1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext2/inode.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Goal-directed block allocation by Stephen Tweedie 17 * (sct@dcs.ed.ac.uk), 1993, 1998 18 * Big-endian to little-endian byte-swapping/bitmaps by 19 * David S. Miller (davem@caip.rutgers.edu), 1995 20 * 64-bit file support on 64-bit platforms by Jakub Jelinek 21 * (jj@sunsite.ms.mff.cuni.cz) 22 * 23 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000 24 */ 25 26 #include <linux/time.h> 27 #include <linux/highuid.h> 28 #include <linux/pagemap.h> 29 #include <linux/blkdev.h> 30 #include <linux/quotaops.h> 31 #include <linux/writeback.h> 32 #include <linux/buffer_head.h> 33 #include <linux/mpage.h> 34 #include <linux/fiemap.h> 35 #include <linux/iomap.h> 36 #include <linux/namei.h> 37 #include <linux/uio.h> 38 #include "ext2.h" 39 #include "acl.h" 40 #include "xattr.h" 41 42 /* 43 * Test whether an inode is a fast symlink. 44 */ 45 static inline int ext2_inode_is_fast_symlink(struct inode *inode) 46 { 47 int ea_blocks = EXT2_I(inode)->i_file_acl ? 48 (inode->i_sb->s_blocksize >> 9) : 0; 49 50 return (S_ISLNK(inode->i_mode) && 51 inode->i_blocks - ea_blocks == 0); 52 } 53 54 static void ext2_truncate_blocks(struct inode *inode, loff_t offset); 55 56 void ext2_write_failed(struct address_space *mapping, loff_t to) 57 { 58 struct inode *inode = mapping->host; 59 60 if (to > inode->i_size) { 61 truncate_pagecache(inode, inode->i_size); 62 ext2_truncate_blocks(inode, inode->i_size); 63 } 64 } 65 66 /* 67 * Called at the last iput() if i_nlink is zero. 68 */ 69 void ext2_evict_inode(struct inode * inode) 70 { 71 struct ext2_block_alloc_info *rsv; 72 int want_delete = 0; 73 74 if (!inode->i_nlink && !is_bad_inode(inode)) { 75 want_delete = 1; 76 dquot_initialize(inode); 77 } else { 78 dquot_drop(inode); 79 } 80 81 truncate_inode_pages_final(&inode->i_data); 82 83 if (want_delete) { 84 sb_start_intwrite(inode->i_sb); 85 /* set dtime */ 86 EXT2_I(inode)->i_dtime = ktime_get_real_seconds(); 87 mark_inode_dirty(inode); 88 sync_inode_metadata(inode, inode_needs_sync(inode)); 89 /* truncate to 0 */ 90 inode->i_size = 0; 91 if (inode->i_blocks) 92 ext2_truncate_blocks(inode, 0); 93 ext2_xattr_delete_inode(inode); 94 } else { 95 mmb_sync(&EXT2_I(inode)->i_metadata_bhs); 96 } 97 mmb_invalidate(&EXT2_I(inode)->i_metadata_bhs); 98 clear_inode(inode); 99 100 ext2_discard_reservation(inode); 101 rsv = EXT2_I(inode)->i_block_alloc_info; 102 EXT2_I(inode)->i_block_alloc_info = NULL; 103 if (unlikely(rsv)) 104 kfree(rsv); 105 106 if (want_delete) { 107 ext2_free_inode(inode); 108 sb_end_intwrite(inode->i_sb); 109 } 110 } 111 112 typedef struct { 113 __le32 *p; 114 __le32 key; 115 struct buffer_head *bh; 116 } Indirect; 117 118 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) 119 { 120 p->key = *(p->p = v); 121 p->bh = bh; 122 } 123 124 static inline int verify_chain(Indirect *from, Indirect *to) 125 { 126 while (from <= to && from->key == *from->p) 127 from++; 128 return (from > to); 129 } 130 131 /** 132 * ext2_block_to_path - parse the block number into array of offsets 133 * @inode: inode in question (we are only interested in its superblock) 134 * @i_block: block number to be parsed 135 * @offsets: array to store the offsets in 136 * @boundary: set this non-zero if the referred-to block is likely to be 137 * followed (on disk) by an indirect block. 138 * To store the locations of file's data ext2 uses a data structure common 139 * for UNIX filesystems - tree of pointers anchored in the inode, with 140 * data blocks at leaves and indirect blocks in intermediate nodes. 141 * This function translates the block number into path in that tree - 142 * return value is the path length and @offsets[n] is the offset of 143 * pointer to (n+1)th node in the nth one. If @block is out of range 144 * (negative or too large) warning is printed and zero returned. 145 * 146 * Note: function doesn't find node addresses, so no IO is needed. All 147 * we need to know is the capacity of indirect blocks (taken from the 148 * inode->i_sb). 149 */ 150 151 /* 152 * Portability note: the last comparison (check that we fit into triple 153 * indirect block) is spelled differently, because otherwise on an 154 * architecture with 32-bit longs and 8Kb pages we might get into trouble 155 * if our filesystem had 8Kb blocks. We might use long long, but that would 156 * kill us on x86. Oh, well, at least the sign propagation does not matter - 157 * i_block would have to be negative in the very beginning, so we would not 158 * get there at all. 159 */ 160 161 static int ext2_block_to_path(struct inode *inode, 162 long i_block, int offsets[4], int *boundary) 163 { 164 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb); 165 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb); 166 const long direct_blocks = EXT2_NDIR_BLOCKS, 167 indirect_blocks = ptrs, 168 double_blocks = (1 << (ptrs_bits * 2)); 169 int n = 0; 170 int final = 0; 171 172 if (i_block < 0) { 173 ext2_msg(inode->i_sb, KERN_WARNING, 174 "warning: %s: block < 0", __func__); 175 } else if (i_block < direct_blocks) { 176 offsets[n++] = i_block; 177 final = direct_blocks; 178 } else if ( (i_block -= direct_blocks) < indirect_blocks) { 179 offsets[n++] = EXT2_IND_BLOCK; 180 offsets[n++] = i_block; 181 final = ptrs; 182 } else if ((i_block -= indirect_blocks) < double_blocks) { 183 offsets[n++] = EXT2_DIND_BLOCK; 184 offsets[n++] = i_block >> ptrs_bits; 185 offsets[n++] = i_block & (ptrs - 1); 186 final = ptrs; 187 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { 188 offsets[n++] = EXT2_TIND_BLOCK; 189 offsets[n++] = i_block >> (ptrs_bits * 2); 190 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); 191 offsets[n++] = i_block & (ptrs - 1); 192 final = ptrs; 193 } else { 194 ext2_msg(inode->i_sb, KERN_WARNING, 195 "warning: %s: block is too big", __func__); 196 } 197 if (boundary) 198 *boundary = final - 1 - (i_block & (ptrs - 1)); 199 200 return n; 201 } 202 203 /** 204 * ext2_get_branch - read the chain of indirect blocks leading to data 205 * @inode: inode in question 206 * @depth: depth of the chain (1 - direct pointer, etc.) 207 * @offsets: offsets of pointers in inode/indirect blocks 208 * @chain: place to store the result 209 * @err: here we store the error value 210 * 211 * Function fills the array of triples <key, p, bh> and returns %NULL 212 * if everything went OK or the pointer to the last filled triple 213 * (incomplete one) otherwise. Upon the return chain[i].key contains 214 * the number of (i+1)-th block in the chain (as it is stored in memory, 215 * i.e. little-endian 32-bit), chain[i].p contains the address of that 216 * number (it points into struct inode for i==0 and into the bh->b_data 217 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect 218 * block for i>0 and NULL for i==0. In other words, it holds the block 219 * numbers of the chain, addresses they were taken from (and where we can 220 * verify that chain did not change) and buffer_heads hosting these 221 * numbers. 222 * 223 * Function stops when it stumbles upon zero pointer (absent block) 224 * (pointer to last triple returned, *@err == 0) 225 * or when it gets an IO error reading an indirect block 226 * (ditto, *@err == -EIO) 227 * or when it notices that chain had been changed while it was reading 228 * (ditto, *@err == -EAGAIN) 229 * or when it reads all @depth-1 indirect blocks successfully and finds 230 * the whole chain, all way to the data (returns %NULL, *err == 0). 231 */ 232 static Indirect *ext2_get_branch(struct inode *inode, 233 int depth, 234 int *offsets, 235 Indirect chain[4], 236 int *err) 237 { 238 struct super_block *sb = inode->i_sb; 239 Indirect *p = chain; 240 struct buffer_head *bh; 241 242 *err = 0; 243 /* i_data is not going away, no lock needed */ 244 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets); 245 if (!p->key) 246 goto no_block; 247 while (--depth) { 248 bh = sb_bread(sb, le32_to_cpu(p->key)); 249 if (!bh) 250 goto failure; 251 read_lock(&EXT2_I(inode)->i_meta_lock); 252 if (!verify_chain(chain, p)) 253 goto changed; 254 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); 255 read_unlock(&EXT2_I(inode)->i_meta_lock); 256 if (!p->key) 257 goto no_block; 258 } 259 return NULL; 260 261 changed: 262 read_unlock(&EXT2_I(inode)->i_meta_lock); 263 brelse(bh); 264 *err = -EAGAIN; 265 goto no_block; 266 failure: 267 *err = -EIO; 268 no_block: 269 return p; 270 } 271 272 /** 273 * ext2_find_near - find a place for allocation with sufficient locality 274 * @inode: owner 275 * @ind: descriptor of indirect block. 276 * 277 * This function returns the preferred place for block allocation. 278 * It is used when heuristic for sequential allocation fails. 279 * Rules are: 280 * + if there is a block to the left of our position - allocate near it. 281 * + if pointer will live in indirect block - allocate near that block. 282 * + if pointer will live in inode - allocate in the same cylinder group. 283 * 284 * In the latter case we colour the starting block by the callers PID to 285 * prevent it from clashing with concurrent allocations for a different inode 286 * in the same block group. The PID is used here so that functionally related 287 * files will be close-by on-disk. 288 * 289 * Caller must make sure that @ind is valid and will stay that way. 290 */ 291 292 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind) 293 { 294 struct ext2_inode_info *ei = EXT2_I(inode); 295 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; 296 __le32 *p; 297 ext2_fsblk_t bg_start; 298 ext2_fsblk_t colour; 299 300 /* Try to find previous block */ 301 for (p = ind->p - 1; p >= start; p--) 302 if (*p) 303 return le32_to_cpu(*p); 304 305 /* No such thing, so let's try location of indirect block */ 306 if (ind->bh) 307 return ind->bh->b_blocknr; 308 309 /* 310 * It is going to be referred from inode itself? OK, just put it into 311 * the same cylinder group then. 312 */ 313 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group); 314 colour = (current->pid % 16) * 315 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16); 316 return bg_start + colour; 317 } 318 319 /** 320 * ext2_find_goal - find a preferred place for allocation. 321 * @inode: owner 322 * @block: block we want 323 * @partial: pointer to the last triple within a chain 324 * 325 * Returns preferred place for a block (the goal). 326 */ 327 328 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block, 329 Indirect *partial) 330 __must_hold(&EXT2_I(inode)->truncate_mutex) 331 { 332 struct ext2_block_alloc_info *block_i; 333 334 block_i = EXT2_I(inode)->i_block_alloc_info; 335 336 /* 337 * try the heuristic for sequential allocation, 338 * failing that at least try to get decent locality. 339 */ 340 if (block_i && (block == block_i->last_alloc_logical_block + 1) 341 && (block_i->last_alloc_physical_block != 0)) { 342 return block_i->last_alloc_physical_block + 1; 343 } 344 345 return ext2_find_near(inode, partial); 346 } 347 348 /** 349 * ext2_blks_to_allocate: Look up the block map and count the number 350 * of direct blocks need to be allocated for the given branch. 351 * 352 * @branch: chain of indirect blocks 353 * @k: number of blocks need for indirect blocks 354 * @blks: number of data blocks to be mapped. 355 * @blocks_to_boundary: the offset in the indirect block 356 * 357 * return the number of direct blocks to allocate. 358 */ 359 static int 360 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks, 361 int blocks_to_boundary) 362 { 363 unsigned long count = 0; 364 365 /* 366 * Simple case, [t,d]Indirect block(s) has not allocated yet 367 * then it's clear blocks on that path have not allocated 368 */ 369 if (k > 0) { 370 /* right now don't hanel cross boundary allocation */ 371 if (blks < blocks_to_boundary + 1) 372 count += blks; 373 else 374 count += blocks_to_boundary + 1; 375 return count; 376 } 377 378 count++; 379 while (count < blks && count <= blocks_to_boundary 380 && le32_to_cpu(*(branch[0].p + count)) == 0) { 381 count++; 382 } 383 return count; 384 } 385 386 /** 387 * ext2_alloc_blocks: Allocate multiple blocks needed for a branch. 388 * @inode: Owner. 389 * @goal: Preferred place for allocation. 390 * @indirect_blks: The number of blocks needed to allocate for indirect blocks. 391 * @blks: The number of blocks need to allocate for direct blocks. 392 * @new_blocks: On return it will store the new block numbers for 393 * the indirect blocks(if needed) and the first direct block. 394 * @err: Error pointer. 395 * 396 * Return: Number of blocks allocated. 397 */ 398 static int ext2_alloc_blocks(struct inode *inode, 399 ext2_fsblk_t goal, int indirect_blks, int blks, 400 ext2_fsblk_t new_blocks[4], int *err) 401 __must_hold(&EXT2_I(inode)->truncate_mutex) 402 { 403 int target, i; 404 unsigned long count = 0; 405 int index = 0; 406 ext2_fsblk_t current_block = 0; 407 int ret = 0; 408 409 /* 410 * Here we try to allocate the requested multiple blocks at once, 411 * on a best-effort basis. 412 * To build a branch, we should allocate blocks for 413 * the indirect blocks(if not allocated yet), and at least 414 * the first direct block of this branch. That's the 415 * minimum number of blocks need to allocate(required) 416 */ 417 target = blks + indirect_blks; 418 419 while (1) { 420 count = target; 421 /* allocating blocks for indirect blocks and direct blocks */ 422 current_block = ext2_new_blocks(inode, goal, &count, err, 0); 423 if (*err) 424 goto failed_out; 425 426 target -= count; 427 /* allocate blocks for indirect blocks */ 428 while (index < indirect_blks && count) { 429 new_blocks[index++] = current_block++; 430 count--; 431 } 432 433 if (count > 0) 434 break; 435 } 436 437 /* save the new block number for the first direct block */ 438 new_blocks[index] = current_block; 439 440 /* total number of blocks allocated for direct blocks */ 441 ret = count; 442 *err = 0; 443 return ret; 444 failed_out: 445 for (i = 0; i <index; i++) 446 ext2_free_blocks(inode, new_blocks[i], 1); 447 if (index) 448 mark_inode_dirty(inode); 449 return ret; 450 } 451 452 /** 453 * ext2_alloc_branch - allocate and set up a chain of blocks. 454 * @inode: owner 455 * @indirect_blks: depth of the chain (number of blocks to allocate) 456 * @blks: number of allocated direct blocks 457 * @goal: preferred place for allocation 458 * @offsets: offsets (in the blocks) to store the pointers to next. 459 * @branch: place to store the chain in. 460 * 461 * This function allocates @num blocks, zeroes out all but the last one, 462 * links them into chain and (if we are synchronous) writes them to disk. 463 * In other words, it prepares a branch that can be spliced onto the 464 * inode. It stores the information about that chain in the branch[], in 465 * the same format as ext2_get_branch() would do. We are calling it after 466 * we had read the existing part of chain and partial points to the last 467 * triple of that (one with zero ->key). Upon the exit we have the same 468 * picture as after the successful ext2_get_block(), except that in one 469 * place chain is disconnected - *branch->p is still zero (we did not 470 * set the last link), but branch->key contains the number that should 471 * be placed into *branch->p to fill that gap. 472 * 473 * If allocation fails we free all blocks we've allocated (and forget 474 * their buffer_heads) and return the error value the from failed 475 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain 476 * as described above and return 0. 477 */ 478 479 static int ext2_alloc_branch(struct inode *inode, 480 int indirect_blks, int *blks, ext2_fsblk_t goal, 481 int *offsets, Indirect *branch) 482 __must_hold(&EXT2_I(inode)->truncate_mutex) 483 { 484 int blocksize = inode->i_sb->s_blocksize; 485 int i, n = 0; 486 int err = 0; 487 struct buffer_head *bh; 488 int num; 489 ext2_fsblk_t new_blocks[4]; 490 ext2_fsblk_t current_block; 491 492 num = ext2_alloc_blocks(inode, goal, indirect_blks, 493 *blks, new_blocks, &err); 494 if (err) 495 return err; 496 497 branch[0].key = cpu_to_le32(new_blocks[0]); 498 /* 499 * metadata blocks and data blocks are allocated. 500 */ 501 for (n = 1; n <= indirect_blks; n++) { 502 /* 503 * Get buffer_head for parent block, zero it out 504 * and set the pointer to new one, then send 505 * parent to disk. 506 */ 507 bh = sb_getblk(inode->i_sb, new_blocks[n-1]); 508 if (unlikely(!bh)) { 509 err = -ENOMEM; 510 goto failed; 511 } 512 branch[n].bh = bh; 513 lock_buffer(bh); 514 memset(bh->b_data, 0, blocksize); 515 branch[n].p = (__le32 *) bh->b_data + offsets[n]; 516 branch[n].key = cpu_to_le32(new_blocks[n]); 517 *branch[n].p = branch[n].key; 518 if ( n == indirect_blks) { 519 current_block = new_blocks[n]; 520 /* 521 * End of chain, update the last new metablock of 522 * the chain to point to the new allocated 523 * data blocks numbers 524 */ 525 for (i=1; i < num; i++) 526 *(branch[n].p + i) = cpu_to_le32(++current_block); 527 } 528 set_buffer_uptodate(bh); 529 unlock_buffer(bh); 530 mmb_mark_buffer_dirty(bh, &EXT2_I(inode)->i_metadata_bhs); 531 /* We used to sync bh here if IS_SYNC(inode). 532 * But we now rely upon generic_write_sync() 533 * and b_inode_buffers. But not for directories. 534 */ 535 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) 536 sync_dirty_buffer(bh); 537 } 538 *blks = num; 539 return err; 540 541 failed: 542 for (i = 1; i < n; i++) 543 bforget(branch[i].bh); 544 for (i = 0; i < indirect_blks; i++) 545 ext2_free_blocks(inode, new_blocks[i], 1); 546 ext2_free_blocks(inode, new_blocks[i], num); 547 return err; 548 } 549 550 /** 551 * ext2_splice_branch - splice the allocated branch onto inode. 552 * @inode: owner 553 * @block: (logical) number of block we are adding 554 * @where: location of missing link 555 * @num: number of indirect blocks we are adding 556 * @blks: number of direct blocks we are adding 557 * 558 * This function fills the missing link and does all housekeeping needed in 559 * inode (->i_blocks, etc.). In case of success we end up with the full 560 * chain to new block and return 0. 561 */ 562 static void ext2_splice_branch(struct inode *inode, 563 long block, Indirect *where, int num, int blks) 564 __must_hold(&EXT2_I(inode)->truncate_mutex) 565 { 566 int i; 567 struct ext2_block_alloc_info *block_i; 568 ext2_fsblk_t current_block; 569 570 block_i = EXT2_I(inode)->i_block_alloc_info; 571 572 /* XXX LOCKING probably should have i_meta_lock ?*/ 573 /* That's it */ 574 575 *where->p = where->key; 576 577 /* 578 * Update the host buffer_head or inode to point to more just allocated 579 * direct blocks blocks 580 */ 581 if (num == 0 && blks > 1) { 582 current_block = le32_to_cpu(where->key) + 1; 583 for (i = 1; i < blks; i++) 584 *(where->p + i ) = cpu_to_le32(current_block++); 585 } 586 587 /* 588 * update the most recently allocated logical & physical block 589 * in i_block_alloc_info, to assist find the proper goal block for next 590 * allocation 591 */ 592 if (block_i) { 593 block_i->last_alloc_logical_block = block + blks - 1; 594 block_i->last_alloc_physical_block = 595 le32_to_cpu(where[num].key) + blks - 1; 596 } 597 598 /* We are done with atomic stuff, now do the rest of housekeeping */ 599 600 /* had we spliced it onto indirect block? */ 601 if (where->bh) 602 mmb_mark_buffer_dirty(where->bh, &EXT2_I(inode)->i_metadata_bhs); 603 604 inode_set_ctime_current(inode); 605 mark_inode_dirty(inode); 606 } 607 608 /* 609 * Allocation strategy is simple: if we have to allocate something, we will 610 * have to go the whole way to leaf. So let's do it before attaching anything 611 * to tree, set linkage between the newborn blocks, write them if sync is 612 * required, recheck the path, free and repeat if check fails, otherwise 613 * set the last missing link (that will protect us from any truncate-generated 614 * removals - all blocks on the path are immune now) and possibly force the 615 * write on the parent block. 616 * That has a nice additional property: no special recovery from the failed 617 * allocations is needed - we simply release blocks and do not touch anything 618 * reachable from inode. 619 * 620 * `handle' can be NULL if create == 0. 621 * 622 * return > 0, # of blocks mapped or allocated. 623 * return = 0, if plain lookup failed. 624 * return < 0, error case. 625 */ 626 static int ext2_get_blocks(struct inode *inode, 627 sector_t iblock, unsigned long maxblocks, 628 u32 *bno, bool *new, bool *boundary, 629 int create) 630 { 631 int err; 632 int offsets[4]; 633 Indirect chain[4]; 634 Indirect *partial; 635 ext2_fsblk_t goal; 636 int indirect_blks; 637 int blocks_to_boundary = 0; 638 int depth; 639 struct ext2_inode_info *ei = EXT2_I(inode); 640 int count = 0; 641 ext2_fsblk_t first_block = 0; 642 643 if (WARN_ON_ONCE(maxblocks == 0)) 644 return -EINVAL; 645 646 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary); 647 648 if (depth == 0) 649 return -EIO; 650 651 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 652 /* Simplest case - block found, no allocation needed */ 653 if (!partial) { 654 first_block = le32_to_cpu(chain[depth - 1].key); 655 count++; 656 /*map more blocks*/ 657 while (count < maxblocks && count <= blocks_to_boundary) { 658 ext2_fsblk_t blk; 659 660 if (!verify_chain(chain, chain + depth - 1)) { 661 /* 662 * Indirect block might be removed by 663 * truncate while we were reading it. 664 * Handling of that case: forget what we've 665 * got now, go to reread. 666 */ 667 err = -EAGAIN; 668 count = 0; 669 partial = chain + depth - 1; 670 break; 671 } 672 blk = le32_to_cpu(*(chain[depth-1].p + count)); 673 if (blk == first_block + count) 674 count++; 675 else 676 break; 677 } 678 if (err != -EAGAIN) 679 goto got_it; 680 } 681 682 /* Next simple case - plain lookup or failed read of indirect block */ 683 if (!create || err == -EIO) 684 goto cleanup; 685 686 mutex_lock(&ei->truncate_mutex); 687 /* 688 * If the indirect block is missing while we are reading 689 * the chain(ext2_get_branch() returns -EAGAIN err), or 690 * if the chain has been changed after we grab the semaphore, 691 * (either because another process truncated this branch, or 692 * another get_block allocated this branch) re-grab the chain to see if 693 * the request block has been allocated or not. 694 * 695 * Since we already block the truncate/other get_block 696 * at this point, we will have the current copy of the chain when we 697 * splice the branch into the tree. 698 */ 699 if (err == -EAGAIN || !verify_chain(chain, partial)) { 700 while (partial > chain) { 701 brelse(partial->bh); 702 partial--; 703 } 704 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 705 if (!partial) { 706 count++; 707 mutex_unlock(&ei->truncate_mutex); 708 goto got_it; 709 } 710 711 if (err) { 712 mutex_unlock(&ei->truncate_mutex); 713 goto cleanup; 714 } 715 } 716 717 /* 718 * Okay, we need to do block allocation. Lazily initialize the block 719 * allocation info here if necessary 720 */ 721 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) 722 ext2_init_block_alloc_info(inode); 723 724 goal = ext2_find_goal(inode, iblock, partial); 725 726 /* the number of blocks need to allocate for [d,t]indirect blocks */ 727 indirect_blks = (chain + depth) - partial - 1; 728 /* 729 * Next look up the indirect map to count the total number of 730 * direct blocks to allocate for this branch. 731 */ 732 count = ext2_blks_to_allocate(partial, indirect_blks, 733 maxblocks, blocks_to_boundary); 734 /* 735 * XXX ???? Block out ext2_truncate while we alter the tree 736 */ 737 err = ext2_alloc_branch(inode, indirect_blks, &count, goal, 738 offsets + (partial - chain), partial); 739 740 if (err) { 741 mutex_unlock(&ei->truncate_mutex); 742 goto cleanup; 743 } 744 745 *new = true; 746 747 ext2_splice_branch(inode, iblock, partial, indirect_blks, count); 748 mutex_unlock(&ei->truncate_mutex); 749 got_it: 750 if (count > blocks_to_boundary) 751 *boundary = true; 752 err = count; 753 /* Clean up and exit */ 754 partial = chain + depth - 1; /* the whole chain */ 755 cleanup: 756 while (partial > chain) { 757 brelse(partial->bh); 758 partial--; 759 } 760 if (err > 0) 761 *bno = le32_to_cpu(chain[depth-1].key); 762 return err; 763 } 764 765 int ext2_get_block(struct inode *inode, sector_t iblock, 766 struct buffer_head *bh_result, int create) 767 { 768 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; 769 bool new = false, boundary = false; 770 u32 bno; 771 int ret; 772 773 ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary, 774 create); 775 if (ret <= 0) 776 return ret; 777 778 map_bh(bh_result, inode->i_sb, bno); 779 bh_result->b_size = (ret << inode->i_blkbits); 780 if (new) 781 set_buffer_new(bh_result); 782 if (boundary) 783 set_buffer_boundary(bh_result); 784 return 0; 785 786 } 787 788 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 789 unsigned flags, struct iomap *iomap, struct iomap *srcmap) 790 { 791 unsigned int blkbits = inode->i_blkbits; 792 unsigned long first_block = offset >> blkbits; 793 unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits; 794 bool new = false, boundary = false; 795 u32 bno; 796 int ret; 797 bool create = flags & IOMAP_WRITE; 798 799 /* 800 * For writes that could fill holes inside i_size on a 801 * DIO_SKIP_HOLES filesystem we forbid block creations: only 802 * overwrites are permitted. 803 */ 804 if ((flags & IOMAP_DIRECT) && 805 (first_block << blkbits) < i_size_read(inode)) 806 create = 0; 807 808 /* 809 * Writes that span EOF might trigger an IO size update on completion, 810 * so consider them to be dirty for the purposes of O_DSYNC even if 811 * there is no other metadata changes pending or have been made here. 812 */ 813 if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode)) 814 iomap->flags |= IOMAP_F_DIRTY; 815 816 ret = ext2_get_blocks(inode, first_block, max_blocks, 817 &bno, &new, &boundary, create); 818 if (ret < 0) 819 return ret; 820 821 iomap->flags = 0; 822 iomap->offset = (u64)first_block << blkbits; 823 iomap->bdev = inode->i_sb->s_bdev; 824 825 if (ret == 0) { 826 /* 827 * Switch to buffered-io for writing to holes in a non-extent 828 * based filesystem to avoid stale data exposure problem. 829 */ 830 if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT)) 831 return -ENOTBLK; 832 iomap->type = IOMAP_HOLE; 833 iomap->addr = IOMAP_NULL_ADDR; 834 iomap->length = 1 << blkbits; 835 } else { 836 iomap->type = IOMAP_MAPPED; 837 iomap->addr = (u64)bno << blkbits; 838 iomap->length = (u64)ret << blkbits; 839 iomap->flags |= IOMAP_F_MERGED; 840 } 841 842 if (new) 843 iomap->flags |= IOMAP_F_NEW; 844 return 0; 845 } 846 847 static int 848 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length, 849 ssize_t written, unsigned flags, struct iomap *iomap) 850 { 851 /* 852 * Switch to buffered-io in case of any error. 853 * Blocks allocated can be used by the buffered-io path. 854 */ 855 if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0) 856 return -ENOTBLK; 857 858 if (iomap->type == IOMAP_MAPPED && 859 written < length && 860 (flags & IOMAP_WRITE)) 861 ext2_write_failed(inode->i_mapping, offset + length); 862 return 0; 863 } 864 865 static DEFINE_IOMAP_ITER_NEXT_END(ext2_iomap_next, ext2_iomap_begin, 866 ext2_iomap_end); 867 868 const struct iomap_ops ext2_iomap_ops = { 869 .iomap_next = ext2_iomap_next, 870 }; 871 872 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 873 u64 start, u64 len) 874 { 875 int ret; 876 loff_t i_size; 877 878 inode_lock(inode); 879 i_size = i_size_read(inode); 880 /* 881 * iomap_fiemap() returns EINVAL for 0 length. Make sure we don't trim 882 * length to 0 but still trim the range as much as possible since 883 * ext2_get_blocks() iterates unmapped space block by block which is 884 * slow. 885 */ 886 if (i_size == 0) 887 i_size = 1; 888 len = min_t(u64, len, i_size); 889 ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops); 890 inode_unlock(inode); 891 892 return ret; 893 } 894 895 static int ext2_read_folio(struct file *file, struct folio *folio) 896 { 897 return mpage_read_folio(folio, ext2_get_block); 898 } 899 900 static void ext2_readahead(struct readahead_control *rac) 901 { 902 mpage_readahead(rac, ext2_get_block); 903 } 904 905 static int 906 ext2_write_begin(const struct kiocb *iocb, struct address_space *mapping, 907 loff_t pos, unsigned len, struct folio **foliop, void **fsdata) 908 { 909 int ret; 910 911 ret = block_write_begin(mapping, pos, len, foliop, ext2_get_block); 912 if (ret < 0) 913 ext2_write_failed(mapping, pos + len); 914 return ret; 915 } 916 917 static int ext2_write_end(const struct kiocb *iocb, 918 struct address_space *mapping, 919 loff_t pos, unsigned len, unsigned copied, 920 struct folio *folio, void *fsdata) 921 { 922 int ret; 923 924 ret = generic_write_end(iocb, mapping, pos, len, copied, folio, fsdata); 925 if (ret < len) 926 ext2_write_failed(mapping, pos + len); 927 return ret; 928 } 929 930 static sector_t ext2_bmap(struct address_space *mapping, sector_t block) 931 { 932 return generic_block_bmap(mapping,block,ext2_get_block); 933 } 934 935 static int 936 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc) 937 { 938 return mpage_writepages(mapping, wbc, ext2_get_block); 939 } 940 941 942 const struct address_space_operations ext2_aops = { 943 .dirty_folio = block_dirty_folio, 944 .invalidate_folio = block_invalidate_folio, 945 .read_folio = ext2_read_folio, 946 .readahead = ext2_readahead, 947 .write_begin = ext2_write_begin, 948 .write_end = ext2_write_end, 949 .bmap = ext2_bmap, 950 .writepages = ext2_writepages, 951 .migrate_folio = buffer_migrate_folio, 952 .is_partially_uptodate = block_is_partially_uptodate, 953 .error_remove_folio = generic_error_remove_folio, 954 }; 955 956 957 /* 958 * Probably it should be a library function... search for first non-zero word 959 * or memcmp with zero_page, whatever is better for particular architecture. 960 * Linus? 961 */ 962 static inline int all_zeroes(__le32 *p, __le32 *q) 963 { 964 while (p < q) 965 if (*p++) 966 return 0; 967 return 1; 968 } 969 970 /** 971 * ext2_find_shared - find the indirect blocks for partial truncation. 972 * @inode: inode in question 973 * @depth: depth of the affected branch 974 * @offsets: offsets of pointers in that branch (see ext2_block_to_path) 975 * @chain: place to store the pointers to partial indirect blocks 976 * @top: place to the (detached) top of branch 977 * 978 * This is a helper function used by ext2_truncate(). 979 * 980 * When we do truncate() we may have to clean the ends of several indirect 981 * blocks but leave the blocks themselves alive. Block is partially 982 * truncated if some data below the new i_size is referred from it (and 983 * it is on the path to the first completely truncated data block, indeed). 984 * We have to free the top of that path along with everything to the right 985 * of the path. Since no allocation past the truncation point is possible 986 * until ext2_truncate() finishes, we may safely do the latter, but top 987 * of branch may require special attention - pageout below the truncation 988 * point might try to populate it. 989 * 990 * We atomically detach the top of branch from the tree, store the block 991 * number of its root in *@top, pointers to buffer_heads of partially 992 * truncated blocks - in @chain[].bh and pointers to their last elements 993 * that should not be removed - in @chain[].p. Return value is the pointer 994 * to last filled element of @chain. 995 * 996 * The work left to caller to do the actual freeing of subtrees: 997 * a) free the subtree starting from *@top 998 * b) free the subtrees whose roots are stored in 999 * (@chain[i].p+1 .. end of @chain[i].bh->b_data) 1000 * c) free the subtrees growing from the inode past the @chain[0].p 1001 * (no partially truncated stuff there). 1002 */ 1003 1004 static Indirect *ext2_find_shared(struct inode *inode, 1005 int depth, 1006 int offsets[4], 1007 Indirect chain[4], 1008 __le32 *top) 1009 __must_hold(&EXT2_I(inode)->truncate_mutex) 1010 { 1011 Indirect *partial, *p; 1012 int k, err; 1013 1014 *top = 0; 1015 for (k = depth; k > 1 && !offsets[k-1]; k--) 1016 ; 1017 partial = ext2_get_branch(inode, k, offsets, chain, &err); 1018 if (!partial) 1019 partial = chain + k-1; 1020 /* 1021 * If the branch acquired continuation since we've looked at it - 1022 * fine, it should all survive and (new) top doesn't belong to us. 1023 */ 1024 write_lock(&EXT2_I(inode)->i_meta_lock); 1025 if (!partial->key && *partial->p) { 1026 write_unlock(&EXT2_I(inode)->i_meta_lock); 1027 goto no_top; 1028 } 1029 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) 1030 ; 1031 /* 1032 * OK, we've found the last block that must survive. The rest of our 1033 * branch should be detached before unlocking. However, if that rest 1034 * of branch is all ours and does not grow immediately from the inode 1035 * it's easier to cheat and just decrement partial->p. 1036 */ 1037 if (p == chain + k - 1 && p > chain) { 1038 p->p--; 1039 } else { 1040 *top = *p->p; 1041 *p->p = 0; 1042 } 1043 write_unlock(&EXT2_I(inode)->i_meta_lock); 1044 1045 while(partial > p) 1046 { 1047 brelse(partial->bh); 1048 partial--; 1049 } 1050 no_top: 1051 return partial; 1052 } 1053 1054 /** 1055 * ext2_free_data - free a list of data blocks 1056 * @inode: inode we are dealing with 1057 * @p: array of block numbers 1058 * @q: points immediately past the end of array 1059 * 1060 * We are freeing all blocks referred from that array (numbers are 1061 * stored as little-endian 32-bit) and updating @inode->i_blocks 1062 * appropriately. 1063 */ 1064 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q) 1065 __must_hold(&EXT2_I(inode)->truncate_mutex) 1066 { 1067 ext2_fsblk_t block_to_free = 0, count = 0; 1068 ext2_fsblk_t nr; 1069 1070 for ( ; p < q ; p++) { 1071 nr = le32_to_cpu(*p); 1072 if (nr) { 1073 *p = 0; 1074 /* accumulate blocks to free if they're contiguous */ 1075 if (count == 0) 1076 goto free_this; 1077 else if (block_to_free == nr - count) 1078 count++; 1079 else { 1080 ext2_free_blocks (inode, block_to_free, count); 1081 mark_inode_dirty(inode); 1082 free_this: 1083 block_to_free = nr; 1084 count = 1; 1085 } 1086 } 1087 } 1088 if (count > 0) { 1089 ext2_free_blocks (inode, block_to_free, count); 1090 mark_inode_dirty(inode); 1091 } 1092 } 1093 1094 /** 1095 * ext2_free_branches - free an array of branches 1096 * @inode: inode we are dealing with 1097 * @p: array of block numbers 1098 * @q: pointer immediately past the end of array 1099 * @depth: depth of the branches to free 1100 * 1101 * We are freeing all blocks referred from these branches (numbers are 1102 * stored as little-endian 32-bit) and updating @inode->i_blocks 1103 * appropriately. 1104 */ 1105 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth) 1106 __must_hold(&EXT2_I(inode)->truncate_mutex) 1107 { 1108 struct buffer_head * bh; 1109 ext2_fsblk_t nr; 1110 1111 if (depth--) { 1112 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 1113 for ( ; p < q ; p++) { 1114 nr = le32_to_cpu(*p); 1115 if (!nr) 1116 continue; 1117 *p = 0; 1118 bh = sb_bread(inode->i_sb, nr); 1119 /* 1120 * A read failure? Report error and clear slot 1121 * (should be rare). 1122 */ 1123 if (!bh) { 1124 ext2_error(inode->i_sb, "ext2_free_branches", 1125 "Read failure, inode=%llu, block=%ld", 1126 inode->i_ino, nr); 1127 continue; 1128 } 1129 ext2_free_branches(inode, 1130 (__le32*)bh->b_data, 1131 (__le32*)bh->b_data + addr_per_block, 1132 depth); 1133 bforget(bh); 1134 ext2_free_blocks(inode, nr, 1); 1135 mark_inode_dirty(inode); 1136 } 1137 } else 1138 ext2_free_data(inode, p, q); 1139 } 1140 1141 /* mapping->invalidate_lock must be held when calling this function */ 1142 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset) 1143 { 1144 __le32 *i_data = EXT2_I(inode)->i_data; 1145 struct ext2_inode_info *ei = EXT2_I(inode); 1146 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 1147 int offsets[4]; 1148 Indirect chain[4]; 1149 Indirect *partial; 1150 __le32 nr = 0; 1151 int n; 1152 long iblock; 1153 unsigned blocksize; 1154 blocksize = inode->i_sb->s_blocksize; 1155 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb); 1156 1157 1158 n = ext2_block_to_path(inode, iblock, offsets, NULL); 1159 if (n == 0) 1160 return; 1161 1162 /* 1163 * From here we block out all ext2_get_block() callers who want to 1164 * modify the block allocation tree. 1165 */ 1166 mutex_lock(&ei->truncate_mutex); 1167 1168 if (n == 1) { 1169 ext2_free_data(inode, i_data+offsets[0], 1170 i_data + EXT2_NDIR_BLOCKS); 1171 goto do_indirects; 1172 } 1173 1174 partial = ext2_find_shared(inode, n, offsets, chain, &nr); 1175 /* Kill the top of shared branch (already detached) */ 1176 if (nr) { 1177 if (partial == chain) 1178 mark_inode_dirty(inode); 1179 else 1180 mmb_mark_buffer_dirty(partial->bh, 1181 &EXT2_I(inode)->i_metadata_bhs); 1182 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial); 1183 } 1184 /* Clear the ends of indirect blocks on the shared branch */ 1185 while (partial > chain) { 1186 ext2_free_branches(inode, 1187 partial->p + 1, 1188 (__le32*)partial->bh->b_data+addr_per_block, 1189 (chain+n-1) - partial); 1190 mmb_mark_buffer_dirty(partial->bh, 1191 &EXT2_I(inode)->i_metadata_bhs); 1192 brelse (partial->bh); 1193 partial--; 1194 } 1195 do_indirects: 1196 /* Kill the remaining (whole) subtrees */ 1197 switch (offsets[0]) { 1198 default: 1199 nr = i_data[EXT2_IND_BLOCK]; 1200 if (nr) { 1201 i_data[EXT2_IND_BLOCK] = 0; 1202 mark_inode_dirty(inode); 1203 ext2_free_branches(inode, &nr, &nr+1, 1); 1204 } 1205 fallthrough; 1206 case EXT2_IND_BLOCK: 1207 nr = i_data[EXT2_DIND_BLOCK]; 1208 if (nr) { 1209 i_data[EXT2_DIND_BLOCK] = 0; 1210 mark_inode_dirty(inode); 1211 ext2_free_branches(inode, &nr, &nr+1, 2); 1212 } 1213 fallthrough; 1214 case EXT2_DIND_BLOCK: 1215 nr = i_data[EXT2_TIND_BLOCK]; 1216 if (nr) { 1217 i_data[EXT2_TIND_BLOCK] = 0; 1218 mark_inode_dirty(inode); 1219 ext2_free_branches(inode, &nr, &nr+1, 3); 1220 } 1221 break; 1222 case EXT2_TIND_BLOCK: 1223 ; 1224 } 1225 1226 ext2_discard_reservation(inode); 1227 1228 mutex_unlock(&ei->truncate_mutex); 1229 } 1230 1231 static void ext2_truncate_blocks(struct inode *inode, loff_t offset) 1232 { 1233 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1234 S_ISLNK(inode->i_mode))) 1235 return; 1236 if (ext2_inode_is_fast_symlink(inode)) 1237 return; 1238 1239 filemap_invalidate_lock(inode->i_mapping); 1240 __ext2_truncate_blocks(inode, offset); 1241 filemap_invalidate_unlock(inode->i_mapping); 1242 } 1243 1244 static int ext2_setsize(struct inode *inode, loff_t newsize) 1245 { 1246 int error; 1247 1248 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1249 S_ISLNK(inode->i_mode))) 1250 return -EINVAL; 1251 if (ext2_inode_is_fast_symlink(inode)) 1252 return -EINVAL; 1253 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 1254 return -EPERM; 1255 1256 inode_dio_wait(inode); 1257 1258 error = block_truncate_page(inode->i_mapping, newsize, ext2_get_block); 1259 if (error) 1260 return error; 1261 1262 filemap_invalidate_lock(inode->i_mapping); 1263 truncate_setsize(inode, newsize); 1264 __ext2_truncate_blocks(inode, newsize); 1265 filemap_invalidate_unlock(inode->i_mapping); 1266 1267 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 1268 mark_inode_dirty(inode); 1269 if (inode_needs_sync(inode)) 1270 sync_inode_metadata(inode, 1); 1271 1272 return 0; 1273 } 1274 1275 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino, 1276 struct buffer_head **p) 1277 { 1278 struct buffer_head * bh; 1279 unsigned long block_group; 1280 unsigned long block; 1281 unsigned long offset; 1282 struct ext2_group_desc * gdp; 1283 1284 *p = NULL; 1285 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) || 1286 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count)) 1287 goto Einval; 1288 1289 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb); 1290 gdp = ext2_get_group_desc(sb, block_group, NULL); 1291 if (!gdp) 1292 goto Egdp; 1293 /* 1294 * Figure out the offset within the block group inode table 1295 */ 1296 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb); 1297 block = le32_to_cpu(gdp->bg_inode_table) + 1298 (offset >> EXT2_BLOCK_SIZE_BITS(sb)); 1299 if (!(bh = sb_bread(sb, block))) 1300 goto Eio; 1301 1302 *p = bh; 1303 offset &= (EXT2_BLOCK_SIZE(sb) - 1); 1304 return (struct ext2_inode *) (bh->b_data + offset); 1305 1306 Einval: 1307 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu", 1308 (unsigned long) ino); 1309 return ERR_PTR(-EINVAL); 1310 Eio: 1311 ext2_error(sb, "ext2_get_inode", 1312 "unable to read inode block - inode=%lu, block=%lu", 1313 (unsigned long) ino, block); 1314 Egdp: 1315 return ERR_PTR(-EIO); 1316 } 1317 1318 void ext2_set_inode_flags(struct inode *inode) 1319 { 1320 unsigned int flags = EXT2_I(inode)->i_flags; 1321 1322 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | 1323 S_DIRSYNC); 1324 if (flags & EXT2_SYNC_FL) 1325 inode->i_flags |= S_SYNC; 1326 if (flags & EXT2_APPEND_FL) 1327 inode->i_flags |= S_APPEND; 1328 if (flags & EXT2_IMMUTABLE_FL) 1329 inode->i_flags |= S_IMMUTABLE; 1330 if (flags & EXT2_NOATIME_FL) 1331 inode->i_flags |= S_NOATIME; 1332 if (flags & EXT2_DIRSYNC_FL) 1333 inode->i_flags |= S_DIRSYNC; 1334 } 1335 1336 void ext2_set_file_ops(struct inode *inode) 1337 { 1338 inode->i_op = &ext2_file_inode_operations; 1339 inode->i_fop = &ext2_file_operations; 1340 inode->i_mapping->a_ops = &ext2_aops; 1341 } 1342 1343 struct inode *ext2_iget (struct super_block *sb, unsigned long ino) 1344 { 1345 struct ext2_inode_info *ei; 1346 struct buffer_head * bh = NULL; 1347 struct ext2_inode *raw_inode; 1348 struct inode *inode; 1349 long ret = -EIO; 1350 int n; 1351 uid_t i_uid; 1352 gid_t i_gid; 1353 1354 inode = iget_locked(sb, ino); 1355 if (!inode) 1356 return ERR_PTR(-ENOMEM); 1357 if (!(inode_state_read_once(inode) & I_NEW)) 1358 return inode; 1359 1360 ei = EXT2_I(inode); 1361 ei->i_block_alloc_info = NULL; 1362 1363 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh); 1364 if (IS_ERR(raw_inode)) { 1365 ret = PTR_ERR(raw_inode); 1366 goto bad_inode; 1367 } 1368 1369 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 1370 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 1371 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 1372 if (!(test_opt (inode->i_sb, NO_UID32))) { 1373 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 1374 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 1375 } 1376 i_uid_write(inode, i_uid); 1377 i_gid_write(inode, i_gid); 1378 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); 1379 inode->i_size = le32_to_cpu(raw_inode->i_size); 1380 inode_set_atime(inode, (signed)le32_to_cpu(raw_inode->i_atime), 0); 1381 inode_set_ctime(inode, (signed)le32_to_cpu(raw_inode->i_ctime), 0); 1382 inode_set_mtime(inode, (signed)le32_to_cpu(raw_inode->i_mtime), 0); 1383 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); 1384 /* We now have enough fields to check if the inode was active or not. 1385 * This is needed because nfsd might try to access dead inodes 1386 * the test is that same one that e2fsck uses 1387 * NeilBrown 1999oct15 1388 */ 1389 if (inode->i_nlink == 0) { 1390 if (inode->i_mode == 0 || ei->i_dtime) { 1391 /* this inode is deleted */ 1392 ret = -ESTALE; 1393 } else { 1394 ext2_error(sb, __func__, 1395 "inode %lu has zero i_nlink with mode 0%o and no dtime, " 1396 "filesystem may be corrupt", 1397 ino, inode->i_mode); 1398 ret = -EFSCORRUPTED; 1399 } 1400 goto bad_inode; 1401 } 1402 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); 1403 ei->i_flags = le32_to_cpu(raw_inode->i_flags); 1404 ext2_set_inode_flags(inode); 1405 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); 1406 ei->i_frag_no = raw_inode->i_frag; 1407 ei->i_frag_size = raw_inode->i_fsize; 1408 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); 1409 ei->i_dir_acl = 0; 1410 1411 if (ei->i_file_acl && 1412 !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) { 1413 ext2_error(sb, "ext2_iget", "bad extended attribute block %u", 1414 ei->i_file_acl); 1415 ret = -EFSCORRUPTED; 1416 goto bad_inode; 1417 } 1418 1419 if (S_ISREG(inode->i_mode)) 1420 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; 1421 else 1422 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); 1423 if (i_size_read(inode) < 0) { 1424 ret = -EFSCORRUPTED; 1425 goto bad_inode; 1426 } 1427 ei->i_dtime = 0; 1428 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 1429 ei->i_state = 0; 1430 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 1431 ei->i_dir_start_lookup = 0; 1432 1433 /* 1434 * NOTE! The in-memory inode i_data array is in little-endian order 1435 * even on big-endian machines: we do NOT byteswap the block numbers! 1436 */ 1437 for (n = 0; n < EXT2_N_BLOCKS; n++) 1438 ei->i_data[n] = raw_inode->i_block[n]; 1439 1440 if (S_ISREG(inode->i_mode)) { 1441 ext2_set_file_ops(inode); 1442 } else if (S_ISDIR(inode->i_mode)) { 1443 inode->i_op = &ext2_dir_inode_operations; 1444 inode->i_fop = &ext2_dir_operations; 1445 inode->i_mapping->a_ops = &ext2_aops; 1446 } else if (S_ISLNK(inode->i_mode)) { 1447 if (ext2_inode_is_fast_symlink(inode)) { 1448 inode->i_link = (char *)ei->i_data; 1449 inode->i_op = &ext2_fast_symlink_inode_operations; 1450 nd_terminate_link(ei->i_data, inode->i_size, 1451 sizeof(ei->i_data) - 1); 1452 } else { 1453 inode->i_op = &ext2_symlink_inode_operations; 1454 inode_nohighmem(inode); 1455 inode->i_mapping->a_ops = &ext2_aops; 1456 } 1457 } else { 1458 inode->i_op = &ext2_special_inode_operations; 1459 if (raw_inode->i_block[0]) 1460 init_special_inode(inode, inode->i_mode, 1461 old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); 1462 else 1463 init_special_inode(inode, inode->i_mode, 1464 new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 1465 } 1466 brelse (bh); 1467 unlock_new_inode(inode); 1468 return inode; 1469 1470 bad_inode: 1471 brelse(bh); 1472 iget_failed(inode); 1473 return ERR_PTR(ret); 1474 } 1475 1476 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc) 1477 { 1478 struct ext2_inode_info *ei = EXT2_I(inode); 1479 struct super_block *sb = inode->i_sb; 1480 ino_t ino = inode->i_ino; 1481 uid_t uid = i_uid_read(inode); 1482 gid_t gid = i_gid_read(inode); 1483 struct buffer_head * bh; 1484 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh); 1485 int n; 1486 int err = 0; 1487 1488 if (IS_ERR(raw_inode)) 1489 return -EIO; 1490 1491 /* For fields not tracking in the in-memory inode, 1492 * initialise them to zero for new inodes. */ 1493 if (ei->i_state & EXT2_STATE_NEW) 1494 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size); 1495 1496 raw_inode->i_mode = cpu_to_le16(inode->i_mode); 1497 if (!(test_opt(sb, NO_UID32))) { 1498 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid)); 1499 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid)); 1500 /* 1501 * Fix up interoperability with old kernels. Otherwise, old inodes get 1502 * re-used with the upper 16 bits of the uid/gid intact 1503 */ 1504 if (!ei->i_dtime) { 1505 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid)); 1506 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid)); 1507 } else { 1508 raw_inode->i_uid_high = 0; 1509 raw_inode->i_gid_high = 0; 1510 } 1511 } else { 1512 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid)); 1513 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid)); 1514 raw_inode->i_uid_high = 0; 1515 raw_inode->i_gid_high = 0; 1516 } 1517 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 1518 raw_inode->i_size = cpu_to_le32(inode->i_size); 1519 raw_inode->i_atime = cpu_to_le32(inode_get_atime_sec(inode)); 1520 raw_inode->i_ctime = cpu_to_le32(inode_get_ctime_sec(inode)); 1521 raw_inode->i_mtime = cpu_to_le32(inode_get_mtime_sec(inode)); 1522 1523 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); 1524 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 1525 raw_inode->i_flags = cpu_to_le32(ei->i_flags); 1526 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); 1527 raw_inode->i_frag = ei->i_frag_no; 1528 raw_inode->i_fsize = ei->i_frag_size; 1529 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); 1530 if (!S_ISREG(inode->i_mode)) 1531 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); 1532 else { 1533 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32); 1534 if (inode->i_size > 0x7fffffffULL) { 1535 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb, 1536 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) || 1537 EXT2_SB(sb)->s_es->s_rev_level == 1538 cpu_to_le32(EXT2_GOOD_OLD_REV)) { 1539 /* If this is the first large file 1540 * created, add a flag to the superblock. 1541 */ 1542 spin_lock(&EXT2_SB(sb)->s_lock); 1543 ext2_update_dynamic_rev(sb); 1544 EXT2_SET_RO_COMPAT_FEATURE(sb, 1545 EXT2_FEATURE_RO_COMPAT_LARGE_FILE); 1546 spin_unlock(&EXT2_SB(sb)->s_lock); 1547 ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1); 1548 } 1549 } 1550 } 1551 1552 raw_inode->i_generation = cpu_to_le32(inode->i_generation); 1553 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { 1554 if (old_valid_dev(inode->i_rdev)) { 1555 raw_inode->i_block[0] = 1556 cpu_to_le32(old_encode_dev(inode->i_rdev)); 1557 raw_inode->i_block[1] = 0; 1558 } else { 1559 raw_inode->i_block[0] = 0; 1560 raw_inode->i_block[1] = 1561 cpu_to_le32(new_encode_dev(inode->i_rdev)); 1562 raw_inode->i_block[2] = 0; 1563 } 1564 } else for (n = 0; n < EXT2_N_BLOCKS; n++) 1565 raw_inode->i_block[n] = ei->i_data[n]; 1566 mark_buffer_dirty(bh); 1567 ei->i_state &= ~EXT2_STATE_NEW; 1568 brelse (bh); 1569 set_inode_metadata_writeback(inode); 1570 return err; 1571 } 1572 1573 int ext2_sync_inode_metadata(struct inode *inode, struct writeback_control *wbc) 1574 { 1575 struct buffer_head *bh; 1576 struct ext2_inode *raw_inode = ext2_get_inode(inode->i_sb, inode->i_ino, 1577 &bh); 1578 int err = 0; 1579 1580 if (IS_ERR(raw_inode)) 1581 return -EIO; 1582 err = mmb_sync(&EXT2_I(inode)->i_metadata_bhs); 1583 if (err) { 1584 ext2_error(inode->i_sb, __func__, 1585 "Error syncing inode metadata ino=%lu\n", 1586 (unsigned long)inode->i_ino); 1587 goto out; 1588 } 1589 sync_dirty_buffer(bh); 1590 if (buffer_write_io_error(bh)) { 1591 ext2_error(inode->i_sb, __func__, 1592 "IO error syncing inode %lu\n", 1593 (unsigned long)inode->i_ino); 1594 err = -EIO; 1595 } 1596 out: 1597 brelse(bh); 1598 return err; 1599 } 1600 1601 int ext2_getattr(const struct mnt_idmap *idmap, const struct path *path, 1602 struct kstat *stat, u32 request_mask, unsigned int query_flags) 1603 { 1604 struct inode *inode = d_inode(path->dentry); 1605 struct ext2_inode_info *ei = EXT2_I(inode); 1606 unsigned int flags; 1607 1608 flags = ei->i_flags & EXT2_FL_USER_VISIBLE; 1609 if (flags & EXT2_APPEND_FL) 1610 stat->attributes |= STATX_ATTR_APPEND; 1611 if (flags & EXT2_COMPR_FL) 1612 stat->attributes |= STATX_ATTR_COMPRESSED; 1613 if (flags & EXT2_IMMUTABLE_FL) 1614 stat->attributes |= STATX_ATTR_IMMUTABLE; 1615 if (flags & EXT2_NODUMP_FL) 1616 stat->attributes |= STATX_ATTR_NODUMP; 1617 stat->attributes_mask |= (STATX_ATTR_APPEND | 1618 STATX_ATTR_COMPRESSED | 1619 STATX_ATTR_ENCRYPTED | 1620 STATX_ATTR_IMMUTABLE | 1621 STATX_ATTR_NODUMP); 1622 1623 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 1624 return 0; 1625 } 1626 1627 int ext2_setattr(const struct mnt_idmap *idmap, struct dentry *dentry, 1628 struct iattr *iattr) 1629 { 1630 struct inode *inode = d_inode(dentry); 1631 int error; 1632 1633 error = setattr_prepare(&nop_mnt_idmap, dentry, iattr); 1634 if (error) 1635 return error; 1636 1637 if (is_quota_modification(&nop_mnt_idmap, inode, iattr)) { 1638 error = dquot_initialize(inode); 1639 if (error) 1640 return error; 1641 } 1642 if (i_uid_needs_update(&nop_mnt_idmap, iattr, inode) || 1643 i_gid_needs_update(&nop_mnt_idmap, iattr, inode)) { 1644 error = dquot_transfer(&nop_mnt_idmap, inode, iattr); 1645 if (error) 1646 return error; 1647 } 1648 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) { 1649 error = ext2_setsize(inode, iattr->ia_size); 1650 if (error) 1651 return error; 1652 } 1653 setattr_copy(&nop_mnt_idmap, inode, iattr); 1654 if (iattr->ia_valid & ATTR_MODE) 1655 error = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode); 1656 mark_inode_dirty(inode); 1657 1658 return error; 1659 } 1660