1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. 4 * Copyright (c) 2013 Red Hat, Inc. 5 * All Rights Reserved. 6 */ 7 #include "xfs.h" 8 #include "xfs_fs.h" 9 #include "xfs_format.h" 10 #include "xfs_log_format.h" 11 #include "xfs_trans_resv.h" 12 #include "xfs_mount.h" 13 #include "xfs_da_format.h" 14 #include "xfs_da_btree.h" 15 #include "xfs_inode.h" 16 #include "xfs_dir2.h" 17 #include "xfs_dir2_priv.h" 18 #include "xfs_error.h" 19 #include "xfs_trans.h" 20 #include "xfs_buf_item.h" 21 #include "xfs_cksum.h" 22 #include "xfs_log.h" 23 24 static xfs_failaddr_t xfs_dir2_data_freefind_verify( 25 struct xfs_dir2_data_hdr *hdr, struct xfs_dir2_data_free *bf, 26 struct xfs_dir2_data_unused *dup, 27 struct xfs_dir2_data_free **bf_ent); 28 29 /* 30 * Check the consistency of the data block. 31 * The input can also be a block-format directory. 32 * Return NULL if the buffer is good, otherwise the address of the error. 33 */ 34 xfs_failaddr_t 35 __xfs_dir3_data_check( 36 struct xfs_inode *dp, /* incore inode pointer */ 37 struct xfs_buf *bp) /* data block's buffer */ 38 { 39 xfs_dir2_dataptr_t addr; /* addr for leaf lookup */ 40 xfs_dir2_data_free_t *bf; /* bestfree table */ 41 xfs_dir2_block_tail_t *btp=NULL; /* block tail */ 42 int count; /* count of entries found */ 43 xfs_dir2_data_hdr_t *hdr; /* data block header */ 44 xfs_dir2_data_entry_t *dep; /* data entry */ 45 xfs_dir2_data_free_t *dfp; /* bestfree entry */ 46 xfs_dir2_data_unused_t *dup; /* unused entry */ 47 char *endp; /* end of useful data */ 48 int freeseen; /* mask of bestfrees seen */ 49 xfs_dahash_t hash; /* hash of current name */ 50 int i; /* leaf index */ 51 int lastfree; /* last entry was unused */ 52 xfs_dir2_leaf_entry_t *lep=NULL; /* block leaf entries */ 53 xfs_mount_t *mp; /* filesystem mount point */ 54 char *p; /* current data position */ 55 int stale; /* count of stale leaves */ 56 struct xfs_name name; 57 const struct xfs_dir_ops *ops; 58 struct xfs_da_geometry *geo; 59 60 mp = bp->b_target->bt_mount; 61 geo = mp->m_dir_geo; 62 63 /* 64 * We can be passed a null dp here from a verifier, so we need to go the 65 * hard way to get them. 66 */ 67 ops = xfs_dir_get_ops(mp, dp); 68 69 /* 70 * If this isn't a directory, or we don't get handed the dir ops, 71 * something is seriously wrong. Bail out. 72 */ 73 if ((dp && !S_ISDIR(VFS_I(dp)->i_mode)) || 74 ops != xfs_dir_get_ops(mp, NULL)) 75 return __this_address; 76 77 hdr = bp->b_addr; 78 p = (char *)ops->data_entry_p(hdr); 79 80 switch (hdr->magic) { 81 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC): 82 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC): 83 btp = xfs_dir2_block_tail_p(geo, hdr); 84 lep = xfs_dir2_block_leaf_p(btp); 85 86 /* 87 * The number of leaf entries is limited by the size of the 88 * block and the amount of space used by the data entries. 89 * We don't know how much space is used by the data entries yet, 90 * so just ensure that the count falls somewhere inside the 91 * block right now. 92 */ 93 if (be32_to_cpu(btp->count) >= 94 ((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry)) 95 return __this_address; 96 break; 97 case cpu_to_be32(XFS_DIR3_DATA_MAGIC): 98 case cpu_to_be32(XFS_DIR2_DATA_MAGIC): 99 break; 100 default: 101 return __this_address; 102 } 103 endp = xfs_dir3_data_endp(geo, hdr); 104 if (!endp) 105 return __this_address; 106 107 /* 108 * Account for zero bestfree entries. 109 */ 110 bf = ops->data_bestfree_p(hdr); 111 count = lastfree = freeseen = 0; 112 if (!bf[0].length) { 113 if (bf[0].offset) 114 return __this_address; 115 freeseen |= 1 << 0; 116 } 117 if (!bf[1].length) { 118 if (bf[1].offset) 119 return __this_address; 120 freeseen |= 1 << 1; 121 } 122 if (!bf[2].length) { 123 if (bf[2].offset) 124 return __this_address; 125 freeseen |= 1 << 2; 126 } 127 128 if (be16_to_cpu(bf[0].length) < be16_to_cpu(bf[1].length)) 129 return __this_address; 130 if (be16_to_cpu(bf[1].length) < be16_to_cpu(bf[2].length)) 131 return __this_address; 132 /* 133 * Loop over the data/unused entries. 134 */ 135 while (p < endp) { 136 dup = (xfs_dir2_data_unused_t *)p; 137 /* 138 * If it's unused, look for the space in the bestfree table. 139 * If we find it, account for that, else make sure it 140 * doesn't need to be there. 141 */ 142 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) { 143 xfs_failaddr_t fa; 144 145 if (lastfree != 0) 146 return __this_address; 147 if (endp < p + be16_to_cpu(dup->length)) 148 return __this_address; 149 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) != 150 (char *)dup - (char *)hdr) 151 return __this_address; 152 fa = xfs_dir2_data_freefind_verify(hdr, bf, dup, &dfp); 153 if (fa) 154 return fa; 155 if (dfp) { 156 i = (int)(dfp - bf); 157 if ((freeseen & (1 << i)) != 0) 158 return __this_address; 159 freeseen |= 1 << i; 160 } else { 161 if (be16_to_cpu(dup->length) > 162 be16_to_cpu(bf[2].length)) 163 return __this_address; 164 } 165 p += be16_to_cpu(dup->length); 166 lastfree = 1; 167 continue; 168 } 169 /* 170 * It's a real entry. Validate the fields. 171 * If this is a block directory then make sure it's 172 * in the leaf section of the block. 173 * The linear search is crude but this is DEBUG code. 174 */ 175 dep = (xfs_dir2_data_entry_t *)p; 176 if (dep->namelen == 0) 177 return __this_address; 178 if (xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber))) 179 return __this_address; 180 if (endp < p + ops->data_entsize(dep->namelen)) 181 return __this_address; 182 if (be16_to_cpu(*ops->data_entry_tag_p(dep)) != 183 (char *)dep - (char *)hdr) 184 return __this_address; 185 if (ops->data_get_ftype(dep) >= XFS_DIR3_FT_MAX) 186 return __this_address; 187 count++; 188 lastfree = 0; 189 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 190 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) { 191 addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, 192 (xfs_dir2_data_aoff_t) 193 ((char *)dep - (char *)hdr)); 194 name.name = dep->name; 195 name.len = dep->namelen; 196 hash = mp->m_dirnameops->hashname(&name); 197 for (i = 0; i < be32_to_cpu(btp->count); i++) { 198 if (be32_to_cpu(lep[i].address) == addr && 199 be32_to_cpu(lep[i].hashval) == hash) 200 break; 201 } 202 if (i >= be32_to_cpu(btp->count)) 203 return __this_address; 204 } 205 p += ops->data_entsize(dep->namelen); 206 } 207 /* 208 * Need to have seen all the entries and all the bestfree slots. 209 */ 210 if (freeseen != 7) 211 return __this_address; 212 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 213 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) { 214 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) { 215 if (lep[i].address == 216 cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) 217 stale++; 218 if (i > 0 && be32_to_cpu(lep[i].hashval) < 219 be32_to_cpu(lep[i - 1].hashval)) 220 return __this_address; 221 } 222 if (count != be32_to_cpu(btp->count) - be32_to_cpu(btp->stale)) 223 return __this_address; 224 if (stale != be32_to_cpu(btp->stale)) 225 return __this_address; 226 } 227 return NULL; 228 } 229 230 #ifdef DEBUG 231 void 232 xfs_dir3_data_check( 233 struct xfs_inode *dp, 234 struct xfs_buf *bp) 235 { 236 xfs_failaddr_t fa; 237 238 fa = __xfs_dir3_data_check(dp, bp); 239 if (!fa) 240 return; 241 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount, 242 bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__, 243 fa); 244 ASSERT(0); 245 } 246 #endif 247 248 static xfs_failaddr_t 249 xfs_dir3_data_verify( 250 struct xfs_buf *bp) 251 { 252 struct xfs_mount *mp = bp->b_target->bt_mount; 253 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr; 254 255 if (xfs_sb_version_hascrc(&mp->m_sb)) { 256 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC)) 257 return __this_address; 258 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid)) 259 return __this_address; 260 if (be64_to_cpu(hdr3->blkno) != bp->b_bn) 261 return __this_address; 262 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn))) 263 return __this_address; 264 } else { 265 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC)) 266 return __this_address; 267 } 268 return __xfs_dir3_data_check(NULL, bp); 269 } 270 271 /* 272 * Readahead of the first block of the directory when it is opened is completely 273 * oblivious to the format of the directory. Hence we can either get a block 274 * format buffer or a data format buffer on readahead. 275 */ 276 static void 277 xfs_dir3_data_reada_verify( 278 struct xfs_buf *bp) 279 { 280 struct xfs_dir2_data_hdr *hdr = bp->b_addr; 281 282 switch (hdr->magic) { 283 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC): 284 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC): 285 bp->b_ops = &xfs_dir3_block_buf_ops; 286 bp->b_ops->verify_read(bp); 287 return; 288 case cpu_to_be32(XFS_DIR2_DATA_MAGIC): 289 case cpu_to_be32(XFS_DIR3_DATA_MAGIC): 290 bp->b_ops = &xfs_dir3_data_buf_ops; 291 bp->b_ops->verify_read(bp); 292 return; 293 default: 294 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address); 295 break; 296 } 297 } 298 299 static void 300 xfs_dir3_data_read_verify( 301 struct xfs_buf *bp) 302 { 303 struct xfs_mount *mp = bp->b_target->bt_mount; 304 xfs_failaddr_t fa; 305 306 if (xfs_sb_version_hascrc(&mp->m_sb) && 307 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF)) 308 xfs_verifier_error(bp, -EFSBADCRC, __this_address); 309 else { 310 fa = xfs_dir3_data_verify(bp); 311 if (fa) 312 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 313 } 314 } 315 316 static void 317 xfs_dir3_data_write_verify( 318 struct xfs_buf *bp) 319 { 320 struct xfs_mount *mp = bp->b_target->bt_mount; 321 struct xfs_buf_log_item *bip = bp->b_log_item; 322 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr; 323 xfs_failaddr_t fa; 324 325 fa = xfs_dir3_data_verify(bp); 326 if (fa) { 327 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 328 return; 329 } 330 331 if (!xfs_sb_version_hascrc(&mp->m_sb)) 332 return; 333 334 if (bip) 335 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn); 336 337 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF); 338 } 339 340 const struct xfs_buf_ops xfs_dir3_data_buf_ops = { 341 .name = "xfs_dir3_data", 342 .verify_read = xfs_dir3_data_read_verify, 343 .verify_write = xfs_dir3_data_write_verify, 344 .verify_struct = xfs_dir3_data_verify, 345 }; 346 347 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = { 348 .name = "xfs_dir3_data_reada", 349 .verify_read = xfs_dir3_data_reada_verify, 350 .verify_write = xfs_dir3_data_write_verify, 351 }; 352 353 354 int 355 xfs_dir3_data_read( 356 struct xfs_trans *tp, 357 struct xfs_inode *dp, 358 xfs_dablk_t bno, 359 xfs_daddr_t mapped_bno, 360 struct xfs_buf **bpp) 361 { 362 int err; 363 364 err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp, 365 XFS_DATA_FORK, &xfs_dir3_data_buf_ops); 366 if (!err && tp && *bpp) 367 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF); 368 return err; 369 } 370 371 int 372 xfs_dir3_data_readahead( 373 struct xfs_inode *dp, 374 xfs_dablk_t bno, 375 xfs_daddr_t mapped_bno) 376 { 377 return xfs_da_reada_buf(dp, bno, mapped_bno, 378 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops); 379 } 380 381 /* 382 * Find the bestfree entry that exactly coincides with unused directory space 383 * or a verifier error because the bestfree data are bad. 384 */ 385 static xfs_failaddr_t 386 xfs_dir2_data_freefind_verify( 387 struct xfs_dir2_data_hdr *hdr, 388 struct xfs_dir2_data_free *bf, 389 struct xfs_dir2_data_unused *dup, 390 struct xfs_dir2_data_free **bf_ent) 391 { 392 struct xfs_dir2_data_free *dfp; 393 xfs_dir2_data_aoff_t off; 394 bool matched = false; 395 bool seenzero = false; 396 397 *bf_ent = NULL; 398 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr); 399 400 /* 401 * Validate some consistency in the bestfree table. 402 * Check order, non-overlapping entries, and if we find the 403 * one we're looking for it has to be exact. 404 */ 405 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) { 406 if (!dfp->offset) { 407 if (dfp->length) 408 return __this_address; 409 seenzero = true; 410 continue; 411 } 412 if (seenzero) 413 return __this_address; 414 if (be16_to_cpu(dfp->offset) == off) { 415 matched = true; 416 if (dfp->length != dup->length) 417 return __this_address; 418 } else if (be16_to_cpu(dfp->offset) > off) { 419 if (off + be16_to_cpu(dup->length) > 420 be16_to_cpu(dfp->offset)) 421 return __this_address; 422 } else { 423 if (be16_to_cpu(dfp->offset) + 424 be16_to_cpu(dfp->length) > off) 425 return __this_address; 426 } 427 if (!matched && 428 be16_to_cpu(dfp->length) < be16_to_cpu(dup->length)) 429 return __this_address; 430 if (dfp > &bf[0] && 431 be16_to_cpu(dfp[-1].length) < be16_to_cpu(dfp[0].length)) 432 return __this_address; 433 } 434 435 /* Looks ok so far; now try to match up with a bestfree entry. */ 436 *bf_ent = xfs_dir2_data_freefind(hdr, bf, dup); 437 return NULL; 438 } 439 440 /* 441 * Given a data block and an unused entry from that block, 442 * return the bestfree entry if any that corresponds to it. 443 */ 444 xfs_dir2_data_free_t * 445 xfs_dir2_data_freefind( 446 struct xfs_dir2_data_hdr *hdr, /* data block header */ 447 struct xfs_dir2_data_free *bf, /* bestfree table pointer */ 448 struct xfs_dir2_data_unused *dup) /* unused space */ 449 { 450 xfs_dir2_data_free_t *dfp; /* bestfree entry */ 451 xfs_dir2_data_aoff_t off; /* offset value needed */ 452 453 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr); 454 455 /* 456 * If this is smaller than the smallest bestfree entry, 457 * it can't be there since they're sorted. 458 */ 459 if (be16_to_cpu(dup->length) < 460 be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length)) 461 return NULL; 462 /* 463 * Look at the three bestfree entries for our guy. 464 */ 465 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) { 466 if (!dfp->offset) 467 return NULL; 468 if (be16_to_cpu(dfp->offset) == off) 469 return dfp; 470 } 471 /* 472 * Didn't find it. This only happens if there are duplicate lengths. 473 */ 474 return NULL; 475 } 476 477 /* 478 * Insert an unused-space entry into the bestfree table. 479 */ 480 xfs_dir2_data_free_t * /* entry inserted */ 481 xfs_dir2_data_freeinsert( 482 struct xfs_dir2_data_hdr *hdr, /* data block pointer */ 483 struct xfs_dir2_data_free *dfp, /* bestfree table pointer */ 484 struct xfs_dir2_data_unused *dup, /* unused space */ 485 int *loghead) /* log the data header (out) */ 486 { 487 xfs_dir2_data_free_t new; /* new bestfree entry */ 488 489 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) || 490 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 491 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || 492 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)); 493 494 new.length = dup->length; 495 new.offset = cpu_to_be16((char *)dup - (char *)hdr); 496 497 /* 498 * Insert at position 0, 1, or 2; or not at all. 499 */ 500 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) { 501 dfp[2] = dfp[1]; 502 dfp[1] = dfp[0]; 503 dfp[0] = new; 504 *loghead = 1; 505 return &dfp[0]; 506 } 507 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) { 508 dfp[2] = dfp[1]; 509 dfp[1] = new; 510 *loghead = 1; 511 return &dfp[1]; 512 } 513 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) { 514 dfp[2] = new; 515 *loghead = 1; 516 return &dfp[2]; 517 } 518 return NULL; 519 } 520 521 /* 522 * Remove a bestfree entry from the table. 523 */ 524 STATIC void 525 xfs_dir2_data_freeremove( 526 struct xfs_dir2_data_hdr *hdr, /* data block header */ 527 struct xfs_dir2_data_free *bf, /* bestfree table pointer */ 528 struct xfs_dir2_data_free *dfp, /* bestfree entry pointer */ 529 int *loghead) /* out: log data header */ 530 { 531 532 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) || 533 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 534 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || 535 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)); 536 537 /* 538 * It's the first entry, slide the next 2 up. 539 */ 540 if (dfp == &bf[0]) { 541 bf[0] = bf[1]; 542 bf[1] = bf[2]; 543 } 544 /* 545 * It's the second entry, slide the 3rd entry up. 546 */ 547 else if (dfp == &bf[1]) 548 bf[1] = bf[2]; 549 /* 550 * Must be the last entry. 551 */ 552 else 553 ASSERT(dfp == &bf[2]); 554 /* 555 * Clear the 3rd entry, must be zero now. 556 */ 557 bf[2].length = 0; 558 bf[2].offset = 0; 559 *loghead = 1; 560 } 561 562 /* 563 * Given a data block, reconstruct its bestfree map. 564 */ 565 void 566 xfs_dir2_data_freescan_int( 567 struct xfs_da_geometry *geo, 568 const struct xfs_dir_ops *ops, 569 struct xfs_dir2_data_hdr *hdr, 570 int *loghead) 571 { 572 xfs_dir2_data_entry_t *dep; /* active data entry */ 573 xfs_dir2_data_unused_t *dup; /* unused data entry */ 574 struct xfs_dir2_data_free *bf; 575 char *endp; /* end of block's data */ 576 char *p; /* current entry pointer */ 577 578 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) || 579 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || 580 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 581 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)); 582 583 /* 584 * Start by clearing the table. 585 */ 586 bf = ops->data_bestfree_p(hdr); 587 memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT); 588 *loghead = 1; 589 /* 590 * Set up pointers. 591 */ 592 p = (char *)ops->data_entry_p(hdr); 593 endp = xfs_dir3_data_endp(geo, hdr); 594 /* 595 * Loop over the block's entries. 596 */ 597 while (p < endp) { 598 dup = (xfs_dir2_data_unused_t *)p; 599 /* 600 * If it's a free entry, insert it. 601 */ 602 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) { 603 ASSERT((char *)dup - (char *)hdr == 604 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup))); 605 xfs_dir2_data_freeinsert(hdr, bf, dup, loghead); 606 p += be16_to_cpu(dup->length); 607 } 608 /* 609 * For active entries, check their tags and skip them. 610 */ 611 else { 612 dep = (xfs_dir2_data_entry_t *)p; 613 ASSERT((char *)dep - (char *)hdr == 614 be16_to_cpu(*ops->data_entry_tag_p(dep))); 615 p += ops->data_entsize(dep->namelen); 616 } 617 } 618 } 619 620 void 621 xfs_dir2_data_freescan( 622 struct xfs_inode *dp, 623 struct xfs_dir2_data_hdr *hdr, 624 int *loghead) 625 { 626 return xfs_dir2_data_freescan_int(dp->i_mount->m_dir_geo, dp->d_ops, 627 hdr, loghead); 628 } 629 630 /* 631 * Initialize a data block at the given block number in the directory. 632 * Give back the buffer for the created block. 633 */ 634 int /* error */ 635 xfs_dir3_data_init( 636 xfs_da_args_t *args, /* directory operation args */ 637 xfs_dir2_db_t blkno, /* logical dir block number */ 638 struct xfs_buf **bpp) /* output block buffer */ 639 { 640 struct xfs_buf *bp; /* block buffer */ 641 xfs_dir2_data_hdr_t *hdr; /* data block header */ 642 xfs_inode_t *dp; /* incore directory inode */ 643 xfs_dir2_data_unused_t *dup; /* unused entry pointer */ 644 struct xfs_dir2_data_free *bf; 645 int error; /* error return value */ 646 int i; /* bestfree index */ 647 xfs_mount_t *mp; /* filesystem mount point */ 648 xfs_trans_t *tp; /* transaction pointer */ 649 int t; /* temp */ 650 651 dp = args->dp; 652 mp = dp->i_mount; 653 tp = args->trans; 654 /* 655 * Get the buffer set up for the block. 656 */ 657 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno), 658 -1, &bp, XFS_DATA_FORK); 659 if (error) 660 return error; 661 bp->b_ops = &xfs_dir3_data_buf_ops; 662 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF); 663 664 /* 665 * Initialize the header. 666 */ 667 hdr = bp->b_addr; 668 if (xfs_sb_version_hascrc(&mp->m_sb)) { 669 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr; 670 671 memset(hdr3, 0, sizeof(*hdr3)); 672 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC); 673 hdr3->blkno = cpu_to_be64(bp->b_bn); 674 hdr3->owner = cpu_to_be64(dp->i_ino); 675 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid); 676 677 } else 678 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC); 679 680 bf = dp->d_ops->data_bestfree_p(hdr); 681 bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset); 682 for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) { 683 bf[i].length = 0; 684 bf[i].offset = 0; 685 } 686 687 /* 688 * Set up an unused entry for the block's body. 689 */ 690 dup = dp->d_ops->data_unused_p(hdr); 691 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG); 692 693 t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset; 694 bf[0].length = cpu_to_be16(t); 695 dup->length = cpu_to_be16(t); 696 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr); 697 /* 698 * Log it and return it. 699 */ 700 xfs_dir2_data_log_header(args, bp); 701 xfs_dir2_data_log_unused(args, bp, dup); 702 *bpp = bp; 703 return 0; 704 } 705 706 /* 707 * Log an active data entry from the block. 708 */ 709 void 710 xfs_dir2_data_log_entry( 711 struct xfs_da_args *args, 712 struct xfs_buf *bp, 713 xfs_dir2_data_entry_t *dep) /* data entry pointer */ 714 { 715 struct xfs_dir2_data_hdr *hdr = bp->b_addr; 716 717 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) || 718 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || 719 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 720 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)); 721 722 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr), 723 (uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) - 724 (char *)hdr - 1)); 725 } 726 727 /* 728 * Log a data block header. 729 */ 730 void 731 xfs_dir2_data_log_header( 732 struct xfs_da_args *args, 733 struct xfs_buf *bp) 734 { 735 #ifdef DEBUG 736 struct xfs_dir2_data_hdr *hdr = bp->b_addr; 737 738 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) || 739 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || 740 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 741 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)); 742 #endif 743 744 xfs_trans_log_buf(args->trans, bp, 0, 745 args->dp->d_ops->data_entry_offset - 1); 746 } 747 748 /* 749 * Log a data unused entry. 750 */ 751 void 752 xfs_dir2_data_log_unused( 753 struct xfs_da_args *args, 754 struct xfs_buf *bp, 755 xfs_dir2_data_unused_t *dup) /* data unused pointer */ 756 { 757 xfs_dir2_data_hdr_t *hdr = bp->b_addr; 758 759 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) || 760 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || 761 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) || 762 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)); 763 764 /* 765 * Log the first part of the unused entry. 766 */ 767 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr), 768 (uint)((char *)&dup->length + sizeof(dup->length) - 769 1 - (char *)hdr)); 770 /* 771 * Log the end (tag) of the unused entry. 772 */ 773 xfs_trans_log_buf(args->trans, bp, 774 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr), 775 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr + 776 sizeof(xfs_dir2_data_off_t) - 1)); 777 } 778 779 /* 780 * Make a byte range in the data block unused. 781 * Its current contents are unimportant. 782 */ 783 void 784 xfs_dir2_data_make_free( 785 struct xfs_da_args *args, 786 struct xfs_buf *bp, 787 xfs_dir2_data_aoff_t offset, /* starting byte offset */ 788 xfs_dir2_data_aoff_t len, /* length in bytes */ 789 int *needlogp, /* out: log header */ 790 int *needscanp) /* out: regen bestfree */ 791 { 792 xfs_dir2_data_hdr_t *hdr; /* data block pointer */ 793 xfs_dir2_data_free_t *dfp; /* bestfree pointer */ 794 char *endptr; /* end of data area */ 795 int needscan; /* need to regen bestfree */ 796 xfs_dir2_data_unused_t *newdup; /* new unused entry */ 797 xfs_dir2_data_unused_t *postdup; /* unused entry after us */ 798 xfs_dir2_data_unused_t *prevdup; /* unused entry before us */ 799 struct xfs_dir2_data_free *bf; 800 801 hdr = bp->b_addr; 802 803 /* 804 * Figure out where the end of the data area is. 805 */ 806 endptr = xfs_dir3_data_endp(args->geo, hdr); 807 ASSERT(endptr != NULL); 808 809 /* 810 * If this isn't the start of the block, then back up to 811 * the previous entry and see if it's free. 812 */ 813 if (offset > args->dp->d_ops->data_entry_offset) { 814 __be16 *tagp; /* tag just before us */ 815 816 tagp = (__be16 *)((char *)hdr + offset) - 1; 817 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp)); 818 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG) 819 prevdup = NULL; 820 } else 821 prevdup = NULL; 822 /* 823 * If this isn't the end of the block, see if the entry after 824 * us is free. 825 */ 826 if ((char *)hdr + offset + len < endptr) { 827 postdup = 828 (xfs_dir2_data_unused_t *)((char *)hdr + offset + len); 829 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG) 830 postdup = NULL; 831 } else 832 postdup = NULL; 833 ASSERT(*needscanp == 0); 834 needscan = 0; 835 /* 836 * Previous and following entries are both free, 837 * merge everything into a single free entry. 838 */ 839 bf = args->dp->d_ops->data_bestfree_p(hdr); 840 if (prevdup && postdup) { 841 xfs_dir2_data_free_t *dfp2; /* another bestfree pointer */ 842 843 /* 844 * See if prevdup and/or postdup are in bestfree table. 845 */ 846 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup); 847 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup); 848 /* 849 * We need a rescan unless there are exactly 2 free entries 850 * namely our two. Then we know what's happening, otherwise 851 * since the third bestfree is there, there might be more 852 * entries. 853 */ 854 needscan = (bf[2].length != 0); 855 /* 856 * Fix up the new big freespace. 857 */ 858 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length)); 859 *xfs_dir2_data_unused_tag_p(prevdup) = 860 cpu_to_be16((char *)prevdup - (char *)hdr); 861 xfs_dir2_data_log_unused(args, bp, prevdup); 862 if (!needscan) { 863 /* 864 * Has to be the case that entries 0 and 1 are 865 * dfp and dfp2 (don't know which is which), and 866 * entry 2 is empty. 867 * Remove entry 1 first then entry 0. 868 */ 869 ASSERT(dfp && dfp2); 870 if (dfp == &bf[1]) { 871 dfp = &bf[0]; 872 ASSERT(dfp2 == dfp); 873 dfp2 = &bf[1]; 874 } 875 xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp); 876 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp); 877 /* 878 * Now insert the new entry. 879 */ 880 dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup, 881 needlogp); 882 ASSERT(dfp == &bf[0]); 883 ASSERT(dfp->length == prevdup->length); 884 ASSERT(!dfp[1].length); 885 ASSERT(!dfp[2].length); 886 } 887 } 888 /* 889 * The entry before us is free, merge with it. 890 */ 891 else if (prevdup) { 892 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup); 893 be16_add_cpu(&prevdup->length, len); 894 *xfs_dir2_data_unused_tag_p(prevdup) = 895 cpu_to_be16((char *)prevdup - (char *)hdr); 896 xfs_dir2_data_log_unused(args, bp, prevdup); 897 /* 898 * If the previous entry was in the table, the new entry 899 * is longer, so it will be in the table too. Remove 900 * the old one and add the new one. 901 */ 902 if (dfp) { 903 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp); 904 xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp); 905 } 906 /* 907 * Otherwise we need a scan if the new entry is big enough. 908 */ 909 else { 910 needscan = be16_to_cpu(prevdup->length) > 911 be16_to_cpu(bf[2].length); 912 } 913 } 914 /* 915 * The following entry is free, merge with it. 916 */ 917 else if (postdup) { 918 dfp = xfs_dir2_data_freefind(hdr, bf, postdup); 919 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset); 920 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG); 921 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length)); 922 *xfs_dir2_data_unused_tag_p(newdup) = 923 cpu_to_be16((char *)newdup - (char *)hdr); 924 xfs_dir2_data_log_unused(args, bp, newdup); 925 /* 926 * If the following entry was in the table, the new entry 927 * is longer, so it will be in the table too. Remove 928 * the old one and add the new one. 929 */ 930 if (dfp) { 931 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp); 932 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp); 933 } 934 /* 935 * Otherwise we need a scan if the new entry is big enough. 936 */ 937 else { 938 needscan = be16_to_cpu(newdup->length) > 939 be16_to_cpu(bf[2].length); 940 } 941 } 942 /* 943 * Neither neighbor is free. Make a new entry. 944 */ 945 else { 946 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset); 947 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG); 948 newdup->length = cpu_to_be16(len); 949 *xfs_dir2_data_unused_tag_p(newdup) = 950 cpu_to_be16((char *)newdup - (char *)hdr); 951 xfs_dir2_data_log_unused(args, bp, newdup); 952 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp); 953 } 954 *needscanp = needscan; 955 } 956 957 /* Check our free data for obvious signs of corruption. */ 958 static inline xfs_failaddr_t 959 xfs_dir2_data_check_free( 960 struct xfs_dir2_data_hdr *hdr, 961 struct xfs_dir2_data_unused *dup, 962 xfs_dir2_data_aoff_t offset, 963 xfs_dir2_data_aoff_t len) 964 { 965 if (hdr->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC) && 966 hdr->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC) && 967 hdr->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) && 968 hdr->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) 969 return __this_address; 970 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG) 971 return __this_address; 972 if (offset < (char *)dup - (char *)hdr) 973 return __this_address; 974 if (offset + len > (char *)dup + be16_to_cpu(dup->length) - (char *)hdr) 975 return __this_address; 976 if ((char *)dup - (char *)hdr != 977 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup))) 978 return __this_address; 979 return NULL; 980 } 981 982 /* Sanity-check a new bestfree entry. */ 983 static inline xfs_failaddr_t 984 xfs_dir2_data_check_new_free( 985 struct xfs_dir2_data_hdr *hdr, 986 struct xfs_dir2_data_free *dfp, 987 struct xfs_dir2_data_unused *newdup) 988 { 989 if (dfp == NULL) 990 return __this_address; 991 if (dfp->length != newdup->length) 992 return __this_address; 993 if (be16_to_cpu(dfp->offset) != (char *)newdup - (char *)hdr) 994 return __this_address; 995 return NULL; 996 } 997 998 /* 999 * Take a byte range out of an existing unused space and make it un-free. 1000 */ 1001 int 1002 xfs_dir2_data_use_free( 1003 struct xfs_da_args *args, 1004 struct xfs_buf *bp, 1005 xfs_dir2_data_unused_t *dup, /* unused entry */ 1006 xfs_dir2_data_aoff_t offset, /* starting offset to use */ 1007 xfs_dir2_data_aoff_t len, /* length to use */ 1008 int *needlogp, /* out: need to log header */ 1009 int *needscanp) /* out: need regen bestfree */ 1010 { 1011 xfs_dir2_data_hdr_t *hdr; /* data block header */ 1012 xfs_dir2_data_free_t *dfp; /* bestfree pointer */ 1013 xfs_dir2_data_unused_t *newdup; /* new unused entry */ 1014 xfs_dir2_data_unused_t *newdup2; /* another new unused entry */ 1015 struct xfs_dir2_data_free *bf; 1016 xfs_failaddr_t fa; 1017 int matchback; /* matches end of freespace */ 1018 int matchfront; /* matches start of freespace */ 1019 int needscan; /* need to regen bestfree */ 1020 int oldlen; /* old unused entry's length */ 1021 1022 hdr = bp->b_addr; 1023 fa = xfs_dir2_data_check_free(hdr, dup, offset, len); 1024 if (fa) 1025 goto corrupt; 1026 /* 1027 * Look up the entry in the bestfree table. 1028 */ 1029 oldlen = be16_to_cpu(dup->length); 1030 bf = args->dp->d_ops->data_bestfree_p(hdr); 1031 dfp = xfs_dir2_data_freefind(hdr, bf, dup); 1032 ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length)); 1033 /* 1034 * Check for alignment with front and back of the entry. 1035 */ 1036 matchfront = (char *)dup - (char *)hdr == offset; 1037 matchback = (char *)dup + oldlen - (char *)hdr == offset + len; 1038 ASSERT(*needscanp == 0); 1039 needscan = 0; 1040 /* 1041 * If we matched it exactly we just need to get rid of it from 1042 * the bestfree table. 1043 */ 1044 if (matchfront && matchback) { 1045 if (dfp) { 1046 needscan = (bf[2].offset != 0); 1047 if (!needscan) 1048 xfs_dir2_data_freeremove(hdr, bf, dfp, 1049 needlogp); 1050 } 1051 } 1052 /* 1053 * We match the first part of the entry. 1054 * Make a new entry with the remaining freespace. 1055 */ 1056 else if (matchfront) { 1057 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len); 1058 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG); 1059 newdup->length = cpu_to_be16(oldlen - len); 1060 *xfs_dir2_data_unused_tag_p(newdup) = 1061 cpu_to_be16((char *)newdup - (char *)hdr); 1062 xfs_dir2_data_log_unused(args, bp, newdup); 1063 /* 1064 * If it was in the table, remove it and add the new one. 1065 */ 1066 if (dfp) { 1067 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp); 1068 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup, 1069 needlogp); 1070 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup); 1071 if (fa) 1072 goto corrupt; 1073 /* 1074 * If we got inserted at the last slot, 1075 * that means we don't know if there was a better 1076 * choice for the last slot, or not. Rescan. 1077 */ 1078 needscan = dfp == &bf[2]; 1079 } 1080 } 1081 /* 1082 * We match the last part of the entry. 1083 * Trim the allocated space off the tail of the entry. 1084 */ 1085 else if (matchback) { 1086 newdup = dup; 1087 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup); 1088 *xfs_dir2_data_unused_tag_p(newdup) = 1089 cpu_to_be16((char *)newdup - (char *)hdr); 1090 xfs_dir2_data_log_unused(args, bp, newdup); 1091 /* 1092 * If it was in the table, remove it and add the new one. 1093 */ 1094 if (dfp) { 1095 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp); 1096 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup, 1097 needlogp); 1098 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup); 1099 if (fa) 1100 goto corrupt; 1101 /* 1102 * If we got inserted at the last slot, 1103 * that means we don't know if there was a better 1104 * choice for the last slot, or not. Rescan. 1105 */ 1106 needscan = dfp == &bf[2]; 1107 } 1108 } 1109 /* 1110 * Poking out the middle of an entry. 1111 * Make two new entries. 1112 */ 1113 else { 1114 newdup = dup; 1115 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup); 1116 *xfs_dir2_data_unused_tag_p(newdup) = 1117 cpu_to_be16((char *)newdup - (char *)hdr); 1118 xfs_dir2_data_log_unused(args, bp, newdup); 1119 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len); 1120 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG); 1121 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length)); 1122 *xfs_dir2_data_unused_tag_p(newdup2) = 1123 cpu_to_be16((char *)newdup2 - (char *)hdr); 1124 xfs_dir2_data_log_unused(args, bp, newdup2); 1125 /* 1126 * If the old entry was in the table, we need to scan 1127 * if the 3rd entry was valid, since these entries 1128 * are smaller than the old one. 1129 * If we don't need to scan that means there were 1 or 2 1130 * entries in the table, and removing the old and adding 1131 * the 2 new will work. 1132 */ 1133 if (dfp) { 1134 needscan = (bf[2].length != 0); 1135 if (!needscan) { 1136 xfs_dir2_data_freeremove(hdr, bf, dfp, 1137 needlogp); 1138 xfs_dir2_data_freeinsert(hdr, bf, newdup, 1139 needlogp); 1140 xfs_dir2_data_freeinsert(hdr, bf, newdup2, 1141 needlogp); 1142 } 1143 } 1144 } 1145 *needscanp = needscan; 1146 return 0; 1147 corrupt: 1148 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, args->dp->i_mount, 1149 hdr, sizeof(*hdr), __FILE__, __LINE__, fa); 1150 return -EFSCORRUPTED; 1151 } 1152 1153 /* Find the end of the entry data in a data/block format dir block. */ 1154 void * 1155 xfs_dir3_data_endp( 1156 struct xfs_da_geometry *geo, 1157 struct xfs_dir2_data_hdr *hdr) 1158 { 1159 switch (hdr->magic) { 1160 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC): 1161 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC): 1162 return xfs_dir2_block_leaf_p(xfs_dir2_block_tail_p(geo, hdr)); 1163 case cpu_to_be32(XFS_DIR3_DATA_MAGIC): 1164 case cpu_to_be32(XFS_DIR2_DATA_MAGIC): 1165 return (char *)hdr + geo->blksize; 1166 default: 1167 return NULL; 1168 } 1169 } 1170