1/* 2 * linux/fs/ext2/inode.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * from 10 * 11 * linux/fs/minix/inode.c 12 * 13 * Copyright (C) 1991, 1992 Linus Torvalds 14 * 15 * Goal-directed block allocation by Stephen Tweedie 16 * (sct@dcs.ed.ac.uk), 1993, 1998 17 * Big-endian to little-endian byte-swapping/bitmaps by 18 * David S. Miller (davem@caip.rutgers.edu), 1995 19 * 64-bit file support on 64-bit platforms by Jakub Jelinek 20 * (jj@sunsite.ms.mff.cuni.cz) 21 * 22 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000 23 */ 24 25#include <linux/fs.h> 26#include <linux/ext2_fs.h> 27#include <linux/locks.h> 28#include <linux/smp_lock.h> 29#include <linux/sched.h> 30#include <linux/highuid.h> 31#include <linux/quotaops.h> 32#include <linux/module.h> 33 34MODULE_AUTHOR("Remy Card and others"); 35MODULE_DESCRIPTION("Second Extended Filesystem"); 36MODULE_LICENSE("GPL"); 37 38 39static int ext2_update_inode(struct inode * inode, int do_sync); 40 41/* 42 * Called at each iput() 43 */ 44void ext2_put_inode (struct inode * inode) 45{ 46 ext2_discard_prealloc (inode); 47} 48 49/* 50 * Called at the last iput() if i_nlink is zero. 51 */ 52void ext2_delete_inode (struct inode * inode) 53{ 54 lock_kernel(); 55 56 if (is_bad_inode(inode) || 57 inode->i_ino == EXT2_ACL_IDX_INO || 58 inode->i_ino == EXT2_ACL_DATA_INO) 59 goto no_delete; 60 inode->u.ext2_i.i_dtime = CURRENT_TIME; 61 mark_inode_dirty(inode); 62 ext2_update_inode(inode, IS_SYNC(inode)); 63 inode->i_size = 0; 64 if (inode->i_blocks) 65 ext2_truncate (inode); 66 ext2_free_inode (inode); 67 68 unlock_kernel(); 69 return; 70no_delete: 71 unlock_kernel(); 72 clear_inode(inode); /* We must guarantee clearing of inode... */ 73} 74 75void ext2_discard_prealloc (struct inode * inode) 76{ 77#ifdef EXT2_PREALLOCATE 78 lock_kernel(); 79 /* Writer: ->i_prealloc* */ 80 if (inode->u.ext2_i.i_prealloc_count) { 81 unsigned short total = inode->u.ext2_i.i_prealloc_count; 82 unsigned long block = inode->u.ext2_i.i_prealloc_block; 83 inode->u.ext2_i.i_prealloc_count = 0; 84 inode->u.ext2_i.i_prealloc_block = 0; 85 /* Writer: end */ 86 ext2_free_blocks (inode, block, total); 87 } 88 unlock_kernel(); 89#endif 90} 91 92static int ext2_alloc_block (struct inode * inode, unsigned long goal, int *err) 93{ 94#ifdef EXT2FS_DEBUG 95 static unsigned long alloc_hits = 0, alloc_attempts = 0; 96#endif 97 unsigned long result; 98 99 100#ifdef EXT2_PREALLOCATE 101 /* Writer: ->i_prealloc* */ 102 if (inode->u.ext2_i.i_prealloc_count && 103 (goal == inode->u.ext2_i.i_prealloc_block || 104 goal + 1 == inode->u.ext2_i.i_prealloc_block)) 105 { 106 result = inode->u.ext2_i.i_prealloc_block++; 107 inode->u.ext2_i.i_prealloc_count--; 108 /* Writer: end */ 109 ext2_debug ("preallocation hit (%lu/%lu).\n", 110 ++alloc_hits, ++alloc_attempts); 111 } else { 112 ext2_discard_prealloc (inode); 113 ext2_debug ("preallocation miss (%lu/%lu).\n", 114 alloc_hits, ++alloc_attempts); 115 if (S_ISREG(inode->i_mode)) 116 result = ext2_new_block (inode, goal, 117 &inode->u.ext2_i.i_prealloc_count, 118 &inode->u.ext2_i.i_prealloc_block, err); 119 else 120 result = ext2_new_block (inode, goal, 0, 0, err); 121 } 122#else 123 result = ext2_new_block (inode, goal, 0, 0, err); 124#endif 125 return result; 126} 127 128typedef struct { 129 u32 *p; 130 u32 key; 131 struct buffer_head *bh; 132} Indirect; 133 134static inline void add_chain(Indirect *p, struct buffer_head *bh, u32 *v) 135{ 136 p->key = *(p->p = v); 137 p->bh = bh; 138} 139 140static inline int verify_chain(Indirect *from, Indirect *to) 141{ 142 while (from <= to && from->key == *from->p) 143 from++; 144 return (from > to); 145} 146 147/** 148 * ext2_block_to_path - parse the block number into array of offsets 149 * @inode: inode in question (we are only interested in its superblock) 150 * @i_block: block number to be parsed 151 * @offsets: array to store the offsets in 152 * 153 * To store the locations of file's data ext2 uses a data structure common 154 * for UNIX filesystems - tree of pointers anchored in the inode, with 155 * data blocks at leaves and indirect blocks in intermediate nodes. 156 * This function translates the block number into path in that tree - 157 * return value is the path length and @offsets[n] is the offset of 158 * pointer to (n+1)th node in the nth one. If @block is out of range 159 * (negative or too large) warning is printed and zero returned. 160 * 161 * Note: function doesn't find node addresses, so no IO is needed. All 162 * we need to know is the capacity of indirect blocks (taken from the 163 * inode->i_sb). 164 */ 165 166/* 167 * Portability note: the last comparison (check that we fit into triple 168 * indirect block) is spelled differently, because otherwise on an 169 * architecture with 32-bit longs and 8Kb pages we might get into trouble 170 * if our filesystem had 8Kb blocks. We might use long long, but that would 171 * kill us on x86. Oh, well, at least the sign propagation does not matter - 172 * i_block would have to be negative in the very beginning, so we would not 173 * get there at all. 174 */ 175 176static int ext2_block_to_path(struct inode *inode, long i_block, int offsets[4]) 177{ 178 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb); 179 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb); 180 const long direct_blocks = EXT2_NDIR_BLOCKS, 181 indirect_blocks = ptrs, 182 double_blocks = (1 << (ptrs_bits * 2)); 183 int n = 0; 184 185 if (i_block < 0) { 186 ext2_warning (inode->i_sb, "ext2_block_to_path", "block < 0"); 187 } else if (i_block < direct_blocks) { 188 offsets[n++] = i_block; 189 } else if ( (i_block -= direct_blocks) < indirect_blocks) { 190 offsets[n++] = EXT2_IND_BLOCK; 191 offsets[n++] = i_block; 192 } else if ((i_block -= indirect_blocks) < double_blocks) { 193 offsets[n++] = EXT2_DIND_BLOCK; 194 offsets[n++] = i_block >> ptrs_bits; 195 offsets[n++] = i_block & (ptrs - 1); 196 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { 197 offsets[n++] = EXT2_TIND_BLOCK; 198 offsets[n++] = i_block >> (ptrs_bits * 2); 199 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); 200 offsets[n++] = i_block & (ptrs - 1); 201 } else { 202 ext2_warning (inode->i_sb, "ext2_block_to_path", "block > big"); 203 } 204 return n; 205} 206 207/** 208 * ext2_get_branch - read the chain of indirect blocks leading to data 209 * @inode: inode in question 210 * @depth: depth of the chain (1 - direct pointer, etc.) 211 * @offsets: offsets of pointers in inode/indirect blocks 212 * @chain: place to store the result 213 * @err: here we store the error value 214 * 215 * Function fills the array of triples <key, p, bh> and returns %NULL 216 * if everything went OK or the pointer to the last filled triple 217 * (incomplete one) otherwise. Upon the return chain[i].key contains 218 * the number of (i+1)-th block in the chain (as it is stored in memory, 219 * i.e. little-endian 32-bit), chain[i].p contains the address of that 220 * number (it points into struct inode for i==0 and into the bh->b_data 221 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect 222 * block for i>0 and NULL for i==0. In other words, it holds the block 223 * numbers of the chain, addresses they were taken from (and where we can 224 * verify that chain did not change) and buffer_heads hosting these 225 * numbers. 226 * 227 * Function stops when it stumbles upon zero pointer (absent block) 228 * (pointer to last triple returned, *@err == 0) 229 * or when it gets an IO error reading an indirect block 230 * (ditto, *@err == -EIO) 231 * or when it notices that chain had been changed while it was reading 232 * (ditto, *@err == -EAGAIN) 233 * or when it reads all @depth-1 indirect blocks successfully and finds 234 * the whole chain, all way to the data (returns %NULL, *err == 0). 235 */ 236static Indirect *ext2_get_branch(struct inode *inode, 237 int depth, 238 int *offsets, 239 Indirect chain[4], 240 int *err) 241{ 242 struct super_block *sb = inode->i_sb; 243 Indirect *p = chain; 244 struct buffer_head *bh; 245 246 *err = 0; 247 /* i_data is not going away, no lock needed */ 248 add_chain (chain, NULL, inode->u.ext2_i.i_data + *offsets); 249 if (!p->key) 250 goto no_block; 251 while (--depth) { 252 bh = sb_bread(sb, le32_to_cpu(p->key)); 253 if (!bh) 254 goto failure; 255 /* Reader: pointers */ 256 if (!verify_chain(chain, p)) 257 goto changed; 258 add_chain(++p, bh, (u32*)bh->b_data + *++offsets); 259 /* Reader: end */ 260 if (!p->key) 261 goto no_block; 262 } 263 return NULL; 264 265changed: 266 *err = -EAGAIN; 267 goto no_block; 268failure: 269 *err = -EIO; 270no_block: 271 return p; 272} 273 274/** 275 * ext2_find_near - find a place for allocation with sufficient locality 276 * @inode: owner 277 * @ind: descriptor of indirect block. 278 * 279 * This function returns the prefered place for block allocation. 280 * It is used when heuristic for sequential allocation fails. 281 * Rules are: 282 * + if there is a block to the left of our position - allocate near it. 283 * + if pointer will live in indirect block - allocate near that block. 284 * + if pointer will live in inode - allocate in the same cylinder group. 285 * Caller must make sure that @ind is valid and will stay that way. 286 */ 287 288static inline unsigned long ext2_find_near(struct inode *inode, Indirect *ind) 289{ 290 u32 *start = ind->bh ? (u32*) ind->bh->b_data : inode->u.ext2_i.i_data; 291 u32 *p; 292 293 /* Try to find previous block */ 294 for (p = ind->p - 1; p >= start; p--) 295 if (*p) 296 return le32_to_cpu(*p); 297 298 /* No such thing, so let's try location of indirect block */ 299 if (ind->bh) 300 return ind->bh->b_blocknr; 301 302 /* 303 * It is going to be refered from inode itself? OK, just put it into 304 * the same cylinder group then. 305 */ 306 return (inode->u.ext2_i.i_block_group * 307 EXT2_BLOCKS_PER_GROUP(inode->i_sb)) + 308 le32_to_cpu(inode->i_sb->u.ext2_sb.s_es->s_first_data_block); 309} 310 311/** 312 * ext2_find_goal - find a prefered place for allocation. 313 * @inode: owner 314 * @block: block we want 315 * @chain: chain of indirect blocks 316 * @partial: pointer to the last triple within a chain 317 * @goal: place to store the result. 318 * 319 * Normally this function find the prefered place for block allocation, 320 * stores it in *@goal and returns zero. If the branch had been changed 321 * under us we return -EAGAIN. 322 */ 323 324static inline int ext2_find_goal(struct inode *inode, 325 long block, 326 Indirect chain[4], 327 Indirect *partial, 328 unsigned long *goal) 329{ 330 /* Writer: ->i_next_alloc* */ 331 if (block == inode->u.ext2_i.i_next_alloc_block + 1) { 332 inode->u.ext2_i.i_next_alloc_block++; 333 inode->u.ext2_i.i_next_alloc_goal++; 334 } 335 /* Writer: end */ 336 /* Reader: pointers, ->i_next_alloc* */ 337 if (verify_chain(chain, partial)) { 338 /* 339 * try the heuristic for sequential allocation, 340 * failing that at least try to get decent locality. 341 */ 342 if (block == inode->u.ext2_i.i_next_alloc_block) 343 *goal = inode->u.ext2_i.i_next_alloc_goal; 344 if (!*goal) 345 *goal = ext2_find_near(inode, partial); 346 return 0; 347 } 348 /* Reader: end */ 349 return -EAGAIN; 350} 351 352/** 353 * ext2_alloc_branch - allocate and set up a chain of blocks. 354 * @inode: owner 355 * @num: depth of the chain (number of blocks to allocate) 356 * @offsets: offsets (in the blocks) to store the pointers to next. 357 * @branch: place to store the chain in. 358 * 359 * This function allocates @num blocks, zeroes out all but the last one, 360 * links them into chain and (if we are synchronous) writes them to disk. 361 * In other words, it prepares a branch that can be spliced onto the 362 * inode. It stores the information about that chain in the branch[], in 363 * the same format as ext2_get_branch() would do. We are calling it after 364 * we had read the existing part of chain and partial points to the last 365 * triple of that (one with zero ->key). Upon the exit we have the same 366 * picture as after the successful ext2_get_block(), excpet that in one 367 * place chain is disconnected - *branch->p is still zero (we did not 368 * set the last link), but branch->key contains the number that should 369 * be placed into *branch->p to fill that gap. 370 * 371 * If allocation fails we free all blocks we've allocated (and forget 372 * their buffer_heads) and return the error value the from failed 373 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain 374 * as described above and return 0. 375 */ 376 377static int ext2_alloc_branch(struct inode *inode, 378 int num, 379 unsigned long goal, 380 int *offsets, 381 Indirect *branch) 382{ 383 int blocksize = inode->i_sb->s_blocksize; 384 int n = 0; 385 int err; 386 int i; 387 int parent = ext2_alloc_block(inode, goal, &err); 388 389 branch[0].key = cpu_to_le32(parent); 390 if (parent) for (n = 1; n < num; n++) { 391 struct buffer_head *bh; 392 /* Allocate the next block */ 393 int nr = ext2_alloc_block(inode, parent, &err); 394 if (!nr) 395 break; 396 branch[n].key = cpu_to_le32(nr); 397 /* 398 * Get buffer_head for parent block, zero it out and set 399 * the pointer to new one, then send parent to disk. 400 */ 401 bh = sb_getblk(inode->i_sb, parent); 402 lock_buffer(bh); 403 memset(bh->b_data, 0, blocksize); 404 branch[n].bh = bh; 405 branch[n].p = (u32*) bh->b_data + offsets[n]; 406 *branch[n].p = branch[n].key; 407 mark_buffer_uptodate(bh, 1); 408 unlock_buffer(bh); 409 mark_buffer_dirty_inode(bh, inode); 410 /* We used to sync bh here if IS_SYNC(inode). 411 * But for S_ISREG files we now rely upon generic_osync_inode() 412 * and b_inode_buffers 413 */ 414 if (S_ISDIR(inode->i_mode) && IS_SYNC(inode)) { 415 ll_rw_block (WRITE, 1, &bh); 416 wait_on_buffer (bh); 417 } 418 parent = nr; 419 } 420 if (n == num) 421 return 0; 422 423 /* Allocation failed, free what we already allocated */ 424 for (i = 1; i < n; i++) 425 bforget(branch[i].bh); 426 for (i = 0; i < n; i++) 427 ext2_free_blocks(inode, le32_to_cpu(branch[i].key), 1); 428 return err; 429} 430 431/** 432 * ext2_splice_branch - splice the allocated branch onto inode. 433 * @inode: owner 434 * @block: (logical) number of block we are adding 435 * @chain: chain of indirect blocks (with a missing link - see 436 * ext2_alloc_branch) 437 * @where: location of missing link 438 * @num: number of blocks we are adding 439 * 440 * This function verifies that chain (up to the missing link) had not 441 * changed, fills the missing link and does all housekeeping needed in 442 * inode (->i_blocks, etc.). In case of success we end up with the full 443 * chain to new block and return 0. Otherwise (== chain had been changed) 444 * we free the new blocks (forgetting their buffer_heads, indeed) and 445 * return -EAGAIN. 446 */ 447 448static inline int ext2_splice_branch(struct inode *inode, 449 long block, 450 Indirect chain[4], 451 Indirect *where, 452 int num) 453{ 454 int i; 455 456 /* Verify that place we are splicing to is still there and vacant */ 457 458 /* Writer: pointers, ->i_next_alloc* */ 459 if (!verify_chain(chain, where-1) || *where->p) 460 /* Writer: end */ 461 goto changed; 462 463 /* That's it */ 464 465 *where->p = where->key; 466 inode->u.ext2_i.i_next_alloc_block = block; 467 inode->u.ext2_i.i_next_alloc_goal = le32_to_cpu(where[num-1].key); 468 469 /* Writer: end */ 470 471 /* We are done with atomic stuff, now do the rest of housekeeping */ 472 473 inode->i_ctime = CURRENT_TIME; 474 475 /* had we spliced it onto indirect block? */ 476 if (where->bh) { 477 mark_buffer_dirty_inode(where->bh, inode); 478 if (S_ISDIR(inode->i_mode) && IS_SYNC(inode)) { 479 ll_rw_block(WRITE, 1, &where->bh); 480 wait_on_buffer(where->bh); 481 } 482 } 483 484 mark_inode_dirty(inode); 485 return 0; 486 487changed: 488 for (i = 1; i < num; i++) 489 bforget(where[i].bh); 490 for (i = 0; i < num; i++) 491 ext2_free_blocks(inode, le32_to_cpu(where[i].key), 1); 492 return -EAGAIN; 493} 494 495/* 496 * Allocation strategy is simple: if we have to allocate something, we will 497 * have to go the whole way to leaf. So let's do it before attaching anything 498 * to tree, set linkage between the newborn blocks, write them if sync is 499 * required, recheck the path, free and repeat if check fails, otherwise 500 * set the last missing link (that will protect us from any truncate-generated 501 * removals - all blocks on the path are immune now) and possibly force the 502 * write on the parent block. 503 * That has a nice additional property: no special recovery from the failed 504 * allocations is needed - we simply release blocks and do not touch anything 505 * reachable from inode. 506 */ 507 508static int ext2_get_block(struct inode *inode, long iblock, struct buffer_head *bh_result, int create) 509{ 510 int err = -EIO; 511 int offsets[4]; 512 Indirect chain[4]; 513 Indirect *partial; 514 unsigned long goal; 515 int left; 516 int depth = ext2_block_to_path(inode, iblock, offsets); 517 518 if (depth == 0) 519 goto out; 520 521 lock_kernel(); 522reread: 523 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 524 525 /* Simplest case - block found, no allocation needed */ 526 if (!partial) { 527got_it: 528 bh_result->b_dev = inode->i_dev; 529 bh_result->b_blocknr = le32_to_cpu(chain[depth-1].key); 530 bh_result->b_state |= (1UL << BH_Mapped); 531 /* Clean up and exit */ 532 partial = chain+depth-1; /* the whole chain */ 533 goto cleanup; 534 } 535 536 /* Next simple case - plain lookup or failed read of indirect block */ 537 if (!create || err == -EIO) { 538cleanup: 539 while (partial > chain) { 540 brelse(partial->bh); 541 partial--; 542 } 543 unlock_kernel(); 544out: 545 return err; 546 } 547 548 /* 549 * Indirect block might be removed by truncate while we were 550 * reading it. Handling of that case (forget what we've got and 551 * reread) is taken out of the main path. 552 */ 553 if (err == -EAGAIN) 554 goto changed; 555 556 if (ext2_find_goal(inode, iblock, chain, partial, &goal) < 0) 557 goto changed; 558 559 left = (chain + depth) - partial; 560 err = ext2_alloc_branch(inode, left, goal, 561 offsets+(partial-chain), partial); 562 if (err) 563 goto cleanup; 564 565 if (ext2_splice_branch(inode, iblock, chain, partial, left) < 0) 566 goto changed; 567 568 bh_result->b_state |= (1UL << BH_New); 569 goto got_it; 570 571changed: 572 while (partial > chain) { 573 brelse(partial->bh); 574 partial--; 575 } 576 goto reread; 577} 578 579static int ext2_writepage(struct page *page) 580{ 581 return block_write_full_page(page,ext2_get_block); 582} 583static int ext2_readpage(struct file *file, struct page *page) 584{ 585 return block_read_full_page(page,ext2_get_block); 586} 587static int ext2_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to) 588{ 589 return block_prepare_write(page,from,to,ext2_get_block); 590} 591static int ext2_bmap(struct address_space *mapping, long block) 592{ 593 return generic_block_bmap(mapping,block,ext2_get_block); 594} 595static int ext2_direct_IO(int rw, struct inode * inode, struct kiobuf * iobuf, unsigned long blocknr, int blocksize) 596{ 597 return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize, ext2_get_block); 598} 599struct address_space_operations ext2_aops = { 600 readpage: ext2_readpage, 601 writepage: ext2_writepage, 602 sync_page: block_sync_page, 603 prepare_write: ext2_prepare_write, 604 commit_write: generic_commit_write, 605 bmap: ext2_bmap, 606 direct_IO: ext2_direct_IO, 607}; 608 609/* 610 * Probably it should be a library function... search for first non-zero word 611 * or memcmp with zero_page, whatever is better for particular architecture. 612 * Linus? 613 */ 614static inline int all_zeroes(u32 *p, u32 *q) 615{ 616 while (p < q) 617 if (*p++) 618 return 0; 619 return 1; 620} 621 622/** 623 * ext2_find_shared - find the indirect blocks for partial truncation. 624 * @inode: inode in question 625 * @depth: depth of the affected branch 626 * @offsets: offsets of pointers in that branch (see ext2_block_to_path) 627 * @chain: place to store the pointers to partial indirect blocks 628 * @top: place to the (detached) top of branch 629 * 630 * This is a helper function used by ext2_truncate(). 631 * 632 * When we do truncate() we may have to clean the ends of several indirect 633 * blocks but leave the blocks themselves alive. Block is partially 634 * truncated if some data below the new i_size is refered from it (and 635 * it is on the path to the first completely truncated data block, indeed). 636 * We have to free the top of that path along with everything to the right 637 * of the path. Since no allocation past the truncation point is possible 638 * until ext2_truncate() finishes, we may safely do the latter, but top 639 * of branch may require special attention - pageout below the truncation 640 * point might try to populate it. 641 * 642 * We atomically detach the top of branch from the tree, store the block 643 * number of its root in *@top, pointers to buffer_heads of partially 644 * truncated blocks - in @chain[].bh and pointers to their last elements 645 * that should not be removed - in @chain[].p. Return value is the pointer 646 * to last filled element of @chain. 647 * 648 * The work left to caller to do the actual freeing of subtrees: 649 * a) free the subtree starting from *@top 650 * b) free the subtrees whose roots are stored in 651 * (@chain[i].p+1 .. end of @chain[i].bh->b_data) 652 * c) free the subtrees growing from the inode past the @chain[0].p 653 * (no partially truncated stuff there). 654 */ 655 656static Indirect *ext2_find_shared(struct inode *inode, 657 int depth, 658 int offsets[4], 659 Indirect chain[4], 660 u32 *top) 661{ 662 Indirect *partial, *p; 663 int k, err; 664 665 *top = 0; 666 for (k = depth; k > 1 && !offsets[k-1]; k--) 667 ; 668 partial = ext2_get_branch(inode, k, offsets, chain, &err); 669 /* Writer: pointers */ 670 if (!partial) 671 partial = chain + k-1; 672 /* 673 * If the branch acquired continuation since we've looked at it - 674 * fine, it should all survive and (new) top doesn't belong to us. 675 */ 676 if (!partial->key && *partial->p) 677 /* Writer: end */ 678 goto no_top; 679 for (p=partial; p>chain && all_zeroes((u32*)p->bh->b_data,p->p); p--) 680 ; 681 /* 682 * OK, we've found the last block that must survive. The rest of our 683 * branch should be detached before unlocking. However, if that rest 684 * of branch is all ours and does not grow immediately from the inode 685 * it's easier to cheat and just decrement partial->p. 686 */ 687 if (p == chain + k - 1 && p > chain) { 688 p->p--; 689 } else { 690 *top = *p->p; 691 *p->p = 0; 692 } 693 /* Writer: end */ 694 695 while(partial > p) 696 { 697 brelse(partial->bh); 698 partial--; 699 } 700no_top: 701 return partial; 702} 703 704/** 705 * ext2_free_data - free a list of data blocks 706 * @inode: inode we are dealing with 707 * @p: array of block numbers 708 * @q: points immediately past the end of array 709 * 710 * We are freeing all blocks refered from that array (numbers are 711 * stored as little-endian 32-bit) and updating @inode->i_blocks 712 * appropriately. 713 */ 714static inline void ext2_free_data(struct inode *inode, u32 *p, u32 *q) 715{ 716 unsigned long block_to_free = 0, count = 0; 717 unsigned long nr; 718 719 for ( ; p < q ; p++) { 720 nr = le32_to_cpu(*p); 721 if (nr) { 722 *p = 0; 723 /* accumulate blocks to free if they're contiguous */ 724 if (count == 0) 725 goto free_this; 726 else if (block_to_free == nr - count) 727 count++; 728 else { 729 mark_inode_dirty(inode); 730 ext2_free_blocks (inode, block_to_free, count); 731 free_this: 732 block_to_free = nr; 733 count = 1; 734 } 735 } 736 } 737 if (count > 0) { 738 mark_inode_dirty(inode); 739 ext2_free_blocks (inode, block_to_free, count); 740 } 741} 742 743/** 744 * ext2_free_branches - free an array of branches 745 * @inode: inode we are dealing with 746 * @p: array of block numbers 747 * @q: pointer immediately past the end of array 748 * @depth: depth of the branches to free 749 * 750 * We are freeing all blocks refered from these branches (numbers are 751 * stored as little-endian 32-bit) and updating @inode->i_blocks 752 * appropriately. 753 */ 754static void ext2_free_branches(struct inode *inode, u32 *p, u32 *q, int depth) 755{ 756 struct buffer_head * bh; 757 unsigned long nr; 758 759 if (depth--) { 760 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 761 for ( ; p < q ; p++) { 762 nr = le32_to_cpu(*p); 763 if (!nr) 764 continue; 765 *p = 0; 766 bh = sb_bread(inode->i_sb, nr); 767 /* 768 * A read failure? Report error and clear slot 769 * (should be rare). 770 */ 771 if (!bh) { 772 ext2_error(inode->i_sb, "ext2_free_branches", 773 "Read failure, inode=%ld, block=%ld", 774 inode->i_ino, nr); 775 continue; 776 } 777 ext2_free_branches(inode, 778 (u32*)bh->b_data, 779 (u32*)bh->b_data + addr_per_block, 780 depth); 781 bforget(bh); 782 ext2_free_blocks(inode, nr, 1); 783 mark_inode_dirty(inode); 784 } 785 } else 786 ext2_free_data(inode, p, q); 787} 788 789void ext2_truncate (struct inode * inode) 790{ 791 u32 *i_data = inode->u.ext2_i.i_data; 792 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 793 int offsets[4]; 794 Indirect chain[4]; 795 Indirect *partial; 796 int nr = 0; 797 int n; 798 long iblock; 799 unsigned blocksize; 800 801 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 802 S_ISLNK(inode->i_mode))) 803 return; 804 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 805 return; 806 807 ext2_discard_prealloc(inode); 808 809 blocksize = inode->i_sb->s_blocksize; 810 iblock = (inode->i_size + blocksize-1) 811 >> EXT2_BLOCK_SIZE_BITS(inode->i_sb); 812 813 block_truncate_page(inode->i_mapping, inode->i_size, ext2_get_block); 814 815 n = ext2_block_to_path(inode, iblock, offsets); 816 if (n == 0) 817 return; 818 819 if (n == 1) { 820 ext2_free_data(inode, i_data+offsets[0], 821 i_data + EXT2_NDIR_BLOCKS); 822 goto do_indirects; 823 } 824 825 partial = ext2_find_shared(inode, n, offsets, chain, &nr); 826 /* Kill the top of shared branch (already detached) */ 827 if (nr) { 828 if (partial == chain) 829 mark_inode_dirty(inode); 830 else 831 mark_buffer_dirty_inode(partial->bh, inode); 832 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial); 833 } 834 /* Clear the ends of indirect blocks on the shared branch */ 835 while (partial > chain) { 836 ext2_free_branches(inode, 837 partial->p + 1, 838 (u32*)partial->bh->b_data + addr_per_block, 839 (chain+n-1) - partial); 840 mark_buffer_dirty_inode(partial->bh, inode); 841 brelse (partial->bh); 842 partial--; 843 } 844do_indirects: 845 /* Kill the remaining (whole) subtrees */ 846 switch (offsets[0]) { 847 default: 848 nr = i_data[EXT2_IND_BLOCK]; 849 if (nr) { 850 i_data[EXT2_IND_BLOCK] = 0; 851 mark_inode_dirty(inode); 852 ext2_free_branches(inode, &nr, &nr+1, 1); 853 } 854 case EXT2_IND_BLOCK: 855 nr = i_data[EXT2_DIND_BLOCK]; 856 if (nr) { 857 i_data[EXT2_DIND_BLOCK] = 0; 858 mark_inode_dirty(inode); 859 ext2_free_branches(inode, &nr, &nr+1, 2); 860 } 861 case EXT2_DIND_BLOCK: 862 nr = i_data[EXT2_TIND_BLOCK]; 863 if (nr) { 864 i_data[EXT2_TIND_BLOCK] = 0; 865 mark_inode_dirty(inode); 866 ext2_free_branches(inode, &nr, &nr+1, 3); 867 } 868 case EXT2_TIND_BLOCK: 869 ; 870 } 871 inode->i_mtime = inode->i_ctime = CURRENT_TIME; 872 if (IS_SYNC(inode)) { 873 fsync_inode_buffers(inode); 874 ext2_sync_inode (inode); 875 } else { 876 mark_inode_dirty(inode); 877 } 878} 879 880void ext2_read_inode (struct inode * inode) 881{ 882 struct buffer_head * bh; 883 struct ext2_inode * raw_inode; 884 unsigned long block_group; 885 unsigned long group_desc; 886 unsigned long desc; 887 unsigned long block; 888 unsigned long offset; 889 struct ext2_group_desc * gdp; 890 891 if ((inode->i_ino != EXT2_ROOT_INO && inode->i_ino != EXT2_ACL_IDX_INO && 892 inode->i_ino != EXT2_ACL_DATA_INO && 893 inode->i_ino < EXT2_FIRST_INO(inode->i_sb)) || 894 inode->i_ino > le32_to_cpu(inode->i_sb->u.ext2_sb.s_es->s_inodes_count)) { 895 ext2_error (inode->i_sb, "ext2_read_inode", 896 "bad inode number: %lu", inode->i_ino); 897 goto bad_inode; 898 } 899 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 900 if (block_group >= inode->i_sb->u.ext2_sb.s_groups_count) { 901 ext2_error (inode->i_sb, "ext2_read_inode", 902 "group >= groups count"); 903 goto bad_inode; 904 } 905 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(inode->i_sb); 906 desc = block_group & (EXT2_DESC_PER_BLOCK(inode->i_sb) - 1); 907 bh = inode->i_sb->u.ext2_sb.s_group_desc[group_desc]; 908 if (!bh) { 909 ext2_error (inode->i_sb, "ext2_read_inode", 910 "Descriptor not loaded"); 911 goto bad_inode; 912 } 913 914 gdp = (struct ext2_group_desc *) bh->b_data; 915 /* 916 * Figure out the offset within the block group inode table 917 */ 918 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) * 919 EXT2_INODE_SIZE(inode->i_sb); 920 block = le32_to_cpu(gdp[desc].bg_inode_table) + 921 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb)); 922 if (!(bh = sb_bread(inode->i_sb, block))) { 923 ext2_error (inode->i_sb, "ext2_read_inode", 924 "unable to read inode block - " 925 "inode=%lu, block=%lu", inode->i_ino, block); 926 goto bad_inode; 927 } 928 offset &= (EXT2_BLOCK_SIZE(inode->i_sb) - 1); 929 raw_inode = (struct ext2_inode *) (bh->b_data + offset); 930 931 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 932 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 933 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 934 if(!(test_opt (inode->i_sb, NO_UID32))) { 935 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 936 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 937 } 938 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); 939 inode->i_size = le32_to_cpu(raw_inode->i_size); 940 inode->i_atime = le32_to_cpu(raw_inode->i_atime); 941 inode->i_ctime = le32_to_cpu(raw_inode->i_ctime); 942 inode->i_mtime = le32_to_cpu(raw_inode->i_mtime); 943 inode->u.ext2_i.i_dtime = le32_to_cpu(raw_inode->i_dtime); 944 /* We now have enough fields to check if the inode was active or not. 945 * This is needed because nfsd might try to access dead inodes 946 * the test is that same one that e2fsck uses 947 * NeilBrown 1999oct15 948 */ 949 if (inode->i_nlink == 0 && (inode->i_mode == 0 || inode->u.ext2_i.i_dtime)) { 950 /* this inode is deleted */ 951 brelse (bh); 952 goto bad_inode; 953 } 954 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size (for stat), not the fs block size */ 955 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); 956 inode->i_version = ++event; 957 inode->u.ext2_i.i_flags = le32_to_cpu(raw_inode->i_flags); 958 inode->u.ext2_i.i_faddr = le32_to_cpu(raw_inode->i_faddr); 959 inode->u.ext2_i.i_frag_no = raw_inode->i_frag; 960 inode->u.ext2_i.i_frag_size = raw_inode->i_fsize; 961 inode->u.ext2_i.i_file_acl = le32_to_cpu(raw_inode->i_file_acl); 962 if (S_ISREG(inode->i_mode)) 963 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; 964 else 965 inode->u.ext2_i.i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); 966 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 967 inode->u.ext2_i.i_prealloc_count = 0; 968 inode->u.ext2_i.i_block_group = block_group; 969 970 /* 971 * NOTE! The in-memory inode i_data array is in little-endian order 972 * even on big-endian machines: we do NOT byteswap the block numbers! 973 */ 974 for (block = 0; block < EXT2_N_BLOCKS; block++) 975 inode->u.ext2_i.i_data[block] = raw_inode->i_block[block]; 976 977 if (inode->i_ino == EXT2_ACL_IDX_INO || 978 inode->i_ino == EXT2_ACL_DATA_INO) 979 /* Nothing to do */ ; 980 else if (S_ISREG(inode->i_mode)) { 981 inode->i_op = &ext2_file_inode_operations; 982 inode->i_fop = &ext2_file_operations; 983 inode->i_mapping->a_ops = &ext2_aops; 984 } else if (S_ISDIR(inode->i_mode)) { 985 inode->i_op = &ext2_dir_inode_operations; 986 inode->i_fop = &ext2_dir_operations; 987 inode->i_mapping->a_ops = &ext2_aops; 988 } else if (S_ISLNK(inode->i_mode)) { 989 if (!inode->i_blocks) 990 inode->i_op = &ext2_fast_symlink_inode_operations; 991 else { 992 inode->i_op = &page_symlink_inode_operations; 993 inode->i_mapping->a_ops = &ext2_aops; 994 } 995 } else 996 init_special_inode(inode, inode->i_mode, 997 le32_to_cpu(raw_inode->i_block[0])); 998 brelse (bh); 999 inode->i_attr_flags = 0; 1000 if (inode->u.ext2_i.i_flags & EXT2_SYNC_FL) { 1001 inode->i_attr_flags |= ATTR_FLAG_SYNCRONOUS; 1002 inode->i_flags |= S_SYNC; 1003 } 1004 if (inode->u.ext2_i.i_flags & EXT2_APPEND_FL) { 1005 inode->i_attr_flags |= ATTR_FLAG_APPEND; 1006 inode->i_flags |= S_APPEND; 1007 } 1008 if (inode->u.ext2_i.i_flags & EXT2_IMMUTABLE_FL) { 1009 inode->i_attr_flags |= ATTR_FLAG_IMMUTABLE; 1010 inode->i_flags |= S_IMMUTABLE; 1011 } 1012 if (inode->u.ext2_i.i_flags & EXT2_NOATIME_FL) { 1013 inode->i_attr_flags |= ATTR_FLAG_NOATIME; 1014 inode->i_flags |= S_NOATIME; 1015 } 1016 return; 1017 1018bad_inode: 1019 make_bad_inode(inode); 1020 return; 1021} 1022 1023static int ext2_update_inode(struct inode * inode, int do_sync) 1024{ 1025 struct buffer_head * bh; 1026 struct ext2_inode * raw_inode; 1027 unsigned long block_group; 1028 unsigned long group_desc; 1029 unsigned long desc; 1030 unsigned long block; 1031 unsigned long offset; 1032 int err = 0; 1033 struct ext2_group_desc * gdp; 1034 1035 if ((inode->i_ino != EXT2_ROOT_INO && 1036 inode->i_ino < EXT2_FIRST_INO(inode->i_sb)) || 1037 inode->i_ino > le32_to_cpu(inode->i_sb->u.ext2_sb.s_es->s_inodes_count)) { 1038 ext2_error (inode->i_sb, "ext2_write_inode", 1039 "bad inode number: %lu", inode->i_ino); 1040 return -EIO; 1041 } 1042 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 1043 if (block_group >= inode->i_sb->u.ext2_sb.s_groups_count) { 1044 ext2_error (inode->i_sb, "ext2_write_inode", 1045 "group >= groups count"); 1046 return -EIO; 1047 } 1048 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(inode->i_sb); 1049 desc = block_group & (EXT2_DESC_PER_BLOCK(inode->i_sb) - 1); 1050 bh = inode->i_sb->u.ext2_sb.s_group_desc[group_desc]; 1051 if (!bh) { 1052 ext2_error (inode->i_sb, "ext2_write_inode", 1053 "Descriptor not loaded"); 1054 return -EIO; 1055 } 1056 gdp = (struct ext2_group_desc *) bh->b_data; 1057 /* 1058 * Figure out the offset within the block group inode table 1059 */ 1060 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) * 1061 EXT2_INODE_SIZE(inode->i_sb); 1062 block = le32_to_cpu(gdp[desc].bg_inode_table) + 1063 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb)); 1064 if (!(bh = sb_bread(inode->i_sb, block))) { 1065 ext2_error (inode->i_sb, "ext2_write_inode", 1066 "unable to read inode block - " 1067 "inode=%lu, block=%lu", inode->i_ino, block); 1068 return -EIO; 1069 } 1070 offset &= EXT2_BLOCK_SIZE(inode->i_sb) - 1; 1071 raw_inode = (struct ext2_inode *) (bh->b_data + offset); 1072 1073 raw_inode->i_mode = cpu_to_le16(inode->i_mode); 1074 if(!(test_opt(inode->i_sb, NO_UID32))) { 1075 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); 1076 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); 1077/* 1078 * Fix up interoperability with old kernels. Otherwise, old inodes get 1079 * re-used with the upper 16 bits of the uid/gid intact 1080 */ 1081 if(!inode->u.ext2_i.i_dtime) { 1082 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(inode->i_uid)); 1083 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(inode->i_gid)); 1084 } else { 1085 raw_inode->i_uid_high = 0; 1086 raw_inode->i_gid_high = 0; 1087 } 1088 } else { 1089 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(inode->i_uid)); 1090 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(inode->i_gid)); 1091 raw_inode->i_uid_high = 0; 1092 raw_inode->i_gid_high = 0; 1093 } 1094 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 1095 raw_inode->i_size = cpu_to_le32(inode->i_size); 1096 raw_inode->i_atime = cpu_to_le32(inode->i_atime); 1097 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime); 1098 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime); 1099 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); 1100 raw_inode->i_dtime = cpu_to_le32(inode->u.ext2_i.i_dtime); 1101 raw_inode->i_flags = cpu_to_le32(inode->u.ext2_i.i_flags); 1102 raw_inode->i_faddr = cpu_to_le32(inode->u.ext2_i.i_faddr); 1103 raw_inode->i_frag = inode->u.ext2_i.i_frag_no; 1104 raw_inode->i_fsize = inode->u.ext2_i.i_frag_size; 1105 raw_inode->i_file_acl = cpu_to_le32(inode->u.ext2_i.i_file_acl); 1106 if (!S_ISREG(inode->i_mode)) 1107 raw_inode->i_dir_acl = cpu_to_le32(inode->u.ext2_i.i_dir_acl); 1108 else { 1109 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32); 1110 if (inode->i_size > 0x7fffffffULL) { 1111 struct super_block *sb = inode->i_sb; 1112 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb, 1113 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) || 1114 EXT2_SB(sb)->s_es->s_rev_level == 1115 cpu_to_le32(EXT2_GOOD_OLD_REV)) { 1116 /* If this is the first large file 1117 * created, add a flag to the superblock. 1118 */ 1119 lock_kernel(); 1120 ext2_update_dynamic_rev(sb); 1121 EXT2_SET_RO_COMPAT_FEATURE(sb, 1122 EXT2_FEATURE_RO_COMPAT_LARGE_FILE); 1123 unlock_kernel(); 1124 ext2_write_super(sb); 1125 } 1126 } 1127 } 1128 1129 raw_inode->i_generation = cpu_to_le32(inode->i_generation); 1130 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) 1131 raw_inode->i_block[0] = cpu_to_le32(kdev_t_to_nr(inode->i_rdev)); 1132 else for (block = 0; block < EXT2_N_BLOCKS; block++) 1133 raw_inode->i_block[block] = inode->u.ext2_i.i_data[block]; 1134 mark_buffer_dirty(bh); 1135 if (do_sync) { 1136 ll_rw_block (WRITE, 1, &bh); 1137 wait_on_buffer (bh); 1138 if (buffer_req(bh) && !buffer_uptodate(bh)) { 1139 printk ("IO error syncing ext2 inode [" 1140 "%s:%08lx]\n", 1141 bdevname(inode->i_dev), inode->i_ino); 1142 err = -EIO; 1143 } 1144 } 1145 brelse (bh); 1146 return err; 1147} 1148 1149void ext2_write_inode (struct inode * inode, int wait) 1150{ 1151 lock_kernel(); 1152 ext2_update_inode (inode, wait); 1153 unlock_kernel(); 1154} 1155 1156int ext2_sync_inode (struct inode *inode) 1157{ 1158 return ext2_update_inode (inode, 1); 1159} 1160