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