1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 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_shared.h" 10 #include "xfs_format.h" 11 #include "xfs_log_format.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_sb.h" 14 #include "xfs_mount.h" 15 #include "xfs_da_format.h" 16 #include "xfs_da_btree.h" 17 #include "xfs_inode.h" 18 #include "xfs_trans.h" 19 #include "xfs_bmap_btree.h" 20 #include "xfs_bmap.h" 21 #include "xfs_attr_sf.h" 22 #include "xfs_attr_remote.h" 23 #include "xfs_attr.h" 24 #include "xfs_attr_leaf.h" 25 #include "xfs_error.h" 26 #include "xfs_trace.h" 27 #include "xfs_buf_item.h" 28 #include "xfs_dir2.h" 29 #include "xfs_log.h" 30 31 32 /* 33 * xfs_attr_leaf.c 34 * 35 * Routines to implement leaf blocks of attributes as Btrees of hashed names. 36 */ 37 38 /*======================================================================== 39 * Function prototypes for the kernel. 40 *========================================================================*/ 41 42 /* 43 * Routines used for growing the Btree. 44 */ 45 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args, 46 xfs_dablk_t which_block, struct xfs_buf **bpp); 47 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer, 48 struct xfs_attr3_icleaf_hdr *ichdr, 49 struct xfs_da_args *args, int freemap_index); 50 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args, 51 struct xfs_attr3_icleaf_hdr *ichdr, 52 struct xfs_buf *leaf_buffer); 53 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state, 54 xfs_da_state_blk_t *blk1, 55 xfs_da_state_blk_t *blk2); 56 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state, 57 xfs_da_state_blk_t *leaf_blk_1, 58 struct xfs_attr3_icleaf_hdr *ichdr1, 59 xfs_da_state_blk_t *leaf_blk_2, 60 struct xfs_attr3_icleaf_hdr *ichdr2, 61 int *number_entries_in_blk1, 62 int *number_usedbytes_in_blk1); 63 64 /* 65 * Utility routines. 66 */ 67 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args, 68 struct xfs_attr_leafblock *src_leaf, 69 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start, 70 struct xfs_attr_leafblock *dst_leaf, 71 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start, 72 int move_count); 73 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index); 74 75 /* 76 * attr3 block 'firstused' conversion helpers. 77 * 78 * firstused refers to the offset of the first used byte of the nameval region 79 * of an attr leaf block. The region starts at the tail of the block and expands 80 * backwards towards the middle. As such, firstused is initialized to the block 81 * size for an empty leaf block and is reduced from there. 82 * 83 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k. 84 * The in-core firstused field is 32-bit and thus supports the maximum fsb size. 85 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this 86 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent 87 * the attr block size. The following helpers manage the conversion between the 88 * in-core and on-disk formats. 89 */ 90 91 static void 92 xfs_attr3_leaf_firstused_from_disk( 93 struct xfs_da_geometry *geo, 94 struct xfs_attr3_icleaf_hdr *to, 95 struct xfs_attr_leafblock *from) 96 { 97 struct xfs_attr3_leaf_hdr *hdr3; 98 99 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) { 100 hdr3 = (struct xfs_attr3_leaf_hdr *) from; 101 to->firstused = be16_to_cpu(hdr3->firstused); 102 } else { 103 to->firstused = be16_to_cpu(from->hdr.firstused); 104 } 105 106 /* 107 * Convert from the magic fsb size value to actual blocksize. This 108 * should only occur for empty blocks when the block size overflows 109 * 16-bits. 110 */ 111 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) { 112 ASSERT(!to->count && !to->usedbytes); 113 ASSERT(geo->blksize > USHRT_MAX); 114 to->firstused = geo->blksize; 115 } 116 } 117 118 static void 119 xfs_attr3_leaf_firstused_to_disk( 120 struct xfs_da_geometry *geo, 121 struct xfs_attr_leafblock *to, 122 struct xfs_attr3_icleaf_hdr *from) 123 { 124 struct xfs_attr3_leaf_hdr *hdr3; 125 uint32_t firstused; 126 127 /* magic value should only be seen on disk */ 128 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF); 129 130 /* 131 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk 132 * value. This only overflows at the max supported value of 64k. Use the 133 * magic on-disk value to represent block size in this case. 134 */ 135 firstused = from->firstused; 136 if (firstused > USHRT_MAX) { 137 ASSERT(from->firstused == geo->blksize); 138 firstused = XFS_ATTR3_LEAF_NULLOFF; 139 } 140 141 if (from->magic == XFS_ATTR3_LEAF_MAGIC) { 142 hdr3 = (struct xfs_attr3_leaf_hdr *) to; 143 hdr3->firstused = cpu_to_be16(firstused); 144 } else { 145 to->hdr.firstused = cpu_to_be16(firstused); 146 } 147 } 148 149 void 150 xfs_attr3_leaf_hdr_from_disk( 151 struct xfs_da_geometry *geo, 152 struct xfs_attr3_icleaf_hdr *to, 153 struct xfs_attr_leafblock *from) 154 { 155 int i; 156 157 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || 158 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); 159 160 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) { 161 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from; 162 163 to->forw = be32_to_cpu(hdr3->info.hdr.forw); 164 to->back = be32_to_cpu(hdr3->info.hdr.back); 165 to->magic = be16_to_cpu(hdr3->info.hdr.magic); 166 to->count = be16_to_cpu(hdr3->count); 167 to->usedbytes = be16_to_cpu(hdr3->usedbytes); 168 xfs_attr3_leaf_firstused_from_disk(geo, to, from); 169 to->holes = hdr3->holes; 170 171 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 172 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base); 173 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size); 174 } 175 return; 176 } 177 to->forw = be32_to_cpu(from->hdr.info.forw); 178 to->back = be32_to_cpu(from->hdr.info.back); 179 to->magic = be16_to_cpu(from->hdr.info.magic); 180 to->count = be16_to_cpu(from->hdr.count); 181 to->usedbytes = be16_to_cpu(from->hdr.usedbytes); 182 xfs_attr3_leaf_firstused_from_disk(geo, to, from); 183 to->holes = from->hdr.holes; 184 185 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 186 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base); 187 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size); 188 } 189 } 190 191 void 192 xfs_attr3_leaf_hdr_to_disk( 193 struct xfs_da_geometry *geo, 194 struct xfs_attr_leafblock *to, 195 struct xfs_attr3_icleaf_hdr *from) 196 { 197 int i; 198 199 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC || 200 from->magic == XFS_ATTR3_LEAF_MAGIC); 201 202 if (from->magic == XFS_ATTR3_LEAF_MAGIC) { 203 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to; 204 205 hdr3->info.hdr.forw = cpu_to_be32(from->forw); 206 hdr3->info.hdr.back = cpu_to_be32(from->back); 207 hdr3->info.hdr.magic = cpu_to_be16(from->magic); 208 hdr3->count = cpu_to_be16(from->count); 209 hdr3->usedbytes = cpu_to_be16(from->usedbytes); 210 xfs_attr3_leaf_firstused_to_disk(geo, to, from); 211 hdr3->holes = from->holes; 212 hdr3->pad1 = 0; 213 214 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 215 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base); 216 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size); 217 } 218 return; 219 } 220 to->hdr.info.forw = cpu_to_be32(from->forw); 221 to->hdr.info.back = cpu_to_be32(from->back); 222 to->hdr.info.magic = cpu_to_be16(from->magic); 223 to->hdr.count = cpu_to_be16(from->count); 224 to->hdr.usedbytes = cpu_to_be16(from->usedbytes); 225 xfs_attr3_leaf_firstused_to_disk(geo, to, from); 226 to->hdr.holes = from->holes; 227 to->hdr.pad1 = 0; 228 229 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 230 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base); 231 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size); 232 } 233 } 234 235 static xfs_failaddr_t 236 xfs_attr3_leaf_verify_entry( 237 struct xfs_mount *mp, 238 char *buf_end, 239 struct xfs_attr_leafblock *leaf, 240 struct xfs_attr3_icleaf_hdr *leafhdr, 241 struct xfs_attr_leaf_entry *ent, 242 int idx, 243 __u32 *last_hashval) 244 { 245 struct xfs_attr_leaf_name_local *lentry; 246 struct xfs_attr_leaf_name_remote *rentry; 247 char *name_end; 248 unsigned int nameidx; 249 unsigned int namesize; 250 __u32 hashval; 251 252 /* hash order check */ 253 hashval = be32_to_cpu(ent->hashval); 254 if (hashval < *last_hashval) 255 return __this_address; 256 *last_hashval = hashval; 257 258 nameidx = be16_to_cpu(ent->nameidx); 259 if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize) 260 return __this_address; 261 262 /* 263 * Check the name information. The namelen fields are u8 so we can't 264 * possibly exceed the maximum name length of 255 bytes. 265 */ 266 if (ent->flags & XFS_ATTR_LOCAL) { 267 lentry = xfs_attr3_leaf_name_local(leaf, idx); 268 namesize = xfs_attr_leaf_entsize_local(lentry->namelen, 269 be16_to_cpu(lentry->valuelen)); 270 name_end = (char *)lentry + namesize; 271 if (lentry->namelen == 0) 272 return __this_address; 273 } else { 274 rentry = xfs_attr3_leaf_name_remote(leaf, idx); 275 namesize = xfs_attr_leaf_entsize_remote(rentry->namelen); 276 name_end = (char *)rentry + namesize; 277 if (rentry->namelen == 0) 278 return __this_address; 279 if (!(ent->flags & XFS_ATTR_INCOMPLETE) && 280 rentry->valueblk == 0) 281 return __this_address; 282 } 283 284 if (name_end > buf_end) 285 return __this_address; 286 287 return NULL; 288 } 289 290 static xfs_failaddr_t 291 xfs_attr3_leaf_verify( 292 struct xfs_buf *bp) 293 { 294 struct xfs_attr3_icleaf_hdr ichdr; 295 struct xfs_mount *mp = bp->b_mount; 296 struct xfs_attr_leafblock *leaf = bp->b_addr; 297 struct xfs_attr_leaf_entry *entries; 298 struct xfs_attr_leaf_entry *ent; 299 char *buf_end; 300 uint32_t end; /* must be 32bit - see below */ 301 __u32 last_hashval = 0; 302 int i; 303 xfs_failaddr_t fa; 304 305 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf); 306 307 fa = xfs_da3_blkinfo_verify(bp, bp->b_addr); 308 if (fa) 309 return fa; 310 311 /* 312 * firstused is the block offset of the first name info structure. 313 * Make sure it doesn't go off the block or crash into the header. 314 */ 315 if (ichdr.firstused > mp->m_attr_geo->blksize) 316 return __this_address; 317 if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf)) 318 return __this_address; 319 320 /* Make sure the entries array doesn't crash into the name info. */ 321 entries = xfs_attr3_leaf_entryp(bp->b_addr); 322 if ((char *)&entries[ichdr.count] > 323 (char *)bp->b_addr + ichdr.firstused) 324 return __this_address; 325 326 /* 327 * NOTE: This verifier historically failed empty leaf buffers because 328 * we expect the fork to be in another format. Empty attr fork format 329 * conversions are possible during xattr set, however, and format 330 * conversion is not atomic with the xattr set that triggers it. We 331 * cannot assume leaf blocks are non-empty until that is addressed. 332 */ 333 buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize; 334 for (i = 0, ent = entries; i < ichdr.count; ent++, i++) { 335 fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr, 336 ent, i, &last_hashval); 337 if (fa) 338 return fa; 339 } 340 341 /* 342 * Quickly check the freemap information. Attribute data has to be 343 * aligned to 4-byte boundaries, and likewise for the free space. 344 * 345 * Note that for 64k block size filesystems, the freemap entries cannot 346 * overflow as they are only be16 fields. However, when checking end 347 * pointer of the freemap, we have to be careful to detect overflows and 348 * so use uint32_t for those checks. 349 */ 350 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 351 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize) 352 return __this_address; 353 if (ichdr.freemap[i].base & 0x3) 354 return __this_address; 355 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize) 356 return __this_address; 357 if (ichdr.freemap[i].size & 0x3) 358 return __this_address; 359 360 /* be care of 16 bit overflows here */ 361 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size; 362 if (end < ichdr.freemap[i].base) 363 return __this_address; 364 if (end > mp->m_attr_geo->blksize) 365 return __this_address; 366 } 367 368 return NULL; 369 } 370 371 static void 372 xfs_attr3_leaf_write_verify( 373 struct xfs_buf *bp) 374 { 375 struct xfs_mount *mp = bp->b_mount; 376 struct xfs_buf_log_item *bip = bp->b_log_item; 377 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr; 378 xfs_failaddr_t fa; 379 380 fa = xfs_attr3_leaf_verify(bp); 381 if (fa) { 382 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 383 return; 384 } 385 386 if (!xfs_sb_version_hascrc(&mp->m_sb)) 387 return; 388 389 if (bip) 390 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); 391 392 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF); 393 } 394 395 /* 396 * leaf/node format detection on trees is sketchy, so a node read can be done on 397 * leaf level blocks when detection identifies the tree as a node format tree 398 * incorrectly. In this case, we need to swap the verifier to match the correct 399 * format of the block being read. 400 */ 401 static void 402 xfs_attr3_leaf_read_verify( 403 struct xfs_buf *bp) 404 { 405 struct xfs_mount *mp = bp->b_mount; 406 xfs_failaddr_t fa; 407 408 if (xfs_sb_version_hascrc(&mp->m_sb) && 409 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF)) 410 xfs_verifier_error(bp, -EFSBADCRC, __this_address); 411 else { 412 fa = xfs_attr3_leaf_verify(bp); 413 if (fa) 414 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 415 } 416 } 417 418 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = { 419 .name = "xfs_attr3_leaf", 420 .magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC), 421 cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) }, 422 .verify_read = xfs_attr3_leaf_read_verify, 423 .verify_write = xfs_attr3_leaf_write_verify, 424 .verify_struct = xfs_attr3_leaf_verify, 425 }; 426 427 int 428 xfs_attr3_leaf_read( 429 struct xfs_trans *tp, 430 struct xfs_inode *dp, 431 xfs_dablk_t bno, 432 struct xfs_buf **bpp) 433 { 434 int err; 435 436 err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK, 437 &xfs_attr3_leaf_buf_ops); 438 if (!err && tp && *bpp) 439 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF); 440 return err; 441 } 442 443 /*======================================================================== 444 * Namespace helper routines 445 *========================================================================*/ 446 447 static bool 448 xfs_attr_match( 449 struct xfs_da_args *args, 450 uint8_t namelen, 451 unsigned char *name, 452 int flags) 453 { 454 if (args->namelen != namelen) 455 return false; 456 if (memcmp(args->name, name, namelen) != 0) 457 return false; 458 /* 459 * If we are looking for incomplete entries, show only those, else only 460 * show complete entries. 461 */ 462 if (args->attr_filter != 463 (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE))) 464 return false; 465 return true; 466 } 467 468 static int 469 xfs_attr_copy_value( 470 struct xfs_da_args *args, 471 unsigned char *value, 472 int valuelen) 473 { 474 /* 475 * No copy if all we have to do is get the length 476 */ 477 if (!args->valuelen) { 478 args->valuelen = valuelen; 479 return 0; 480 } 481 482 /* 483 * No copy if the length of the existing buffer is too small 484 */ 485 if (args->valuelen < valuelen) { 486 args->valuelen = valuelen; 487 return -ERANGE; 488 } 489 490 if (!args->value) { 491 args->value = kmem_alloc_large(valuelen, KM_NOLOCKDEP); 492 if (!args->value) 493 return -ENOMEM; 494 } 495 args->valuelen = valuelen; 496 497 /* remote block xattr requires IO for copy-in */ 498 if (args->rmtblkno) 499 return xfs_attr_rmtval_get(args); 500 501 /* 502 * This is to prevent a GCC warning because the remote xattr case 503 * doesn't have a value to pass in. In that case, we never reach here, 504 * but GCC can't work that out and so throws a "passing NULL to 505 * memcpy" warning. 506 */ 507 if (!value) 508 return -EINVAL; 509 memcpy(args->value, value, valuelen); 510 return 0; 511 } 512 513 /*======================================================================== 514 * External routines when attribute fork size < XFS_LITINO(mp). 515 *========================================================================*/ 516 517 /* 518 * Query whether the total requested number of attr fork bytes of extended 519 * attribute space will be able to fit inline. 520 * 521 * Returns zero if not, else the i_forkoff fork offset to be used in the 522 * literal area for attribute data once the new bytes have been added. 523 * 524 * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value; 525 * special case for dev/uuid inodes, they have fixed size data forks. 526 */ 527 int 528 xfs_attr_shortform_bytesfit( 529 struct xfs_inode *dp, 530 int bytes) 531 { 532 struct xfs_mount *mp = dp->i_mount; 533 int64_t dsize; 534 int minforkoff; 535 int maxforkoff; 536 int offset; 537 538 /* 539 * Check if the new size could fit at all first: 540 */ 541 if (bytes > XFS_LITINO(mp)) 542 return 0; 543 544 /* rounded down */ 545 offset = (XFS_LITINO(mp) - bytes) >> 3; 546 547 if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) { 548 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; 549 return (offset >= minforkoff) ? minforkoff : 0; 550 } 551 552 /* 553 * If the requested numbers of bytes is smaller or equal to the 554 * current attribute fork size we can always proceed. 555 * 556 * Note that if_bytes in the data fork might actually be larger than 557 * the current data fork size is due to delalloc extents. In that 558 * case either the extent count will go down when they are converted 559 * to real extents, or the delalloc conversion will take care of the 560 * literal area rebalancing. 561 */ 562 if (bytes <= XFS_IFORK_ASIZE(dp)) 563 return dp->i_forkoff; 564 565 /* 566 * For attr2 we can try to move the forkoff if there is space in the 567 * literal area, but for the old format we are done if there is no 568 * space in the fixed attribute fork. 569 */ 570 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) 571 return 0; 572 573 dsize = dp->i_df.if_bytes; 574 575 switch (dp->i_df.if_format) { 576 case XFS_DINODE_FMT_EXTENTS: 577 /* 578 * If there is no attr fork and the data fork is extents, 579 * determine if creating the default attr fork will result 580 * in the extents form migrating to btree. If so, the 581 * minimum offset only needs to be the space required for 582 * the btree root. 583 */ 584 if (!dp->i_forkoff && dp->i_df.if_bytes > 585 xfs_default_attroffset(dp)) 586 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS); 587 break; 588 case XFS_DINODE_FMT_BTREE: 589 /* 590 * If we have a data btree then keep forkoff if we have one, 591 * otherwise we are adding a new attr, so then we set 592 * minforkoff to where the btree root can finish so we have 593 * plenty of room for attrs 594 */ 595 if (dp->i_forkoff) { 596 if (offset < dp->i_forkoff) 597 return 0; 598 return dp->i_forkoff; 599 } 600 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot); 601 break; 602 } 603 604 /* 605 * A data fork btree root must have space for at least 606 * MINDBTPTRS key/ptr pairs if the data fork is small or empty. 607 */ 608 minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS)); 609 minforkoff = roundup(minforkoff, 8) >> 3; 610 611 /* attr fork btree root can have at least this many key/ptr pairs */ 612 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS); 613 maxforkoff = maxforkoff >> 3; /* rounded down */ 614 615 if (offset >= maxforkoff) 616 return maxforkoff; 617 if (offset >= minforkoff) 618 return offset; 619 return 0; 620 } 621 622 /* 623 * Switch on the ATTR2 superblock bit (implies also FEATURES2) 624 */ 625 STATIC void 626 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp) 627 { 628 if ((mp->m_flags & XFS_MOUNT_ATTR2) && 629 !(xfs_sb_version_hasattr2(&mp->m_sb))) { 630 spin_lock(&mp->m_sb_lock); 631 if (!xfs_sb_version_hasattr2(&mp->m_sb)) { 632 xfs_sb_version_addattr2(&mp->m_sb); 633 spin_unlock(&mp->m_sb_lock); 634 xfs_log_sb(tp); 635 } else 636 spin_unlock(&mp->m_sb_lock); 637 } 638 } 639 640 /* 641 * Create the initial contents of a shortform attribute list. 642 */ 643 void 644 xfs_attr_shortform_create( 645 struct xfs_da_args *args) 646 { 647 struct xfs_inode *dp = args->dp; 648 struct xfs_ifork *ifp = dp->i_afp; 649 struct xfs_attr_sf_hdr *hdr; 650 651 trace_xfs_attr_sf_create(args); 652 653 ASSERT(ifp->if_bytes == 0); 654 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS) { 655 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */ 656 ifp->if_format = XFS_DINODE_FMT_LOCAL; 657 } 658 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK); 659 hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data; 660 memset(hdr, 0, sizeof(*hdr)); 661 hdr->totsize = cpu_to_be16(sizeof(*hdr)); 662 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA); 663 } 664 665 /* 666 * Return -EEXIST if attr is found, or -ENOATTR if not 667 * args: args containing attribute name and namelen 668 * sfep: If not null, pointer will be set to the last attr entry found on 669 -EEXIST. On -ENOATTR pointer is left at the last entry in the list 670 * basep: If not null, pointer is set to the byte offset of the entry in the 671 * list on -EEXIST. On -ENOATTR, pointer is left at the byte offset of 672 * the last entry in the list 673 */ 674 int 675 xfs_attr_sf_findname( 676 struct xfs_da_args *args, 677 struct xfs_attr_sf_entry **sfep, 678 unsigned int *basep) 679 { 680 struct xfs_attr_shortform *sf; 681 struct xfs_attr_sf_entry *sfe; 682 unsigned int base = sizeof(struct xfs_attr_sf_hdr); 683 int size = 0; 684 int end; 685 int i; 686 687 sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data; 688 sfe = &sf->list[0]; 689 end = sf->hdr.count; 690 for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe), 691 base += size, i++) { 692 size = xfs_attr_sf_entsize(sfe); 693 if (!xfs_attr_match(args, sfe->namelen, sfe->nameval, 694 sfe->flags)) 695 continue; 696 break; 697 } 698 699 if (sfep != NULL) 700 *sfep = sfe; 701 702 if (basep != NULL) 703 *basep = base; 704 705 if (i == end) 706 return -ENOATTR; 707 return -EEXIST; 708 } 709 710 /* 711 * Add a name/value pair to the shortform attribute list. 712 * Overflow from the inode has already been checked for. 713 */ 714 void 715 xfs_attr_shortform_add( 716 struct xfs_da_args *args, 717 int forkoff) 718 { 719 struct xfs_attr_shortform *sf; 720 struct xfs_attr_sf_entry *sfe; 721 int offset, size; 722 struct xfs_mount *mp; 723 struct xfs_inode *dp; 724 struct xfs_ifork *ifp; 725 726 trace_xfs_attr_sf_add(args); 727 728 dp = args->dp; 729 mp = dp->i_mount; 730 dp->i_forkoff = forkoff; 731 732 ifp = dp->i_afp; 733 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL); 734 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 735 if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST) 736 ASSERT(0); 737 738 offset = (char *)sfe - (char *)sf; 739 size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen); 740 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); 741 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 742 sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset); 743 744 sfe->namelen = args->namelen; 745 sfe->valuelen = args->valuelen; 746 sfe->flags = args->attr_filter; 747 memcpy(sfe->nameval, args->name, args->namelen); 748 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen); 749 sf->hdr.count++; 750 be16_add_cpu(&sf->hdr.totsize, size); 751 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA); 752 753 xfs_sbversion_add_attr2(mp, args->trans); 754 } 755 756 /* 757 * After the last attribute is removed revert to original inode format, 758 * making all literal area available to the data fork once more. 759 */ 760 void 761 xfs_attr_fork_remove( 762 struct xfs_inode *ip, 763 struct xfs_trans *tp) 764 { 765 ASSERT(ip->i_afp->if_nextents == 0); 766 767 xfs_idestroy_fork(ip->i_afp); 768 kmem_cache_free(xfs_ifork_zone, ip->i_afp); 769 ip->i_afp = NULL; 770 ip->i_forkoff = 0; 771 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 772 } 773 774 /* 775 * Remove an attribute from the shortform attribute list structure. 776 */ 777 int 778 xfs_attr_shortform_remove( 779 struct xfs_da_args *args) 780 { 781 struct xfs_attr_shortform *sf; 782 struct xfs_attr_sf_entry *sfe; 783 int size = 0, end, totsize; 784 unsigned int base; 785 struct xfs_mount *mp; 786 struct xfs_inode *dp; 787 int error; 788 789 trace_xfs_attr_sf_remove(args); 790 791 dp = args->dp; 792 mp = dp->i_mount; 793 sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data; 794 795 error = xfs_attr_sf_findname(args, &sfe, &base); 796 if (error != -EEXIST) 797 return error; 798 size = xfs_attr_sf_entsize(sfe); 799 800 /* 801 * Fix up the attribute fork data, covering the hole 802 */ 803 end = base + size; 804 totsize = be16_to_cpu(sf->hdr.totsize); 805 if (end != totsize) 806 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end); 807 sf->hdr.count--; 808 be16_add_cpu(&sf->hdr.totsize, -size); 809 810 /* 811 * Fix up the start offset of the attribute fork 812 */ 813 totsize -= size; 814 if (totsize == sizeof(xfs_attr_sf_hdr_t) && 815 (mp->m_flags & XFS_MOUNT_ATTR2) && 816 (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) && 817 !(args->op_flags & XFS_DA_OP_ADDNAME)) { 818 xfs_attr_fork_remove(dp, args->trans); 819 } else { 820 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK); 821 dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize); 822 ASSERT(dp->i_forkoff); 823 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || 824 (args->op_flags & XFS_DA_OP_ADDNAME) || 825 !(mp->m_flags & XFS_MOUNT_ATTR2) || 826 dp->i_df.if_format == XFS_DINODE_FMT_BTREE); 827 xfs_trans_log_inode(args->trans, dp, 828 XFS_ILOG_CORE | XFS_ILOG_ADATA); 829 } 830 831 xfs_sbversion_add_attr2(mp, args->trans); 832 833 return 0; 834 } 835 836 /* 837 * Look up a name in a shortform attribute list structure. 838 */ 839 /*ARGSUSED*/ 840 int 841 xfs_attr_shortform_lookup(xfs_da_args_t *args) 842 { 843 struct xfs_attr_shortform *sf; 844 struct xfs_attr_sf_entry *sfe; 845 int i; 846 struct xfs_ifork *ifp; 847 848 trace_xfs_attr_sf_lookup(args); 849 850 ifp = args->dp->i_afp; 851 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL); 852 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 853 sfe = &sf->list[0]; 854 for (i = 0; i < sf->hdr.count; 855 sfe = xfs_attr_sf_nextentry(sfe), i++) { 856 if (xfs_attr_match(args, sfe->namelen, sfe->nameval, 857 sfe->flags)) 858 return -EEXIST; 859 } 860 return -ENOATTR; 861 } 862 863 /* 864 * Retrieve the attribute value and length. 865 * 866 * If args->valuelen is zero, only the length needs to be returned. Unlike a 867 * lookup, we only return an error if the attribute does not exist or we can't 868 * retrieve the value. 869 */ 870 int 871 xfs_attr_shortform_getvalue( 872 struct xfs_da_args *args) 873 { 874 struct xfs_attr_shortform *sf; 875 struct xfs_attr_sf_entry *sfe; 876 int i; 877 878 ASSERT(args->dp->i_afp->if_format == XFS_DINODE_FMT_LOCAL); 879 sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data; 880 sfe = &sf->list[0]; 881 for (i = 0; i < sf->hdr.count; 882 sfe = xfs_attr_sf_nextentry(sfe), i++) { 883 if (xfs_attr_match(args, sfe->namelen, sfe->nameval, 884 sfe->flags)) 885 return xfs_attr_copy_value(args, 886 &sfe->nameval[args->namelen], sfe->valuelen); 887 } 888 return -ENOATTR; 889 } 890 891 /* 892 * Convert from using the shortform to the leaf. On success, return the 893 * buffer so that we can keep it locked until we're totally done with it. 894 */ 895 int 896 xfs_attr_shortform_to_leaf( 897 struct xfs_da_args *args, 898 struct xfs_buf **leaf_bp) 899 { 900 struct xfs_inode *dp; 901 struct xfs_attr_shortform *sf; 902 struct xfs_attr_sf_entry *sfe; 903 struct xfs_da_args nargs; 904 char *tmpbuffer; 905 int error, i, size; 906 xfs_dablk_t blkno; 907 struct xfs_buf *bp; 908 struct xfs_ifork *ifp; 909 910 trace_xfs_attr_sf_to_leaf(args); 911 912 dp = args->dp; 913 ifp = dp->i_afp; 914 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 915 size = be16_to_cpu(sf->hdr.totsize); 916 tmpbuffer = kmem_alloc(size, 0); 917 ASSERT(tmpbuffer != NULL); 918 memcpy(tmpbuffer, ifp->if_u1.if_data, size); 919 sf = (struct xfs_attr_shortform *)tmpbuffer; 920 921 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK); 922 xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK); 923 924 bp = NULL; 925 error = xfs_da_grow_inode(args, &blkno); 926 if (error) 927 goto out; 928 929 ASSERT(blkno == 0); 930 error = xfs_attr3_leaf_create(args, blkno, &bp); 931 if (error) 932 goto out; 933 934 memset((char *)&nargs, 0, sizeof(nargs)); 935 nargs.dp = dp; 936 nargs.geo = args->geo; 937 nargs.total = args->total; 938 nargs.whichfork = XFS_ATTR_FORK; 939 nargs.trans = args->trans; 940 nargs.op_flags = XFS_DA_OP_OKNOENT; 941 942 sfe = &sf->list[0]; 943 for (i = 0; i < sf->hdr.count; i++) { 944 nargs.name = sfe->nameval; 945 nargs.namelen = sfe->namelen; 946 nargs.value = &sfe->nameval[nargs.namelen]; 947 nargs.valuelen = sfe->valuelen; 948 nargs.hashval = xfs_da_hashname(sfe->nameval, 949 sfe->namelen); 950 nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK; 951 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */ 952 ASSERT(error == -ENOATTR); 953 error = xfs_attr3_leaf_add(bp, &nargs); 954 ASSERT(error != -ENOSPC); 955 if (error) 956 goto out; 957 sfe = xfs_attr_sf_nextentry(sfe); 958 } 959 error = 0; 960 *leaf_bp = bp; 961 out: 962 kmem_free(tmpbuffer); 963 return error; 964 } 965 966 /* 967 * Check a leaf attribute block to see if all the entries would fit into 968 * a shortform attribute list. 969 */ 970 int 971 xfs_attr_shortform_allfit( 972 struct xfs_buf *bp, 973 struct xfs_inode *dp) 974 { 975 struct xfs_attr_leafblock *leaf; 976 struct xfs_attr_leaf_entry *entry; 977 xfs_attr_leaf_name_local_t *name_loc; 978 struct xfs_attr3_icleaf_hdr leafhdr; 979 int bytes; 980 int i; 981 struct xfs_mount *mp = bp->b_mount; 982 983 leaf = bp->b_addr; 984 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf); 985 entry = xfs_attr3_leaf_entryp(leaf); 986 987 bytes = sizeof(struct xfs_attr_sf_hdr); 988 for (i = 0; i < leafhdr.count; entry++, i++) { 989 if (entry->flags & XFS_ATTR_INCOMPLETE) 990 continue; /* don't copy partial entries */ 991 if (!(entry->flags & XFS_ATTR_LOCAL)) 992 return 0; 993 name_loc = xfs_attr3_leaf_name_local(leaf, i); 994 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX) 995 return 0; 996 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX) 997 return 0; 998 bytes += xfs_attr_sf_entsize_byname(name_loc->namelen, 999 be16_to_cpu(name_loc->valuelen)); 1000 } 1001 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) && 1002 (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) && 1003 (bytes == sizeof(struct xfs_attr_sf_hdr))) 1004 return -1; 1005 return xfs_attr_shortform_bytesfit(dp, bytes); 1006 } 1007 1008 /* Verify the consistency of an inline attribute fork. */ 1009 xfs_failaddr_t 1010 xfs_attr_shortform_verify( 1011 struct xfs_inode *ip) 1012 { 1013 struct xfs_attr_shortform *sfp; 1014 struct xfs_attr_sf_entry *sfep; 1015 struct xfs_attr_sf_entry *next_sfep; 1016 char *endp; 1017 struct xfs_ifork *ifp; 1018 int i; 1019 int64_t size; 1020 1021 ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL); 1022 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK); 1023 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 1024 size = ifp->if_bytes; 1025 1026 /* 1027 * Give up if the attribute is way too short. 1028 */ 1029 if (size < sizeof(struct xfs_attr_sf_hdr)) 1030 return __this_address; 1031 1032 endp = (char *)sfp + size; 1033 1034 /* Check all reported entries */ 1035 sfep = &sfp->list[0]; 1036 for (i = 0; i < sfp->hdr.count; i++) { 1037 /* 1038 * struct xfs_attr_sf_entry has a variable length. 1039 * Check the fixed-offset parts of the structure are 1040 * within the data buffer. 1041 * xfs_attr_sf_entry is defined with a 1-byte variable 1042 * array at the end, so we must subtract that off. 1043 */ 1044 if (((char *)sfep + sizeof(*sfep)) >= endp) 1045 return __this_address; 1046 1047 /* Don't allow names with known bad length. */ 1048 if (sfep->namelen == 0) 1049 return __this_address; 1050 1051 /* 1052 * Check that the variable-length part of the structure is 1053 * within the data buffer. The next entry starts after the 1054 * name component, so nextentry is an acceptable test. 1055 */ 1056 next_sfep = xfs_attr_sf_nextentry(sfep); 1057 if ((char *)next_sfep > endp) 1058 return __this_address; 1059 1060 /* 1061 * Check for unknown flags. Short form doesn't support 1062 * the incomplete or local bits, so we can use the namespace 1063 * mask here. 1064 */ 1065 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK) 1066 return __this_address; 1067 1068 /* 1069 * Check for invalid namespace combinations. We only allow 1070 * one namespace flag per xattr, so we can just count the 1071 * bits (i.e. hweight) here. 1072 */ 1073 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1) 1074 return __this_address; 1075 1076 sfep = next_sfep; 1077 } 1078 if ((void *)sfep != (void *)endp) 1079 return __this_address; 1080 1081 return NULL; 1082 } 1083 1084 /* 1085 * Convert a leaf attribute list to shortform attribute list 1086 */ 1087 int 1088 xfs_attr3_leaf_to_shortform( 1089 struct xfs_buf *bp, 1090 struct xfs_da_args *args, 1091 int forkoff) 1092 { 1093 struct xfs_attr_leafblock *leaf; 1094 struct xfs_attr3_icleaf_hdr ichdr; 1095 struct xfs_attr_leaf_entry *entry; 1096 struct xfs_attr_leaf_name_local *name_loc; 1097 struct xfs_da_args nargs; 1098 struct xfs_inode *dp = args->dp; 1099 char *tmpbuffer; 1100 int error; 1101 int i; 1102 1103 trace_xfs_attr_leaf_to_sf(args); 1104 1105 tmpbuffer = kmem_alloc(args->geo->blksize, 0); 1106 if (!tmpbuffer) 1107 return -ENOMEM; 1108 1109 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize); 1110 1111 leaf = (xfs_attr_leafblock_t *)tmpbuffer; 1112 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 1113 entry = xfs_attr3_leaf_entryp(leaf); 1114 1115 /* XXX (dgc): buffer is about to be marked stale - why zero it? */ 1116 memset(bp->b_addr, 0, args->geo->blksize); 1117 1118 /* 1119 * Clean out the prior contents of the attribute list. 1120 */ 1121 error = xfs_da_shrink_inode(args, 0, bp); 1122 if (error) 1123 goto out; 1124 1125 if (forkoff == -1) { 1126 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2); 1127 ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE); 1128 xfs_attr_fork_remove(dp, args->trans); 1129 goto out; 1130 } 1131 1132 xfs_attr_shortform_create(args); 1133 1134 /* 1135 * Copy the attributes 1136 */ 1137 memset((char *)&nargs, 0, sizeof(nargs)); 1138 nargs.geo = args->geo; 1139 nargs.dp = dp; 1140 nargs.total = args->total; 1141 nargs.whichfork = XFS_ATTR_FORK; 1142 nargs.trans = args->trans; 1143 nargs.op_flags = XFS_DA_OP_OKNOENT; 1144 1145 for (i = 0; i < ichdr.count; entry++, i++) { 1146 if (entry->flags & XFS_ATTR_INCOMPLETE) 1147 continue; /* don't copy partial entries */ 1148 if (!entry->nameidx) 1149 continue; 1150 ASSERT(entry->flags & XFS_ATTR_LOCAL); 1151 name_loc = xfs_attr3_leaf_name_local(leaf, i); 1152 nargs.name = name_loc->nameval; 1153 nargs.namelen = name_loc->namelen; 1154 nargs.value = &name_loc->nameval[nargs.namelen]; 1155 nargs.valuelen = be16_to_cpu(name_loc->valuelen); 1156 nargs.hashval = be32_to_cpu(entry->hashval); 1157 nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK; 1158 xfs_attr_shortform_add(&nargs, forkoff); 1159 } 1160 error = 0; 1161 1162 out: 1163 kmem_free(tmpbuffer); 1164 return error; 1165 } 1166 1167 /* 1168 * Convert from using a single leaf to a root node and a leaf. 1169 */ 1170 int 1171 xfs_attr3_leaf_to_node( 1172 struct xfs_da_args *args) 1173 { 1174 struct xfs_attr_leafblock *leaf; 1175 struct xfs_attr3_icleaf_hdr icleafhdr; 1176 struct xfs_attr_leaf_entry *entries; 1177 struct xfs_da3_icnode_hdr icnodehdr; 1178 struct xfs_da_intnode *node; 1179 struct xfs_inode *dp = args->dp; 1180 struct xfs_mount *mp = dp->i_mount; 1181 struct xfs_buf *bp1 = NULL; 1182 struct xfs_buf *bp2 = NULL; 1183 xfs_dablk_t blkno; 1184 int error; 1185 1186 trace_xfs_attr_leaf_to_node(args); 1187 1188 error = xfs_da_grow_inode(args, &blkno); 1189 if (error) 1190 goto out; 1191 error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1); 1192 if (error) 1193 goto out; 1194 1195 error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK); 1196 if (error) 1197 goto out; 1198 1199 /* copy leaf to new buffer, update identifiers */ 1200 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF); 1201 bp2->b_ops = bp1->b_ops; 1202 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize); 1203 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1204 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr; 1205 hdr3->blkno = cpu_to_be64(bp2->b_bn); 1206 } 1207 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1); 1208 1209 /* 1210 * Set up the new root node. 1211 */ 1212 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK); 1213 if (error) 1214 goto out; 1215 node = bp1->b_addr; 1216 xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node); 1217 1218 leaf = bp2->b_addr; 1219 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf); 1220 entries = xfs_attr3_leaf_entryp(leaf); 1221 1222 /* both on-disk, don't endian-flip twice */ 1223 icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval; 1224 icnodehdr.btree[0].before = cpu_to_be32(blkno); 1225 icnodehdr.count = 1; 1226 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr); 1227 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1); 1228 error = 0; 1229 out: 1230 return error; 1231 } 1232 1233 /*======================================================================== 1234 * Routines used for growing the Btree. 1235 *========================================================================*/ 1236 1237 /* 1238 * Create the initial contents of a leaf attribute list 1239 * or a leaf in a node attribute list. 1240 */ 1241 STATIC int 1242 xfs_attr3_leaf_create( 1243 struct xfs_da_args *args, 1244 xfs_dablk_t blkno, 1245 struct xfs_buf **bpp) 1246 { 1247 struct xfs_attr_leafblock *leaf; 1248 struct xfs_attr3_icleaf_hdr ichdr; 1249 struct xfs_inode *dp = args->dp; 1250 struct xfs_mount *mp = dp->i_mount; 1251 struct xfs_buf *bp; 1252 int error; 1253 1254 trace_xfs_attr_leaf_create(args); 1255 1256 error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp, 1257 XFS_ATTR_FORK); 1258 if (error) 1259 return error; 1260 bp->b_ops = &xfs_attr3_leaf_buf_ops; 1261 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF); 1262 leaf = bp->b_addr; 1263 memset(leaf, 0, args->geo->blksize); 1264 1265 memset(&ichdr, 0, sizeof(ichdr)); 1266 ichdr.firstused = args->geo->blksize; 1267 1268 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1269 struct xfs_da3_blkinfo *hdr3 = bp->b_addr; 1270 1271 ichdr.magic = XFS_ATTR3_LEAF_MAGIC; 1272 1273 hdr3->blkno = cpu_to_be64(bp->b_bn); 1274 hdr3->owner = cpu_to_be64(dp->i_ino); 1275 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid); 1276 1277 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr); 1278 } else { 1279 ichdr.magic = XFS_ATTR_LEAF_MAGIC; 1280 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr); 1281 } 1282 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base; 1283 1284 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 1285 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1); 1286 1287 *bpp = bp; 1288 return 0; 1289 } 1290 1291 /* 1292 * Split the leaf node, rebalance, then add the new entry. 1293 */ 1294 int 1295 xfs_attr3_leaf_split( 1296 struct xfs_da_state *state, 1297 struct xfs_da_state_blk *oldblk, 1298 struct xfs_da_state_blk *newblk) 1299 { 1300 xfs_dablk_t blkno; 1301 int error; 1302 1303 trace_xfs_attr_leaf_split(state->args); 1304 1305 /* 1306 * Allocate space for a new leaf node. 1307 */ 1308 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC); 1309 error = xfs_da_grow_inode(state->args, &blkno); 1310 if (error) 1311 return error; 1312 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp); 1313 if (error) 1314 return error; 1315 newblk->blkno = blkno; 1316 newblk->magic = XFS_ATTR_LEAF_MAGIC; 1317 1318 /* 1319 * Rebalance the entries across the two leaves. 1320 * NOTE: rebalance() currently depends on the 2nd block being empty. 1321 */ 1322 xfs_attr3_leaf_rebalance(state, oldblk, newblk); 1323 error = xfs_da3_blk_link(state, oldblk, newblk); 1324 if (error) 1325 return error; 1326 1327 /* 1328 * Save info on "old" attribute for "atomic rename" ops, leaf_add() 1329 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the 1330 * "new" attrs info. Will need the "old" info to remove it later. 1331 * 1332 * Insert the "new" entry in the correct block. 1333 */ 1334 if (state->inleaf) { 1335 trace_xfs_attr_leaf_add_old(state->args); 1336 error = xfs_attr3_leaf_add(oldblk->bp, state->args); 1337 } else { 1338 trace_xfs_attr_leaf_add_new(state->args); 1339 error = xfs_attr3_leaf_add(newblk->bp, state->args); 1340 } 1341 1342 /* 1343 * Update last hashval in each block since we added the name. 1344 */ 1345 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL); 1346 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL); 1347 return error; 1348 } 1349 1350 /* 1351 * Add a name to the leaf attribute list structure. 1352 */ 1353 int 1354 xfs_attr3_leaf_add( 1355 struct xfs_buf *bp, 1356 struct xfs_da_args *args) 1357 { 1358 struct xfs_attr_leafblock *leaf; 1359 struct xfs_attr3_icleaf_hdr ichdr; 1360 int tablesize; 1361 int entsize; 1362 int sum; 1363 int tmp; 1364 int i; 1365 1366 trace_xfs_attr_leaf_add(args); 1367 1368 leaf = bp->b_addr; 1369 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 1370 ASSERT(args->index >= 0 && args->index <= ichdr.count); 1371 entsize = xfs_attr_leaf_newentsize(args, NULL); 1372 1373 /* 1374 * Search through freemap for first-fit on new name length. 1375 * (may need to figure in size of entry struct too) 1376 */ 1377 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t) 1378 + xfs_attr3_leaf_hdr_size(leaf); 1379 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) { 1380 if (tablesize > ichdr.firstused) { 1381 sum += ichdr.freemap[i].size; 1382 continue; 1383 } 1384 if (!ichdr.freemap[i].size) 1385 continue; /* no space in this map */ 1386 tmp = entsize; 1387 if (ichdr.freemap[i].base < ichdr.firstused) 1388 tmp += sizeof(xfs_attr_leaf_entry_t); 1389 if (ichdr.freemap[i].size >= tmp) { 1390 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i); 1391 goto out_log_hdr; 1392 } 1393 sum += ichdr.freemap[i].size; 1394 } 1395 1396 /* 1397 * If there are no holes in the address space of the block, 1398 * and we don't have enough freespace, then compaction will do us 1399 * no good and we should just give up. 1400 */ 1401 if (!ichdr.holes && sum < entsize) 1402 return -ENOSPC; 1403 1404 /* 1405 * Compact the entries to coalesce free space. 1406 * This may change the hdr->count via dropping INCOMPLETE entries. 1407 */ 1408 xfs_attr3_leaf_compact(args, &ichdr, bp); 1409 1410 /* 1411 * After compaction, the block is guaranteed to have only one 1412 * free region, in freemap[0]. If it is not big enough, give up. 1413 */ 1414 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) { 1415 tmp = -ENOSPC; 1416 goto out_log_hdr; 1417 } 1418 1419 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0); 1420 1421 out_log_hdr: 1422 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 1423 xfs_trans_log_buf(args->trans, bp, 1424 XFS_DA_LOGRANGE(leaf, &leaf->hdr, 1425 xfs_attr3_leaf_hdr_size(leaf))); 1426 return tmp; 1427 } 1428 1429 /* 1430 * Add a name to a leaf attribute list structure. 1431 */ 1432 STATIC int 1433 xfs_attr3_leaf_add_work( 1434 struct xfs_buf *bp, 1435 struct xfs_attr3_icleaf_hdr *ichdr, 1436 struct xfs_da_args *args, 1437 int mapindex) 1438 { 1439 struct xfs_attr_leafblock *leaf; 1440 struct xfs_attr_leaf_entry *entry; 1441 struct xfs_attr_leaf_name_local *name_loc; 1442 struct xfs_attr_leaf_name_remote *name_rmt; 1443 struct xfs_mount *mp; 1444 int tmp; 1445 int i; 1446 1447 trace_xfs_attr_leaf_add_work(args); 1448 1449 leaf = bp->b_addr; 1450 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE); 1451 ASSERT(args->index >= 0 && args->index <= ichdr->count); 1452 1453 /* 1454 * Force open some space in the entry array and fill it in. 1455 */ 1456 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 1457 if (args->index < ichdr->count) { 1458 tmp = ichdr->count - args->index; 1459 tmp *= sizeof(xfs_attr_leaf_entry_t); 1460 memmove(entry + 1, entry, tmp); 1461 xfs_trans_log_buf(args->trans, bp, 1462 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry))); 1463 } 1464 ichdr->count++; 1465 1466 /* 1467 * Allocate space for the new string (at the end of the run). 1468 */ 1469 mp = args->trans->t_mountp; 1470 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize); 1471 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0); 1472 ASSERT(ichdr->freemap[mapindex].size >= 1473 xfs_attr_leaf_newentsize(args, NULL)); 1474 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize); 1475 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0); 1476 1477 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp); 1478 1479 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base + 1480 ichdr->freemap[mapindex].size); 1481 entry->hashval = cpu_to_be32(args->hashval); 1482 entry->flags = args->attr_filter; 1483 if (tmp) 1484 entry->flags |= XFS_ATTR_LOCAL; 1485 if (args->op_flags & XFS_DA_OP_RENAME) { 1486 entry->flags |= XFS_ATTR_INCOMPLETE; 1487 if ((args->blkno2 == args->blkno) && 1488 (args->index2 <= args->index)) { 1489 args->index2++; 1490 } 1491 } 1492 xfs_trans_log_buf(args->trans, bp, 1493 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 1494 ASSERT((args->index == 0) || 1495 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval))); 1496 ASSERT((args->index == ichdr->count - 1) || 1497 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval))); 1498 1499 /* 1500 * For "remote" attribute values, simply note that we need to 1501 * allocate space for the "remote" value. We can't actually 1502 * allocate the extents in this transaction, and we can't decide 1503 * which blocks they should be as we might allocate more blocks 1504 * as part of this transaction (a split operation for example). 1505 */ 1506 if (entry->flags & XFS_ATTR_LOCAL) { 1507 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 1508 name_loc->namelen = args->namelen; 1509 name_loc->valuelen = cpu_to_be16(args->valuelen); 1510 memcpy((char *)name_loc->nameval, args->name, args->namelen); 1511 memcpy((char *)&name_loc->nameval[args->namelen], args->value, 1512 be16_to_cpu(name_loc->valuelen)); 1513 } else { 1514 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 1515 name_rmt->namelen = args->namelen; 1516 memcpy((char *)name_rmt->name, args->name, args->namelen); 1517 entry->flags |= XFS_ATTR_INCOMPLETE; 1518 /* just in case */ 1519 name_rmt->valuelen = 0; 1520 name_rmt->valueblk = 0; 1521 args->rmtblkno = 1; 1522 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen); 1523 args->rmtvaluelen = args->valuelen; 1524 } 1525 xfs_trans_log_buf(args->trans, bp, 1526 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index), 1527 xfs_attr_leaf_entsize(leaf, args->index))); 1528 1529 /* 1530 * Update the control info for this leaf node 1531 */ 1532 if (be16_to_cpu(entry->nameidx) < ichdr->firstused) 1533 ichdr->firstused = be16_to_cpu(entry->nameidx); 1534 1535 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t) 1536 + xfs_attr3_leaf_hdr_size(leaf)); 1537 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t) 1538 + xfs_attr3_leaf_hdr_size(leaf); 1539 1540 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 1541 if (ichdr->freemap[i].base == tmp) { 1542 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t); 1543 ichdr->freemap[i].size -= 1544 min_t(uint16_t, ichdr->freemap[i].size, 1545 sizeof(xfs_attr_leaf_entry_t)); 1546 } 1547 } 1548 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index); 1549 return 0; 1550 } 1551 1552 /* 1553 * Garbage collect a leaf attribute list block by copying it to a new buffer. 1554 */ 1555 STATIC void 1556 xfs_attr3_leaf_compact( 1557 struct xfs_da_args *args, 1558 struct xfs_attr3_icleaf_hdr *ichdr_dst, 1559 struct xfs_buf *bp) 1560 { 1561 struct xfs_attr_leafblock *leaf_src; 1562 struct xfs_attr_leafblock *leaf_dst; 1563 struct xfs_attr3_icleaf_hdr ichdr_src; 1564 struct xfs_trans *trans = args->trans; 1565 char *tmpbuffer; 1566 1567 trace_xfs_attr_leaf_compact(args); 1568 1569 tmpbuffer = kmem_alloc(args->geo->blksize, 0); 1570 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize); 1571 memset(bp->b_addr, 0, args->geo->blksize); 1572 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer; 1573 leaf_dst = bp->b_addr; 1574 1575 /* 1576 * Copy the on-disk header back into the destination buffer to ensure 1577 * all the information in the header that is not part of the incore 1578 * header structure is preserved. 1579 */ 1580 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src)); 1581 1582 /* Initialise the incore headers */ 1583 ichdr_src = *ichdr_dst; /* struct copy */ 1584 ichdr_dst->firstused = args->geo->blksize; 1585 ichdr_dst->usedbytes = 0; 1586 ichdr_dst->count = 0; 1587 ichdr_dst->holes = 0; 1588 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src); 1589 ichdr_dst->freemap[0].size = ichdr_dst->firstused - 1590 ichdr_dst->freemap[0].base; 1591 1592 /* write the header back to initialise the underlying buffer */ 1593 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst); 1594 1595 /* 1596 * Copy all entry's in the same (sorted) order, 1597 * but allocate name/value pairs packed and in sequence. 1598 */ 1599 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0, 1600 leaf_dst, ichdr_dst, 0, ichdr_src.count); 1601 /* 1602 * this logs the entire buffer, but the caller must write the header 1603 * back to the buffer when it is finished modifying it. 1604 */ 1605 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1); 1606 1607 kmem_free(tmpbuffer); 1608 } 1609 1610 /* 1611 * Compare two leaf blocks "order". 1612 * Return 0 unless leaf2 should go before leaf1. 1613 */ 1614 static int 1615 xfs_attr3_leaf_order( 1616 struct xfs_buf *leaf1_bp, 1617 struct xfs_attr3_icleaf_hdr *leaf1hdr, 1618 struct xfs_buf *leaf2_bp, 1619 struct xfs_attr3_icleaf_hdr *leaf2hdr) 1620 { 1621 struct xfs_attr_leaf_entry *entries1; 1622 struct xfs_attr_leaf_entry *entries2; 1623 1624 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr); 1625 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr); 1626 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 && 1627 ((be32_to_cpu(entries2[0].hashval) < 1628 be32_to_cpu(entries1[0].hashval)) || 1629 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) < 1630 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) { 1631 return 1; 1632 } 1633 return 0; 1634 } 1635 1636 int 1637 xfs_attr_leaf_order( 1638 struct xfs_buf *leaf1_bp, 1639 struct xfs_buf *leaf2_bp) 1640 { 1641 struct xfs_attr3_icleaf_hdr ichdr1; 1642 struct xfs_attr3_icleaf_hdr ichdr2; 1643 struct xfs_mount *mp = leaf1_bp->b_mount; 1644 1645 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr); 1646 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr); 1647 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2); 1648 } 1649 1650 /* 1651 * Redistribute the attribute list entries between two leaf nodes, 1652 * taking into account the size of the new entry. 1653 * 1654 * NOTE: if new block is empty, then it will get the upper half of the 1655 * old block. At present, all (one) callers pass in an empty second block. 1656 * 1657 * This code adjusts the args->index/blkno and args->index2/blkno2 fields 1658 * to match what it is doing in splitting the attribute leaf block. Those 1659 * values are used in "atomic rename" operations on attributes. Note that 1660 * the "new" and "old" values can end up in different blocks. 1661 */ 1662 STATIC void 1663 xfs_attr3_leaf_rebalance( 1664 struct xfs_da_state *state, 1665 struct xfs_da_state_blk *blk1, 1666 struct xfs_da_state_blk *blk2) 1667 { 1668 struct xfs_da_args *args; 1669 struct xfs_attr_leafblock *leaf1; 1670 struct xfs_attr_leafblock *leaf2; 1671 struct xfs_attr3_icleaf_hdr ichdr1; 1672 struct xfs_attr3_icleaf_hdr ichdr2; 1673 struct xfs_attr_leaf_entry *entries1; 1674 struct xfs_attr_leaf_entry *entries2; 1675 int count; 1676 int totallen; 1677 int max; 1678 int space; 1679 int swap; 1680 1681 /* 1682 * Set up environment. 1683 */ 1684 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC); 1685 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC); 1686 leaf1 = blk1->bp->b_addr; 1687 leaf2 = blk2->bp->b_addr; 1688 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1); 1689 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2); 1690 ASSERT(ichdr2.count == 0); 1691 args = state->args; 1692 1693 trace_xfs_attr_leaf_rebalance(args); 1694 1695 /* 1696 * Check ordering of blocks, reverse if it makes things simpler. 1697 * 1698 * NOTE: Given that all (current) callers pass in an empty 1699 * second block, this code should never set "swap". 1700 */ 1701 swap = 0; 1702 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) { 1703 swap(blk1, blk2); 1704 1705 /* swap structures rather than reconverting them */ 1706 swap(ichdr1, ichdr2); 1707 1708 leaf1 = blk1->bp->b_addr; 1709 leaf2 = blk2->bp->b_addr; 1710 swap = 1; 1711 } 1712 1713 /* 1714 * Examine entries until we reduce the absolute difference in 1715 * byte usage between the two blocks to a minimum. Then get 1716 * the direction to copy and the number of elements to move. 1717 * 1718 * "inleaf" is true if the new entry should be inserted into blk1. 1719 * If "swap" is also true, then reverse the sense of "inleaf". 1720 */ 1721 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1, 1722 blk2, &ichdr2, 1723 &count, &totallen); 1724 if (swap) 1725 state->inleaf = !state->inleaf; 1726 1727 /* 1728 * Move any entries required from leaf to leaf: 1729 */ 1730 if (count < ichdr1.count) { 1731 /* 1732 * Figure the total bytes to be added to the destination leaf. 1733 */ 1734 /* number entries being moved */ 1735 count = ichdr1.count - count; 1736 space = ichdr1.usedbytes - totallen; 1737 space += count * sizeof(xfs_attr_leaf_entry_t); 1738 1739 /* 1740 * leaf2 is the destination, compact it if it looks tight. 1741 */ 1742 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1); 1743 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t); 1744 if (space > max) 1745 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp); 1746 1747 /* 1748 * Move high entries from leaf1 to low end of leaf2. 1749 */ 1750 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1, 1751 ichdr1.count - count, leaf2, &ichdr2, 0, count); 1752 1753 } else if (count > ichdr1.count) { 1754 /* 1755 * I assert that since all callers pass in an empty 1756 * second buffer, this code should never execute. 1757 */ 1758 ASSERT(0); 1759 1760 /* 1761 * Figure the total bytes to be added to the destination leaf. 1762 */ 1763 /* number entries being moved */ 1764 count -= ichdr1.count; 1765 space = totallen - ichdr1.usedbytes; 1766 space += count * sizeof(xfs_attr_leaf_entry_t); 1767 1768 /* 1769 * leaf1 is the destination, compact it if it looks tight. 1770 */ 1771 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1); 1772 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t); 1773 if (space > max) 1774 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp); 1775 1776 /* 1777 * Move low entries from leaf2 to high end of leaf1. 1778 */ 1779 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1, 1780 ichdr1.count, count); 1781 } 1782 1783 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1); 1784 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2); 1785 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1); 1786 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1); 1787 1788 /* 1789 * Copy out last hashval in each block for B-tree code. 1790 */ 1791 entries1 = xfs_attr3_leaf_entryp(leaf1); 1792 entries2 = xfs_attr3_leaf_entryp(leaf2); 1793 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval); 1794 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval); 1795 1796 /* 1797 * Adjust the expected index for insertion. 1798 * NOTE: this code depends on the (current) situation that the 1799 * second block was originally empty. 1800 * 1801 * If the insertion point moved to the 2nd block, we must adjust 1802 * the index. We must also track the entry just following the 1803 * new entry for use in an "atomic rename" operation, that entry 1804 * is always the "old" entry and the "new" entry is what we are 1805 * inserting. The index/blkno fields refer to the "old" entry, 1806 * while the index2/blkno2 fields refer to the "new" entry. 1807 */ 1808 if (blk1->index > ichdr1.count) { 1809 ASSERT(state->inleaf == 0); 1810 blk2->index = blk1->index - ichdr1.count; 1811 args->index = args->index2 = blk2->index; 1812 args->blkno = args->blkno2 = blk2->blkno; 1813 } else if (blk1->index == ichdr1.count) { 1814 if (state->inleaf) { 1815 args->index = blk1->index; 1816 args->blkno = blk1->blkno; 1817 args->index2 = 0; 1818 args->blkno2 = blk2->blkno; 1819 } else { 1820 /* 1821 * On a double leaf split, the original attr location 1822 * is already stored in blkno2/index2, so don't 1823 * overwrite it overwise we corrupt the tree. 1824 */ 1825 blk2->index = blk1->index - ichdr1.count; 1826 args->index = blk2->index; 1827 args->blkno = blk2->blkno; 1828 if (!state->extravalid) { 1829 /* 1830 * set the new attr location to match the old 1831 * one and let the higher level split code 1832 * decide where in the leaf to place it. 1833 */ 1834 args->index2 = blk2->index; 1835 args->blkno2 = blk2->blkno; 1836 } 1837 } 1838 } else { 1839 ASSERT(state->inleaf == 1); 1840 args->index = args->index2 = blk1->index; 1841 args->blkno = args->blkno2 = blk1->blkno; 1842 } 1843 } 1844 1845 /* 1846 * Examine entries until we reduce the absolute difference in 1847 * byte usage between the two blocks to a minimum. 1848 * GROT: Is this really necessary? With other than a 512 byte blocksize, 1849 * GROT: there will always be enough room in either block for a new entry. 1850 * GROT: Do a double-split for this case? 1851 */ 1852 STATIC int 1853 xfs_attr3_leaf_figure_balance( 1854 struct xfs_da_state *state, 1855 struct xfs_da_state_blk *blk1, 1856 struct xfs_attr3_icleaf_hdr *ichdr1, 1857 struct xfs_da_state_blk *blk2, 1858 struct xfs_attr3_icleaf_hdr *ichdr2, 1859 int *countarg, 1860 int *usedbytesarg) 1861 { 1862 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr; 1863 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr; 1864 struct xfs_attr_leaf_entry *entry; 1865 int count; 1866 int max; 1867 int index; 1868 int totallen = 0; 1869 int half; 1870 int lastdelta; 1871 int foundit = 0; 1872 int tmp; 1873 1874 /* 1875 * Examine entries until we reduce the absolute difference in 1876 * byte usage between the two blocks to a minimum. 1877 */ 1878 max = ichdr1->count + ichdr2->count; 1879 half = (max + 1) * sizeof(*entry); 1880 half += ichdr1->usedbytes + ichdr2->usedbytes + 1881 xfs_attr_leaf_newentsize(state->args, NULL); 1882 half /= 2; 1883 lastdelta = state->args->geo->blksize; 1884 entry = xfs_attr3_leaf_entryp(leaf1); 1885 for (count = index = 0; count < max; entry++, index++, count++) { 1886 1887 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A)) 1888 /* 1889 * The new entry is in the first block, account for it. 1890 */ 1891 if (count == blk1->index) { 1892 tmp = totallen + sizeof(*entry) + 1893 xfs_attr_leaf_newentsize(state->args, NULL); 1894 if (XFS_ATTR_ABS(half - tmp) > lastdelta) 1895 break; 1896 lastdelta = XFS_ATTR_ABS(half - tmp); 1897 totallen = tmp; 1898 foundit = 1; 1899 } 1900 1901 /* 1902 * Wrap around into the second block if necessary. 1903 */ 1904 if (count == ichdr1->count) { 1905 leaf1 = leaf2; 1906 entry = xfs_attr3_leaf_entryp(leaf1); 1907 index = 0; 1908 } 1909 1910 /* 1911 * Figure out if next leaf entry would be too much. 1912 */ 1913 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1, 1914 index); 1915 if (XFS_ATTR_ABS(half - tmp) > lastdelta) 1916 break; 1917 lastdelta = XFS_ATTR_ABS(half - tmp); 1918 totallen = tmp; 1919 #undef XFS_ATTR_ABS 1920 } 1921 1922 /* 1923 * Calculate the number of usedbytes that will end up in lower block. 1924 * If new entry not in lower block, fix up the count. 1925 */ 1926 totallen -= count * sizeof(*entry); 1927 if (foundit) { 1928 totallen -= sizeof(*entry) + 1929 xfs_attr_leaf_newentsize(state->args, NULL); 1930 } 1931 1932 *countarg = count; 1933 *usedbytesarg = totallen; 1934 return foundit; 1935 } 1936 1937 /*======================================================================== 1938 * Routines used for shrinking the Btree. 1939 *========================================================================*/ 1940 1941 /* 1942 * Check a leaf block and its neighbors to see if the block should be 1943 * collapsed into one or the other neighbor. Always keep the block 1944 * with the smaller block number. 1945 * If the current block is over 50% full, don't try to join it, return 0. 1946 * If the block is empty, fill in the state structure and return 2. 1947 * If it can be collapsed, fill in the state structure and return 1. 1948 * If nothing can be done, return 0. 1949 * 1950 * GROT: allow for INCOMPLETE entries in calculation. 1951 */ 1952 int 1953 xfs_attr3_leaf_toosmall( 1954 struct xfs_da_state *state, 1955 int *action) 1956 { 1957 struct xfs_attr_leafblock *leaf; 1958 struct xfs_da_state_blk *blk; 1959 struct xfs_attr3_icleaf_hdr ichdr; 1960 struct xfs_buf *bp; 1961 xfs_dablk_t blkno; 1962 int bytes; 1963 int forward; 1964 int error; 1965 int retval; 1966 int i; 1967 1968 trace_xfs_attr_leaf_toosmall(state->args); 1969 1970 /* 1971 * Check for the degenerate case of the block being over 50% full. 1972 * If so, it's not worth even looking to see if we might be able 1973 * to coalesce with a sibling. 1974 */ 1975 blk = &state->path.blk[ state->path.active-1 ]; 1976 leaf = blk->bp->b_addr; 1977 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf); 1978 bytes = xfs_attr3_leaf_hdr_size(leaf) + 1979 ichdr.count * sizeof(xfs_attr_leaf_entry_t) + 1980 ichdr.usedbytes; 1981 if (bytes > (state->args->geo->blksize >> 1)) { 1982 *action = 0; /* blk over 50%, don't try to join */ 1983 return 0; 1984 } 1985 1986 /* 1987 * Check for the degenerate case of the block being empty. 1988 * If the block is empty, we'll simply delete it, no need to 1989 * coalesce it with a sibling block. We choose (arbitrarily) 1990 * to merge with the forward block unless it is NULL. 1991 */ 1992 if (ichdr.count == 0) { 1993 /* 1994 * Make altpath point to the block we want to keep and 1995 * path point to the block we want to drop (this one). 1996 */ 1997 forward = (ichdr.forw != 0); 1998 memcpy(&state->altpath, &state->path, sizeof(state->path)); 1999 error = xfs_da3_path_shift(state, &state->altpath, forward, 2000 0, &retval); 2001 if (error) 2002 return error; 2003 if (retval) { 2004 *action = 0; 2005 } else { 2006 *action = 2; 2007 } 2008 return 0; 2009 } 2010 2011 /* 2012 * Examine each sibling block to see if we can coalesce with 2013 * at least 25% free space to spare. We need to figure out 2014 * whether to merge with the forward or the backward block. 2015 * We prefer coalescing with the lower numbered sibling so as 2016 * to shrink an attribute list over time. 2017 */ 2018 /* start with smaller blk num */ 2019 forward = ichdr.forw < ichdr.back; 2020 for (i = 0; i < 2; forward = !forward, i++) { 2021 struct xfs_attr3_icleaf_hdr ichdr2; 2022 if (forward) 2023 blkno = ichdr.forw; 2024 else 2025 blkno = ichdr.back; 2026 if (blkno == 0) 2027 continue; 2028 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp, 2029 blkno, &bp); 2030 if (error) 2031 return error; 2032 2033 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr); 2034 2035 bytes = state->args->geo->blksize - 2036 (state->args->geo->blksize >> 2) - 2037 ichdr.usedbytes - ichdr2.usedbytes - 2038 ((ichdr.count + ichdr2.count) * 2039 sizeof(xfs_attr_leaf_entry_t)) - 2040 xfs_attr3_leaf_hdr_size(leaf); 2041 2042 xfs_trans_brelse(state->args->trans, bp); 2043 if (bytes >= 0) 2044 break; /* fits with at least 25% to spare */ 2045 } 2046 if (i >= 2) { 2047 *action = 0; 2048 return 0; 2049 } 2050 2051 /* 2052 * Make altpath point to the block we want to keep (the lower 2053 * numbered block) and path point to the block we want to drop. 2054 */ 2055 memcpy(&state->altpath, &state->path, sizeof(state->path)); 2056 if (blkno < blk->blkno) { 2057 error = xfs_da3_path_shift(state, &state->altpath, forward, 2058 0, &retval); 2059 } else { 2060 error = xfs_da3_path_shift(state, &state->path, forward, 2061 0, &retval); 2062 } 2063 if (error) 2064 return error; 2065 if (retval) { 2066 *action = 0; 2067 } else { 2068 *action = 1; 2069 } 2070 return 0; 2071 } 2072 2073 /* 2074 * Remove a name from the leaf attribute list structure. 2075 * 2076 * Return 1 if leaf is less than 37% full, 0 if >= 37% full. 2077 * If two leaves are 37% full, when combined they will leave 25% free. 2078 */ 2079 int 2080 xfs_attr3_leaf_remove( 2081 struct xfs_buf *bp, 2082 struct xfs_da_args *args) 2083 { 2084 struct xfs_attr_leafblock *leaf; 2085 struct xfs_attr3_icleaf_hdr ichdr; 2086 struct xfs_attr_leaf_entry *entry; 2087 int before; 2088 int after; 2089 int smallest; 2090 int entsize; 2091 int tablesize; 2092 int tmp; 2093 int i; 2094 2095 trace_xfs_attr_leaf_remove(args); 2096 2097 leaf = bp->b_addr; 2098 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2099 2100 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8); 2101 ASSERT(args->index >= 0 && args->index < ichdr.count); 2102 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) + 2103 xfs_attr3_leaf_hdr_size(leaf)); 2104 2105 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2106 2107 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused); 2108 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize); 2109 2110 /* 2111 * Scan through free region table: 2112 * check for adjacency of free'd entry with an existing one, 2113 * find smallest free region in case we need to replace it, 2114 * adjust any map that borders the entry table, 2115 */ 2116 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t) 2117 + xfs_attr3_leaf_hdr_size(leaf); 2118 tmp = ichdr.freemap[0].size; 2119 before = after = -1; 2120 smallest = XFS_ATTR_LEAF_MAPSIZE - 1; 2121 entsize = xfs_attr_leaf_entsize(leaf, args->index); 2122 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 2123 ASSERT(ichdr.freemap[i].base < args->geo->blksize); 2124 ASSERT(ichdr.freemap[i].size < args->geo->blksize); 2125 if (ichdr.freemap[i].base == tablesize) { 2126 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t); 2127 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t); 2128 } 2129 2130 if (ichdr.freemap[i].base + ichdr.freemap[i].size == 2131 be16_to_cpu(entry->nameidx)) { 2132 before = i; 2133 } else if (ichdr.freemap[i].base == 2134 (be16_to_cpu(entry->nameidx) + entsize)) { 2135 after = i; 2136 } else if (ichdr.freemap[i].size < tmp) { 2137 tmp = ichdr.freemap[i].size; 2138 smallest = i; 2139 } 2140 } 2141 2142 /* 2143 * Coalesce adjacent freemap regions, 2144 * or replace the smallest region. 2145 */ 2146 if ((before >= 0) || (after >= 0)) { 2147 if ((before >= 0) && (after >= 0)) { 2148 ichdr.freemap[before].size += entsize; 2149 ichdr.freemap[before].size += ichdr.freemap[after].size; 2150 ichdr.freemap[after].base = 0; 2151 ichdr.freemap[after].size = 0; 2152 } else if (before >= 0) { 2153 ichdr.freemap[before].size += entsize; 2154 } else { 2155 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx); 2156 ichdr.freemap[after].size += entsize; 2157 } 2158 } else { 2159 /* 2160 * Replace smallest region (if it is smaller than free'd entry) 2161 */ 2162 if (ichdr.freemap[smallest].size < entsize) { 2163 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx); 2164 ichdr.freemap[smallest].size = entsize; 2165 } 2166 } 2167 2168 /* 2169 * Did we remove the first entry? 2170 */ 2171 if (be16_to_cpu(entry->nameidx) == ichdr.firstused) 2172 smallest = 1; 2173 else 2174 smallest = 0; 2175 2176 /* 2177 * Compress the remaining entries and zero out the removed stuff. 2178 */ 2179 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize); 2180 ichdr.usedbytes -= entsize; 2181 xfs_trans_log_buf(args->trans, bp, 2182 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index), 2183 entsize)); 2184 2185 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t); 2186 memmove(entry, entry + 1, tmp); 2187 ichdr.count--; 2188 xfs_trans_log_buf(args->trans, bp, 2189 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t))); 2190 2191 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count]; 2192 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t)); 2193 2194 /* 2195 * If we removed the first entry, re-find the first used byte 2196 * in the name area. Note that if the entry was the "firstused", 2197 * then we don't have a "hole" in our block resulting from 2198 * removing the name. 2199 */ 2200 if (smallest) { 2201 tmp = args->geo->blksize; 2202 entry = xfs_attr3_leaf_entryp(leaf); 2203 for (i = ichdr.count - 1; i >= 0; entry++, i--) { 2204 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused); 2205 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize); 2206 2207 if (be16_to_cpu(entry->nameidx) < tmp) 2208 tmp = be16_to_cpu(entry->nameidx); 2209 } 2210 ichdr.firstused = tmp; 2211 ASSERT(ichdr.firstused != 0); 2212 } else { 2213 ichdr.holes = 1; /* mark as needing compaction */ 2214 } 2215 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 2216 xfs_trans_log_buf(args->trans, bp, 2217 XFS_DA_LOGRANGE(leaf, &leaf->hdr, 2218 xfs_attr3_leaf_hdr_size(leaf))); 2219 2220 /* 2221 * Check if leaf is less than 50% full, caller may want to 2222 * "join" the leaf with a sibling if so. 2223 */ 2224 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) + 2225 ichdr.count * sizeof(xfs_attr_leaf_entry_t); 2226 2227 return tmp < args->geo->magicpct; /* leaf is < 37% full */ 2228 } 2229 2230 /* 2231 * Move all the attribute list entries from drop_leaf into save_leaf. 2232 */ 2233 void 2234 xfs_attr3_leaf_unbalance( 2235 struct xfs_da_state *state, 2236 struct xfs_da_state_blk *drop_blk, 2237 struct xfs_da_state_blk *save_blk) 2238 { 2239 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr; 2240 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr; 2241 struct xfs_attr3_icleaf_hdr drophdr; 2242 struct xfs_attr3_icleaf_hdr savehdr; 2243 struct xfs_attr_leaf_entry *entry; 2244 2245 trace_xfs_attr_leaf_unbalance(state->args); 2246 2247 drop_leaf = drop_blk->bp->b_addr; 2248 save_leaf = save_blk->bp->b_addr; 2249 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf); 2250 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf); 2251 entry = xfs_attr3_leaf_entryp(drop_leaf); 2252 2253 /* 2254 * Save last hashval from dying block for later Btree fixup. 2255 */ 2256 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval); 2257 2258 /* 2259 * Check if we need a temp buffer, or can we do it in place. 2260 * Note that we don't check "leaf" for holes because we will 2261 * always be dropping it, toosmall() decided that for us already. 2262 */ 2263 if (savehdr.holes == 0) { 2264 /* 2265 * dest leaf has no holes, so we add there. May need 2266 * to make some room in the entry array. 2267 */ 2268 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr, 2269 drop_blk->bp, &drophdr)) { 2270 xfs_attr3_leaf_moveents(state->args, 2271 drop_leaf, &drophdr, 0, 2272 save_leaf, &savehdr, 0, 2273 drophdr.count); 2274 } else { 2275 xfs_attr3_leaf_moveents(state->args, 2276 drop_leaf, &drophdr, 0, 2277 save_leaf, &savehdr, 2278 savehdr.count, drophdr.count); 2279 } 2280 } else { 2281 /* 2282 * Destination has holes, so we make a temporary copy 2283 * of the leaf and add them both to that. 2284 */ 2285 struct xfs_attr_leafblock *tmp_leaf; 2286 struct xfs_attr3_icleaf_hdr tmphdr; 2287 2288 tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0); 2289 2290 /* 2291 * Copy the header into the temp leaf so that all the stuff 2292 * not in the incore header is present and gets copied back in 2293 * once we've moved all the entries. 2294 */ 2295 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf)); 2296 2297 memset(&tmphdr, 0, sizeof(tmphdr)); 2298 tmphdr.magic = savehdr.magic; 2299 tmphdr.forw = savehdr.forw; 2300 tmphdr.back = savehdr.back; 2301 tmphdr.firstused = state->args->geo->blksize; 2302 2303 /* write the header to the temp buffer to initialise it */ 2304 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr); 2305 2306 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr, 2307 drop_blk->bp, &drophdr)) { 2308 xfs_attr3_leaf_moveents(state->args, 2309 drop_leaf, &drophdr, 0, 2310 tmp_leaf, &tmphdr, 0, 2311 drophdr.count); 2312 xfs_attr3_leaf_moveents(state->args, 2313 save_leaf, &savehdr, 0, 2314 tmp_leaf, &tmphdr, tmphdr.count, 2315 savehdr.count); 2316 } else { 2317 xfs_attr3_leaf_moveents(state->args, 2318 save_leaf, &savehdr, 0, 2319 tmp_leaf, &tmphdr, 0, 2320 savehdr.count); 2321 xfs_attr3_leaf_moveents(state->args, 2322 drop_leaf, &drophdr, 0, 2323 tmp_leaf, &tmphdr, tmphdr.count, 2324 drophdr.count); 2325 } 2326 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize); 2327 savehdr = tmphdr; /* struct copy */ 2328 kmem_free(tmp_leaf); 2329 } 2330 2331 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr); 2332 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0, 2333 state->args->geo->blksize - 1); 2334 2335 /* 2336 * Copy out last hashval in each block for B-tree code. 2337 */ 2338 entry = xfs_attr3_leaf_entryp(save_leaf); 2339 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval); 2340 } 2341 2342 /*======================================================================== 2343 * Routines used for finding things in the Btree. 2344 *========================================================================*/ 2345 2346 /* 2347 * Look up a name in a leaf attribute list structure. 2348 * This is the internal routine, it uses the caller's buffer. 2349 * 2350 * Note that duplicate keys are allowed, but only check within the 2351 * current leaf node. The Btree code must check in adjacent leaf nodes. 2352 * 2353 * Return in args->index the index into the entry[] array of either 2354 * the found entry, or where the entry should have been (insert before 2355 * that entry). 2356 * 2357 * Don't change the args->value unless we find the attribute. 2358 */ 2359 int 2360 xfs_attr3_leaf_lookup_int( 2361 struct xfs_buf *bp, 2362 struct xfs_da_args *args) 2363 { 2364 struct xfs_attr_leafblock *leaf; 2365 struct xfs_attr3_icleaf_hdr ichdr; 2366 struct xfs_attr_leaf_entry *entry; 2367 struct xfs_attr_leaf_entry *entries; 2368 struct xfs_attr_leaf_name_local *name_loc; 2369 struct xfs_attr_leaf_name_remote *name_rmt; 2370 xfs_dahash_t hashval; 2371 int probe; 2372 int span; 2373 2374 trace_xfs_attr_leaf_lookup(args); 2375 2376 leaf = bp->b_addr; 2377 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2378 entries = xfs_attr3_leaf_entryp(leaf); 2379 if (ichdr.count >= args->geo->blksize / 8) { 2380 xfs_buf_mark_corrupt(bp); 2381 return -EFSCORRUPTED; 2382 } 2383 2384 /* 2385 * Binary search. (note: small blocks will skip this loop) 2386 */ 2387 hashval = args->hashval; 2388 probe = span = ichdr.count / 2; 2389 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) { 2390 span /= 2; 2391 if (be32_to_cpu(entry->hashval) < hashval) 2392 probe += span; 2393 else if (be32_to_cpu(entry->hashval) > hashval) 2394 probe -= span; 2395 else 2396 break; 2397 } 2398 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) { 2399 xfs_buf_mark_corrupt(bp); 2400 return -EFSCORRUPTED; 2401 } 2402 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) { 2403 xfs_buf_mark_corrupt(bp); 2404 return -EFSCORRUPTED; 2405 } 2406 2407 /* 2408 * Since we may have duplicate hashval's, find the first matching 2409 * hashval in the leaf. 2410 */ 2411 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) { 2412 entry--; 2413 probe--; 2414 } 2415 while (probe < ichdr.count && 2416 be32_to_cpu(entry->hashval) < hashval) { 2417 entry++; 2418 probe++; 2419 } 2420 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) { 2421 args->index = probe; 2422 return -ENOATTR; 2423 } 2424 2425 /* 2426 * Duplicate keys may be present, so search all of them for a match. 2427 */ 2428 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval); 2429 entry++, probe++) { 2430 /* 2431 * GROT: Add code to remove incomplete entries. 2432 */ 2433 if (entry->flags & XFS_ATTR_LOCAL) { 2434 name_loc = xfs_attr3_leaf_name_local(leaf, probe); 2435 if (!xfs_attr_match(args, name_loc->namelen, 2436 name_loc->nameval, entry->flags)) 2437 continue; 2438 args->index = probe; 2439 return -EEXIST; 2440 } else { 2441 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe); 2442 if (!xfs_attr_match(args, name_rmt->namelen, 2443 name_rmt->name, entry->flags)) 2444 continue; 2445 args->index = probe; 2446 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen); 2447 args->rmtblkno = be32_to_cpu(name_rmt->valueblk); 2448 args->rmtblkcnt = xfs_attr3_rmt_blocks( 2449 args->dp->i_mount, 2450 args->rmtvaluelen); 2451 return -EEXIST; 2452 } 2453 } 2454 args->index = probe; 2455 return -ENOATTR; 2456 } 2457 2458 /* 2459 * Get the value associated with an attribute name from a leaf attribute 2460 * list structure. 2461 * 2462 * If args->valuelen is zero, only the length needs to be returned. Unlike a 2463 * lookup, we only return an error if the attribute does not exist or we can't 2464 * retrieve the value. 2465 */ 2466 int 2467 xfs_attr3_leaf_getvalue( 2468 struct xfs_buf *bp, 2469 struct xfs_da_args *args) 2470 { 2471 struct xfs_attr_leafblock *leaf; 2472 struct xfs_attr3_icleaf_hdr ichdr; 2473 struct xfs_attr_leaf_entry *entry; 2474 struct xfs_attr_leaf_name_local *name_loc; 2475 struct xfs_attr_leaf_name_remote *name_rmt; 2476 2477 leaf = bp->b_addr; 2478 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2479 ASSERT(ichdr.count < args->geo->blksize / 8); 2480 ASSERT(args->index < ichdr.count); 2481 2482 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2483 if (entry->flags & XFS_ATTR_LOCAL) { 2484 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 2485 ASSERT(name_loc->namelen == args->namelen); 2486 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0); 2487 return xfs_attr_copy_value(args, 2488 &name_loc->nameval[args->namelen], 2489 be16_to_cpu(name_loc->valuelen)); 2490 } 2491 2492 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2493 ASSERT(name_rmt->namelen == args->namelen); 2494 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0); 2495 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen); 2496 args->rmtblkno = be32_to_cpu(name_rmt->valueblk); 2497 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount, 2498 args->rmtvaluelen); 2499 return xfs_attr_copy_value(args, NULL, args->rmtvaluelen); 2500 } 2501 2502 /*======================================================================== 2503 * Utility routines. 2504 *========================================================================*/ 2505 2506 /* 2507 * Move the indicated entries from one leaf to another. 2508 * NOTE: this routine modifies both source and destination leaves. 2509 */ 2510 /*ARGSUSED*/ 2511 STATIC void 2512 xfs_attr3_leaf_moveents( 2513 struct xfs_da_args *args, 2514 struct xfs_attr_leafblock *leaf_s, 2515 struct xfs_attr3_icleaf_hdr *ichdr_s, 2516 int start_s, 2517 struct xfs_attr_leafblock *leaf_d, 2518 struct xfs_attr3_icleaf_hdr *ichdr_d, 2519 int start_d, 2520 int count) 2521 { 2522 struct xfs_attr_leaf_entry *entry_s; 2523 struct xfs_attr_leaf_entry *entry_d; 2524 int desti; 2525 int tmp; 2526 int i; 2527 2528 /* 2529 * Check for nothing to do. 2530 */ 2531 if (count == 0) 2532 return; 2533 2534 /* 2535 * Set up environment. 2536 */ 2537 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC || 2538 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC); 2539 ASSERT(ichdr_s->magic == ichdr_d->magic); 2540 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8); 2541 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s)) 2542 + xfs_attr3_leaf_hdr_size(leaf_s)); 2543 ASSERT(ichdr_d->count < args->geo->blksize / 8); 2544 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d)) 2545 + xfs_attr3_leaf_hdr_size(leaf_d)); 2546 2547 ASSERT(start_s < ichdr_s->count); 2548 ASSERT(start_d <= ichdr_d->count); 2549 ASSERT(count <= ichdr_s->count); 2550 2551 2552 /* 2553 * Move the entries in the destination leaf up to make a hole? 2554 */ 2555 if (start_d < ichdr_d->count) { 2556 tmp = ichdr_d->count - start_d; 2557 tmp *= sizeof(xfs_attr_leaf_entry_t); 2558 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d]; 2559 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count]; 2560 memmove(entry_d, entry_s, tmp); 2561 } 2562 2563 /* 2564 * Copy all entry's in the same (sorted) order, 2565 * but allocate attribute info packed and in sequence. 2566 */ 2567 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2568 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d]; 2569 desti = start_d; 2570 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) { 2571 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused); 2572 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i); 2573 #ifdef GROT 2574 /* 2575 * Code to drop INCOMPLETE entries. Difficult to use as we 2576 * may also need to change the insertion index. Code turned 2577 * off for 6.2, should be revisited later. 2578 */ 2579 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */ 2580 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp); 2581 ichdr_s->usedbytes -= tmp; 2582 ichdr_s->count -= 1; 2583 entry_d--; /* to compensate for ++ in loop hdr */ 2584 desti--; 2585 if ((start_s + i) < offset) 2586 result++; /* insertion index adjustment */ 2587 } else { 2588 #endif /* GROT */ 2589 ichdr_d->firstused -= tmp; 2590 /* both on-disk, don't endian flip twice */ 2591 entry_d->hashval = entry_s->hashval; 2592 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused); 2593 entry_d->flags = entry_s->flags; 2594 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp 2595 <= args->geo->blksize); 2596 memmove(xfs_attr3_leaf_name(leaf_d, desti), 2597 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp); 2598 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp 2599 <= args->geo->blksize); 2600 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp); 2601 ichdr_s->usedbytes -= tmp; 2602 ichdr_d->usedbytes += tmp; 2603 ichdr_s->count -= 1; 2604 ichdr_d->count += 1; 2605 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t) 2606 + xfs_attr3_leaf_hdr_size(leaf_d); 2607 ASSERT(ichdr_d->firstused >= tmp); 2608 #ifdef GROT 2609 } 2610 #endif /* GROT */ 2611 } 2612 2613 /* 2614 * Zero out the entries we just copied. 2615 */ 2616 if (start_s == ichdr_s->count) { 2617 tmp = count * sizeof(xfs_attr_leaf_entry_t); 2618 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2619 ASSERT(((char *)entry_s + tmp) <= 2620 ((char *)leaf_s + args->geo->blksize)); 2621 memset(entry_s, 0, tmp); 2622 } else { 2623 /* 2624 * Move the remaining entries down to fill the hole, 2625 * then zero the entries at the top. 2626 */ 2627 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t); 2628 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count]; 2629 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2630 memmove(entry_d, entry_s, tmp); 2631 2632 tmp = count * sizeof(xfs_attr_leaf_entry_t); 2633 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count]; 2634 ASSERT(((char *)entry_s + tmp) <= 2635 ((char *)leaf_s + args->geo->blksize)); 2636 memset(entry_s, 0, tmp); 2637 } 2638 2639 /* 2640 * Fill in the freemap information 2641 */ 2642 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d); 2643 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t); 2644 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base; 2645 ichdr_d->freemap[1].base = 0; 2646 ichdr_d->freemap[2].base = 0; 2647 ichdr_d->freemap[1].size = 0; 2648 ichdr_d->freemap[2].size = 0; 2649 ichdr_s->holes = 1; /* leaf may not be compact */ 2650 } 2651 2652 /* 2653 * Pick up the last hashvalue from a leaf block. 2654 */ 2655 xfs_dahash_t 2656 xfs_attr_leaf_lasthash( 2657 struct xfs_buf *bp, 2658 int *count) 2659 { 2660 struct xfs_attr3_icleaf_hdr ichdr; 2661 struct xfs_attr_leaf_entry *entries; 2662 struct xfs_mount *mp = bp->b_mount; 2663 2664 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr); 2665 entries = xfs_attr3_leaf_entryp(bp->b_addr); 2666 if (count) 2667 *count = ichdr.count; 2668 if (!ichdr.count) 2669 return 0; 2670 return be32_to_cpu(entries[ichdr.count - 1].hashval); 2671 } 2672 2673 /* 2674 * Calculate the number of bytes used to store the indicated attribute 2675 * (whether local or remote only calculate bytes in this block). 2676 */ 2677 STATIC int 2678 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index) 2679 { 2680 struct xfs_attr_leaf_entry *entries; 2681 xfs_attr_leaf_name_local_t *name_loc; 2682 xfs_attr_leaf_name_remote_t *name_rmt; 2683 int size; 2684 2685 entries = xfs_attr3_leaf_entryp(leaf); 2686 if (entries[index].flags & XFS_ATTR_LOCAL) { 2687 name_loc = xfs_attr3_leaf_name_local(leaf, index); 2688 size = xfs_attr_leaf_entsize_local(name_loc->namelen, 2689 be16_to_cpu(name_loc->valuelen)); 2690 } else { 2691 name_rmt = xfs_attr3_leaf_name_remote(leaf, index); 2692 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen); 2693 } 2694 return size; 2695 } 2696 2697 /* 2698 * Calculate the number of bytes that would be required to store the new 2699 * attribute (whether local or remote only calculate bytes in this block). 2700 * This routine decides as a side effect whether the attribute will be 2701 * a "local" or a "remote" attribute. 2702 */ 2703 int 2704 xfs_attr_leaf_newentsize( 2705 struct xfs_da_args *args, 2706 int *local) 2707 { 2708 int size; 2709 2710 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen); 2711 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) { 2712 if (local) 2713 *local = 1; 2714 return size; 2715 } 2716 if (local) 2717 *local = 0; 2718 return xfs_attr_leaf_entsize_remote(args->namelen); 2719 } 2720 2721 2722 /*======================================================================== 2723 * Manage the INCOMPLETE flag in a leaf entry 2724 *========================================================================*/ 2725 2726 /* 2727 * Clear the INCOMPLETE flag on an entry in a leaf block. 2728 */ 2729 int 2730 xfs_attr3_leaf_clearflag( 2731 struct xfs_da_args *args) 2732 { 2733 struct xfs_attr_leafblock *leaf; 2734 struct xfs_attr_leaf_entry *entry; 2735 struct xfs_attr_leaf_name_remote *name_rmt; 2736 struct xfs_buf *bp; 2737 int error; 2738 #ifdef DEBUG 2739 struct xfs_attr3_icleaf_hdr ichdr; 2740 xfs_attr_leaf_name_local_t *name_loc; 2741 int namelen; 2742 char *name; 2743 #endif /* DEBUG */ 2744 2745 trace_xfs_attr_leaf_clearflag(args); 2746 /* 2747 * Set up the operation. 2748 */ 2749 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp); 2750 if (error) 2751 return error; 2752 2753 leaf = bp->b_addr; 2754 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2755 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE); 2756 2757 #ifdef DEBUG 2758 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2759 ASSERT(args->index < ichdr.count); 2760 ASSERT(args->index >= 0); 2761 2762 if (entry->flags & XFS_ATTR_LOCAL) { 2763 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 2764 namelen = name_loc->namelen; 2765 name = (char *)name_loc->nameval; 2766 } else { 2767 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2768 namelen = name_rmt->namelen; 2769 name = (char *)name_rmt->name; 2770 } 2771 ASSERT(be32_to_cpu(entry->hashval) == args->hashval); 2772 ASSERT(namelen == args->namelen); 2773 ASSERT(memcmp(name, args->name, namelen) == 0); 2774 #endif /* DEBUG */ 2775 2776 entry->flags &= ~XFS_ATTR_INCOMPLETE; 2777 xfs_trans_log_buf(args->trans, bp, 2778 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 2779 2780 if (args->rmtblkno) { 2781 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0); 2782 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2783 name_rmt->valueblk = cpu_to_be32(args->rmtblkno); 2784 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen); 2785 xfs_trans_log_buf(args->trans, bp, 2786 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt))); 2787 } 2788 2789 return 0; 2790 } 2791 2792 /* 2793 * Set the INCOMPLETE flag on an entry in a leaf block. 2794 */ 2795 int 2796 xfs_attr3_leaf_setflag( 2797 struct xfs_da_args *args) 2798 { 2799 struct xfs_attr_leafblock *leaf; 2800 struct xfs_attr_leaf_entry *entry; 2801 struct xfs_attr_leaf_name_remote *name_rmt; 2802 struct xfs_buf *bp; 2803 int error; 2804 #ifdef DEBUG 2805 struct xfs_attr3_icleaf_hdr ichdr; 2806 #endif 2807 2808 trace_xfs_attr_leaf_setflag(args); 2809 2810 /* 2811 * Set up the operation. 2812 */ 2813 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp); 2814 if (error) 2815 return error; 2816 2817 leaf = bp->b_addr; 2818 #ifdef DEBUG 2819 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2820 ASSERT(args->index < ichdr.count); 2821 ASSERT(args->index >= 0); 2822 #endif 2823 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2824 2825 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0); 2826 entry->flags |= XFS_ATTR_INCOMPLETE; 2827 xfs_trans_log_buf(args->trans, bp, 2828 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 2829 if ((entry->flags & XFS_ATTR_LOCAL) == 0) { 2830 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2831 name_rmt->valueblk = 0; 2832 name_rmt->valuelen = 0; 2833 xfs_trans_log_buf(args->trans, bp, 2834 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt))); 2835 } 2836 2837 return 0; 2838 } 2839 2840 /* 2841 * In a single transaction, clear the INCOMPLETE flag on the leaf entry 2842 * given by args->blkno/index and set the INCOMPLETE flag on the leaf 2843 * entry given by args->blkno2/index2. 2844 * 2845 * Note that they could be in different blocks, or in the same block. 2846 */ 2847 int 2848 xfs_attr3_leaf_flipflags( 2849 struct xfs_da_args *args) 2850 { 2851 struct xfs_attr_leafblock *leaf1; 2852 struct xfs_attr_leafblock *leaf2; 2853 struct xfs_attr_leaf_entry *entry1; 2854 struct xfs_attr_leaf_entry *entry2; 2855 struct xfs_attr_leaf_name_remote *name_rmt; 2856 struct xfs_buf *bp1; 2857 struct xfs_buf *bp2; 2858 int error; 2859 #ifdef DEBUG 2860 struct xfs_attr3_icleaf_hdr ichdr1; 2861 struct xfs_attr3_icleaf_hdr ichdr2; 2862 xfs_attr_leaf_name_local_t *name_loc; 2863 int namelen1, namelen2; 2864 char *name1, *name2; 2865 #endif /* DEBUG */ 2866 2867 trace_xfs_attr_leaf_flipflags(args); 2868 2869 /* 2870 * Read the block containing the "old" attr 2871 */ 2872 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1); 2873 if (error) 2874 return error; 2875 2876 /* 2877 * Read the block containing the "new" attr, if it is different 2878 */ 2879 if (args->blkno2 != args->blkno) { 2880 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2, 2881 &bp2); 2882 if (error) 2883 return error; 2884 } else { 2885 bp2 = bp1; 2886 } 2887 2888 leaf1 = bp1->b_addr; 2889 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index]; 2890 2891 leaf2 = bp2->b_addr; 2892 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2]; 2893 2894 #ifdef DEBUG 2895 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1); 2896 ASSERT(args->index < ichdr1.count); 2897 ASSERT(args->index >= 0); 2898 2899 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2); 2900 ASSERT(args->index2 < ichdr2.count); 2901 ASSERT(args->index2 >= 0); 2902 2903 if (entry1->flags & XFS_ATTR_LOCAL) { 2904 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index); 2905 namelen1 = name_loc->namelen; 2906 name1 = (char *)name_loc->nameval; 2907 } else { 2908 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index); 2909 namelen1 = name_rmt->namelen; 2910 name1 = (char *)name_rmt->name; 2911 } 2912 if (entry2->flags & XFS_ATTR_LOCAL) { 2913 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2); 2914 namelen2 = name_loc->namelen; 2915 name2 = (char *)name_loc->nameval; 2916 } else { 2917 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2); 2918 namelen2 = name_rmt->namelen; 2919 name2 = (char *)name_rmt->name; 2920 } 2921 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval)); 2922 ASSERT(namelen1 == namelen2); 2923 ASSERT(memcmp(name1, name2, namelen1) == 0); 2924 #endif /* DEBUG */ 2925 2926 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE); 2927 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0); 2928 2929 entry1->flags &= ~XFS_ATTR_INCOMPLETE; 2930 xfs_trans_log_buf(args->trans, bp1, 2931 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1))); 2932 if (args->rmtblkno) { 2933 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0); 2934 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index); 2935 name_rmt->valueblk = cpu_to_be32(args->rmtblkno); 2936 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen); 2937 xfs_trans_log_buf(args->trans, bp1, 2938 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt))); 2939 } 2940 2941 entry2->flags |= XFS_ATTR_INCOMPLETE; 2942 xfs_trans_log_buf(args->trans, bp2, 2943 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2))); 2944 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) { 2945 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2); 2946 name_rmt->valueblk = 0; 2947 name_rmt->valuelen = 0; 2948 xfs_trans_log_buf(args->trans, bp2, 2949 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt))); 2950 } 2951 2952 return 0; 2953 } 2954