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/fs.h> 20 #include <linux/pagemap.h> 21 #include <linux/highmem.h> 22 #include <linux/time.h> 23 #include <linux/init.h> 24 #include <linux/string.h> 25 #include <linux/backing-dev.h> 26 #include <linux/mpage.h> 27 #include <linux/swap.h> 28 #include <linux/writeback.h> 29 #include <linux/statfs.h> 30 #include <linux/compat.h> 31 #include "ctree.h" 32 #include "disk-io.h" 33 #include "transaction.h" 34 #include "btrfs_inode.h" 35 #include "ioctl.h" 36 #include "print-tree.h" 37 #include "tree-log.h" 38 #include "locking.h" 39 #include "compat.h" 40 41 42 /* simple helper to fault in pages and copy. This should go away 43 * and be replaced with calls into generic code. 44 */ 45 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, 46 int write_bytes, 47 struct page **prepared_pages, 48 const char __user *buf) 49 { 50 long page_fault = 0; 51 int i; 52 int offset = pos & (PAGE_CACHE_SIZE - 1); 53 54 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) { 55 size_t count = min_t(size_t, 56 PAGE_CACHE_SIZE - offset, write_bytes); 57 struct page *page = prepared_pages[i]; 58 fault_in_pages_readable(buf, count); 59 60 /* Copy data from userspace to the current page */ 61 kmap(page); 62 page_fault = __copy_from_user(page_address(page) + offset, 63 buf, count); 64 /* Flush processor's dcache for this page */ 65 flush_dcache_page(page); 66 kunmap(page); 67 buf += count; 68 write_bytes -= count; 69 70 if (page_fault) 71 break; 72 } 73 return page_fault ? -EFAULT : 0; 74 } 75 76 /* 77 * unlocks pages after btrfs_file_write is done with them 78 */ 79 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages) 80 { 81 size_t i; 82 for (i = 0; i < num_pages; i++) { 83 if (!pages[i]) 84 break; 85 /* page checked is some magic around finding pages that 86 * have been modified without going through btrfs_set_page_dirty 87 * clear it here 88 */ 89 ClearPageChecked(pages[i]); 90 unlock_page(pages[i]); 91 mark_page_accessed(pages[i]); 92 page_cache_release(pages[i]); 93 } 94 } 95 96 /* 97 * after copy_from_user, pages need to be dirtied and we need to make 98 * sure holes are created between the current EOF and the start of 99 * any next extents (if required). 100 * 101 * this also makes the decision about creating an inline extent vs 102 * doing real data extents, marking pages dirty and delalloc as required. 103 */ 104 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans, 105 struct btrfs_root *root, 106 struct file *file, 107 struct page **pages, 108 size_t num_pages, 109 loff_t pos, 110 size_t write_bytes) 111 { 112 int err = 0; 113 int i; 114 struct inode *inode = fdentry(file)->d_inode; 115 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 116 u64 hint_byte; 117 u64 num_bytes; 118 u64 start_pos; 119 u64 end_of_last_block; 120 u64 end_pos = pos + write_bytes; 121 loff_t isize = i_size_read(inode); 122 123 start_pos = pos & ~((u64)root->sectorsize - 1); 124 num_bytes = (write_bytes + pos - start_pos + 125 root->sectorsize - 1) & ~((u64)root->sectorsize - 1); 126 127 end_of_last_block = start_pos + num_bytes - 1; 128 129 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS); 130 trans = btrfs_join_transaction(root, 1); 131 if (!trans) { 132 err = -ENOMEM; 133 goto out_unlock; 134 } 135 btrfs_set_trans_block_group(trans, inode); 136 hint_byte = 0; 137 138 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS); 139 140 /* check for reserved extents on each page, we don't want 141 * to reset the delalloc bit on things that already have 142 * extents reserved. 143 */ 144 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block); 145 for (i = 0; i < num_pages; i++) { 146 struct page *p = pages[i]; 147 SetPageUptodate(p); 148 ClearPageChecked(p); 149 set_page_dirty(p); 150 } 151 if (end_pos > isize) { 152 i_size_write(inode, end_pos); 153 /* we've only changed i_size in ram, and we haven't updated 154 * the disk i_size. There is no need to log the inode 155 * at this time. 156 */ 157 } 158 err = btrfs_end_transaction(trans, root); 159 out_unlock: 160 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS); 161 return err; 162 } 163 164 /* 165 * this drops all the extents in the cache that intersect the range 166 * [start, end]. Existing extents are split as required. 167 */ 168 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, 169 int skip_pinned) 170 { 171 struct extent_map *em; 172 struct extent_map *split = NULL; 173 struct extent_map *split2 = NULL; 174 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 175 u64 len = end - start + 1; 176 int ret; 177 int testend = 1; 178 unsigned long flags; 179 int compressed = 0; 180 181 WARN_ON(end < start); 182 if (end == (u64)-1) { 183 len = (u64)-1; 184 testend = 0; 185 } 186 while (1) { 187 if (!split) 188 split = alloc_extent_map(GFP_NOFS); 189 if (!split2) 190 split2 = alloc_extent_map(GFP_NOFS); 191 192 spin_lock(&em_tree->lock); 193 em = lookup_extent_mapping(em_tree, start, len); 194 if (!em) { 195 spin_unlock(&em_tree->lock); 196 break; 197 } 198 flags = em->flags; 199 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { 200 spin_unlock(&em_tree->lock); 201 if (em->start <= start && 202 (!testend || em->start + em->len >= start + len)) { 203 free_extent_map(em); 204 break; 205 } 206 if (start < em->start) { 207 len = em->start - start; 208 } else { 209 len = start + len - (em->start + em->len); 210 start = em->start + em->len; 211 } 212 free_extent_map(em); 213 continue; 214 } 215 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 216 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 217 remove_extent_mapping(em_tree, em); 218 219 if (em->block_start < EXTENT_MAP_LAST_BYTE && 220 em->start < start) { 221 split->start = em->start; 222 split->len = start - em->start; 223 split->orig_start = em->orig_start; 224 split->block_start = em->block_start; 225 226 if (compressed) 227 split->block_len = em->block_len; 228 else 229 split->block_len = split->len; 230 231 split->bdev = em->bdev; 232 split->flags = flags; 233 ret = add_extent_mapping(em_tree, split); 234 BUG_ON(ret); 235 free_extent_map(split); 236 split = split2; 237 split2 = NULL; 238 } 239 if (em->block_start < EXTENT_MAP_LAST_BYTE && 240 testend && em->start + em->len > start + len) { 241 u64 diff = start + len - em->start; 242 243 split->start = start + len; 244 split->len = em->start + em->len - (start + len); 245 split->bdev = em->bdev; 246 split->flags = flags; 247 248 if (compressed) { 249 split->block_len = em->block_len; 250 split->block_start = em->block_start; 251 split->orig_start = em->orig_start; 252 } else { 253 split->block_len = split->len; 254 split->block_start = em->block_start + diff; 255 split->orig_start = split->start; 256 } 257 258 ret = add_extent_mapping(em_tree, split); 259 BUG_ON(ret); 260 free_extent_map(split); 261 split = NULL; 262 } 263 spin_unlock(&em_tree->lock); 264 265 /* once for us */ 266 free_extent_map(em); 267 /* once for the tree*/ 268 free_extent_map(em); 269 } 270 if (split) 271 free_extent_map(split); 272 if (split2) 273 free_extent_map(split2); 274 return 0; 275 } 276 277 /* 278 * this is very complex, but the basic idea is to drop all extents 279 * in the range start - end. hint_block is filled in with a block number 280 * that would be a good hint to the block allocator for this file. 281 * 282 * If an extent intersects the range but is not entirely inside the range 283 * it is either truncated or split. Anything entirely inside the range 284 * is deleted from the tree. 285 * 286 * inline_limit is used to tell this code which offsets in the file to keep 287 * if they contain inline extents. 288 */ 289 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans, 290 struct btrfs_root *root, struct inode *inode, 291 u64 start, u64 end, u64 locked_end, 292 u64 inline_limit, u64 *hint_byte) 293 { 294 u64 extent_end = 0; 295 u64 search_start = start; 296 u64 ram_bytes = 0; 297 u64 disk_bytenr = 0; 298 u64 orig_locked_end = locked_end; 299 u8 compression; 300 u8 encryption; 301 u16 other_encoding = 0; 302 struct extent_buffer *leaf; 303 struct btrfs_file_extent_item *extent; 304 struct btrfs_path *path; 305 struct btrfs_key key; 306 struct btrfs_file_extent_item old; 307 int keep; 308 int slot; 309 int bookend; 310 int found_type = 0; 311 int found_extent; 312 int found_inline; 313 int recow; 314 int ret; 315 316 inline_limit = 0; 317 btrfs_drop_extent_cache(inode, start, end - 1, 0); 318 319 path = btrfs_alloc_path(); 320 if (!path) 321 return -ENOMEM; 322 while (1) { 323 recow = 0; 324 btrfs_release_path(root, path); 325 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, 326 search_start, -1); 327 if (ret < 0) 328 goto out; 329 if (ret > 0) { 330 if (path->slots[0] == 0) { 331 ret = 0; 332 goto out; 333 } 334 path->slots[0]--; 335 } 336 next_slot: 337 keep = 0; 338 bookend = 0; 339 found_extent = 0; 340 found_inline = 0; 341 compression = 0; 342 encryption = 0; 343 extent = NULL; 344 leaf = path->nodes[0]; 345 slot = path->slots[0]; 346 ret = 0; 347 btrfs_item_key_to_cpu(leaf, &key, slot); 348 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY && 349 key.offset >= end) { 350 goto out; 351 } 352 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY || 353 key.objectid != inode->i_ino) { 354 goto out; 355 } 356 if (recow) { 357 search_start = max(key.offset, start); 358 continue; 359 } 360 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) { 361 extent = btrfs_item_ptr(leaf, slot, 362 struct btrfs_file_extent_item); 363 found_type = btrfs_file_extent_type(leaf, extent); 364 compression = btrfs_file_extent_compression(leaf, 365 extent); 366 encryption = btrfs_file_extent_encryption(leaf, 367 extent); 368 other_encoding = btrfs_file_extent_other_encoding(leaf, 369 extent); 370 if (found_type == BTRFS_FILE_EXTENT_REG || 371 found_type == BTRFS_FILE_EXTENT_PREALLOC) { 372 extent_end = 373 btrfs_file_extent_disk_bytenr(leaf, 374 extent); 375 if (extent_end) 376 *hint_byte = extent_end; 377 378 extent_end = key.offset + 379 btrfs_file_extent_num_bytes(leaf, extent); 380 ram_bytes = btrfs_file_extent_ram_bytes(leaf, 381 extent); 382 found_extent = 1; 383 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { 384 found_inline = 1; 385 extent_end = key.offset + 386 btrfs_file_extent_inline_len(leaf, extent); 387 } 388 } else { 389 extent_end = search_start; 390 } 391 392 /* we found nothing we can drop */ 393 if ((!found_extent && !found_inline) || 394 search_start >= extent_end) { 395 int nextret; 396 u32 nritems; 397 nritems = btrfs_header_nritems(leaf); 398 if (slot >= nritems - 1) { 399 nextret = btrfs_next_leaf(root, path); 400 if (nextret) 401 goto out; 402 recow = 1; 403 } else { 404 path->slots[0]++; 405 } 406 goto next_slot; 407 } 408 409 if (end <= extent_end && start >= key.offset && found_inline) 410 *hint_byte = EXTENT_MAP_INLINE; 411 412 if (found_extent) { 413 read_extent_buffer(leaf, &old, (unsigned long)extent, 414 sizeof(old)); 415 } 416 417 if (end < extent_end && end >= key.offset) { 418 bookend = 1; 419 if (found_inline && start <= key.offset) 420 keep = 1; 421 } 422 423 if (bookend && found_extent) { 424 if (locked_end < extent_end) { 425 ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 426 locked_end, extent_end - 1, 427 GFP_NOFS); 428 if (!ret) { 429 btrfs_release_path(root, path); 430 lock_extent(&BTRFS_I(inode)->io_tree, 431 locked_end, extent_end - 1, 432 GFP_NOFS); 433 locked_end = extent_end; 434 continue; 435 } 436 locked_end = extent_end; 437 } 438 disk_bytenr = le64_to_cpu(old.disk_bytenr); 439 if (disk_bytenr != 0) { 440 ret = btrfs_inc_extent_ref(trans, root, 441 disk_bytenr, 442 le64_to_cpu(old.disk_num_bytes), 0, 443 root->root_key.objectid, 444 key.objectid, key.offset - 445 le64_to_cpu(old.offset)); 446 BUG_ON(ret); 447 } 448 } 449 450 if (found_inline) { 451 u64 mask = root->sectorsize - 1; 452 search_start = (extent_end + mask) & ~mask; 453 } else 454 search_start = extent_end; 455 456 /* truncate existing extent */ 457 if (start > key.offset) { 458 u64 new_num; 459 u64 old_num; 460 keep = 1; 461 WARN_ON(start & (root->sectorsize - 1)); 462 if (found_extent) { 463 new_num = start - key.offset; 464 old_num = btrfs_file_extent_num_bytes(leaf, 465 extent); 466 *hint_byte = 467 btrfs_file_extent_disk_bytenr(leaf, 468 extent); 469 if (btrfs_file_extent_disk_bytenr(leaf, 470 extent)) { 471 inode_sub_bytes(inode, old_num - 472 new_num); 473 } 474 btrfs_set_file_extent_num_bytes(leaf, 475 extent, new_num); 476 btrfs_mark_buffer_dirty(leaf); 477 } else if (key.offset < inline_limit && 478 (end > extent_end) && 479 (inline_limit < extent_end)) { 480 u32 new_size; 481 new_size = btrfs_file_extent_calc_inline_size( 482 inline_limit - key.offset); 483 inode_sub_bytes(inode, extent_end - 484 inline_limit); 485 btrfs_set_file_extent_ram_bytes(leaf, extent, 486 new_size); 487 if (!compression && !encryption) { 488 btrfs_truncate_item(trans, root, path, 489 new_size, 1); 490 } 491 } 492 } 493 /* delete the entire extent */ 494 if (!keep) { 495 if (found_inline) 496 inode_sub_bytes(inode, extent_end - 497 key.offset); 498 ret = btrfs_del_item(trans, root, path); 499 /* TODO update progress marker and return */ 500 BUG_ON(ret); 501 extent = NULL; 502 btrfs_release_path(root, path); 503 /* the extent will be freed later */ 504 } 505 if (bookend && found_inline && start <= key.offset) { 506 u32 new_size; 507 new_size = btrfs_file_extent_calc_inline_size( 508 extent_end - end); 509 inode_sub_bytes(inode, end - key.offset); 510 btrfs_set_file_extent_ram_bytes(leaf, extent, 511 new_size); 512 if (!compression && !encryption) 513 ret = btrfs_truncate_item(trans, root, path, 514 new_size, 0); 515 BUG_ON(ret); 516 } 517 /* create bookend, splitting the extent in two */ 518 if (bookend && found_extent) { 519 struct btrfs_key ins; 520 ins.objectid = inode->i_ino; 521 ins.offset = end; 522 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY); 523 524 btrfs_release_path(root, path); 525 path->leave_spinning = 1; 526 ret = btrfs_insert_empty_item(trans, root, path, &ins, 527 sizeof(*extent)); 528 BUG_ON(ret); 529 530 leaf = path->nodes[0]; 531 extent = btrfs_item_ptr(leaf, path->slots[0], 532 struct btrfs_file_extent_item); 533 write_extent_buffer(leaf, &old, 534 (unsigned long)extent, sizeof(old)); 535 536 btrfs_set_file_extent_compression(leaf, extent, 537 compression); 538 btrfs_set_file_extent_encryption(leaf, extent, 539 encryption); 540 btrfs_set_file_extent_other_encoding(leaf, extent, 541 other_encoding); 542 btrfs_set_file_extent_offset(leaf, extent, 543 le64_to_cpu(old.offset) + end - key.offset); 544 WARN_ON(le64_to_cpu(old.num_bytes) < 545 (extent_end - end)); 546 btrfs_set_file_extent_num_bytes(leaf, extent, 547 extent_end - end); 548 549 /* 550 * set the ram bytes to the size of the full extent 551 * before splitting. This is a worst case flag, 552 * but its the best we can do because we don't know 553 * how splitting affects compression 554 */ 555 btrfs_set_file_extent_ram_bytes(leaf, extent, 556 ram_bytes); 557 btrfs_set_file_extent_type(leaf, extent, found_type); 558 559 btrfs_unlock_up_safe(path, 1); 560 btrfs_mark_buffer_dirty(path->nodes[0]); 561 btrfs_set_lock_blocking(path->nodes[0]); 562 563 path->leave_spinning = 0; 564 btrfs_release_path(root, path); 565 if (disk_bytenr != 0) 566 inode_add_bytes(inode, extent_end - end); 567 } 568 569 if (found_extent && !keep) { 570 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr); 571 572 if (old_disk_bytenr != 0) { 573 inode_sub_bytes(inode, 574 le64_to_cpu(old.num_bytes)); 575 ret = btrfs_free_extent(trans, root, 576 old_disk_bytenr, 577 le64_to_cpu(old.disk_num_bytes), 578 0, root->root_key.objectid, 579 key.objectid, key.offset - 580 le64_to_cpu(old.offset)); 581 BUG_ON(ret); 582 *hint_byte = old_disk_bytenr; 583 } 584 } 585 586 if (search_start >= end) { 587 ret = 0; 588 goto out; 589 } 590 } 591 out: 592 btrfs_free_path(path); 593 if (locked_end > orig_locked_end) { 594 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end, 595 locked_end - 1, GFP_NOFS); 596 } 597 return ret; 598 } 599 600 static int extent_mergeable(struct extent_buffer *leaf, int slot, 601 u64 objectid, u64 bytenr, u64 *start, u64 *end) 602 { 603 struct btrfs_file_extent_item *fi; 604 struct btrfs_key key; 605 u64 extent_end; 606 607 if (slot < 0 || slot >= btrfs_header_nritems(leaf)) 608 return 0; 609 610 btrfs_item_key_to_cpu(leaf, &key, slot); 611 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) 612 return 0; 613 614 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 615 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || 616 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || 617 btrfs_file_extent_compression(leaf, fi) || 618 btrfs_file_extent_encryption(leaf, fi) || 619 btrfs_file_extent_other_encoding(leaf, fi)) 620 return 0; 621 622 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); 623 if ((*start && *start != key.offset) || (*end && *end != extent_end)) 624 return 0; 625 626 *start = key.offset; 627 *end = extent_end; 628 return 1; 629 } 630 631 /* 632 * Mark extent in the range start - end as written. 633 * 634 * This changes extent type from 'pre-allocated' to 'regular'. If only 635 * part of extent is marked as written, the extent will be split into 636 * two or three. 637 */ 638 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, 639 struct btrfs_root *root, 640 struct inode *inode, u64 start, u64 end) 641 { 642 struct extent_buffer *leaf; 643 struct btrfs_path *path; 644 struct btrfs_file_extent_item *fi; 645 struct btrfs_key key; 646 u64 bytenr; 647 u64 num_bytes; 648 u64 extent_end; 649 u64 orig_offset; 650 u64 other_start; 651 u64 other_end; 652 u64 split = start; 653 u64 locked_end = end; 654 int extent_type; 655 int split_end = 1; 656 int ret; 657 658 btrfs_drop_extent_cache(inode, start, end - 1, 0); 659 660 path = btrfs_alloc_path(); 661 BUG_ON(!path); 662 again: 663 key.objectid = inode->i_ino; 664 key.type = BTRFS_EXTENT_DATA_KEY; 665 if (split == start) 666 key.offset = split; 667 else 668 key.offset = split - 1; 669 670 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 671 if (ret > 0 && path->slots[0] > 0) 672 path->slots[0]--; 673 674 leaf = path->nodes[0]; 675 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 676 BUG_ON(key.objectid != inode->i_ino || 677 key.type != BTRFS_EXTENT_DATA_KEY); 678 fi = btrfs_item_ptr(leaf, path->slots[0], 679 struct btrfs_file_extent_item); 680 extent_type = btrfs_file_extent_type(leaf, fi); 681 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC); 682 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); 683 BUG_ON(key.offset > start || extent_end < end); 684 685 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 686 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 687 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); 688 689 if (key.offset == start) 690 split = end; 691 692 if (key.offset == start && extent_end == end) { 693 int del_nr = 0; 694 int del_slot = 0; 695 other_start = end; 696 other_end = 0; 697 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino, 698 bytenr, &other_start, &other_end)) { 699 extent_end = other_end; 700 del_slot = path->slots[0] + 1; 701 del_nr++; 702 ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 703 0, root->root_key.objectid, 704 inode->i_ino, orig_offset); 705 BUG_ON(ret); 706 } 707 other_start = 0; 708 other_end = start; 709 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino, 710 bytenr, &other_start, &other_end)) { 711 key.offset = other_start; 712 del_slot = path->slots[0]; 713 del_nr++; 714 ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 715 0, root->root_key.objectid, 716 inode->i_ino, orig_offset); 717 BUG_ON(ret); 718 } 719 split_end = 0; 720 if (del_nr == 0) { 721 btrfs_set_file_extent_type(leaf, fi, 722 BTRFS_FILE_EXTENT_REG); 723 goto done; 724 } 725 726 fi = btrfs_item_ptr(leaf, del_slot - 1, 727 struct btrfs_file_extent_item); 728 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG); 729 btrfs_set_file_extent_num_bytes(leaf, fi, 730 extent_end - key.offset); 731 btrfs_mark_buffer_dirty(leaf); 732 733 ret = btrfs_del_items(trans, root, path, del_slot, del_nr); 734 BUG_ON(ret); 735 goto release; 736 } else if (split == start) { 737 if (locked_end < extent_end) { 738 ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 739 locked_end, extent_end - 1, GFP_NOFS); 740 if (!ret) { 741 btrfs_release_path(root, path); 742 lock_extent(&BTRFS_I(inode)->io_tree, 743 locked_end, extent_end - 1, GFP_NOFS); 744 locked_end = extent_end; 745 goto again; 746 } 747 locked_end = extent_end; 748 } 749 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset); 750 } else { 751 BUG_ON(key.offset != start); 752 key.offset = split; 753 btrfs_set_file_extent_offset(leaf, fi, key.offset - 754 orig_offset); 755 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split); 756 btrfs_set_item_key_safe(trans, root, path, &key); 757 extent_end = split; 758 } 759 760 if (extent_end == end) { 761 split_end = 0; 762 extent_type = BTRFS_FILE_EXTENT_REG; 763 } 764 if (extent_end == end && split == start) { 765 other_start = end; 766 other_end = 0; 767 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino, 768 bytenr, &other_start, &other_end)) { 769 path->slots[0]++; 770 fi = btrfs_item_ptr(leaf, path->slots[0], 771 struct btrfs_file_extent_item); 772 key.offset = split; 773 btrfs_set_item_key_safe(trans, root, path, &key); 774 btrfs_set_file_extent_offset(leaf, fi, key.offset - 775 orig_offset); 776 btrfs_set_file_extent_num_bytes(leaf, fi, 777 other_end - split); 778 goto done; 779 } 780 } 781 if (extent_end == end && split == end) { 782 other_start = 0; 783 other_end = start; 784 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino, 785 bytenr, &other_start, &other_end)) { 786 path->slots[0]--; 787 fi = btrfs_item_ptr(leaf, path->slots[0], 788 struct btrfs_file_extent_item); 789 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - 790 other_start); 791 goto done; 792 } 793 } 794 795 btrfs_mark_buffer_dirty(leaf); 796 797 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, 798 root->root_key.objectid, 799 inode->i_ino, orig_offset); 800 BUG_ON(ret); 801 btrfs_release_path(root, path); 802 803 key.offset = start; 804 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi)); 805 BUG_ON(ret); 806 807 leaf = path->nodes[0]; 808 fi = btrfs_item_ptr(leaf, path->slots[0], 809 struct btrfs_file_extent_item); 810 btrfs_set_file_extent_generation(leaf, fi, trans->transid); 811 btrfs_set_file_extent_type(leaf, fi, extent_type); 812 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr); 813 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes); 814 btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset); 815 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset); 816 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); 817 btrfs_set_file_extent_compression(leaf, fi, 0); 818 btrfs_set_file_extent_encryption(leaf, fi, 0); 819 btrfs_set_file_extent_other_encoding(leaf, fi, 0); 820 done: 821 btrfs_mark_buffer_dirty(leaf); 822 823 release: 824 btrfs_release_path(root, path); 825 if (split_end && split == start) { 826 split = end; 827 goto again; 828 } 829 if (locked_end > end) { 830 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1, 831 GFP_NOFS); 832 } 833 btrfs_free_path(path); 834 return 0; 835 } 836 837 /* 838 * this gets pages into the page cache and locks them down, it also properly 839 * waits for data=ordered extents to finish before allowing the pages to be 840 * modified. 841 */ 842 static noinline int prepare_pages(struct btrfs_root *root, struct file *file, 843 struct page **pages, size_t num_pages, 844 loff_t pos, unsigned long first_index, 845 unsigned long last_index, size_t write_bytes) 846 { 847 int i; 848 unsigned long index = pos >> PAGE_CACHE_SHIFT; 849 struct inode *inode = fdentry(file)->d_inode; 850 int err = 0; 851 u64 start_pos; 852 u64 last_pos; 853 854 start_pos = pos & ~((u64)root->sectorsize - 1); 855 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; 856 857 if (start_pos > inode->i_size) { 858 err = btrfs_cont_expand(inode, start_pos); 859 if (err) 860 return err; 861 } 862 863 memset(pages, 0, num_pages * sizeof(struct page *)); 864 again: 865 for (i = 0; i < num_pages; i++) { 866 pages[i] = grab_cache_page(inode->i_mapping, index + i); 867 if (!pages[i]) { 868 err = -ENOMEM; 869 BUG_ON(1); 870 } 871 wait_on_page_writeback(pages[i]); 872 } 873 if (start_pos < inode->i_size) { 874 struct btrfs_ordered_extent *ordered; 875 lock_extent(&BTRFS_I(inode)->io_tree, 876 start_pos, last_pos - 1, GFP_NOFS); 877 ordered = btrfs_lookup_first_ordered_extent(inode, 878 last_pos - 1); 879 if (ordered && 880 ordered->file_offset + ordered->len > start_pos && 881 ordered->file_offset < last_pos) { 882 btrfs_put_ordered_extent(ordered); 883 unlock_extent(&BTRFS_I(inode)->io_tree, 884 start_pos, last_pos - 1, GFP_NOFS); 885 for (i = 0; i < num_pages; i++) { 886 unlock_page(pages[i]); 887 page_cache_release(pages[i]); 888 } 889 btrfs_wait_ordered_range(inode, start_pos, 890 last_pos - start_pos); 891 goto again; 892 } 893 if (ordered) 894 btrfs_put_ordered_extent(ordered); 895 896 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos, 897 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC, 898 GFP_NOFS); 899 unlock_extent(&BTRFS_I(inode)->io_tree, 900 start_pos, last_pos - 1, GFP_NOFS); 901 } 902 for (i = 0; i < num_pages; i++) { 903 clear_page_dirty_for_io(pages[i]); 904 set_page_extent_mapped(pages[i]); 905 WARN_ON(!PageLocked(pages[i])); 906 } 907 return 0; 908 } 909 910 static ssize_t btrfs_file_write(struct file *file, const char __user *buf, 911 size_t count, loff_t *ppos) 912 { 913 loff_t pos; 914 loff_t start_pos; 915 ssize_t num_written = 0; 916 ssize_t err = 0; 917 int ret = 0; 918 struct inode *inode = fdentry(file)->d_inode; 919 struct btrfs_root *root = BTRFS_I(inode)->root; 920 struct page **pages = NULL; 921 int nrptrs; 922 struct page *pinned[2]; 923 unsigned long first_index; 924 unsigned long last_index; 925 int will_write; 926 927 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) || 928 (file->f_flags & O_DIRECT)); 929 930 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE, 931 PAGE_CACHE_SIZE / (sizeof(struct page *))); 932 pinned[0] = NULL; 933 pinned[1] = NULL; 934 935 pos = *ppos; 936 start_pos = pos; 937 938 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 939 current->backing_dev_info = inode->i_mapping->backing_dev_info; 940 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 941 if (err) 942 goto out_nolock; 943 if (count == 0) 944 goto out_nolock; 945 946 err = file_remove_suid(file); 947 if (err) 948 goto out_nolock; 949 file_update_time(file); 950 951 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); 952 953 mutex_lock(&inode->i_mutex); 954 BTRFS_I(inode)->sequence++; 955 first_index = pos >> PAGE_CACHE_SHIFT; 956 last_index = (pos + count) >> PAGE_CACHE_SHIFT; 957 958 /* 959 * there are lots of better ways to do this, but this code 960 * makes sure the first and last page in the file range are 961 * up to date and ready for cow 962 */ 963 if ((pos & (PAGE_CACHE_SIZE - 1))) { 964 pinned[0] = grab_cache_page(inode->i_mapping, first_index); 965 if (!PageUptodate(pinned[0])) { 966 ret = btrfs_readpage(NULL, pinned[0]); 967 BUG_ON(ret); 968 wait_on_page_locked(pinned[0]); 969 } else { 970 unlock_page(pinned[0]); 971 } 972 } 973 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) { 974 pinned[1] = grab_cache_page(inode->i_mapping, last_index); 975 if (!PageUptodate(pinned[1])) { 976 ret = btrfs_readpage(NULL, pinned[1]); 977 BUG_ON(ret); 978 wait_on_page_locked(pinned[1]); 979 } else { 980 unlock_page(pinned[1]); 981 } 982 } 983 984 while (count > 0) { 985 size_t offset = pos & (PAGE_CACHE_SIZE - 1); 986 size_t write_bytes = min(count, nrptrs * 987 (size_t)PAGE_CACHE_SIZE - 988 offset); 989 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >> 990 PAGE_CACHE_SHIFT; 991 992 WARN_ON(num_pages > nrptrs); 993 memset(pages, 0, sizeof(struct page *) * nrptrs); 994 995 ret = btrfs_check_data_free_space(root, inode, write_bytes); 996 if (ret) 997 goto out; 998 999 ret = prepare_pages(root, file, pages, num_pages, 1000 pos, first_index, last_index, 1001 write_bytes); 1002 if (ret) { 1003 btrfs_free_reserved_data_space(root, inode, 1004 write_bytes); 1005 goto out; 1006 } 1007 1008 ret = btrfs_copy_from_user(pos, num_pages, 1009 write_bytes, pages, buf); 1010 if (ret) { 1011 btrfs_free_reserved_data_space(root, inode, 1012 write_bytes); 1013 btrfs_drop_pages(pages, num_pages); 1014 goto out; 1015 } 1016 1017 ret = dirty_and_release_pages(NULL, root, file, pages, 1018 num_pages, pos, write_bytes); 1019 btrfs_drop_pages(pages, num_pages); 1020 if (ret) { 1021 btrfs_free_reserved_data_space(root, inode, 1022 write_bytes); 1023 goto out; 1024 } 1025 1026 if (will_write) { 1027 btrfs_fdatawrite_range(inode->i_mapping, pos, 1028 pos + write_bytes - 1, 1029 WB_SYNC_ALL); 1030 } else { 1031 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1032 num_pages); 1033 if (num_pages < 1034 (root->leafsize >> PAGE_CACHE_SHIFT) + 1) 1035 btrfs_btree_balance_dirty(root, 1); 1036 btrfs_throttle(root); 1037 } 1038 1039 buf += write_bytes; 1040 count -= write_bytes; 1041 pos += write_bytes; 1042 num_written += write_bytes; 1043 1044 cond_resched(); 1045 } 1046 out: 1047 mutex_unlock(&inode->i_mutex); 1048 if (ret) 1049 err = ret; 1050 1051 out_nolock: 1052 kfree(pages); 1053 if (pinned[0]) 1054 page_cache_release(pinned[0]); 1055 if (pinned[1]) 1056 page_cache_release(pinned[1]); 1057 *ppos = pos; 1058 1059 /* 1060 * we want to make sure fsync finds this change 1061 * but we haven't joined a transaction running right now. 1062 * 1063 * Later on, someone is sure to update the inode and get the 1064 * real transid recorded. 1065 * 1066 * We set last_trans now to the fs_info generation + 1, 1067 * this will either be one more than the running transaction 1068 * or the generation used for the next transaction if there isn't 1069 * one running right now. 1070 */ 1071 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; 1072 1073 if (num_written > 0 && will_write) { 1074 struct btrfs_trans_handle *trans; 1075 1076 err = btrfs_wait_ordered_range(inode, start_pos, num_written); 1077 if (err) 1078 num_written = err; 1079 1080 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) { 1081 trans = btrfs_start_transaction(root, 1); 1082 ret = btrfs_log_dentry_safe(trans, root, 1083 file->f_dentry); 1084 if (ret == 0) { 1085 ret = btrfs_sync_log(trans, root); 1086 if (ret == 0) 1087 btrfs_end_transaction(trans, root); 1088 else 1089 btrfs_commit_transaction(trans, root); 1090 } else { 1091 btrfs_commit_transaction(trans, root); 1092 } 1093 } 1094 if (file->f_flags & O_DIRECT) { 1095 invalidate_mapping_pages(inode->i_mapping, 1096 start_pos >> PAGE_CACHE_SHIFT, 1097 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT); 1098 } 1099 } 1100 current->backing_dev_info = NULL; 1101 return num_written ? num_written : err; 1102 } 1103 1104 int btrfs_release_file(struct inode *inode, struct file *filp) 1105 { 1106 /* 1107 * ordered_data_close is set by settattr when we are about to truncate 1108 * a file from a non-zero size to a zero size. This tries to 1109 * flush down new bytes that may have been written if the 1110 * application were using truncate to replace a file in place. 1111 */ 1112 if (BTRFS_I(inode)->ordered_data_close) { 1113 BTRFS_I(inode)->ordered_data_close = 0; 1114 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); 1115 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) 1116 filemap_flush(inode->i_mapping); 1117 } 1118 if (filp->private_data) 1119 btrfs_ioctl_trans_end(filp); 1120 return 0; 1121 } 1122 1123 /* 1124 * fsync call for both files and directories. This logs the inode into 1125 * the tree log instead of forcing full commits whenever possible. 1126 * 1127 * It needs to call filemap_fdatawait so that all ordered extent updates are 1128 * in the metadata btree are up to date for copying to the log. 1129 * 1130 * It drops the inode mutex before doing the tree log commit. This is an 1131 * important optimization for directories because holding the mutex prevents 1132 * new operations on the dir while we write to disk. 1133 */ 1134 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) 1135 { 1136 struct inode *inode = dentry->d_inode; 1137 struct btrfs_root *root = BTRFS_I(inode)->root; 1138 int ret = 0; 1139 struct btrfs_trans_handle *trans; 1140 1141 /* 1142 * check the transaction that last modified this inode 1143 * and see if its already been committed 1144 */ 1145 if (!BTRFS_I(inode)->last_trans) 1146 goto out; 1147 1148 mutex_lock(&root->fs_info->trans_mutex); 1149 if (BTRFS_I(inode)->last_trans <= 1150 root->fs_info->last_trans_committed) { 1151 BTRFS_I(inode)->last_trans = 0; 1152 mutex_unlock(&root->fs_info->trans_mutex); 1153 goto out; 1154 } 1155 mutex_unlock(&root->fs_info->trans_mutex); 1156 1157 root->log_batch++; 1158 filemap_fdatawrite(inode->i_mapping); 1159 btrfs_wait_ordered_range(inode, 0, (u64)-1); 1160 root->log_batch++; 1161 1162 if (datasync && !(inode->i_state & I_DIRTY_PAGES)) 1163 goto out; 1164 /* 1165 * ok we haven't committed the transaction yet, lets do a commit 1166 */ 1167 if (file && file->private_data) 1168 btrfs_ioctl_trans_end(file); 1169 1170 trans = btrfs_start_transaction(root, 1); 1171 if (!trans) { 1172 ret = -ENOMEM; 1173 goto out; 1174 } 1175 1176 ret = btrfs_log_dentry_safe(trans, root, dentry); 1177 if (ret < 0) 1178 goto out; 1179 1180 /* we've logged all the items and now have a consistent 1181 * version of the file in the log. It is possible that 1182 * someone will come in and modify the file, but that's 1183 * fine because the log is consistent on disk, and we 1184 * have references to all of the file's extents 1185 * 1186 * It is possible that someone will come in and log the 1187 * file again, but that will end up using the synchronization 1188 * inside btrfs_sync_log to keep things safe. 1189 */ 1190 mutex_unlock(&dentry->d_inode->i_mutex); 1191 1192 if (ret > 0) { 1193 ret = btrfs_commit_transaction(trans, root); 1194 } else { 1195 ret = btrfs_sync_log(trans, root); 1196 if (ret == 0) 1197 ret = btrfs_end_transaction(trans, root); 1198 else 1199 ret = btrfs_commit_transaction(trans, root); 1200 } 1201 mutex_lock(&dentry->d_inode->i_mutex); 1202 out: 1203 return ret > 0 ? EIO : ret; 1204 } 1205 1206 static struct vm_operations_struct btrfs_file_vm_ops = { 1207 .fault = filemap_fault, 1208 .page_mkwrite = btrfs_page_mkwrite, 1209 }; 1210 1211 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) 1212 { 1213 vma->vm_ops = &btrfs_file_vm_ops; 1214 file_accessed(filp); 1215 return 0; 1216 } 1217 1218 struct file_operations btrfs_file_operations = { 1219 .llseek = generic_file_llseek, 1220 .read = do_sync_read, 1221 .aio_read = generic_file_aio_read, 1222 .splice_read = generic_file_splice_read, 1223 .write = btrfs_file_write, 1224 .mmap = btrfs_file_mmap, 1225 .open = generic_file_open, 1226 .release = btrfs_release_file, 1227 .fsync = btrfs_sync_file, 1228 .unlocked_ioctl = btrfs_ioctl, 1229 #ifdef CONFIG_COMPAT 1230 .compat_ioctl = btrfs_ioctl, 1231 #endif 1232 }; 1233