extents.c revision dd54567a83057821a667109a5cadec6eed0295de
1/* 2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com 3 * Written by Alex Tomas <alex@clusterfs.com> 4 * 5 * Architecture independence: 6 * Copyright (c) 2005, Bull S.A. 7 * Written by Pierre Peiffer <pierre.peiffer@bull.net> 8 * 9 * This program is free software; you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License version 2 as 11 * published by the Free Software Foundation. 12 * 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public Licens 19 * along with this program; if not, write to the Free Software 20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- 21 */ 22 23/* 24 * Extents support for EXT4 25 * 26 * TODO: 27 * - ext4*_error() should be used in some situations 28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate 29 * - smart tree reduction 30 */ 31 32#include <linux/module.h> 33#include <linux/fs.h> 34#include <linux/time.h> 35#include <linux/ext4_jbd2.h> 36#include <linux/jbd.h> 37#include <linux/highuid.h> 38#include <linux/pagemap.h> 39#include <linux/quotaops.h> 40#include <linux/string.h> 41#include <linux/slab.h> 42#include <linux/falloc.h> 43#include <linux/ext4_fs_extents.h> 44#include <asm/uaccess.h> 45 46 47/* 48 * ext_pblock: 49 * combine low and high parts of physical block number into ext4_fsblk_t 50 */ 51static ext4_fsblk_t ext_pblock(struct ext4_extent *ex) 52{ 53 ext4_fsblk_t block; 54 55 block = le32_to_cpu(ex->ee_start); 56 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1; 57 return block; 58} 59 60/* 61 * idx_pblock: 62 * combine low and high parts of a leaf physical block number into ext4_fsblk_t 63 */ 64static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix) 65{ 66 ext4_fsblk_t block; 67 68 block = le32_to_cpu(ix->ei_leaf); 69 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1; 70 return block; 71} 72 73/* 74 * ext4_ext_store_pblock: 75 * stores a large physical block number into an extent struct, 76 * breaking it into parts 77 */ 78static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb) 79{ 80 ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff)); 81 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); 82} 83 84/* 85 * ext4_idx_store_pblock: 86 * stores a large physical block number into an index struct, 87 * breaking it into parts 88 */ 89static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb) 90{ 91 ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff)); 92 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); 93} 94 95static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed) 96{ 97 int err; 98 99 if (handle->h_buffer_credits > needed) 100 return handle; 101 if (!ext4_journal_extend(handle, needed)) 102 return handle; 103 err = ext4_journal_restart(handle, needed); 104 105 return handle; 106} 107 108/* 109 * could return: 110 * - EROFS 111 * - ENOMEM 112 */ 113static int ext4_ext_get_access(handle_t *handle, struct inode *inode, 114 struct ext4_ext_path *path) 115{ 116 if (path->p_bh) { 117 /* path points to block */ 118 return ext4_journal_get_write_access(handle, path->p_bh); 119 } 120 /* path points to leaf/index in inode body */ 121 /* we use in-core data, no need to protect them */ 122 return 0; 123} 124 125/* 126 * could return: 127 * - EROFS 128 * - ENOMEM 129 * - EIO 130 */ 131static int ext4_ext_dirty(handle_t *handle, struct inode *inode, 132 struct ext4_ext_path *path) 133{ 134 int err; 135 if (path->p_bh) { 136 /* path points to block */ 137 err = ext4_journal_dirty_metadata(handle, path->p_bh); 138 } else { 139 /* path points to leaf/index in inode body */ 140 err = ext4_mark_inode_dirty(handle, inode); 141 } 142 return err; 143} 144 145static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, 146 struct ext4_ext_path *path, 147 ext4_fsblk_t block) 148{ 149 struct ext4_inode_info *ei = EXT4_I(inode); 150 ext4_fsblk_t bg_start; 151 ext4_grpblk_t colour; 152 int depth; 153 154 if (path) { 155 struct ext4_extent *ex; 156 depth = path->p_depth; 157 158 /* try to predict block placement */ 159 ex = path[depth].p_ext; 160 if (ex) 161 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block)); 162 163 /* it looks like index is empty; 164 * try to find starting block from index itself */ 165 if (path[depth].p_bh) 166 return path[depth].p_bh->b_blocknr; 167 } 168 169 /* OK. use inode's group */ 170 bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) + 171 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block); 172 colour = (current->pid % 16) * 173 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); 174 return bg_start + colour + block; 175} 176 177static ext4_fsblk_t 178ext4_ext_new_block(handle_t *handle, struct inode *inode, 179 struct ext4_ext_path *path, 180 struct ext4_extent *ex, int *err) 181{ 182 ext4_fsblk_t goal, newblock; 183 184 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); 185 newblock = ext4_new_block(handle, inode, goal, err); 186 return newblock; 187} 188 189static int ext4_ext_space_block(struct inode *inode) 190{ 191 int size; 192 193 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) 194 / sizeof(struct ext4_extent); 195#ifdef AGGRESSIVE_TEST 196 if (size > 6) 197 size = 6; 198#endif 199 return size; 200} 201 202static int ext4_ext_space_block_idx(struct inode *inode) 203{ 204 int size; 205 206 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) 207 / sizeof(struct ext4_extent_idx); 208#ifdef AGGRESSIVE_TEST 209 if (size > 5) 210 size = 5; 211#endif 212 return size; 213} 214 215static int ext4_ext_space_root(struct inode *inode) 216{ 217 int size; 218 219 size = sizeof(EXT4_I(inode)->i_data); 220 size -= sizeof(struct ext4_extent_header); 221 size /= sizeof(struct ext4_extent); 222#ifdef AGGRESSIVE_TEST 223 if (size > 3) 224 size = 3; 225#endif 226 return size; 227} 228 229static int ext4_ext_space_root_idx(struct inode *inode) 230{ 231 int size; 232 233 size = sizeof(EXT4_I(inode)->i_data); 234 size -= sizeof(struct ext4_extent_header); 235 size /= sizeof(struct ext4_extent_idx); 236#ifdef AGGRESSIVE_TEST 237 if (size > 4) 238 size = 4; 239#endif 240 return size; 241} 242 243static int 244ext4_ext_max_entries(struct inode *inode, int depth) 245{ 246 int max; 247 248 if (depth == ext_depth(inode)) { 249 if (depth == 0) 250 max = ext4_ext_space_root(inode); 251 else 252 max = ext4_ext_space_root_idx(inode); 253 } else { 254 if (depth == 0) 255 max = ext4_ext_space_block(inode); 256 else 257 max = ext4_ext_space_block_idx(inode); 258 } 259 260 return max; 261} 262 263static int __ext4_ext_check_header(const char *function, struct inode *inode, 264 struct ext4_extent_header *eh, 265 int depth) 266{ 267 const char *error_msg; 268 int max = 0; 269 270 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { 271 error_msg = "invalid magic"; 272 goto corrupted; 273 } 274 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { 275 error_msg = "unexpected eh_depth"; 276 goto corrupted; 277 } 278 if (unlikely(eh->eh_max == 0)) { 279 error_msg = "invalid eh_max"; 280 goto corrupted; 281 } 282 max = ext4_ext_max_entries(inode, depth); 283 if (unlikely(le16_to_cpu(eh->eh_max) > max)) { 284 error_msg = "too large eh_max"; 285 goto corrupted; 286 } 287 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { 288 error_msg = "invalid eh_entries"; 289 goto corrupted; 290 } 291 return 0; 292 293corrupted: 294 ext4_error(inode->i_sb, function, 295 "bad header in inode #%lu: %s - magic %x, " 296 "entries %u, max %u(%u), depth %u(%u)", 297 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), 298 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), 299 max, le16_to_cpu(eh->eh_depth), depth); 300 301 return -EIO; 302} 303 304#define ext4_ext_check_header(inode, eh, depth) \ 305 __ext4_ext_check_header(__FUNCTION__, inode, eh, depth) 306 307#ifdef EXT_DEBUG 308static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) 309{ 310 int k, l = path->p_depth; 311 312 ext_debug("path:"); 313 for (k = 0; k <= l; k++, path++) { 314 if (path->p_idx) { 315 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), 316 idx_pblock(path->p_idx)); 317 } else if (path->p_ext) { 318 ext_debug(" %d:%d:%llu ", 319 le32_to_cpu(path->p_ext->ee_block), 320 ext4_ext_get_actual_len(path->p_ext), 321 ext_pblock(path->p_ext)); 322 } else 323 ext_debug(" []"); 324 } 325 ext_debug("\n"); 326} 327 328static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) 329{ 330 int depth = ext_depth(inode); 331 struct ext4_extent_header *eh; 332 struct ext4_extent *ex; 333 int i; 334 335 if (!path) 336 return; 337 338 eh = path[depth].p_hdr; 339 ex = EXT_FIRST_EXTENT(eh); 340 341 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { 342 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block), 343 ext4_ext_get_actual_len(ex), ext_pblock(ex)); 344 } 345 ext_debug("\n"); 346} 347#else 348#define ext4_ext_show_path(inode,path) 349#define ext4_ext_show_leaf(inode,path) 350#endif 351 352static void ext4_ext_drop_refs(struct ext4_ext_path *path) 353{ 354 int depth = path->p_depth; 355 int i; 356 357 for (i = 0; i <= depth; i++, path++) 358 if (path->p_bh) { 359 brelse(path->p_bh); 360 path->p_bh = NULL; 361 } 362} 363 364/* 365 * ext4_ext_binsearch_idx: 366 * binary search for the closest index of the given block 367 * the header must be checked before calling this 368 */ 369static void 370ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block) 371{ 372 struct ext4_extent_header *eh = path->p_hdr; 373 struct ext4_extent_idx *r, *l, *m; 374 375 376 ext_debug("binsearch for %d(idx): ", block); 377 378 l = EXT_FIRST_INDEX(eh) + 1; 379 r = EXT_LAST_INDEX(eh); 380 while (l <= r) { 381 m = l + (r - l) / 2; 382 if (block < le32_to_cpu(m->ei_block)) 383 r = m - 1; 384 else 385 l = m + 1; 386 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), 387 m, le32_to_cpu(m->ei_block), 388 r, le32_to_cpu(r->ei_block)); 389 } 390 391 path->p_idx = l - 1; 392 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), 393 idx_pblock(path->p_idx)); 394 395#ifdef CHECK_BINSEARCH 396 { 397 struct ext4_extent_idx *chix, *ix; 398 int k; 399 400 chix = ix = EXT_FIRST_INDEX(eh); 401 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { 402 if (k != 0 && 403 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { 404 printk("k=%d, ix=0x%p, first=0x%p\n", k, 405 ix, EXT_FIRST_INDEX(eh)); 406 printk("%u <= %u\n", 407 le32_to_cpu(ix->ei_block), 408 le32_to_cpu(ix[-1].ei_block)); 409 } 410 BUG_ON(k && le32_to_cpu(ix->ei_block) 411 <= le32_to_cpu(ix[-1].ei_block)); 412 if (block < le32_to_cpu(ix->ei_block)) 413 break; 414 chix = ix; 415 } 416 BUG_ON(chix != path->p_idx); 417 } 418#endif 419 420} 421 422/* 423 * ext4_ext_binsearch: 424 * binary search for closest extent of the given block 425 * the header must be checked before calling this 426 */ 427static void 428ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) 429{ 430 struct ext4_extent_header *eh = path->p_hdr; 431 struct ext4_extent *r, *l, *m; 432 433 if (eh->eh_entries == 0) { 434 /* 435 * this leaf is empty: 436 * we get such a leaf in split/add case 437 */ 438 return; 439 } 440 441 ext_debug("binsearch for %d: ", block); 442 443 l = EXT_FIRST_EXTENT(eh) + 1; 444 r = EXT_LAST_EXTENT(eh); 445 446 while (l <= r) { 447 m = l + (r - l) / 2; 448 if (block < le32_to_cpu(m->ee_block)) 449 r = m - 1; 450 else 451 l = m + 1; 452 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), 453 m, le32_to_cpu(m->ee_block), 454 r, le32_to_cpu(r->ee_block)); 455 } 456 457 path->p_ext = l - 1; 458 ext_debug(" -> %d:%llu:%d ", 459 le32_to_cpu(path->p_ext->ee_block), 460 ext_pblock(path->p_ext), 461 ext4_ext_get_actual_len(path->p_ext)); 462 463#ifdef CHECK_BINSEARCH 464 { 465 struct ext4_extent *chex, *ex; 466 int k; 467 468 chex = ex = EXT_FIRST_EXTENT(eh); 469 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { 470 BUG_ON(k && le32_to_cpu(ex->ee_block) 471 <= le32_to_cpu(ex[-1].ee_block)); 472 if (block < le32_to_cpu(ex->ee_block)) 473 break; 474 chex = ex; 475 } 476 BUG_ON(chex != path->p_ext); 477 } 478#endif 479 480} 481 482int ext4_ext_tree_init(handle_t *handle, struct inode *inode) 483{ 484 struct ext4_extent_header *eh; 485 486 eh = ext_inode_hdr(inode); 487 eh->eh_depth = 0; 488 eh->eh_entries = 0; 489 eh->eh_magic = EXT4_EXT_MAGIC; 490 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode)); 491 ext4_mark_inode_dirty(handle, inode); 492 ext4_ext_invalidate_cache(inode); 493 return 0; 494} 495 496struct ext4_ext_path * 497ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path) 498{ 499 struct ext4_extent_header *eh; 500 struct buffer_head *bh; 501 short int depth, i, ppos = 0, alloc = 0; 502 503 eh = ext_inode_hdr(inode); 504 depth = ext_depth(inode); 505 if (ext4_ext_check_header(inode, eh, depth)) 506 return ERR_PTR(-EIO); 507 508 509 /* account possible depth increase */ 510 if (!path) { 511 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), 512 GFP_NOFS); 513 if (!path) 514 return ERR_PTR(-ENOMEM); 515 alloc = 1; 516 } 517 path[0].p_hdr = eh; 518 519 i = depth; 520 /* walk through the tree */ 521 while (i) { 522 ext_debug("depth %d: num %d, max %d\n", 523 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); 524 525 ext4_ext_binsearch_idx(inode, path + ppos, block); 526 path[ppos].p_block = idx_pblock(path[ppos].p_idx); 527 path[ppos].p_depth = i; 528 path[ppos].p_ext = NULL; 529 530 bh = sb_bread(inode->i_sb, path[ppos].p_block); 531 if (!bh) 532 goto err; 533 534 eh = ext_block_hdr(bh); 535 ppos++; 536 BUG_ON(ppos > depth); 537 path[ppos].p_bh = bh; 538 path[ppos].p_hdr = eh; 539 i--; 540 541 if (ext4_ext_check_header(inode, eh, i)) 542 goto err; 543 } 544 545 path[ppos].p_depth = i; 546 path[ppos].p_hdr = eh; 547 path[ppos].p_ext = NULL; 548 path[ppos].p_idx = NULL; 549 550 /* find extent */ 551 ext4_ext_binsearch(inode, path + ppos, block); 552 553 ext4_ext_show_path(inode, path); 554 555 return path; 556 557err: 558 ext4_ext_drop_refs(path); 559 if (alloc) 560 kfree(path); 561 return ERR_PTR(-EIO); 562} 563 564/* 565 * ext4_ext_insert_index: 566 * insert new index [@logical;@ptr] into the block at @curp; 567 * check where to insert: before @curp or after @curp 568 */ 569static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, 570 struct ext4_ext_path *curp, 571 int logical, ext4_fsblk_t ptr) 572{ 573 struct ext4_extent_idx *ix; 574 int len, err; 575 576 err = ext4_ext_get_access(handle, inode, curp); 577 if (err) 578 return err; 579 580 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block)); 581 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; 582 if (logical > le32_to_cpu(curp->p_idx->ei_block)) { 583 /* insert after */ 584 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { 585 len = (len - 1) * sizeof(struct ext4_extent_idx); 586 len = len < 0 ? 0 : len; 587 ext_debug("insert new index %d after: %llu. " 588 "move %d from 0x%p to 0x%p\n", 589 logical, ptr, len, 590 (curp->p_idx + 1), (curp->p_idx + 2)); 591 memmove(curp->p_idx + 2, curp->p_idx + 1, len); 592 } 593 ix = curp->p_idx + 1; 594 } else { 595 /* insert before */ 596 len = len * sizeof(struct ext4_extent_idx); 597 len = len < 0 ? 0 : len; 598 ext_debug("insert new index %d before: %llu. " 599 "move %d from 0x%p to 0x%p\n", 600 logical, ptr, len, 601 curp->p_idx, (curp->p_idx + 1)); 602 memmove(curp->p_idx + 1, curp->p_idx, len); 603 ix = curp->p_idx; 604 } 605 606 ix->ei_block = cpu_to_le32(logical); 607 ext4_idx_store_pblock(ix, ptr); 608 curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1); 609 610 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries) 611 > le16_to_cpu(curp->p_hdr->eh_max)); 612 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr)); 613 614 err = ext4_ext_dirty(handle, inode, curp); 615 ext4_std_error(inode->i_sb, err); 616 617 return err; 618} 619 620/* 621 * ext4_ext_split: 622 * inserts new subtree into the path, using free index entry 623 * at depth @at: 624 * - allocates all needed blocks (new leaf and all intermediate index blocks) 625 * - makes decision where to split 626 * - moves remaining extents and index entries (right to the split point) 627 * into the newly allocated blocks 628 * - initializes subtree 629 */ 630static int ext4_ext_split(handle_t *handle, struct inode *inode, 631 struct ext4_ext_path *path, 632 struct ext4_extent *newext, int at) 633{ 634 struct buffer_head *bh = NULL; 635 int depth = ext_depth(inode); 636 struct ext4_extent_header *neh; 637 struct ext4_extent_idx *fidx; 638 struct ext4_extent *ex; 639 int i = at, k, m, a; 640 ext4_fsblk_t newblock, oldblock; 641 __le32 border; 642 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ 643 int err = 0; 644 645 /* make decision: where to split? */ 646 /* FIXME: now decision is simplest: at current extent */ 647 648 /* if current leaf will be split, then we should use 649 * border from split point */ 650 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr)); 651 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { 652 border = path[depth].p_ext[1].ee_block; 653 ext_debug("leaf will be split." 654 " next leaf starts at %d\n", 655 le32_to_cpu(border)); 656 } else { 657 border = newext->ee_block; 658 ext_debug("leaf will be added." 659 " next leaf starts at %d\n", 660 le32_to_cpu(border)); 661 } 662 663 /* 664 * If error occurs, then we break processing 665 * and mark filesystem read-only. index won't 666 * be inserted and tree will be in consistent 667 * state. Next mount will repair buffers too. 668 */ 669 670 /* 671 * Get array to track all allocated blocks. 672 * We need this to handle errors and free blocks 673 * upon them. 674 */ 675 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); 676 if (!ablocks) 677 return -ENOMEM; 678 679 /* allocate all needed blocks */ 680 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); 681 for (a = 0; a < depth - at; a++) { 682 newblock = ext4_ext_new_block(handle, inode, path, newext, &err); 683 if (newblock == 0) 684 goto cleanup; 685 ablocks[a] = newblock; 686 } 687 688 /* initialize new leaf */ 689 newblock = ablocks[--a]; 690 BUG_ON(newblock == 0); 691 bh = sb_getblk(inode->i_sb, newblock); 692 if (!bh) { 693 err = -EIO; 694 goto cleanup; 695 } 696 lock_buffer(bh); 697 698 err = ext4_journal_get_create_access(handle, bh); 699 if (err) 700 goto cleanup; 701 702 neh = ext_block_hdr(bh); 703 neh->eh_entries = 0; 704 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); 705 neh->eh_magic = EXT4_EXT_MAGIC; 706 neh->eh_depth = 0; 707 ex = EXT_FIRST_EXTENT(neh); 708 709 /* move remainder of path[depth] to the new leaf */ 710 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max); 711 /* start copy from next extent */ 712 /* TODO: we could do it by single memmove */ 713 m = 0; 714 path[depth].p_ext++; 715 while (path[depth].p_ext <= 716 EXT_MAX_EXTENT(path[depth].p_hdr)) { 717 ext_debug("move %d:%llu:%d in new leaf %llu\n", 718 le32_to_cpu(path[depth].p_ext->ee_block), 719 ext_pblock(path[depth].p_ext), 720 ext4_ext_get_actual_len(path[depth].p_ext), 721 newblock); 722 /*memmove(ex++, path[depth].p_ext++, 723 sizeof(struct ext4_extent)); 724 neh->eh_entries++;*/ 725 path[depth].p_ext++; 726 m++; 727 } 728 if (m) { 729 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); 730 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m); 731 } 732 733 set_buffer_uptodate(bh); 734 unlock_buffer(bh); 735 736 err = ext4_journal_dirty_metadata(handle, bh); 737 if (err) 738 goto cleanup; 739 brelse(bh); 740 bh = NULL; 741 742 /* correct old leaf */ 743 if (m) { 744 err = ext4_ext_get_access(handle, inode, path + depth); 745 if (err) 746 goto cleanup; 747 path[depth].p_hdr->eh_entries = 748 cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m); 749 err = ext4_ext_dirty(handle, inode, path + depth); 750 if (err) 751 goto cleanup; 752 753 } 754 755 /* create intermediate indexes */ 756 k = depth - at - 1; 757 BUG_ON(k < 0); 758 if (k) 759 ext_debug("create %d intermediate indices\n", k); 760 /* insert new index into current index block */ 761 /* current depth stored in i var */ 762 i = depth - 1; 763 while (k--) { 764 oldblock = newblock; 765 newblock = ablocks[--a]; 766 bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock); 767 if (!bh) { 768 err = -EIO; 769 goto cleanup; 770 } 771 lock_buffer(bh); 772 773 err = ext4_journal_get_create_access(handle, bh); 774 if (err) 775 goto cleanup; 776 777 neh = ext_block_hdr(bh); 778 neh->eh_entries = cpu_to_le16(1); 779 neh->eh_magic = EXT4_EXT_MAGIC; 780 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); 781 neh->eh_depth = cpu_to_le16(depth - i); 782 fidx = EXT_FIRST_INDEX(neh); 783 fidx->ei_block = border; 784 ext4_idx_store_pblock(fidx, oldblock); 785 786 ext_debug("int.index at %d (block %llu): %lu -> %llu\n", i, 787 newblock, (unsigned long) le32_to_cpu(border), 788 oldblock); 789 /* copy indexes */ 790 m = 0; 791 path[i].p_idx++; 792 793 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, 794 EXT_MAX_INDEX(path[i].p_hdr)); 795 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != 796 EXT_LAST_INDEX(path[i].p_hdr)); 797 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { 798 ext_debug("%d: move %d:%llu in new index %llu\n", i, 799 le32_to_cpu(path[i].p_idx->ei_block), 800 idx_pblock(path[i].p_idx), 801 newblock); 802 /*memmove(++fidx, path[i].p_idx++, 803 sizeof(struct ext4_extent_idx)); 804 neh->eh_entries++; 805 BUG_ON(neh->eh_entries > neh->eh_max);*/ 806 path[i].p_idx++; 807 m++; 808 } 809 if (m) { 810 memmove(++fidx, path[i].p_idx - m, 811 sizeof(struct ext4_extent_idx) * m); 812 neh->eh_entries = 813 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m); 814 } 815 set_buffer_uptodate(bh); 816 unlock_buffer(bh); 817 818 err = ext4_journal_dirty_metadata(handle, bh); 819 if (err) 820 goto cleanup; 821 brelse(bh); 822 bh = NULL; 823 824 /* correct old index */ 825 if (m) { 826 err = ext4_ext_get_access(handle, inode, path + i); 827 if (err) 828 goto cleanup; 829 path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m); 830 err = ext4_ext_dirty(handle, inode, path + i); 831 if (err) 832 goto cleanup; 833 } 834 835 i--; 836 } 837 838 /* insert new index */ 839 err = ext4_ext_insert_index(handle, inode, path + at, 840 le32_to_cpu(border), newblock); 841 842cleanup: 843 if (bh) { 844 if (buffer_locked(bh)) 845 unlock_buffer(bh); 846 brelse(bh); 847 } 848 849 if (err) { 850 /* free all allocated blocks in error case */ 851 for (i = 0; i < depth; i++) { 852 if (!ablocks[i]) 853 continue; 854 ext4_free_blocks(handle, inode, ablocks[i], 1); 855 } 856 } 857 kfree(ablocks); 858 859 return err; 860} 861 862/* 863 * ext4_ext_grow_indepth: 864 * implements tree growing procedure: 865 * - allocates new block 866 * - moves top-level data (index block or leaf) into the new block 867 * - initializes new top-level, creating index that points to the 868 * just created block 869 */ 870static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, 871 struct ext4_ext_path *path, 872 struct ext4_extent *newext) 873{ 874 struct ext4_ext_path *curp = path; 875 struct ext4_extent_header *neh; 876 struct ext4_extent_idx *fidx; 877 struct buffer_head *bh; 878 ext4_fsblk_t newblock; 879 int err = 0; 880 881 newblock = ext4_ext_new_block(handle, inode, path, newext, &err); 882 if (newblock == 0) 883 return err; 884 885 bh = sb_getblk(inode->i_sb, newblock); 886 if (!bh) { 887 err = -EIO; 888 ext4_std_error(inode->i_sb, err); 889 return err; 890 } 891 lock_buffer(bh); 892 893 err = ext4_journal_get_create_access(handle, bh); 894 if (err) { 895 unlock_buffer(bh); 896 goto out; 897 } 898 899 /* move top-level index/leaf into new block */ 900 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); 901 902 /* set size of new block */ 903 neh = ext_block_hdr(bh); 904 /* old root could have indexes or leaves 905 * so calculate e_max right way */ 906 if (ext_depth(inode)) 907 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); 908 else 909 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); 910 neh->eh_magic = EXT4_EXT_MAGIC; 911 set_buffer_uptodate(bh); 912 unlock_buffer(bh); 913 914 err = ext4_journal_dirty_metadata(handle, bh); 915 if (err) 916 goto out; 917 918 /* create index in new top-level index: num,max,pointer */ 919 err = ext4_ext_get_access(handle, inode, curp); 920 if (err) 921 goto out; 922 923 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; 924 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); 925 curp->p_hdr->eh_entries = cpu_to_le16(1); 926 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); 927 928 if (path[0].p_hdr->eh_depth) 929 curp->p_idx->ei_block = 930 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; 931 else 932 curp->p_idx->ei_block = 933 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; 934 ext4_idx_store_pblock(curp->p_idx, newblock); 935 936 neh = ext_inode_hdr(inode); 937 fidx = EXT_FIRST_INDEX(neh); 938 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", 939 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), 940 le32_to_cpu(fidx->ei_block), idx_pblock(fidx)); 941 942 neh->eh_depth = cpu_to_le16(path->p_depth + 1); 943 err = ext4_ext_dirty(handle, inode, curp); 944out: 945 brelse(bh); 946 947 return err; 948} 949 950/* 951 * ext4_ext_create_new_leaf: 952 * finds empty index and adds new leaf. 953 * if no free index is found, then it requests in-depth growing. 954 */ 955static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, 956 struct ext4_ext_path *path, 957 struct ext4_extent *newext) 958{ 959 struct ext4_ext_path *curp; 960 int depth, i, err = 0; 961 962repeat: 963 i = depth = ext_depth(inode); 964 965 /* walk up to the tree and look for free index entry */ 966 curp = path + depth; 967 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { 968 i--; 969 curp--; 970 } 971 972 /* we use already allocated block for index block, 973 * so subsequent data blocks should be contiguous */ 974 if (EXT_HAS_FREE_INDEX(curp)) { 975 /* if we found index with free entry, then use that 976 * entry: create all needed subtree and add new leaf */ 977 err = ext4_ext_split(handle, inode, path, newext, i); 978 979 /* refill path */ 980 ext4_ext_drop_refs(path); 981 path = ext4_ext_find_extent(inode, 982 le32_to_cpu(newext->ee_block), 983 path); 984 if (IS_ERR(path)) 985 err = PTR_ERR(path); 986 } else { 987 /* tree is full, time to grow in depth */ 988 err = ext4_ext_grow_indepth(handle, inode, path, newext); 989 if (err) 990 goto out; 991 992 /* refill path */ 993 ext4_ext_drop_refs(path); 994 path = ext4_ext_find_extent(inode, 995 le32_to_cpu(newext->ee_block), 996 path); 997 if (IS_ERR(path)) { 998 err = PTR_ERR(path); 999 goto out; 1000 } 1001 1002 /* 1003 * only first (depth 0 -> 1) produces free space; 1004 * in all other cases we have to split the grown tree 1005 */ 1006 depth = ext_depth(inode); 1007 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { 1008 /* now we need to split */ 1009 goto repeat; 1010 } 1011 } 1012 1013out: 1014 return err; 1015} 1016 1017/* 1018 * ext4_ext_next_allocated_block: 1019 * returns allocated block in subsequent extent or EXT_MAX_BLOCK. 1020 * NOTE: it considers block number from index entry as 1021 * allocated block. Thus, index entries have to be consistent 1022 * with leaves. 1023 */ 1024static unsigned long 1025ext4_ext_next_allocated_block(struct ext4_ext_path *path) 1026{ 1027 int depth; 1028 1029 BUG_ON(path == NULL); 1030 depth = path->p_depth; 1031 1032 if (depth == 0 && path->p_ext == NULL) 1033 return EXT_MAX_BLOCK; 1034 1035 while (depth >= 0) { 1036 if (depth == path->p_depth) { 1037 /* leaf */ 1038 if (path[depth].p_ext != 1039 EXT_LAST_EXTENT(path[depth].p_hdr)) 1040 return le32_to_cpu(path[depth].p_ext[1].ee_block); 1041 } else { 1042 /* index */ 1043 if (path[depth].p_idx != 1044 EXT_LAST_INDEX(path[depth].p_hdr)) 1045 return le32_to_cpu(path[depth].p_idx[1].ei_block); 1046 } 1047 depth--; 1048 } 1049 1050 return EXT_MAX_BLOCK; 1051} 1052 1053/* 1054 * ext4_ext_next_leaf_block: 1055 * returns first allocated block from next leaf or EXT_MAX_BLOCK 1056 */ 1057static unsigned ext4_ext_next_leaf_block(struct inode *inode, 1058 struct ext4_ext_path *path) 1059{ 1060 int depth; 1061 1062 BUG_ON(path == NULL); 1063 depth = path->p_depth; 1064 1065 /* zero-tree has no leaf blocks at all */ 1066 if (depth == 0) 1067 return EXT_MAX_BLOCK; 1068 1069 /* go to index block */ 1070 depth--; 1071 1072 while (depth >= 0) { 1073 if (path[depth].p_idx != 1074 EXT_LAST_INDEX(path[depth].p_hdr)) 1075 return le32_to_cpu(path[depth].p_idx[1].ei_block); 1076 depth--; 1077 } 1078 1079 return EXT_MAX_BLOCK; 1080} 1081 1082/* 1083 * ext4_ext_correct_indexes: 1084 * if leaf gets modified and modified extent is first in the leaf, 1085 * then we have to correct all indexes above. 1086 * TODO: do we need to correct tree in all cases? 1087 */ 1088int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, 1089 struct ext4_ext_path *path) 1090{ 1091 struct ext4_extent_header *eh; 1092 int depth = ext_depth(inode); 1093 struct ext4_extent *ex; 1094 __le32 border; 1095 int k, err = 0; 1096 1097 eh = path[depth].p_hdr; 1098 ex = path[depth].p_ext; 1099 BUG_ON(ex == NULL); 1100 BUG_ON(eh == NULL); 1101 1102 if (depth == 0) { 1103 /* there is no tree at all */ 1104 return 0; 1105 } 1106 1107 if (ex != EXT_FIRST_EXTENT(eh)) { 1108 /* we correct tree if first leaf got modified only */ 1109 return 0; 1110 } 1111 1112 /* 1113 * TODO: we need correction if border is smaller than current one 1114 */ 1115 k = depth - 1; 1116 border = path[depth].p_ext->ee_block; 1117 err = ext4_ext_get_access(handle, inode, path + k); 1118 if (err) 1119 return err; 1120 path[k].p_idx->ei_block = border; 1121 err = ext4_ext_dirty(handle, inode, path + k); 1122 if (err) 1123 return err; 1124 1125 while (k--) { 1126 /* change all left-side indexes */ 1127 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) 1128 break; 1129 err = ext4_ext_get_access(handle, inode, path + k); 1130 if (err) 1131 break; 1132 path[k].p_idx->ei_block = border; 1133 err = ext4_ext_dirty(handle, inode, path + k); 1134 if (err) 1135 break; 1136 } 1137 1138 return err; 1139} 1140 1141static int 1142ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, 1143 struct ext4_extent *ex2) 1144{ 1145 unsigned short ext1_ee_len, ext2_ee_len, max_len; 1146 1147 /* 1148 * Make sure that either both extents are uninitialized, or 1149 * both are _not_. 1150 */ 1151 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) 1152 return 0; 1153 1154 if (ext4_ext_is_uninitialized(ex1)) 1155 max_len = EXT_UNINIT_MAX_LEN; 1156 else 1157 max_len = EXT_INIT_MAX_LEN; 1158 1159 ext1_ee_len = ext4_ext_get_actual_len(ex1); 1160 ext2_ee_len = ext4_ext_get_actual_len(ex2); 1161 1162 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != 1163 le32_to_cpu(ex2->ee_block)) 1164 return 0; 1165 1166 /* 1167 * To allow future support for preallocated extents to be added 1168 * as an RO_COMPAT feature, refuse to merge to extents if 1169 * this can result in the top bit of ee_len being set. 1170 */ 1171 if (ext1_ee_len + ext2_ee_len > max_len) 1172 return 0; 1173#ifdef AGGRESSIVE_TEST 1174 if (le16_to_cpu(ex1->ee_len) >= 4) 1175 return 0; 1176#endif 1177 1178 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) 1179 return 1; 1180 return 0; 1181} 1182 1183/* 1184 * This function tries to merge the "ex" extent to the next extent in the tree. 1185 * It always tries to merge towards right. If you want to merge towards 1186 * left, pass "ex - 1" as argument instead of "ex". 1187 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns 1188 * 1 if they got merged. 1189 */ 1190int ext4_ext_try_to_merge(struct inode *inode, 1191 struct ext4_ext_path *path, 1192 struct ext4_extent *ex) 1193{ 1194 struct ext4_extent_header *eh; 1195 unsigned int depth, len; 1196 int merge_done = 0; 1197 int uninitialized = 0; 1198 1199 depth = ext_depth(inode); 1200 BUG_ON(path[depth].p_hdr == NULL); 1201 eh = path[depth].p_hdr; 1202 1203 while (ex < EXT_LAST_EXTENT(eh)) { 1204 if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) 1205 break; 1206 /* merge with next extent! */ 1207 if (ext4_ext_is_uninitialized(ex)) 1208 uninitialized = 1; 1209 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) 1210 + ext4_ext_get_actual_len(ex + 1)); 1211 if (uninitialized) 1212 ext4_ext_mark_uninitialized(ex); 1213 1214 if (ex + 1 < EXT_LAST_EXTENT(eh)) { 1215 len = (EXT_LAST_EXTENT(eh) - ex - 1) 1216 * sizeof(struct ext4_extent); 1217 memmove(ex + 1, ex + 2, len); 1218 } 1219 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1); 1220 merge_done = 1; 1221 WARN_ON(eh->eh_entries == 0); 1222 if (!eh->eh_entries) 1223 ext4_error(inode->i_sb, "ext4_ext_try_to_merge", 1224 "inode#%lu, eh->eh_entries = 0!", inode->i_ino); 1225 } 1226 1227 return merge_done; 1228} 1229 1230/* 1231 * check if a portion of the "newext" extent overlaps with an 1232 * existing extent. 1233 * 1234 * If there is an overlap discovered, it updates the length of the newext 1235 * such that there will be no overlap, and then returns 1. 1236 * If there is no overlap found, it returns 0. 1237 */ 1238unsigned int ext4_ext_check_overlap(struct inode *inode, 1239 struct ext4_extent *newext, 1240 struct ext4_ext_path *path) 1241{ 1242 unsigned long b1, b2; 1243 unsigned int depth, len1; 1244 unsigned int ret = 0; 1245 1246 b1 = le32_to_cpu(newext->ee_block); 1247 len1 = ext4_ext_get_actual_len(newext); 1248 depth = ext_depth(inode); 1249 if (!path[depth].p_ext) 1250 goto out; 1251 b2 = le32_to_cpu(path[depth].p_ext->ee_block); 1252 1253 /* 1254 * get the next allocated block if the extent in the path 1255 * is before the requested block(s) 1256 */ 1257 if (b2 < b1) { 1258 b2 = ext4_ext_next_allocated_block(path); 1259 if (b2 == EXT_MAX_BLOCK) 1260 goto out; 1261 } 1262 1263 /* check for wrap through zero */ 1264 if (b1 + len1 < b1) { 1265 len1 = EXT_MAX_BLOCK - b1; 1266 newext->ee_len = cpu_to_le16(len1); 1267 ret = 1; 1268 } 1269 1270 /* check for overlap */ 1271 if (b1 + len1 > b2) { 1272 newext->ee_len = cpu_to_le16(b2 - b1); 1273 ret = 1; 1274 } 1275out: 1276 return ret; 1277} 1278 1279/* 1280 * ext4_ext_insert_extent: 1281 * tries to merge requsted extent into the existing extent or 1282 * inserts requested extent as new one into the tree, 1283 * creating new leaf in the no-space case. 1284 */ 1285int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, 1286 struct ext4_ext_path *path, 1287 struct ext4_extent *newext) 1288{ 1289 struct ext4_extent_header * eh; 1290 struct ext4_extent *ex, *fex; 1291 struct ext4_extent *nearex; /* nearest extent */ 1292 struct ext4_ext_path *npath = NULL; 1293 int depth, len, err, next; 1294 unsigned uninitialized = 0; 1295 1296 BUG_ON(ext4_ext_get_actual_len(newext) == 0); 1297 depth = ext_depth(inode); 1298 ex = path[depth].p_ext; 1299 BUG_ON(path[depth].p_hdr == NULL); 1300 1301 /* try to insert block into found extent and return */ 1302 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { 1303 ext_debug("append %d block to %d:%d (from %llu)\n", 1304 ext4_ext_get_actual_len(newext), 1305 le32_to_cpu(ex->ee_block), 1306 ext4_ext_get_actual_len(ex), ext_pblock(ex)); 1307 err = ext4_ext_get_access(handle, inode, path + depth); 1308 if (err) 1309 return err; 1310 1311 /* 1312 * ext4_can_extents_be_merged should have checked that either 1313 * both extents are uninitialized, or both aren't. Thus we 1314 * need to check only one of them here. 1315 */ 1316 if (ext4_ext_is_uninitialized(ex)) 1317 uninitialized = 1; 1318 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) 1319 + ext4_ext_get_actual_len(newext)); 1320 if (uninitialized) 1321 ext4_ext_mark_uninitialized(ex); 1322 eh = path[depth].p_hdr; 1323 nearex = ex; 1324 goto merge; 1325 } 1326 1327repeat: 1328 depth = ext_depth(inode); 1329 eh = path[depth].p_hdr; 1330 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) 1331 goto has_space; 1332 1333 /* probably next leaf has space for us? */ 1334 fex = EXT_LAST_EXTENT(eh); 1335 next = ext4_ext_next_leaf_block(inode, path); 1336 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) 1337 && next != EXT_MAX_BLOCK) { 1338 ext_debug("next leaf block - %d\n", next); 1339 BUG_ON(npath != NULL); 1340 npath = ext4_ext_find_extent(inode, next, NULL); 1341 if (IS_ERR(npath)) 1342 return PTR_ERR(npath); 1343 BUG_ON(npath->p_depth != path->p_depth); 1344 eh = npath[depth].p_hdr; 1345 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { 1346 ext_debug("next leaf isnt full(%d)\n", 1347 le16_to_cpu(eh->eh_entries)); 1348 path = npath; 1349 goto repeat; 1350 } 1351 ext_debug("next leaf has no free space(%d,%d)\n", 1352 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); 1353 } 1354 1355 /* 1356 * There is no free space in the found leaf. 1357 * We're gonna add a new leaf in the tree. 1358 */ 1359 err = ext4_ext_create_new_leaf(handle, inode, path, newext); 1360 if (err) 1361 goto cleanup; 1362 depth = ext_depth(inode); 1363 eh = path[depth].p_hdr; 1364 1365has_space: 1366 nearex = path[depth].p_ext; 1367 1368 err = ext4_ext_get_access(handle, inode, path + depth); 1369 if (err) 1370 goto cleanup; 1371 1372 if (!nearex) { 1373 /* there is no extent in this leaf, create first one */ 1374 ext_debug("first extent in the leaf: %d:%llu:%d\n", 1375 le32_to_cpu(newext->ee_block), 1376 ext_pblock(newext), 1377 ext4_ext_get_actual_len(newext)); 1378 path[depth].p_ext = EXT_FIRST_EXTENT(eh); 1379 } else if (le32_to_cpu(newext->ee_block) 1380 > le32_to_cpu(nearex->ee_block)) { 1381/* BUG_ON(newext->ee_block == nearex->ee_block); */ 1382 if (nearex != EXT_LAST_EXTENT(eh)) { 1383 len = EXT_MAX_EXTENT(eh) - nearex; 1384 len = (len - 1) * sizeof(struct ext4_extent); 1385 len = len < 0 ? 0 : len; 1386 ext_debug("insert %d:%llu:%d after: nearest 0x%p, " 1387 "move %d from 0x%p to 0x%p\n", 1388 le32_to_cpu(newext->ee_block), 1389 ext_pblock(newext), 1390 ext4_ext_get_actual_len(newext), 1391 nearex, len, nearex + 1, nearex + 2); 1392 memmove(nearex + 2, nearex + 1, len); 1393 } 1394 path[depth].p_ext = nearex + 1; 1395 } else { 1396 BUG_ON(newext->ee_block == nearex->ee_block); 1397 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); 1398 len = len < 0 ? 0 : len; 1399 ext_debug("insert %d:%llu:%d before: nearest 0x%p, " 1400 "move %d from 0x%p to 0x%p\n", 1401 le32_to_cpu(newext->ee_block), 1402 ext_pblock(newext), 1403 ext4_ext_get_actual_len(newext), 1404 nearex, len, nearex + 1, nearex + 2); 1405 memmove(nearex + 1, nearex, len); 1406 path[depth].p_ext = nearex; 1407 } 1408 1409 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1); 1410 nearex = path[depth].p_ext; 1411 nearex->ee_block = newext->ee_block; 1412 nearex->ee_start = newext->ee_start; 1413 nearex->ee_start_hi = newext->ee_start_hi; 1414 nearex->ee_len = newext->ee_len; 1415 1416merge: 1417 /* try to merge extents to the right */ 1418 ext4_ext_try_to_merge(inode, path, nearex); 1419 1420 /* try to merge extents to the left */ 1421 1422 /* time to correct all indexes above */ 1423 err = ext4_ext_correct_indexes(handle, inode, path); 1424 if (err) 1425 goto cleanup; 1426 1427 err = ext4_ext_dirty(handle, inode, path + depth); 1428 1429cleanup: 1430 if (npath) { 1431 ext4_ext_drop_refs(npath); 1432 kfree(npath); 1433 } 1434 ext4_ext_tree_changed(inode); 1435 ext4_ext_invalidate_cache(inode); 1436 return err; 1437} 1438 1439int ext4_ext_walk_space(struct inode *inode, unsigned long block, 1440 unsigned long num, ext_prepare_callback func, 1441 void *cbdata) 1442{ 1443 struct ext4_ext_path *path = NULL; 1444 struct ext4_ext_cache cbex; 1445 struct ext4_extent *ex; 1446 unsigned long next, start = 0, end = 0; 1447 unsigned long last = block + num; 1448 int depth, exists, err = 0; 1449 1450 BUG_ON(func == NULL); 1451 BUG_ON(inode == NULL); 1452 1453 while (block < last && block != EXT_MAX_BLOCK) { 1454 num = last - block; 1455 /* find extent for this block */ 1456 path = ext4_ext_find_extent(inode, block, path); 1457 if (IS_ERR(path)) { 1458 err = PTR_ERR(path); 1459 path = NULL; 1460 break; 1461 } 1462 1463 depth = ext_depth(inode); 1464 BUG_ON(path[depth].p_hdr == NULL); 1465 ex = path[depth].p_ext; 1466 next = ext4_ext_next_allocated_block(path); 1467 1468 exists = 0; 1469 if (!ex) { 1470 /* there is no extent yet, so try to allocate 1471 * all requested space */ 1472 start = block; 1473 end = block + num; 1474 } else if (le32_to_cpu(ex->ee_block) > block) { 1475 /* need to allocate space before found extent */ 1476 start = block; 1477 end = le32_to_cpu(ex->ee_block); 1478 if (block + num < end) 1479 end = block + num; 1480 } else if (block >= le32_to_cpu(ex->ee_block) 1481 + ext4_ext_get_actual_len(ex)) { 1482 /* need to allocate space after found extent */ 1483 start = block; 1484 end = block + num; 1485 if (end >= next) 1486 end = next; 1487 } else if (block >= le32_to_cpu(ex->ee_block)) { 1488 /* 1489 * some part of requested space is covered 1490 * by found extent 1491 */ 1492 start = block; 1493 end = le32_to_cpu(ex->ee_block) 1494 + ext4_ext_get_actual_len(ex); 1495 if (block + num < end) 1496 end = block + num; 1497 exists = 1; 1498 } else { 1499 BUG(); 1500 } 1501 BUG_ON(end <= start); 1502 1503 if (!exists) { 1504 cbex.ec_block = start; 1505 cbex.ec_len = end - start; 1506 cbex.ec_start = 0; 1507 cbex.ec_type = EXT4_EXT_CACHE_GAP; 1508 } else { 1509 cbex.ec_block = le32_to_cpu(ex->ee_block); 1510 cbex.ec_len = ext4_ext_get_actual_len(ex); 1511 cbex.ec_start = ext_pblock(ex); 1512 cbex.ec_type = EXT4_EXT_CACHE_EXTENT; 1513 } 1514 1515 BUG_ON(cbex.ec_len == 0); 1516 err = func(inode, path, &cbex, cbdata); 1517 ext4_ext_drop_refs(path); 1518 1519 if (err < 0) 1520 break; 1521 if (err == EXT_REPEAT) 1522 continue; 1523 else if (err == EXT_BREAK) { 1524 err = 0; 1525 break; 1526 } 1527 1528 if (ext_depth(inode) != depth) { 1529 /* depth was changed. we have to realloc path */ 1530 kfree(path); 1531 path = NULL; 1532 } 1533 1534 block = cbex.ec_block + cbex.ec_len; 1535 } 1536 1537 if (path) { 1538 ext4_ext_drop_refs(path); 1539 kfree(path); 1540 } 1541 1542 return err; 1543} 1544 1545static void 1546ext4_ext_put_in_cache(struct inode *inode, __u32 block, 1547 __u32 len, ext4_fsblk_t start, int type) 1548{ 1549 struct ext4_ext_cache *cex; 1550 BUG_ON(len == 0); 1551 cex = &EXT4_I(inode)->i_cached_extent; 1552 cex->ec_type = type; 1553 cex->ec_block = block; 1554 cex->ec_len = len; 1555 cex->ec_start = start; 1556} 1557 1558/* 1559 * ext4_ext_put_gap_in_cache: 1560 * calculate boundaries of the gap that the requested block fits into 1561 * and cache this gap 1562 */ 1563static void 1564ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, 1565 unsigned long block) 1566{ 1567 int depth = ext_depth(inode); 1568 unsigned long lblock, len; 1569 struct ext4_extent *ex; 1570 1571 ex = path[depth].p_ext; 1572 if (ex == NULL) { 1573 /* there is no extent yet, so gap is [0;-] */ 1574 lblock = 0; 1575 len = EXT_MAX_BLOCK; 1576 ext_debug("cache gap(whole file):"); 1577 } else if (block < le32_to_cpu(ex->ee_block)) { 1578 lblock = block; 1579 len = le32_to_cpu(ex->ee_block) - block; 1580 ext_debug("cache gap(before): %lu [%lu:%lu]", 1581 (unsigned long) block, 1582 (unsigned long) le32_to_cpu(ex->ee_block), 1583 (unsigned long) ext4_ext_get_actual_len(ex)); 1584 } else if (block >= le32_to_cpu(ex->ee_block) 1585 + ext4_ext_get_actual_len(ex)) { 1586 lblock = le32_to_cpu(ex->ee_block) 1587 + ext4_ext_get_actual_len(ex); 1588 len = ext4_ext_next_allocated_block(path); 1589 ext_debug("cache gap(after): [%lu:%lu] %lu", 1590 (unsigned long) le32_to_cpu(ex->ee_block), 1591 (unsigned long) ext4_ext_get_actual_len(ex), 1592 (unsigned long) block); 1593 BUG_ON(len == lblock); 1594 len = len - lblock; 1595 } else { 1596 lblock = len = 0; 1597 BUG(); 1598 } 1599 1600 ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len); 1601 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP); 1602} 1603 1604static int 1605ext4_ext_in_cache(struct inode *inode, unsigned long block, 1606 struct ext4_extent *ex) 1607{ 1608 struct ext4_ext_cache *cex; 1609 1610 cex = &EXT4_I(inode)->i_cached_extent; 1611 1612 /* has cache valid data? */ 1613 if (cex->ec_type == EXT4_EXT_CACHE_NO) 1614 return EXT4_EXT_CACHE_NO; 1615 1616 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP && 1617 cex->ec_type != EXT4_EXT_CACHE_EXTENT); 1618 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) { 1619 ex->ee_block = cpu_to_le32(cex->ec_block); 1620 ext4_ext_store_pblock(ex, cex->ec_start); 1621 ex->ee_len = cpu_to_le16(cex->ec_len); 1622 ext_debug("%lu cached by %lu:%lu:%llu\n", 1623 (unsigned long) block, 1624 (unsigned long) cex->ec_block, 1625 (unsigned long) cex->ec_len, 1626 cex->ec_start); 1627 return cex->ec_type; 1628 } 1629 1630 /* not in cache */ 1631 return EXT4_EXT_CACHE_NO; 1632} 1633 1634/* 1635 * ext4_ext_rm_idx: 1636 * removes index from the index block. 1637 * It's used in truncate case only, thus all requests are for 1638 * last index in the block only. 1639 */ 1640int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, 1641 struct ext4_ext_path *path) 1642{ 1643 struct buffer_head *bh; 1644 int err; 1645 ext4_fsblk_t leaf; 1646 1647 /* free index block */ 1648 path--; 1649 leaf = idx_pblock(path->p_idx); 1650 BUG_ON(path->p_hdr->eh_entries == 0); 1651 err = ext4_ext_get_access(handle, inode, path); 1652 if (err) 1653 return err; 1654 path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1); 1655 err = ext4_ext_dirty(handle, inode, path); 1656 if (err) 1657 return err; 1658 ext_debug("index is empty, remove it, free block %llu\n", leaf); 1659 bh = sb_find_get_block(inode->i_sb, leaf); 1660 ext4_forget(handle, 1, inode, bh, leaf); 1661 ext4_free_blocks(handle, inode, leaf, 1); 1662 return err; 1663} 1664 1665/* 1666 * ext4_ext_calc_credits_for_insert: 1667 * This routine returns max. credits that the extent tree can consume. 1668 * It should be OK for low-performance paths like ->writepage() 1669 * To allow many writing processes to fit into a single transaction, 1670 * the caller should calculate credits under truncate_mutex and 1671 * pass the actual path. 1672 */ 1673int ext4_ext_calc_credits_for_insert(struct inode *inode, 1674 struct ext4_ext_path *path) 1675{ 1676 int depth, needed; 1677 1678 if (path) { 1679 /* probably there is space in leaf? */ 1680 depth = ext_depth(inode); 1681 if (le16_to_cpu(path[depth].p_hdr->eh_entries) 1682 < le16_to_cpu(path[depth].p_hdr->eh_max)) 1683 return 1; 1684 } 1685 1686 /* 1687 * given 32-bit logical block (4294967296 blocks), max. tree 1688 * can be 4 levels in depth -- 4 * 340^4 == 53453440000. 1689 * Let's also add one more level for imbalance. 1690 */ 1691 depth = 5; 1692 1693 /* allocation of new data block(s) */ 1694 needed = 2; 1695 1696 /* 1697 * tree can be full, so it would need to grow in depth: 1698 * we need one credit to modify old root, credits for 1699 * new root will be added in split accounting 1700 */ 1701 needed += 1; 1702 1703 /* 1704 * Index split can happen, we would need: 1705 * allocate intermediate indexes (bitmap + group) 1706 * + change two blocks at each level, but root (already included) 1707 */ 1708 needed += (depth * 2) + (depth * 2); 1709 1710 /* any allocation modifies superblock */ 1711 needed += 1; 1712 1713 return needed; 1714} 1715 1716static int ext4_remove_blocks(handle_t *handle, struct inode *inode, 1717 struct ext4_extent *ex, 1718 unsigned long from, unsigned long to) 1719{ 1720 struct buffer_head *bh; 1721 unsigned short ee_len = ext4_ext_get_actual_len(ex); 1722 int i; 1723 1724#ifdef EXTENTS_STATS 1725 { 1726 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1727 spin_lock(&sbi->s_ext_stats_lock); 1728 sbi->s_ext_blocks += ee_len; 1729 sbi->s_ext_extents++; 1730 if (ee_len < sbi->s_ext_min) 1731 sbi->s_ext_min = ee_len; 1732 if (ee_len > sbi->s_ext_max) 1733 sbi->s_ext_max = ee_len; 1734 if (ext_depth(inode) > sbi->s_depth_max) 1735 sbi->s_depth_max = ext_depth(inode); 1736 spin_unlock(&sbi->s_ext_stats_lock); 1737 } 1738#endif 1739 if (from >= le32_to_cpu(ex->ee_block) 1740 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { 1741 /* tail removal */ 1742 unsigned long num; 1743 ext4_fsblk_t start; 1744 num = le32_to_cpu(ex->ee_block) + ee_len - from; 1745 start = ext_pblock(ex) + ee_len - num; 1746 ext_debug("free last %lu blocks starting %llu\n", num, start); 1747 for (i = 0; i < num; i++) { 1748 bh = sb_find_get_block(inode->i_sb, start + i); 1749 ext4_forget(handle, 0, inode, bh, start + i); 1750 } 1751 ext4_free_blocks(handle, inode, start, num); 1752 } else if (from == le32_to_cpu(ex->ee_block) 1753 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { 1754 printk("strange request: removal %lu-%lu from %u:%u\n", 1755 from, to, le32_to_cpu(ex->ee_block), ee_len); 1756 } else { 1757 printk("strange request: removal(2) %lu-%lu from %u:%u\n", 1758 from, to, le32_to_cpu(ex->ee_block), ee_len); 1759 } 1760 return 0; 1761} 1762 1763static int 1764ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, 1765 struct ext4_ext_path *path, unsigned long start) 1766{ 1767 int err = 0, correct_index = 0; 1768 int depth = ext_depth(inode), credits; 1769 struct ext4_extent_header *eh; 1770 unsigned a, b, block, num; 1771 unsigned long ex_ee_block; 1772 unsigned short ex_ee_len; 1773 unsigned uninitialized = 0; 1774 struct ext4_extent *ex; 1775 1776 /* the header must be checked already in ext4_ext_remove_space() */ 1777 ext_debug("truncate since %lu in leaf\n", start); 1778 if (!path[depth].p_hdr) 1779 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); 1780 eh = path[depth].p_hdr; 1781 BUG_ON(eh == NULL); 1782 1783 /* find where to start removing */ 1784 ex = EXT_LAST_EXTENT(eh); 1785 1786 ex_ee_block = le32_to_cpu(ex->ee_block); 1787 if (ext4_ext_is_uninitialized(ex)) 1788 uninitialized = 1; 1789 ex_ee_len = ext4_ext_get_actual_len(ex); 1790 1791 while (ex >= EXT_FIRST_EXTENT(eh) && 1792 ex_ee_block + ex_ee_len > start) { 1793 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len); 1794 path[depth].p_ext = ex; 1795 1796 a = ex_ee_block > start ? ex_ee_block : start; 1797 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? 1798 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; 1799 1800 ext_debug(" border %u:%u\n", a, b); 1801 1802 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { 1803 block = 0; 1804 num = 0; 1805 BUG(); 1806 } else if (a != ex_ee_block) { 1807 /* remove tail of the extent */ 1808 block = ex_ee_block; 1809 num = a - block; 1810 } else if (b != ex_ee_block + ex_ee_len - 1) { 1811 /* remove head of the extent */ 1812 block = a; 1813 num = b - a; 1814 /* there is no "make a hole" API yet */ 1815 BUG(); 1816 } else { 1817 /* remove whole extent: excellent! */ 1818 block = ex_ee_block; 1819 num = 0; 1820 BUG_ON(a != ex_ee_block); 1821 BUG_ON(b != ex_ee_block + ex_ee_len - 1); 1822 } 1823 1824 /* at present, extent can't cross block group: */ 1825 /* leaf + bitmap + group desc + sb + inode */ 1826 credits = 5; 1827 if (ex == EXT_FIRST_EXTENT(eh)) { 1828 correct_index = 1; 1829 credits += (ext_depth(inode)) + 1; 1830 } 1831#ifdef CONFIG_QUOTA 1832 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); 1833#endif 1834 1835 handle = ext4_ext_journal_restart(handle, credits); 1836 if (IS_ERR(handle)) { 1837 err = PTR_ERR(handle); 1838 goto out; 1839 } 1840 1841 err = ext4_ext_get_access(handle, inode, path + depth); 1842 if (err) 1843 goto out; 1844 1845 err = ext4_remove_blocks(handle, inode, ex, a, b); 1846 if (err) 1847 goto out; 1848 1849 if (num == 0) { 1850 /* this extent is removed; mark slot entirely unused */ 1851 ext4_ext_store_pblock(ex, 0); 1852 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1); 1853 } 1854 1855 ex->ee_block = cpu_to_le32(block); 1856 ex->ee_len = cpu_to_le16(num); 1857 /* 1858 * Do not mark uninitialized if all the blocks in the 1859 * extent have been removed. 1860 */ 1861 if (uninitialized && num) 1862 ext4_ext_mark_uninitialized(ex); 1863 1864 err = ext4_ext_dirty(handle, inode, path + depth); 1865 if (err) 1866 goto out; 1867 1868 ext_debug("new extent: %u:%u:%llu\n", block, num, 1869 ext_pblock(ex)); 1870 ex--; 1871 ex_ee_block = le32_to_cpu(ex->ee_block); 1872 ex_ee_len = ext4_ext_get_actual_len(ex); 1873 } 1874 1875 if (correct_index && eh->eh_entries) 1876 err = ext4_ext_correct_indexes(handle, inode, path); 1877 1878 /* if this leaf is free, then we should 1879 * remove it from index block above */ 1880 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) 1881 err = ext4_ext_rm_idx(handle, inode, path + depth); 1882 1883out: 1884 return err; 1885} 1886 1887/* 1888 * ext4_ext_more_to_rm: 1889 * returns 1 if current index has to be freed (even partial) 1890 */ 1891static int 1892ext4_ext_more_to_rm(struct ext4_ext_path *path) 1893{ 1894 BUG_ON(path->p_idx == NULL); 1895 1896 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) 1897 return 0; 1898 1899 /* 1900 * if truncate on deeper level happened, it wasn't partial, 1901 * so we have to consider current index for truncation 1902 */ 1903 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) 1904 return 0; 1905 return 1; 1906} 1907 1908int ext4_ext_remove_space(struct inode *inode, unsigned long start) 1909{ 1910 struct super_block *sb = inode->i_sb; 1911 int depth = ext_depth(inode); 1912 struct ext4_ext_path *path; 1913 handle_t *handle; 1914 int i = 0, err = 0; 1915 1916 ext_debug("truncate since %lu\n", start); 1917 1918 /* probably first extent we're gonna free will be last in block */ 1919 handle = ext4_journal_start(inode, depth + 1); 1920 if (IS_ERR(handle)) 1921 return PTR_ERR(handle); 1922 1923 ext4_ext_invalidate_cache(inode); 1924 1925 /* 1926 * We start scanning from right side, freeing all the blocks 1927 * after i_size and walking into the tree depth-wise. 1928 */ 1929 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL); 1930 if (path == NULL) { 1931 ext4_journal_stop(handle); 1932 return -ENOMEM; 1933 } 1934 path[0].p_hdr = ext_inode_hdr(inode); 1935 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) { 1936 err = -EIO; 1937 goto out; 1938 } 1939 path[0].p_depth = depth; 1940 1941 while (i >= 0 && err == 0) { 1942 if (i == depth) { 1943 /* this is leaf block */ 1944 err = ext4_ext_rm_leaf(handle, inode, path, start); 1945 /* root level has p_bh == NULL, brelse() eats this */ 1946 brelse(path[i].p_bh); 1947 path[i].p_bh = NULL; 1948 i--; 1949 continue; 1950 } 1951 1952 /* this is index block */ 1953 if (!path[i].p_hdr) { 1954 ext_debug("initialize header\n"); 1955 path[i].p_hdr = ext_block_hdr(path[i].p_bh); 1956 } 1957 1958 if (!path[i].p_idx) { 1959 /* this level hasn't been touched yet */ 1960 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); 1961 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; 1962 ext_debug("init index ptr: hdr 0x%p, num %d\n", 1963 path[i].p_hdr, 1964 le16_to_cpu(path[i].p_hdr->eh_entries)); 1965 } else { 1966 /* we were already here, see at next index */ 1967 path[i].p_idx--; 1968 } 1969 1970 ext_debug("level %d - index, first 0x%p, cur 0x%p\n", 1971 i, EXT_FIRST_INDEX(path[i].p_hdr), 1972 path[i].p_idx); 1973 if (ext4_ext_more_to_rm(path + i)) { 1974 struct buffer_head *bh; 1975 /* go to the next level */ 1976 ext_debug("move to level %d (block %llu)\n", 1977 i + 1, idx_pblock(path[i].p_idx)); 1978 memset(path + i + 1, 0, sizeof(*path)); 1979 bh = sb_bread(sb, idx_pblock(path[i].p_idx)); 1980 if (!bh) { 1981 /* should we reset i_size? */ 1982 err = -EIO; 1983 break; 1984 } 1985 if (WARN_ON(i + 1 > depth)) { 1986 err = -EIO; 1987 break; 1988 } 1989 if (ext4_ext_check_header(inode, ext_block_hdr(bh), 1990 depth - i - 1)) { 1991 err = -EIO; 1992 break; 1993 } 1994 path[i + 1].p_bh = bh; 1995 1996 /* save actual number of indexes since this 1997 * number is changed at the next iteration */ 1998 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); 1999 i++; 2000 } else { 2001 /* we finished processing this index, go up */ 2002 if (path[i].p_hdr->eh_entries == 0 && i > 0) { 2003 /* index is empty, remove it; 2004 * handle must be already prepared by the 2005 * truncatei_leaf() */ 2006 err = ext4_ext_rm_idx(handle, inode, path + i); 2007 } 2008 /* root level has p_bh == NULL, brelse() eats this */ 2009 brelse(path[i].p_bh); 2010 path[i].p_bh = NULL; 2011 i--; 2012 ext_debug("return to level %d\n", i); 2013 } 2014 } 2015 2016 /* TODO: flexible tree reduction should be here */ 2017 if (path->p_hdr->eh_entries == 0) { 2018 /* 2019 * truncate to zero freed all the tree, 2020 * so we need to correct eh_depth 2021 */ 2022 err = ext4_ext_get_access(handle, inode, path); 2023 if (err == 0) { 2024 ext_inode_hdr(inode)->eh_depth = 0; 2025 ext_inode_hdr(inode)->eh_max = 2026 cpu_to_le16(ext4_ext_space_root(inode)); 2027 err = ext4_ext_dirty(handle, inode, path); 2028 } 2029 } 2030out: 2031 ext4_ext_tree_changed(inode); 2032 ext4_ext_drop_refs(path); 2033 kfree(path); 2034 ext4_journal_stop(handle); 2035 2036 return err; 2037} 2038 2039/* 2040 * called at mount time 2041 */ 2042void ext4_ext_init(struct super_block *sb) 2043{ 2044 /* 2045 * possible initialization would be here 2046 */ 2047 2048 if (test_opt(sb, EXTENTS)) { 2049 printk("EXT4-fs: file extents enabled"); 2050#ifdef AGGRESSIVE_TEST 2051 printk(", aggressive tests"); 2052#endif 2053#ifdef CHECK_BINSEARCH 2054 printk(", check binsearch"); 2055#endif 2056#ifdef EXTENTS_STATS 2057 printk(", stats"); 2058#endif 2059 printk("\n"); 2060#ifdef EXTENTS_STATS 2061 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); 2062 EXT4_SB(sb)->s_ext_min = 1 << 30; 2063 EXT4_SB(sb)->s_ext_max = 0; 2064#endif 2065 } 2066} 2067 2068/* 2069 * called at umount time 2070 */ 2071void ext4_ext_release(struct super_block *sb) 2072{ 2073 if (!test_opt(sb, EXTENTS)) 2074 return; 2075 2076#ifdef EXTENTS_STATS 2077 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { 2078 struct ext4_sb_info *sbi = EXT4_SB(sb); 2079 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", 2080 sbi->s_ext_blocks, sbi->s_ext_extents, 2081 sbi->s_ext_blocks / sbi->s_ext_extents); 2082 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", 2083 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); 2084 } 2085#endif 2086} 2087 2088/* 2089 * This function is called by ext4_ext_get_blocks() if someone tries to write 2090 * to an uninitialized extent. It may result in splitting the uninitialized 2091 * extent into multiple extents (upto three - one initialized and two 2092 * uninitialized). 2093 * There are three possibilities: 2094 * a> There is no split required: Entire extent should be initialized 2095 * b> Splits in two extents: Write is happening at either end of the extent 2096 * c> Splits in three extents: Somone is writing in middle of the extent 2097 */ 2098int ext4_ext_convert_to_initialized(handle_t *handle, struct inode *inode, 2099 struct ext4_ext_path *path, 2100 ext4_fsblk_t iblock, 2101 unsigned long max_blocks) 2102{ 2103 struct ext4_extent *ex, newex; 2104 struct ext4_extent *ex1 = NULL; 2105 struct ext4_extent *ex2 = NULL; 2106 struct ext4_extent *ex3 = NULL; 2107 struct ext4_extent_header *eh; 2108 unsigned int allocated, ee_block, ee_len, depth; 2109 ext4_fsblk_t newblock; 2110 int err = 0; 2111 int ret = 0; 2112 2113 depth = ext_depth(inode); 2114 eh = path[depth].p_hdr; 2115 ex = path[depth].p_ext; 2116 ee_block = le32_to_cpu(ex->ee_block); 2117 ee_len = ext4_ext_get_actual_len(ex); 2118 allocated = ee_len - (iblock - ee_block); 2119 newblock = iblock - ee_block + ext_pblock(ex); 2120 ex2 = ex; 2121 2122 /* ex1: ee_block to iblock - 1 : uninitialized */ 2123 if (iblock > ee_block) { 2124 ex1 = ex; 2125 ex1->ee_len = cpu_to_le16(iblock - ee_block); 2126 ext4_ext_mark_uninitialized(ex1); 2127 ex2 = &newex; 2128 } 2129 /* 2130 * for sanity, update the length of the ex2 extent before 2131 * we insert ex3, if ex1 is NULL. This is to avoid temporary 2132 * overlap of blocks. 2133 */ 2134 if (!ex1 && allocated > max_blocks) 2135 ex2->ee_len = cpu_to_le16(max_blocks); 2136 /* ex3: to ee_block + ee_len : uninitialised */ 2137 if (allocated > max_blocks) { 2138 unsigned int newdepth; 2139 ex3 = &newex; 2140 ex3->ee_block = cpu_to_le32(iblock + max_blocks); 2141 ext4_ext_store_pblock(ex3, newblock + max_blocks); 2142 ex3->ee_len = cpu_to_le16(allocated - max_blocks); 2143 ext4_ext_mark_uninitialized(ex3); 2144 err = ext4_ext_insert_extent(handle, inode, path, ex3); 2145 if (err) 2146 goto out; 2147 /* 2148 * The depth, and hence eh & ex might change 2149 * as part of the insert above. 2150 */ 2151 newdepth = ext_depth(inode); 2152 if (newdepth != depth) { 2153 depth = newdepth; 2154 path = ext4_ext_find_extent(inode, iblock, NULL); 2155 if (IS_ERR(path)) { 2156 err = PTR_ERR(path); 2157 path = NULL; 2158 goto out; 2159 } 2160 eh = path[depth].p_hdr; 2161 ex = path[depth].p_ext; 2162 if (ex2 != &newex) 2163 ex2 = ex; 2164 } 2165 allocated = max_blocks; 2166 } 2167 /* 2168 * If there was a change of depth as part of the 2169 * insertion of ex3 above, we need to update the length 2170 * of the ex1 extent again here 2171 */ 2172 if (ex1 && ex1 != ex) { 2173 ex1 = ex; 2174 ex1->ee_len = cpu_to_le16(iblock - ee_block); 2175 ext4_ext_mark_uninitialized(ex1); 2176 ex2 = &newex; 2177 } 2178 /* ex2: iblock to iblock + maxblocks-1 : initialised */ 2179 ex2->ee_block = cpu_to_le32(iblock); 2180 ex2->ee_start = cpu_to_le32(newblock); 2181 ext4_ext_store_pblock(ex2, newblock); 2182 ex2->ee_len = cpu_to_le16(allocated); 2183 if (ex2 != ex) 2184 goto insert; 2185 err = ext4_ext_get_access(handle, inode, path + depth); 2186 if (err) 2187 goto out; 2188 /* 2189 * New (initialized) extent starts from the first block 2190 * in the current extent. i.e., ex2 == ex 2191 * We have to see if it can be merged with the extent 2192 * on the left. 2193 */ 2194 if (ex2 > EXT_FIRST_EXTENT(eh)) { 2195 /* 2196 * To merge left, pass "ex2 - 1" to try_to_merge(), 2197 * since it merges towards right _only_. 2198 */ 2199 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1); 2200 if (ret) { 2201 err = ext4_ext_correct_indexes(handle, inode, path); 2202 if (err) 2203 goto out; 2204 depth = ext_depth(inode); 2205 ex2--; 2206 } 2207 } 2208 /* 2209 * Try to Merge towards right. This might be required 2210 * only when the whole extent is being written to. 2211 * i.e. ex2 == ex and ex3 == NULL. 2212 */ 2213 if (!ex3) { 2214 ret = ext4_ext_try_to_merge(inode, path, ex2); 2215 if (ret) { 2216 err = ext4_ext_correct_indexes(handle, inode, path); 2217 if (err) 2218 goto out; 2219 } 2220 } 2221 /* Mark modified extent as dirty */ 2222 err = ext4_ext_dirty(handle, inode, path + depth); 2223 goto out; 2224insert: 2225 err = ext4_ext_insert_extent(handle, inode, path, &newex); 2226out: 2227 return err ? err : allocated; 2228} 2229 2230int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, 2231 ext4_fsblk_t iblock, 2232 unsigned long max_blocks, struct buffer_head *bh_result, 2233 int create, int extend_disksize) 2234{ 2235 struct ext4_ext_path *path = NULL; 2236 struct ext4_extent_header *eh; 2237 struct ext4_extent newex, *ex; 2238 ext4_fsblk_t goal, newblock; 2239 int err = 0, depth, ret; 2240 unsigned long allocated = 0; 2241 2242 __clear_bit(BH_New, &bh_result->b_state); 2243 ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock, 2244 max_blocks, (unsigned) inode->i_ino); 2245 mutex_lock(&EXT4_I(inode)->truncate_mutex); 2246 2247 /* check in cache */ 2248 goal = ext4_ext_in_cache(inode, iblock, &newex); 2249 if (goal) { 2250 if (goal == EXT4_EXT_CACHE_GAP) { 2251 if (!create) { 2252 /* 2253 * block isn't allocated yet and 2254 * user doesn't want to allocate it 2255 */ 2256 goto out2; 2257 } 2258 /* we should allocate requested block */ 2259 } else if (goal == EXT4_EXT_CACHE_EXTENT) { 2260 /* block is already allocated */ 2261 newblock = iblock 2262 - le32_to_cpu(newex.ee_block) 2263 + ext_pblock(&newex); 2264 /* number of remaining blocks in the extent */ 2265 allocated = le16_to_cpu(newex.ee_len) - 2266 (iblock - le32_to_cpu(newex.ee_block)); 2267 goto out; 2268 } else { 2269 BUG(); 2270 } 2271 } 2272 2273 /* find extent for this block */ 2274 path = ext4_ext_find_extent(inode, iblock, NULL); 2275 if (IS_ERR(path)) { 2276 err = PTR_ERR(path); 2277 path = NULL; 2278 goto out2; 2279 } 2280 2281 depth = ext_depth(inode); 2282 2283 /* 2284 * consistent leaf must not be empty; 2285 * this situation is possible, though, _during_ tree modification; 2286 * this is why assert can't be put in ext4_ext_find_extent() 2287 */ 2288 BUG_ON(path[depth].p_ext == NULL && depth != 0); 2289 eh = path[depth].p_hdr; 2290 2291 ex = path[depth].p_ext; 2292 if (ex) { 2293 unsigned long ee_block = le32_to_cpu(ex->ee_block); 2294 ext4_fsblk_t ee_start = ext_pblock(ex); 2295 unsigned short ee_len; 2296 2297 /* 2298 * Uninitialized extents are treated as holes, except that 2299 * we split out initialized portions during a write. 2300 */ 2301 ee_len = ext4_ext_get_actual_len(ex); 2302 /* if found extent covers block, simply return it */ 2303 if (iblock >= ee_block && iblock < ee_block + ee_len) { 2304 newblock = iblock - ee_block + ee_start; 2305 /* number of remaining blocks in the extent */ 2306 allocated = ee_len - (iblock - ee_block); 2307 ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock, 2308 ee_block, ee_len, newblock); 2309 2310 /* Do not put uninitialized extent in the cache */ 2311 if (!ext4_ext_is_uninitialized(ex)) { 2312 ext4_ext_put_in_cache(inode, ee_block, 2313 ee_len, ee_start, 2314 EXT4_EXT_CACHE_EXTENT); 2315 goto out; 2316 } 2317 if (create == EXT4_CREATE_UNINITIALIZED_EXT) 2318 goto out; 2319 if (!create) 2320 goto out2; 2321 2322 ret = ext4_ext_convert_to_initialized(handle, inode, 2323 path, iblock, 2324 max_blocks); 2325 if (ret <= 0) 2326 goto out2; 2327 else 2328 allocated = ret; 2329 goto outnew; 2330 } 2331 } 2332 2333 /* 2334 * requested block isn't allocated yet; 2335 * we couldn't try to create block if create flag is zero 2336 */ 2337 if (!create) { 2338 /* 2339 * put just found gap into cache to speed up 2340 * subsequent requests 2341 */ 2342 ext4_ext_put_gap_in_cache(inode, path, iblock); 2343 goto out2; 2344 } 2345 /* 2346 * Okay, we need to do block allocation. Lazily initialize the block 2347 * allocation info here if necessary. 2348 */ 2349 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info)) 2350 ext4_init_block_alloc_info(inode); 2351 2352 /* allocate new block */ 2353 goal = ext4_ext_find_goal(inode, path, iblock); 2354 2355 /* 2356 * See if request is beyond maximum number of blocks we can have in 2357 * a single extent. For an initialized extent this limit is 2358 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is 2359 * EXT_UNINIT_MAX_LEN. 2360 */ 2361 if (max_blocks > EXT_INIT_MAX_LEN && 2362 create != EXT4_CREATE_UNINITIALIZED_EXT) 2363 max_blocks = EXT_INIT_MAX_LEN; 2364 else if (max_blocks > EXT_UNINIT_MAX_LEN && 2365 create == EXT4_CREATE_UNINITIALIZED_EXT) 2366 max_blocks = EXT_UNINIT_MAX_LEN; 2367 2368 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ 2369 newex.ee_block = cpu_to_le32(iblock); 2370 newex.ee_len = cpu_to_le16(max_blocks); 2371 err = ext4_ext_check_overlap(inode, &newex, path); 2372 if (err) 2373 allocated = le16_to_cpu(newex.ee_len); 2374 else 2375 allocated = max_blocks; 2376 newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err); 2377 if (!newblock) 2378 goto out2; 2379 ext_debug("allocate new block: goal %llu, found %llu/%lu\n", 2380 goal, newblock, allocated); 2381 2382 /* try to insert new extent into found leaf and return */ 2383 ext4_ext_store_pblock(&newex, newblock); 2384 newex.ee_len = cpu_to_le16(allocated); 2385 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */ 2386 ext4_ext_mark_uninitialized(&newex); 2387 err = ext4_ext_insert_extent(handle, inode, path, &newex); 2388 if (err) { 2389 /* free data blocks we just allocated */ 2390 ext4_free_blocks(handle, inode, ext_pblock(&newex), 2391 le16_to_cpu(newex.ee_len)); 2392 goto out2; 2393 } 2394 2395 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize) 2396 EXT4_I(inode)->i_disksize = inode->i_size; 2397 2398 /* previous routine could use block we allocated */ 2399 newblock = ext_pblock(&newex); 2400outnew: 2401 __set_bit(BH_New, &bh_result->b_state); 2402 2403 /* Cache only when it is _not_ an uninitialized extent */ 2404 if (create != EXT4_CREATE_UNINITIALIZED_EXT) 2405 ext4_ext_put_in_cache(inode, iblock, allocated, newblock, 2406 EXT4_EXT_CACHE_EXTENT); 2407out: 2408 if (allocated > max_blocks) 2409 allocated = max_blocks; 2410 ext4_ext_show_leaf(inode, path); 2411 __set_bit(BH_Mapped, &bh_result->b_state); 2412 bh_result->b_bdev = inode->i_sb->s_bdev; 2413 bh_result->b_blocknr = newblock; 2414out2: 2415 if (path) { 2416 ext4_ext_drop_refs(path); 2417 kfree(path); 2418 } 2419 mutex_unlock(&EXT4_I(inode)->truncate_mutex); 2420 2421 return err ? err : allocated; 2422} 2423 2424void ext4_ext_truncate(struct inode * inode, struct page *page) 2425{ 2426 struct address_space *mapping = inode->i_mapping; 2427 struct super_block *sb = inode->i_sb; 2428 unsigned long last_block; 2429 handle_t *handle; 2430 int err = 0; 2431 2432 /* 2433 * probably first extent we're gonna free will be last in block 2434 */ 2435 err = ext4_writepage_trans_blocks(inode) + 3; 2436 handle = ext4_journal_start(inode, err); 2437 if (IS_ERR(handle)) { 2438 if (page) { 2439 clear_highpage(page); 2440 flush_dcache_page(page); 2441 unlock_page(page); 2442 page_cache_release(page); 2443 } 2444 return; 2445 } 2446 2447 if (page) 2448 ext4_block_truncate_page(handle, page, mapping, inode->i_size); 2449 2450 mutex_lock(&EXT4_I(inode)->truncate_mutex); 2451 ext4_ext_invalidate_cache(inode); 2452 2453 /* 2454 * TODO: optimization is possible here. 2455 * Probably we need not scan at all, 2456 * because page truncation is enough. 2457 */ 2458 if (ext4_orphan_add(handle, inode)) 2459 goto out_stop; 2460 2461 /* we have to know where to truncate from in crash case */ 2462 EXT4_I(inode)->i_disksize = inode->i_size; 2463 ext4_mark_inode_dirty(handle, inode); 2464 2465 last_block = (inode->i_size + sb->s_blocksize - 1) 2466 >> EXT4_BLOCK_SIZE_BITS(sb); 2467 err = ext4_ext_remove_space(inode, last_block); 2468 2469 /* In a multi-transaction truncate, we only make the final 2470 * transaction synchronous. 2471 */ 2472 if (IS_SYNC(inode)) 2473 handle->h_sync = 1; 2474 2475out_stop: 2476 /* 2477 * If this was a simple ftruncate() and the file will remain alive, 2478 * then we need to clear up the orphan record which we created above. 2479 * However, if this was a real unlink then we were called by 2480 * ext4_delete_inode(), and we allow that function to clean up the 2481 * orphan info for us. 2482 */ 2483 if (inode->i_nlink) 2484 ext4_orphan_del(handle, inode); 2485 2486 mutex_unlock(&EXT4_I(inode)->truncate_mutex); 2487 ext4_journal_stop(handle); 2488} 2489 2490/* 2491 * ext4_ext_writepage_trans_blocks: 2492 * calculate max number of blocks we could modify 2493 * in order to allocate new block for an inode 2494 */ 2495int ext4_ext_writepage_trans_blocks(struct inode *inode, int num) 2496{ 2497 int needed; 2498 2499 needed = ext4_ext_calc_credits_for_insert(inode, NULL); 2500 2501 /* caller wants to allocate num blocks, but note it includes sb */ 2502 needed = needed * num - (num - 1); 2503 2504#ifdef CONFIG_QUOTA 2505 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); 2506#endif 2507 2508 return needed; 2509} 2510 2511/* 2512 * preallocate space for a file. This implements ext4's fallocate inode 2513 * operation, which gets called from sys_fallocate system call. 2514 * For block-mapped files, posix_fallocate should fall back to the method 2515 * of writing zeroes to the required new blocks (the same behavior which is 2516 * expected for file systems which do not support fallocate() system call). 2517 */ 2518long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) 2519{ 2520 handle_t *handle; 2521 ext4_fsblk_t block, max_blocks; 2522 ext4_fsblk_t nblocks = 0; 2523 int ret = 0; 2524 int ret2 = 0; 2525 int retries = 0; 2526 struct buffer_head map_bh; 2527 unsigned int credits, blkbits = inode->i_blkbits; 2528 2529 /* 2530 * currently supporting (pre)allocate mode for extent-based 2531 * files _only_ 2532 */ 2533 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 2534 return -EOPNOTSUPP; 2535 2536 /* preallocation to directories is currently not supported */ 2537 if (S_ISDIR(inode->i_mode)) 2538 return -ENODEV; 2539 2540 block = offset >> blkbits; 2541 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) 2542 - block; 2543 2544 /* 2545 * credits to insert 1 extent into extent tree + buffers to be able to 2546 * modify 1 super block, 1 block bitmap and 1 group descriptor. 2547 */ 2548 credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3; 2549retry: 2550 while (ret >= 0 && ret < max_blocks) { 2551 block = block + ret; 2552 max_blocks = max_blocks - ret; 2553 handle = ext4_journal_start(inode, credits); 2554 if (IS_ERR(handle)) { 2555 ret = PTR_ERR(handle); 2556 break; 2557 } 2558 2559 ret = ext4_ext_get_blocks(handle, inode, block, 2560 max_blocks, &map_bh, 2561 EXT4_CREATE_UNINITIALIZED_EXT, 0); 2562 WARN_ON(!ret); 2563 if (!ret) { 2564 ext4_error(inode->i_sb, "ext4_fallocate", 2565 "ext4_ext_get_blocks returned 0! inode#%lu" 2566 ", block=%llu, max_blocks=%llu", 2567 inode->i_ino, block, max_blocks); 2568 ret = -EIO; 2569 ext4_mark_inode_dirty(handle, inode); 2570 ret2 = ext4_journal_stop(handle); 2571 break; 2572 } 2573 if (ret > 0) { 2574 /* check wrap through sign-bit/zero here */ 2575 if ((block + ret) < 0 || (block + ret) < block) { 2576 ret = -EIO; 2577 ext4_mark_inode_dirty(handle, inode); 2578 ret2 = ext4_journal_stop(handle); 2579 break; 2580 } 2581 if (buffer_new(&map_bh) && ((block + ret) > 2582 (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits) 2583 >> blkbits))) 2584 nblocks = nblocks + ret; 2585 } 2586 2587 /* Update ctime if new blocks get allocated */ 2588 if (nblocks) { 2589 struct timespec now; 2590 2591 now = current_fs_time(inode->i_sb); 2592 if (!timespec_equal(&inode->i_ctime, &now)) 2593 inode->i_ctime = now; 2594 } 2595 2596 ext4_mark_inode_dirty(handle, inode); 2597 ret2 = ext4_journal_stop(handle); 2598 if (ret2) 2599 break; 2600 } 2601 2602 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 2603 goto retry; 2604 2605 /* 2606 * Time to update the file size. 2607 * Update only when preallocation was requested beyond the file size. 2608 */ 2609 if (!(mode & FALLOC_FL_KEEP_SIZE) && 2610 (offset + len) > i_size_read(inode)) { 2611 if (ret > 0) { 2612 /* 2613 * if no error, we assume preallocation succeeded 2614 * completely 2615 */ 2616 mutex_lock(&inode->i_mutex); 2617 i_size_write(inode, offset + len); 2618 EXT4_I(inode)->i_disksize = i_size_read(inode); 2619 mutex_unlock(&inode->i_mutex); 2620 } else if (ret < 0 && nblocks) { 2621 /* Handle partial allocation scenario */ 2622 loff_t newsize; 2623 2624 mutex_lock(&inode->i_mutex); 2625 newsize = (nblocks << blkbits) + i_size_read(inode); 2626 i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits)); 2627 EXT4_I(inode)->i_disksize = i_size_read(inode); 2628 mutex_unlock(&inode->i_mutex); 2629 } 2630 } 2631 2632 return ret > 0 ? ret2 : ret; 2633} 2634