| 831 | spin_lock(&tree->lock); |
| 832 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
| 833 | if (disk_bytenr >= ordered_sum->bytenr) { |
| 834 | num_sectors = ordered_sum->len / sectorsize; |
| 835 | sector_sums = ordered_sum->sums; |
| 836 | for (i = 0; i < num_sectors; i++) { |
| 837 | if (sector_sums[i].bytenr == disk_bytenr) { |
| 838 | *sum = sector_sums[i].sum; |
| 839 | ret = 0; |
| 840 | goto out; |
| 841 | } |
| 842 | } |
| 843 | } |
| 844 | } |
| 845 | out: |
| 846 | spin_unlock(&tree->lock); |
| 847 | btrfs_put_ordered_extent(ordered); |
| 848 | return ret; |
| 849 | } |
| 850 | |
| 851 | |
| 852 | /* |
| 853 | * add a given inode to the list of inodes that must be fully on |
| 854 | * disk before a transaction commit finishes. |
| 855 | * |
| 856 | * This basically gives us the ext3 style data=ordered mode, and it is mostly |
| 857 | * used to make sure renamed files are fully on disk. |
| 858 | * |
| 859 | * It is a noop if the inode is already fully on disk. |
| 860 | * |
| 861 | * If trans is not null, we'll do a friendly check for a transaction that |
| 862 | * is already flushing things and force the IO down ourselves. |
| 863 | */ |
| 864 | int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, |
| 865 | struct btrfs_root *root, |
| 866 | struct inode *inode) |
| 867 | { |
| 868 | u64 last_mod; |
| 869 | |
| 870 | last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); |
| 871 | |
| 872 | /* |
| 873 | * if this file hasn't been changed since the last transaction |
| 874 | * commit, we can safely return without doing anything |
| 875 | */ |
| 876 | if (last_mod < root->fs_info->last_trans_committed) |
| 877 | return 0; |
| 878 | |
| 879 | /* |
| 880 | * the transaction is already committing. Just start the IO and |
| 881 | * don't bother with all of this list nonsense |
| 882 | */ |
| 883 | if (trans && root->fs_info->running_transaction->blocked) { |
| 884 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
| 885 | return 0; |
| 886 | } |
| 887 | |
| 888 | spin_lock(&root->fs_info->ordered_extent_lock); |
| 889 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) { |
| 890 | list_add_tail(&BTRFS_I(inode)->ordered_operations, |
| 891 | &root->fs_info->ordered_operations); |
| 892 | } |
| 893 | spin_unlock(&root->fs_info->ordered_extent_lock); |
| 894 | |
| 895 | return 0; |
| 896 | } |