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/dax.h> 30 #include <linux/blkdev.h> 31 #include <linux/quotaops.h> 32 #include <linux/writeback.h> 33 #include <linux/buffer_head.h> 34 #include <linux/mpage.h> 35 #include <linux/fiemap.h> 36 #include <linux/iomap.h> 37 #include <linux/namei.h> 38 #include <linux/uio.h> 39 #include "ext2.h" 40 #include "acl.h" 41 #include "xattr.h" 42 43 static int __ext2_write_inode(struct inode *inode, int do_sync); 44 45 /* 46 * Test whether an inode is a fast symlink. 47 */ 48 static inline int ext2_inode_is_fast_symlink(struct inode *inode) 49 { 50 int ea_blocks = EXT2_I(inode)->i_file_acl ? 51 (inode->i_sb->s_blocksize >> 9) : 0; 52 53 return (S_ISLNK(inode->i_mode) && 54 inode->i_blocks - ea_blocks == 0); 55 } 56 57 static void ext2_truncate_blocks(struct inode *inode, loff_t offset); 58 59 void ext2_write_failed(struct address_space *mapping, loff_t to) 60 { 61 struct inode *inode = mapping->host; 62 63 if (to > inode->i_size) { 64 truncate_pagecache(inode, inode->i_size); 65 ext2_truncate_blocks(inode, inode->i_size); 66 } 67 } 68 69 /* 70 * Called at the last iput() if i_nlink is zero. 71 */ 72 void ext2_evict_inode(struct inode * inode) 73 { 74 struct ext2_block_alloc_info *rsv; 75 int want_delete = 0; 76 77 if (!inode->i_nlink && !is_bad_inode(inode)) { 78 want_delete = 1; 79 dquot_initialize(inode); 80 } else { 81 dquot_drop(inode); 82 } 83 84 truncate_inode_pages_final(&inode->i_data); 85 86 if (want_delete) { 87 sb_start_intwrite(inode->i_sb); 88 /* set dtime */ 89 EXT2_I(inode)->i_dtime = ktime_get_real_seconds(); 90 mark_inode_dirty(inode); 91 __ext2_write_inode(inode, inode_needs_sync(inode)); 92 /* truncate to 0 */ 93 inode->i_size = 0; 94 if (inode->i_blocks) 95 ext2_truncate_blocks(inode, 0); 96 ext2_xattr_delete_inode(inode); 97 } else { 98 mmb_sync(&EXT2_I(inode)->i_metadata_bhs); 99 } 100 mmb_invalidate(&EXT2_I(inode)->i_metadata_bhs); 101 clear_inode(inode); 102 103 ext2_discard_reservation(inode); 104 rsv = EXT2_I(inode)->i_block_alloc_info; 105 EXT2_I(inode)->i_block_alloc_info = NULL; 106 if (unlikely(rsv)) 107 kfree(rsv); 108 109 if (want_delete) { 110 ext2_free_inode(inode); 111 sb_end_intwrite(inode->i_sb); 112 } 113 } 114 115 typedef struct { 116 __le32 *p; 117 __le32 key; 118 struct buffer_head *bh; 119 } Indirect; 120 121 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) 122 { 123 p->key = *(p->p = v); 124 p->bh = bh; 125 } 126 127 static inline int verify_chain(Indirect *from, Indirect *to) 128 { 129 while (from <= to && from->key == *from->p) 130 from++; 131 return (from > to); 132 } 133 134 /** 135 * ext2_block_to_path - parse the block number into array of offsets 136 * @inode: inode in question (we are only interested in its superblock) 137 * @i_block: block number to be parsed 138 * @offsets: array to store the offsets in 139 * @boundary: set this non-zero if the referred-to block is likely to be 140 * followed (on disk) by an indirect block. 141 * To store the locations of file's data ext2 uses a data structure common 142 * for UNIX filesystems - tree of pointers anchored in the inode, with 143 * data blocks at leaves and indirect blocks in intermediate nodes. 144 * This function translates the block number into path in that tree - 145 * return value is the path length and @offsets[n] is the offset of 146 * pointer to (n+1)th node in the nth one. If @block is out of range 147 * (negative or too large) warning is printed and zero returned. 148 * 149 * Note: function doesn't find node addresses, so no IO is needed. All 150 * we need to know is the capacity of indirect blocks (taken from the 151 * inode->i_sb). 152 */ 153 154 /* 155 * Portability note: the last comparison (check that we fit into triple 156 * indirect block) is spelled differently, because otherwise on an 157 * architecture with 32-bit longs and 8Kb pages we might get into trouble 158 * if our filesystem had 8Kb blocks. We might use long long, but that would 159 * kill us on x86. Oh, well, at least the sign propagation does not matter - 160 * i_block would have to be negative in the very beginning, so we would not 161 * get there at all. 162 */ 163 164 static int ext2_block_to_path(struct inode *inode, 165 long i_block, int offsets[4], int *boundary) 166 { 167 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb); 168 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb); 169 const long direct_blocks = EXT2_NDIR_BLOCKS, 170 indirect_blocks = ptrs, 171 double_blocks = (1 << (ptrs_bits * 2)); 172 int n = 0; 173 int final = 0; 174 175 if (i_block < 0) { 176 ext2_msg(inode->i_sb, KERN_WARNING, 177 "warning: %s: block < 0", __func__); 178 } else if (i_block < direct_blocks) { 179 offsets[n++] = i_block; 180 final = direct_blocks; 181 } else if ( (i_block -= direct_blocks) < indirect_blocks) { 182 offsets[n++] = EXT2_IND_BLOCK; 183 offsets[n++] = i_block; 184 final = ptrs; 185 } else if ((i_block -= indirect_blocks) < double_blocks) { 186 offsets[n++] = EXT2_DIND_BLOCK; 187 offsets[n++] = i_block >> ptrs_bits; 188 offsets[n++] = i_block & (ptrs - 1); 189 final = ptrs; 190 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { 191 offsets[n++] = EXT2_TIND_BLOCK; 192 offsets[n++] = i_block >> (ptrs_bits * 2); 193 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); 194 offsets[n++] = i_block & (ptrs - 1); 195 final = ptrs; 196 } else { 197 ext2_msg(inode->i_sb, KERN_WARNING, 198 "warning: %s: block is too big", __func__); 199 } 200 if (boundary) 201 *boundary = final - 1 - (i_block & (ptrs - 1)); 202 203 return n; 204 } 205 206 /** 207 * ext2_get_branch - read the chain of indirect blocks leading to data 208 * @inode: inode in question 209 * @depth: depth of the chain (1 - direct pointer, etc.) 210 * @offsets: offsets of pointers in inode/indirect blocks 211 * @chain: place to store the result 212 * @err: here we store the error value 213 * 214 * Function fills the array of triples <key, p, bh> and returns %NULL 215 * if everything went OK or the pointer to the last filled triple 216 * (incomplete one) otherwise. Upon the return chain[i].key contains 217 * the number of (i+1)-th block in the chain (as it is stored in memory, 218 * i.e. little-endian 32-bit), chain[i].p contains the address of that 219 * number (it points into struct inode for i==0 and into the bh->b_data 220 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect 221 * block for i>0 and NULL for i==0. In other words, it holds the block 222 * numbers of the chain, addresses they were taken from (and where we can 223 * verify that chain did not change) and buffer_heads hosting these 224 * numbers. 225 * 226 * Function stops when it stumbles upon zero pointer (absent block) 227 * (pointer to last triple returned, *@err == 0) 228 * or when it gets an IO error reading an indirect block 229 * (ditto, *@err == -EIO) 230 * or when it notices that chain had been changed while it was reading 231 * (ditto, *@err == -EAGAIN) 232 * or when it reads all @depth-1 indirect blocks successfully and finds 233 * the whole chain, all way to the data (returns %NULL, *err == 0). 234 */ 235 static Indirect *ext2_get_branch(struct inode *inode, 236 int depth, 237 int *offsets, 238 Indirect chain[4], 239 int *err) 240 { 241 struct super_block *sb = inode->i_sb; 242 Indirect *p = chain; 243 struct buffer_head *bh; 244 245 *err = 0; 246 /* i_data is not going away, no lock needed */ 247 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets); 248 if (!p->key) 249 goto no_block; 250 while (--depth) { 251 bh = sb_bread(sb, le32_to_cpu(p->key)); 252 if (!bh) 253 goto failure; 254 read_lock(&EXT2_I(inode)->i_meta_lock); 255 if (!verify_chain(chain, p)) 256 goto changed; 257 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); 258 read_unlock(&EXT2_I(inode)->i_meta_lock); 259 if (!p->key) 260 goto no_block; 261 } 262 return NULL; 263 264 changed: 265 read_unlock(&EXT2_I(inode)->i_meta_lock); 266 brelse(bh); 267 *err = -EAGAIN; 268 goto no_block; 269 failure: 270 *err = -EIO; 271 no_block: 272 return p; 273 } 274 275 /** 276 * ext2_find_near - find a place for allocation with sufficient locality 277 * @inode: owner 278 * @ind: descriptor of indirect block. 279 * 280 * This function returns the preferred place for block allocation. 281 * It is used when heuristic for sequential allocation fails. 282 * Rules are: 283 * + if there is a block to the left of our position - allocate near it. 284 * + if pointer will live in indirect block - allocate near that block. 285 * + if pointer will live in inode - allocate in the same cylinder group. 286 * 287 * In the latter case we colour the starting block by the callers PID to 288 * prevent it from clashing with concurrent allocations for a different inode 289 * in the same block group. The PID is used here so that functionally related 290 * files will be close-by on-disk. 291 * 292 * Caller must make sure that @ind is valid and will stay that way. 293 */ 294 295 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind) 296 { 297 struct ext2_inode_info *ei = EXT2_I(inode); 298 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; 299 __le32 *p; 300 ext2_fsblk_t bg_start; 301 ext2_fsblk_t colour; 302 303 /* Try to find previous block */ 304 for (p = ind->p - 1; p >= start; p--) 305 if (*p) 306 return le32_to_cpu(*p); 307 308 /* No such thing, so let's try location of indirect block */ 309 if (ind->bh) 310 return ind->bh->b_blocknr; 311 312 /* 313 * It is going to be referred from inode itself? OK, just put it into 314 * the same cylinder group then. 315 */ 316 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group); 317 colour = (current->pid % 16) * 318 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16); 319 return bg_start + colour; 320 } 321 322 /** 323 * ext2_find_goal - find a preferred place for allocation. 324 * @inode: owner 325 * @block: block we want 326 * @partial: pointer to the last triple within a chain 327 * 328 * Returns preferred place for a block (the goal). 329 */ 330 331 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block, 332 Indirect *partial) 333 { 334 struct ext2_block_alloc_info *block_i; 335 336 block_i = EXT2_I(inode)->i_block_alloc_info; 337 338 /* 339 * try the heuristic for sequential allocation, 340 * failing that at least try to get decent locality. 341 */ 342 if (block_i && (block == block_i->last_alloc_logical_block + 1) 343 && (block_i->last_alloc_physical_block != 0)) { 344 return block_i->last_alloc_physical_block + 1; 345 } 346 347 return ext2_find_near(inode, partial); 348 } 349 350 /** 351 * ext2_blks_to_allocate: Look up the block map and count the number 352 * of direct blocks need to be allocated for the given branch. 353 * 354 * @branch: chain of indirect blocks 355 * @k: number of blocks need for indirect blocks 356 * @blks: number of data blocks to be mapped. 357 * @blocks_to_boundary: the offset in the indirect block 358 * 359 * return the number of direct blocks to allocate. 360 */ 361 static int 362 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks, 363 int blocks_to_boundary) 364 { 365 unsigned long count = 0; 366 367 /* 368 * Simple case, [t,d]Indirect block(s) has not allocated yet 369 * then it's clear blocks on that path have not allocated 370 */ 371 if (k > 0) { 372 /* right now don't hanel cross boundary allocation */ 373 if (blks < blocks_to_boundary + 1) 374 count += blks; 375 else 376 count += blocks_to_boundary + 1; 377 return count; 378 } 379 380 count++; 381 while (count < blks && count <= blocks_to_boundary 382 && le32_to_cpu(*(branch[0].p + count)) == 0) { 383 count++; 384 } 385 return count; 386 } 387 388 /** 389 * ext2_alloc_blocks: Allocate multiple blocks needed for a branch. 390 * @inode: Owner. 391 * @goal: Preferred place for allocation. 392 * @indirect_blks: The number of blocks needed to allocate for indirect blocks. 393 * @blks: The number of blocks need to allocate for direct blocks. 394 * @new_blocks: On return it will store the new block numbers for 395 * the indirect blocks(if needed) and the first direct block. 396 * @err: Error pointer. 397 * 398 * Return: Number of blocks allocated. 399 */ 400 static int ext2_alloc_blocks(struct inode *inode, 401 ext2_fsblk_t goal, int indirect_blks, int blks, 402 ext2_fsblk_t new_blocks[4], int *err) 403 { 404 int target, i; 405 unsigned long count = 0; 406 int index = 0; 407 ext2_fsblk_t current_block = 0; 408 int ret = 0; 409 410 /* 411 * Here we try to allocate the requested multiple blocks at once, 412 * on a best-effort basis. 413 * To build a branch, we should allocate blocks for 414 * the indirect blocks(if not allocated yet), and at least 415 * the first direct block of this branch. That's the 416 * minimum number of blocks need to allocate(required) 417 */ 418 target = blks + indirect_blks; 419 420 while (1) { 421 count = target; 422 /* allocating blocks for indirect blocks and direct blocks */ 423 current_block = ext2_new_blocks(inode, goal, &count, err, 0); 424 if (*err) 425 goto failed_out; 426 427 target -= count; 428 /* allocate blocks for indirect blocks */ 429 while (index < indirect_blks && count) { 430 new_blocks[index++] = current_block++; 431 count--; 432 } 433 434 if (count > 0) 435 break; 436 } 437 438 /* save the new block number for the first direct block */ 439 new_blocks[index] = current_block; 440 441 /* total number of blocks allocated for direct blocks */ 442 ret = count; 443 *err = 0; 444 return ret; 445 failed_out: 446 for (i = 0; i <index; i++) 447 ext2_free_blocks(inode, new_blocks[i], 1); 448 if (index) 449 mark_inode_dirty(inode); 450 return ret; 451 } 452 453 /** 454 * ext2_alloc_branch - allocate and set up a chain of blocks. 455 * @inode: owner 456 * @indirect_blks: depth of the chain (number of blocks to allocate) 457 * @blks: number of allocated direct blocks 458 * @goal: preferred place for allocation 459 * @offsets: offsets (in the blocks) to store the pointers to next. 460 * @branch: place to store the chain in. 461 * 462 * This function allocates @num blocks, zeroes out all but the last one, 463 * links them into chain and (if we are synchronous) writes them to disk. 464 * In other words, it prepares a branch that can be spliced onto the 465 * inode. It stores the information about that chain in the branch[], in 466 * the same format as ext2_get_branch() would do. We are calling it after 467 * we had read the existing part of chain and partial points to the last 468 * triple of that (one with zero ->key). Upon the exit we have the same 469 * picture as after the successful ext2_get_block(), except that in one 470 * place chain is disconnected - *branch->p is still zero (we did not 471 * set the last link), but branch->key contains the number that should 472 * be placed into *branch->p to fill that gap. 473 * 474 * If allocation fails we free all blocks we've allocated (and forget 475 * their buffer_heads) and return the error value the from failed 476 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain 477 * as described above and return 0. 478 */ 479 480 static int ext2_alloc_branch(struct inode *inode, 481 int indirect_blks, int *blks, ext2_fsblk_t goal, 482 int *offsets, Indirect *branch) 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 { 565 int i; 566 struct ext2_block_alloc_info *block_i; 567 ext2_fsblk_t current_block; 568 569 block_i = EXT2_I(inode)->i_block_alloc_info; 570 571 /* XXX LOCKING probably should have i_meta_lock ?*/ 572 /* That's it */ 573 574 *where->p = where->key; 575 576 /* 577 * Update the host buffer_head or inode to point to more just allocated 578 * direct blocks blocks 579 */ 580 if (num == 0 && blks > 1) { 581 current_block = le32_to_cpu(where->key) + 1; 582 for (i = 1; i < blks; i++) 583 *(where->p + i ) = cpu_to_le32(current_block++); 584 } 585 586 /* 587 * update the most recently allocated logical & physical block 588 * in i_block_alloc_info, to assist find the proper goal block for next 589 * allocation 590 */ 591 if (block_i) { 592 block_i->last_alloc_logical_block = block + blks - 1; 593 block_i->last_alloc_physical_block = 594 le32_to_cpu(where[num].key) + blks - 1; 595 } 596 597 /* We are done with atomic stuff, now do the rest of housekeeping */ 598 599 /* had we spliced it onto indirect block? */ 600 if (where->bh) 601 mmb_mark_buffer_dirty(where->bh, &EXT2_I(inode)->i_metadata_bhs); 602 603 inode_set_ctime_current(inode); 604 mark_inode_dirty(inode); 605 } 606 607 /* 608 * Allocation strategy is simple: if we have to allocate something, we will 609 * have to go the whole way to leaf. So let's do it before attaching anything 610 * to tree, set linkage between the newborn blocks, write them if sync is 611 * required, recheck the path, free and repeat if check fails, otherwise 612 * set the last missing link (that will protect us from any truncate-generated 613 * removals - all blocks on the path are immune now) and possibly force the 614 * write on the parent block. 615 * That has a nice additional property: no special recovery from the failed 616 * allocations is needed - we simply release blocks and do not touch anything 617 * reachable from inode. 618 * 619 * `handle' can be NULL if create == 0. 620 * 621 * return > 0, # of blocks mapped or allocated. 622 * return = 0, if plain lookup failed. 623 * return < 0, error case. 624 */ 625 static int ext2_get_blocks(struct inode *inode, 626 sector_t iblock, unsigned long maxblocks, 627 u32 *bno, bool *new, bool *boundary, 628 int create) 629 { 630 int err; 631 int offsets[4]; 632 Indirect chain[4]; 633 Indirect *partial; 634 ext2_fsblk_t goal; 635 int indirect_blks; 636 int blocks_to_boundary = 0; 637 int depth; 638 struct ext2_inode_info *ei = EXT2_I(inode); 639 int count = 0; 640 ext2_fsblk_t first_block = 0; 641 642 if (WARN_ON_ONCE(maxblocks == 0)) 643 return -EINVAL; 644 645 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary); 646 647 if (depth == 0) 648 return -EIO; 649 650 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 651 /* Simplest case - block found, no allocation needed */ 652 if (!partial) { 653 first_block = le32_to_cpu(chain[depth - 1].key); 654 count++; 655 /*map more blocks*/ 656 while (count < maxblocks && count <= blocks_to_boundary) { 657 ext2_fsblk_t blk; 658 659 if (!verify_chain(chain, chain + depth - 1)) { 660 /* 661 * Indirect block might be removed by 662 * truncate while we were reading it. 663 * Handling of that case: forget what we've 664 * got now, go to reread. 665 */ 666 err = -EAGAIN; 667 count = 0; 668 partial = chain + depth - 1; 669 break; 670 } 671 blk = le32_to_cpu(*(chain[depth-1].p + count)); 672 if (blk == first_block + count) 673 count++; 674 else 675 break; 676 } 677 if (err != -EAGAIN) 678 goto got_it; 679 } 680 681 /* Next simple case - plain lookup or failed read of indirect block */ 682 if (!create || err == -EIO) 683 goto cleanup; 684 685 mutex_lock(&ei->truncate_mutex); 686 /* 687 * If the indirect block is missing while we are reading 688 * the chain(ext2_get_branch() returns -EAGAIN err), or 689 * if the chain has been changed after we grab the semaphore, 690 * (either because another process truncated this branch, or 691 * another get_block allocated this branch) re-grab the chain to see if 692 * the request block has been allocated or not. 693 * 694 * Since we already block the truncate/other get_block 695 * at this point, we will have the current copy of the chain when we 696 * splice the branch into the tree. 697 */ 698 if (err == -EAGAIN || !verify_chain(chain, partial)) { 699 while (partial > chain) { 700 brelse(partial->bh); 701 partial--; 702 } 703 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 704 if (!partial) { 705 count++; 706 mutex_unlock(&ei->truncate_mutex); 707 goto got_it; 708 } 709 710 if (err) { 711 mutex_unlock(&ei->truncate_mutex); 712 goto cleanup; 713 } 714 } 715 716 /* 717 * Okay, we need to do block allocation. Lazily initialize the block 718 * allocation info here if necessary 719 */ 720 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) 721 ext2_init_block_alloc_info(inode); 722 723 goal = ext2_find_goal(inode, iblock, partial); 724 725 /* the number of blocks need to allocate for [d,t]indirect blocks */ 726 indirect_blks = (chain + depth) - partial - 1; 727 /* 728 * Next look up the indirect map to count the total number of 729 * direct blocks to allocate for this branch. 730 */ 731 count = ext2_blks_to_allocate(partial, indirect_blks, 732 maxblocks, blocks_to_boundary); 733 /* 734 * XXX ???? Block out ext2_truncate while we alter the tree 735 */ 736 err = ext2_alloc_branch(inode, indirect_blks, &count, goal, 737 offsets + (partial - chain), partial); 738 739 if (err) { 740 mutex_unlock(&ei->truncate_mutex); 741 goto cleanup; 742 } 743 744 if (IS_DAX(inode)) { 745 /* 746 * We must unmap blocks before zeroing so that writeback cannot 747 * overwrite zeros with stale data from block device page cache. 748 */ 749 clean_bdev_aliases(inode->i_sb->s_bdev, 750 le32_to_cpu(chain[depth-1].key), 751 count); 752 /* 753 * block must be initialised before we put it in the tree 754 * so that it's not found by another thread before it's 755 * initialised 756 */ 757 err = sb_issue_zeroout(inode->i_sb, 758 le32_to_cpu(chain[depth-1].key), count, 759 GFP_KERNEL); 760 if (err) { 761 mutex_unlock(&ei->truncate_mutex); 762 goto cleanup; 763 } 764 } 765 *new = true; 766 767 ext2_splice_branch(inode, iblock, partial, indirect_blks, count); 768 mutex_unlock(&ei->truncate_mutex); 769 got_it: 770 if (count > blocks_to_boundary) 771 *boundary = true; 772 err = count; 773 /* Clean up and exit */ 774 partial = chain + depth - 1; /* the whole chain */ 775 cleanup: 776 while (partial > chain) { 777 brelse(partial->bh); 778 partial--; 779 } 780 if (err > 0) 781 *bno = le32_to_cpu(chain[depth-1].key); 782 return err; 783 } 784 785 int ext2_get_block(struct inode *inode, sector_t iblock, 786 struct buffer_head *bh_result, int create) 787 { 788 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; 789 bool new = false, boundary = false; 790 u32 bno; 791 int ret; 792 793 ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary, 794 create); 795 if (ret <= 0) 796 return ret; 797 798 map_bh(bh_result, inode->i_sb, bno); 799 bh_result->b_size = (ret << inode->i_blkbits); 800 if (new) 801 set_buffer_new(bh_result); 802 if (boundary) 803 set_buffer_boundary(bh_result); 804 return 0; 805 806 } 807 808 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 809 unsigned flags, struct iomap *iomap, struct iomap *srcmap) 810 { 811 unsigned int blkbits = inode->i_blkbits; 812 unsigned long first_block = offset >> blkbits; 813 unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits; 814 struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb); 815 bool new = false, boundary = false; 816 u32 bno; 817 int ret; 818 bool create = flags & IOMAP_WRITE; 819 820 /* 821 * For writes that could fill holes inside i_size on a 822 * DIO_SKIP_HOLES filesystem we forbid block creations: only 823 * overwrites are permitted. 824 */ 825 if ((flags & IOMAP_DIRECT) && 826 (first_block << blkbits) < i_size_read(inode)) 827 create = 0; 828 829 /* 830 * Writes that span EOF might trigger an IO size update on completion, 831 * so consider them to be dirty for the purposes of O_DSYNC even if 832 * there is no other metadata changes pending or have been made here. 833 */ 834 if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode)) 835 iomap->flags |= IOMAP_F_DIRTY; 836 837 ret = ext2_get_blocks(inode, first_block, max_blocks, 838 &bno, &new, &boundary, create); 839 if (ret < 0) 840 return ret; 841 842 iomap->flags = 0; 843 iomap->offset = (u64)first_block << blkbits; 844 if (flags & IOMAP_DAX) 845 iomap->dax_dev = sbi->s_daxdev; 846 else 847 iomap->bdev = inode->i_sb->s_bdev; 848 849 if (ret == 0) { 850 /* 851 * Switch to buffered-io for writing to holes in a non-extent 852 * based filesystem to avoid stale data exposure problem. 853 */ 854 if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT)) 855 return -ENOTBLK; 856 iomap->type = IOMAP_HOLE; 857 iomap->addr = IOMAP_NULL_ADDR; 858 iomap->length = 1 << blkbits; 859 } else { 860 iomap->type = IOMAP_MAPPED; 861 iomap->addr = (u64)bno << blkbits; 862 if (flags & IOMAP_DAX) 863 iomap->addr += sbi->s_dax_part_off; 864 iomap->length = (u64)ret << blkbits; 865 iomap->flags |= IOMAP_F_MERGED; 866 } 867 868 if (new) 869 iomap->flags |= IOMAP_F_NEW; 870 return 0; 871 } 872 873 static int 874 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length, 875 ssize_t written, unsigned flags, struct iomap *iomap) 876 { 877 /* 878 * Switch to buffered-io in case of any error. 879 * Blocks allocated can be used by the buffered-io path. 880 */ 881 if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0) 882 return -ENOTBLK; 883 884 if (iomap->type == IOMAP_MAPPED && 885 written < length && 886 (flags & IOMAP_WRITE)) 887 ext2_write_failed(inode->i_mapping, offset + length); 888 return 0; 889 } 890 891 const struct iomap_ops ext2_iomap_ops = { 892 .iomap_begin = ext2_iomap_begin, 893 .iomap_end = ext2_iomap_end, 894 }; 895 896 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 897 u64 start, u64 len) 898 { 899 int ret; 900 loff_t i_size; 901 902 inode_lock(inode); 903 i_size = i_size_read(inode); 904 /* 905 * iomap_fiemap() returns EINVAL for 0 length. Make sure we don't trim 906 * length to 0 but still trim the range as much as possible since 907 * ext2_get_blocks() iterates unmapped space block by block which is 908 * slow. 909 */ 910 if (i_size == 0) 911 i_size = 1; 912 len = min_t(u64, len, i_size); 913 ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops); 914 inode_unlock(inode); 915 916 return ret; 917 } 918 919 static int ext2_read_folio(struct file *file, struct folio *folio) 920 { 921 return mpage_read_folio(folio, ext2_get_block); 922 } 923 924 static void ext2_readahead(struct readahead_control *rac) 925 { 926 mpage_readahead(rac, ext2_get_block); 927 } 928 929 static int 930 ext2_write_begin(const struct kiocb *iocb, struct address_space *mapping, 931 loff_t pos, unsigned len, struct folio **foliop, void **fsdata) 932 { 933 int ret; 934 935 ret = block_write_begin(mapping, pos, len, foliop, ext2_get_block); 936 if (ret < 0) 937 ext2_write_failed(mapping, pos + len); 938 return ret; 939 } 940 941 static int ext2_write_end(const struct kiocb *iocb, 942 struct address_space *mapping, 943 loff_t pos, unsigned len, unsigned copied, 944 struct folio *folio, void *fsdata) 945 { 946 int ret; 947 948 ret = generic_write_end(iocb, mapping, pos, len, copied, folio, fsdata); 949 if (ret < len) 950 ext2_write_failed(mapping, pos + len); 951 return ret; 952 } 953 954 static sector_t ext2_bmap(struct address_space *mapping, sector_t block) 955 { 956 return generic_block_bmap(mapping,block,ext2_get_block); 957 } 958 959 static int 960 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc) 961 { 962 return mpage_writepages(mapping, wbc, ext2_get_block); 963 } 964 965 static int 966 ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc) 967 { 968 struct ext2_sb_info *sbi = EXT2_SB(mapping->host->i_sb); 969 970 return dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); 971 } 972 973 const struct address_space_operations ext2_aops = { 974 .dirty_folio = block_dirty_folio, 975 .invalidate_folio = block_invalidate_folio, 976 .read_folio = ext2_read_folio, 977 .readahead = ext2_readahead, 978 .write_begin = ext2_write_begin, 979 .write_end = ext2_write_end, 980 .bmap = ext2_bmap, 981 .writepages = ext2_writepages, 982 .migrate_folio = buffer_migrate_folio, 983 .is_partially_uptodate = block_is_partially_uptodate, 984 .error_remove_folio = generic_error_remove_folio, 985 }; 986 987 static const struct address_space_operations ext2_dax_aops = { 988 .writepages = ext2_dax_writepages, 989 .dirty_folio = noop_dirty_folio, 990 }; 991 992 /* 993 * Probably it should be a library function... search for first non-zero word 994 * or memcmp with zero_page, whatever is better for particular architecture. 995 * Linus? 996 */ 997 static inline int all_zeroes(__le32 *p, __le32 *q) 998 { 999 while (p < q) 1000 if (*p++) 1001 return 0; 1002 return 1; 1003 } 1004 1005 /** 1006 * ext2_find_shared - find the indirect blocks for partial truncation. 1007 * @inode: inode in question 1008 * @depth: depth of the affected branch 1009 * @offsets: offsets of pointers in that branch (see ext2_block_to_path) 1010 * @chain: place to store the pointers to partial indirect blocks 1011 * @top: place to the (detached) top of branch 1012 * 1013 * This is a helper function used by ext2_truncate(). 1014 * 1015 * When we do truncate() we may have to clean the ends of several indirect 1016 * blocks but leave the blocks themselves alive. Block is partially 1017 * truncated if some data below the new i_size is referred from it (and 1018 * it is on the path to the first completely truncated data block, indeed). 1019 * We have to free the top of that path along with everything to the right 1020 * of the path. Since no allocation past the truncation point is possible 1021 * until ext2_truncate() finishes, we may safely do the latter, but top 1022 * of branch may require special attention - pageout below the truncation 1023 * point might try to populate it. 1024 * 1025 * We atomically detach the top of branch from the tree, store the block 1026 * number of its root in *@top, pointers to buffer_heads of partially 1027 * truncated blocks - in @chain[].bh and pointers to their last elements 1028 * that should not be removed - in @chain[].p. Return value is the pointer 1029 * to last filled element of @chain. 1030 * 1031 * The work left to caller to do the actual freeing of subtrees: 1032 * a) free the subtree starting from *@top 1033 * b) free the subtrees whose roots are stored in 1034 * (@chain[i].p+1 .. end of @chain[i].bh->b_data) 1035 * c) free the subtrees growing from the inode past the @chain[0].p 1036 * (no partially truncated stuff there). 1037 */ 1038 1039 static Indirect *ext2_find_shared(struct inode *inode, 1040 int depth, 1041 int offsets[4], 1042 Indirect chain[4], 1043 __le32 *top) 1044 { 1045 Indirect *partial, *p; 1046 int k, err; 1047 1048 *top = 0; 1049 for (k = depth; k > 1 && !offsets[k-1]; k--) 1050 ; 1051 partial = ext2_get_branch(inode, k, offsets, chain, &err); 1052 if (!partial) 1053 partial = chain + k-1; 1054 /* 1055 * If the branch acquired continuation since we've looked at it - 1056 * fine, it should all survive and (new) top doesn't belong to us. 1057 */ 1058 write_lock(&EXT2_I(inode)->i_meta_lock); 1059 if (!partial->key && *partial->p) { 1060 write_unlock(&EXT2_I(inode)->i_meta_lock); 1061 goto no_top; 1062 } 1063 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) 1064 ; 1065 /* 1066 * OK, we've found the last block that must survive. The rest of our 1067 * branch should be detached before unlocking. However, if that rest 1068 * of branch is all ours and does not grow immediately from the inode 1069 * it's easier to cheat and just decrement partial->p. 1070 */ 1071 if (p == chain + k - 1 && p > chain) { 1072 p->p--; 1073 } else { 1074 *top = *p->p; 1075 *p->p = 0; 1076 } 1077 write_unlock(&EXT2_I(inode)->i_meta_lock); 1078 1079 while(partial > p) 1080 { 1081 brelse(partial->bh); 1082 partial--; 1083 } 1084 no_top: 1085 return partial; 1086 } 1087 1088 /** 1089 * ext2_free_data - free a list of data blocks 1090 * @inode: inode we are dealing with 1091 * @p: array of block numbers 1092 * @q: points immediately past the end of array 1093 * 1094 * We are freeing all blocks referred from that array (numbers are 1095 * stored as little-endian 32-bit) and updating @inode->i_blocks 1096 * appropriately. 1097 */ 1098 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q) 1099 { 1100 ext2_fsblk_t block_to_free = 0, count = 0; 1101 ext2_fsblk_t nr; 1102 1103 for ( ; p < q ; p++) { 1104 nr = le32_to_cpu(*p); 1105 if (nr) { 1106 *p = 0; 1107 /* accumulate blocks to free if they're contiguous */ 1108 if (count == 0) 1109 goto free_this; 1110 else if (block_to_free == nr - count) 1111 count++; 1112 else { 1113 ext2_free_blocks (inode, block_to_free, count); 1114 mark_inode_dirty(inode); 1115 free_this: 1116 block_to_free = nr; 1117 count = 1; 1118 } 1119 } 1120 } 1121 if (count > 0) { 1122 ext2_free_blocks (inode, block_to_free, count); 1123 mark_inode_dirty(inode); 1124 } 1125 } 1126 1127 /** 1128 * ext2_free_branches - free an array of branches 1129 * @inode: inode we are dealing with 1130 * @p: array of block numbers 1131 * @q: pointer immediately past the end of array 1132 * @depth: depth of the branches to free 1133 * 1134 * We are freeing all blocks referred from these branches (numbers are 1135 * stored as little-endian 32-bit) and updating @inode->i_blocks 1136 * appropriately. 1137 */ 1138 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth) 1139 { 1140 struct buffer_head * bh; 1141 ext2_fsblk_t nr; 1142 1143 if (depth--) { 1144 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 1145 for ( ; p < q ; p++) { 1146 nr = le32_to_cpu(*p); 1147 if (!nr) 1148 continue; 1149 *p = 0; 1150 bh = sb_bread(inode->i_sb, nr); 1151 /* 1152 * A read failure? Report error and clear slot 1153 * (should be rare). 1154 */ 1155 if (!bh) { 1156 ext2_error(inode->i_sb, "ext2_free_branches", 1157 "Read failure, inode=%llu, block=%ld", 1158 inode->i_ino, nr); 1159 continue; 1160 } 1161 ext2_free_branches(inode, 1162 (__le32*)bh->b_data, 1163 (__le32*)bh->b_data + addr_per_block, 1164 depth); 1165 bforget(bh); 1166 ext2_free_blocks(inode, nr, 1); 1167 mark_inode_dirty(inode); 1168 } 1169 } else 1170 ext2_free_data(inode, p, q); 1171 } 1172 1173 /* mapping->invalidate_lock must be held when calling this function */ 1174 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset) 1175 { 1176 __le32 *i_data = EXT2_I(inode)->i_data; 1177 struct ext2_inode_info *ei = EXT2_I(inode); 1178 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 1179 int offsets[4]; 1180 Indirect chain[4]; 1181 Indirect *partial; 1182 __le32 nr = 0; 1183 int n; 1184 long iblock; 1185 unsigned blocksize; 1186 blocksize = inode->i_sb->s_blocksize; 1187 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb); 1188 1189 #ifdef CONFIG_FS_DAX 1190 WARN_ON(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)); 1191 #endif 1192 1193 n = ext2_block_to_path(inode, iblock, offsets, NULL); 1194 if (n == 0) 1195 return; 1196 1197 /* 1198 * From here we block out all ext2_get_block() callers who want to 1199 * modify the block allocation tree. 1200 */ 1201 mutex_lock(&ei->truncate_mutex); 1202 1203 if (n == 1) { 1204 ext2_free_data(inode, i_data+offsets[0], 1205 i_data + EXT2_NDIR_BLOCKS); 1206 goto do_indirects; 1207 } 1208 1209 partial = ext2_find_shared(inode, n, offsets, chain, &nr); 1210 /* Kill the top of shared branch (already detached) */ 1211 if (nr) { 1212 if (partial == chain) 1213 mark_inode_dirty(inode); 1214 else 1215 mmb_mark_buffer_dirty(partial->bh, 1216 &EXT2_I(inode)->i_metadata_bhs); 1217 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial); 1218 } 1219 /* Clear the ends of indirect blocks on the shared branch */ 1220 while (partial > chain) { 1221 ext2_free_branches(inode, 1222 partial->p + 1, 1223 (__le32*)partial->bh->b_data+addr_per_block, 1224 (chain+n-1) - partial); 1225 mmb_mark_buffer_dirty(partial->bh, 1226 &EXT2_I(inode)->i_metadata_bhs); 1227 brelse (partial->bh); 1228 partial--; 1229 } 1230 do_indirects: 1231 /* Kill the remaining (whole) subtrees */ 1232 switch (offsets[0]) { 1233 default: 1234 nr = i_data[EXT2_IND_BLOCK]; 1235 if (nr) { 1236 i_data[EXT2_IND_BLOCK] = 0; 1237 mark_inode_dirty(inode); 1238 ext2_free_branches(inode, &nr, &nr+1, 1); 1239 } 1240 fallthrough; 1241 case EXT2_IND_BLOCK: 1242 nr = i_data[EXT2_DIND_BLOCK]; 1243 if (nr) { 1244 i_data[EXT2_DIND_BLOCK] = 0; 1245 mark_inode_dirty(inode); 1246 ext2_free_branches(inode, &nr, &nr+1, 2); 1247 } 1248 fallthrough; 1249 case EXT2_DIND_BLOCK: 1250 nr = i_data[EXT2_TIND_BLOCK]; 1251 if (nr) { 1252 i_data[EXT2_TIND_BLOCK] = 0; 1253 mark_inode_dirty(inode); 1254 ext2_free_branches(inode, &nr, &nr+1, 3); 1255 } 1256 break; 1257 case EXT2_TIND_BLOCK: 1258 ; 1259 } 1260 1261 ext2_discard_reservation(inode); 1262 1263 mutex_unlock(&ei->truncate_mutex); 1264 } 1265 1266 static void ext2_truncate_blocks(struct inode *inode, loff_t offset) 1267 { 1268 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1269 S_ISLNK(inode->i_mode))) 1270 return; 1271 if (ext2_inode_is_fast_symlink(inode)) 1272 return; 1273 1274 filemap_invalidate_lock(inode->i_mapping); 1275 __ext2_truncate_blocks(inode, offset); 1276 filemap_invalidate_unlock(inode->i_mapping); 1277 } 1278 1279 static int ext2_setsize(struct inode *inode, loff_t newsize) 1280 { 1281 int error; 1282 1283 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1284 S_ISLNK(inode->i_mode))) 1285 return -EINVAL; 1286 if (ext2_inode_is_fast_symlink(inode)) 1287 return -EINVAL; 1288 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 1289 return -EPERM; 1290 1291 inode_dio_wait(inode); 1292 1293 if (IS_DAX(inode)) 1294 error = dax_truncate_page(inode, newsize, NULL, 1295 &ext2_iomap_ops); 1296 else 1297 error = block_truncate_page(inode->i_mapping, 1298 newsize, ext2_get_block); 1299 if (error) 1300 return error; 1301 1302 filemap_invalidate_lock(inode->i_mapping); 1303 truncate_setsize(inode, newsize); 1304 __ext2_truncate_blocks(inode, newsize); 1305 filemap_invalidate_unlock(inode->i_mapping); 1306 1307 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 1308 if (inode_needs_sync(inode)) { 1309 mmb_sync(&EXT2_I(inode)->i_metadata_bhs); 1310 sync_inode_metadata(inode, 1); 1311 } else { 1312 mark_inode_dirty(inode); 1313 } 1314 1315 return 0; 1316 } 1317 1318 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino, 1319 struct buffer_head **p) 1320 { 1321 struct buffer_head * bh; 1322 unsigned long block_group; 1323 unsigned long block; 1324 unsigned long offset; 1325 struct ext2_group_desc * gdp; 1326 1327 *p = NULL; 1328 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) || 1329 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count)) 1330 goto Einval; 1331 1332 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb); 1333 gdp = ext2_get_group_desc(sb, block_group, NULL); 1334 if (!gdp) 1335 goto Egdp; 1336 /* 1337 * Figure out the offset within the block group inode table 1338 */ 1339 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb); 1340 block = le32_to_cpu(gdp->bg_inode_table) + 1341 (offset >> EXT2_BLOCK_SIZE_BITS(sb)); 1342 if (!(bh = sb_bread(sb, block))) 1343 goto Eio; 1344 1345 *p = bh; 1346 offset &= (EXT2_BLOCK_SIZE(sb) - 1); 1347 return (struct ext2_inode *) (bh->b_data + offset); 1348 1349 Einval: 1350 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu", 1351 (unsigned long) ino); 1352 return ERR_PTR(-EINVAL); 1353 Eio: 1354 ext2_error(sb, "ext2_get_inode", 1355 "unable to read inode block - inode=%lu, block=%lu", 1356 (unsigned long) ino, block); 1357 Egdp: 1358 return ERR_PTR(-EIO); 1359 } 1360 1361 void ext2_set_inode_flags(struct inode *inode) 1362 { 1363 unsigned int flags = EXT2_I(inode)->i_flags; 1364 1365 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | 1366 S_DIRSYNC | S_DAX); 1367 if (flags & EXT2_SYNC_FL) 1368 inode->i_flags |= S_SYNC; 1369 if (flags & EXT2_APPEND_FL) 1370 inode->i_flags |= S_APPEND; 1371 if (flags & EXT2_IMMUTABLE_FL) 1372 inode->i_flags |= S_IMMUTABLE; 1373 if (flags & EXT2_NOATIME_FL) 1374 inode->i_flags |= S_NOATIME; 1375 if (flags & EXT2_DIRSYNC_FL) 1376 inode->i_flags |= S_DIRSYNC; 1377 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode)) 1378 inode->i_flags |= S_DAX; 1379 } 1380 1381 void ext2_set_file_ops(struct inode *inode) 1382 { 1383 inode->i_op = &ext2_file_inode_operations; 1384 inode->i_fop = &ext2_file_operations; 1385 if (IS_DAX(inode)) 1386 inode->i_mapping->a_ops = &ext2_dax_aops; 1387 else 1388 inode->i_mapping->a_ops = &ext2_aops; 1389 } 1390 1391 struct inode *ext2_iget (struct super_block *sb, unsigned long ino) 1392 { 1393 struct ext2_inode_info *ei; 1394 struct buffer_head * bh = NULL; 1395 struct ext2_inode *raw_inode; 1396 struct inode *inode; 1397 long ret = -EIO; 1398 int n; 1399 uid_t i_uid; 1400 gid_t i_gid; 1401 1402 inode = iget_locked(sb, ino); 1403 if (!inode) 1404 return ERR_PTR(-ENOMEM); 1405 if (!(inode_state_read_once(inode) & I_NEW)) 1406 return inode; 1407 1408 ei = EXT2_I(inode); 1409 ei->i_block_alloc_info = NULL; 1410 1411 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh); 1412 if (IS_ERR(raw_inode)) { 1413 ret = PTR_ERR(raw_inode); 1414 goto bad_inode; 1415 } 1416 1417 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 1418 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 1419 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 1420 if (!(test_opt (inode->i_sb, NO_UID32))) { 1421 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 1422 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 1423 } 1424 i_uid_write(inode, i_uid); 1425 i_gid_write(inode, i_gid); 1426 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); 1427 inode->i_size = le32_to_cpu(raw_inode->i_size); 1428 inode_set_atime(inode, (signed)le32_to_cpu(raw_inode->i_atime), 0); 1429 inode_set_ctime(inode, (signed)le32_to_cpu(raw_inode->i_ctime), 0); 1430 inode_set_mtime(inode, (signed)le32_to_cpu(raw_inode->i_mtime), 0); 1431 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); 1432 /* We now have enough fields to check if the inode was active or not. 1433 * This is needed because nfsd might try to access dead inodes 1434 * the test is that same one that e2fsck uses 1435 * NeilBrown 1999oct15 1436 */ 1437 if (inode->i_nlink == 0) { 1438 if (inode->i_mode == 0 || ei->i_dtime) { 1439 /* this inode is deleted */ 1440 ret = -ESTALE; 1441 } else { 1442 ext2_error(sb, __func__, 1443 "inode %lu has zero i_nlink with mode 0%o and no dtime, " 1444 "filesystem may be corrupt", 1445 ino, inode->i_mode); 1446 ret = -EFSCORRUPTED; 1447 } 1448 goto bad_inode; 1449 } 1450 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); 1451 ei->i_flags = le32_to_cpu(raw_inode->i_flags); 1452 ext2_set_inode_flags(inode); 1453 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); 1454 ei->i_frag_no = raw_inode->i_frag; 1455 ei->i_frag_size = raw_inode->i_fsize; 1456 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); 1457 ei->i_dir_acl = 0; 1458 1459 if (ei->i_file_acl && 1460 !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) { 1461 ext2_error(sb, "ext2_iget", "bad extended attribute block %u", 1462 ei->i_file_acl); 1463 ret = -EFSCORRUPTED; 1464 goto bad_inode; 1465 } 1466 1467 if (S_ISREG(inode->i_mode)) 1468 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; 1469 else 1470 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); 1471 if (i_size_read(inode) < 0) { 1472 ret = -EFSCORRUPTED; 1473 goto bad_inode; 1474 } 1475 ei->i_dtime = 0; 1476 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 1477 ei->i_state = 0; 1478 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 1479 ei->i_dir_start_lookup = 0; 1480 1481 /* 1482 * NOTE! The in-memory inode i_data array is in little-endian order 1483 * even on big-endian machines: we do NOT byteswap the block numbers! 1484 */ 1485 for (n = 0; n < EXT2_N_BLOCKS; n++) 1486 ei->i_data[n] = raw_inode->i_block[n]; 1487 1488 if (S_ISREG(inode->i_mode)) { 1489 ext2_set_file_ops(inode); 1490 } else if (S_ISDIR(inode->i_mode)) { 1491 inode->i_op = &ext2_dir_inode_operations; 1492 inode->i_fop = &ext2_dir_operations; 1493 inode->i_mapping->a_ops = &ext2_aops; 1494 } else if (S_ISLNK(inode->i_mode)) { 1495 if (ext2_inode_is_fast_symlink(inode)) { 1496 inode->i_link = (char *)ei->i_data; 1497 inode->i_op = &ext2_fast_symlink_inode_operations; 1498 nd_terminate_link(ei->i_data, inode->i_size, 1499 sizeof(ei->i_data) - 1); 1500 } else { 1501 inode->i_op = &ext2_symlink_inode_operations; 1502 inode_nohighmem(inode); 1503 inode->i_mapping->a_ops = &ext2_aops; 1504 } 1505 } else { 1506 inode->i_op = &ext2_special_inode_operations; 1507 if (raw_inode->i_block[0]) 1508 init_special_inode(inode, inode->i_mode, 1509 old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); 1510 else 1511 init_special_inode(inode, inode->i_mode, 1512 new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 1513 } 1514 brelse (bh); 1515 unlock_new_inode(inode); 1516 return inode; 1517 1518 bad_inode: 1519 brelse(bh); 1520 iget_failed(inode); 1521 return ERR_PTR(ret); 1522 } 1523 1524 static int __ext2_write_inode(struct inode *inode, int do_sync) 1525 { 1526 struct ext2_inode_info *ei = EXT2_I(inode); 1527 struct super_block *sb = inode->i_sb; 1528 ino_t ino = inode->i_ino; 1529 uid_t uid = i_uid_read(inode); 1530 gid_t gid = i_gid_read(inode); 1531 struct buffer_head * bh; 1532 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh); 1533 int n; 1534 int err = 0; 1535 1536 if (IS_ERR(raw_inode)) 1537 return -EIO; 1538 1539 /* For fields not tracking in the in-memory inode, 1540 * initialise them to zero for new inodes. */ 1541 if (ei->i_state & EXT2_STATE_NEW) 1542 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size); 1543 1544 raw_inode->i_mode = cpu_to_le16(inode->i_mode); 1545 if (!(test_opt(sb, NO_UID32))) { 1546 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid)); 1547 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid)); 1548 /* 1549 * Fix up interoperability with old kernels. Otherwise, old inodes get 1550 * re-used with the upper 16 bits of the uid/gid intact 1551 */ 1552 if (!ei->i_dtime) { 1553 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid)); 1554 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid)); 1555 } else { 1556 raw_inode->i_uid_high = 0; 1557 raw_inode->i_gid_high = 0; 1558 } 1559 } else { 1560 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid)); 1561 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid)); 1562 raw_inode->i_uid_high = 0; 1563 raw_inode->i_gid_high = 0; 1564 } 1565 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 1566 raw_inode->i_size = cpu_to_le32(inode->i_size); 1567 raw_inode->i_atime = cpu_to_le32(inode_get_atime_sec(inode)); 1568 raw_inode->i_ctime = cpu_to_le32(inode_get_ctime_sec(inode)); 1569 raw_inode->i_mtime = cpu_to_le32(inode_get_mtime_sec(inode)); 1570 1571 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); 1572 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 1573 raw_inode->i_flags = cpu_to_le32(ei->i_flags); 1574 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); 1575 raw_inode->i_frag = ei->i_frag_no; 1576 raw_inode->i_fsize = ei->i_frag_size; 1577 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); 1578 if (!S_ISREG(inode->i_mode)) 1579 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); 1580 else { 1581 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32); 1582 if (inode->i_size > 0x7fffffffULL) { 1583 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb, 1584 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) || 1585 EXT2_SB(sb)->s_es->s_rev_level == 1586 cpu_to_le32(EXT2_GOOD_OLD_REV)) { 1587 /* If this is the first large file 1588 * created, add a flag to the superblock. 1589 */ 1590 spin_lock(&EXT2_SB(sb)->s_lock); 1591 ext2_update_dynamic_rev(sb); 1592 EXT2_SET_RO_COMPAT_FEATURE(sb, 1593 EXT2_FEATURE_RO_COMPAT_LARGE_FILE); 1594 spin_unlock(&EXT2_SB(sb)->s_lock); 1595 ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1); 1596 } 1597 } 1598 } 1599 1600 raw_inode->i_generation = cpu_to_le32(inode->i_generation); 1601 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { 1602 if (old_valid_dev(inode->i_rdev)) { 1603 raw_inode->i_block[0] = 1604 cpu_to_le32(old_encode_dev(inode->i_rdev)); 1605 raw_inode->i_block[1] = 0; 1606 } else { 1607 raw_inode->i_block[0] = 0; 1608 raw_inode->i_block[1] = 1609 cpu_to_le32(new_encode_dev(inode->i_rdev)); 1610 raw_inode->i_block[2] = 0; 1611 } 1612 } else for (n = 0; n < EXT2_N_BLOCKS; n++) 1613 raw_inode->i_block[n] = ei->i_data[n]; 1614 mark_buffer_dirty(bh); 1615 if (do_sync) { 1616 sync_dirty_buffer(bh); 1617 if (buffer_req(bh) && !buffer_uptodate(bh)) { 1618 printk ("IO error syncing ext2 inode [%s:%08lx]\n", 1619 sb->s_id, (unsigned long) ino); 1620 err = -EIO; 1621 } 1622 } 1623 ei->i_state &= ~EXT2_STATE_NEW; 1624 brelse (bh); 1625 return err; 1626 } 1627 1628 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc) 1629 { 1630 return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL); 1631 } 1632 1633 int ext2_getattr(struct mnt_idmap *idmap, const struct path *path, 1634 struct kstat *stat, u32 request_mask, unsigned int query_flags) 1635 { 1636 struct inode *inode = d_inode(path->dentry); 1637 struct ext2_inode_info *ei = EXT2_I(inode); 1638 unsigned int flags; 1639 1640 flags = ei->i_flags & EXT2_FL_USER_VISIBLE; 1641 if (flags & EXT2_APPEND_FL) 1642 stat->attributes |= STATX_ATTR_APPEND; 1643 if (flags & EXT2_COMPR_FL) 1644 stat->attributes |= STATX_ATTR_COMPRESSED; 1645 if (flags & EXT2_IMMUTABLE_FL) 1646 stat->attributes |= STATX_ATTR_IMMUTABLE; 1647 if (flags & EXT2_NODUMP_FL) 1648 stat->attributes |= STATX_ATTR_NODUMP; 1649 stat->attributes_mask |= (STATX_ATTR_APPEND | 1650 STATX_ATTR_COMPRESSED | 1651 STATX_ATTR_ENCRYPTED | 1652 STATX_ATTR_IMMUTABLE | 1653 STATX_ATTR_NODUMP); 1654 1655 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 1656 return 0; 1657 } 1658 1659 int ext2_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 1660 struct iattr *iattr) 1661 { 1662 struct inode *inode = d_inode(dentry); 1663 int error; 1664 1665 error = setattr_prepare(&nop_mnt_idmap, dentry, iattr); 1666 if (error) 1667 return error; 1668 1669 if (is_quota_modification(&nop_mnt_idmap, inode, iattr)) { 1670 error = dquot_initialize(inode); 1671 if (error) 1672 return error; 1673 } 1674 if (i_uid_needs_update(&nop_mnt_idmap, iattr, inode) || 1675 i_gid_needs_update(&nop_mnt_idmap, iattr, inode)) { 1676 error = dquot_transfer(&nop_mnt_idmap, inode, iattr); 1677 if (error) 1678 return error; 1679 } 1680 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) { 1681 error = ext2_setsize(inode, iattr->ia_size); 1682 if (error) 1683 return error; 1684 } 1685 setattr_copy(&nop_mnt_idmap, inode, iattr); 1686 if (iattr->ia_valid & ATTR_MODE) 1687 error = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode); 1688 mark_inode_dirty(inode); 1689 1690 return error; 1691 } 1692