1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * Copyright (c) 2013 Red Hat, Inc. 5 * All Rights Reserved. 6 */ 7 #include "xfs.h" 8 #include "xfs_fs.h" 9 #include "xfs_shared.h" 10 #include "xfs_format.h" 11 #include "xfs_log_format.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_sb.h" 14 #include "xfs_mount.h" 15 #include "xfs_da_format.h" 16 #include "xfs_da_btree.h" 17 #include "xfs_inode.h" 18 #include "xfs_trans.h" 19 #include "xfs_bmap_btree.h" 20 #include "xfs_bmap.h" 21 #include "xfs_attr_sf.h" 22 #include "xfs_attr_remote.h" 23 #include "xfs_attr.h" 24 #include "xfs_attr_leaf.h" 25 #include "xfs_error.h" 26 #include "xfs_trace.h" 27 #include "xfs_buf_item.h" 28 #include "xfs_dir2.h" 29 #include "xfs_log.h" 30 31 32 /* 33 * xfs_attr_leaf.c 34 * 35 * Routines to implement leaf blocks of attributes as Btrees of hashed names. 36 */ 37 38 /*======================================================================== 39 * Function prototypes for the kernel. 40 *========================================================================*/ 41 42 /* 43 * Routines used for growing the Btree. 44 */ 45 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args, 46 xfs_dablk_t which_block, struct xfs_buf **bpp); 47 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer, 48 struct xfs_attr3_icleaf_hdr *ichdr, 49 struct xfs_da_args *args, int freemap_index); 50 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args, 51 struct xfs_attr3_icleaf_hdr *ichdr, 52 struct xfs_buf *leaf_buffer); 53 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state, 54 xfs_da_state_blk_t *blk1, 55 xfs_da_state_blk_t *blk2); 56 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state, 57 xfs_da_state_blk_t *leaf_blk_1, 58 struct xfs_attr3_icleaf_hdr *ichdr1, 59 xfs_da_state_blk_t *leaf_blk_2, 60 struct xfs_attr3_icleaf_hdr *ichdr2, 61 int *number_entries_in_blk1, 62 int *number_usedbytes_in_blk1); 63 64 /* 65 * Utility routines. 66 */ 67 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args, 68 struct xfs_attr_leafblock *src_leaf, 69 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start, 70 struct xfs_attr_leafblock *dst_leaf, 71 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start, 72 int move_count); 73 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index); 74 75 /* 76 * attr3 block 'firstused' conversion helpers. 77 * 78 * firstused refers to the offset of the first used byte of the nameval region 79 * of an attr leaf block. The region starts at the tail of the block and expands 80 * backwards towards the middle. As such, firstused is initialized to the block 81 * size for an empty leaf block and is reduced from there. 82 * 83 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k. 84 * The in-core firstused field is 32-bit and thus supports the maximum fsb size. 85 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this 86 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent 87 * the attr block size. The following helpers manage the conversion between the 88 * in-core and on-disk formats. 89 */ 90 91 static void 92 xfs_attr3_leaf_firstused_from_disk( 93 struct xfs_da_geometry *geo, 94 struct xfs_attr3_icleaf_hdr *to, 95 struct xfs_attr_leafblock *from) 96 { 97 struct xfs_attr3_leaf_hdr *hdr3; 98 99 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) { 100 hdr3 = (struct xfs_attr3_leaf_hdr *) from; 101 to->firstused = be16_to_cpu(hdr3->firstused); 102 } else { 103 to->firstused = be16_to_cpu(from->hdr.firstused); 104 } 105 106 /* 107 * Convert from the magic fsb size value to actual blocksize. This 108 * should only occur for empty blocks when the block size overflows 109 * 16-bits. 110 */ 111 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) { 112 ASSERT(!to->count && !to->usedbytes); 113 ASSERT(geo->blksize > USHRT_MAX); 114 to->firstused = geo->blksize; 115 } 116 } 117 118 static void 119 xfs_attr3_leaf_firstused_to_disk( 120 struct xfs_da_geometry *geo, 121 struct xfs_attr_leafblock *to, 122 struct xfs_attr3_icleaf_hdr *from) 123 { 124 struct xfs_attr3_leaf_hdr *hdr3; 125 uint32_t firstused; 126 127 /* magic value should only be seen on disk */ 128 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF); 129 130 /* 131 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk 132 * value. This only overflows at the max supported value of 64k. Use the 133 * magic on-disk value to represent block size in this case. 134 */ 135 firstused = from->firstused; 136 if (firstused > USHRT_MAX) { 137 ASSERT(from->firstused == geo->blksize); 138 firstused = XFS_ATTR3_LEAF_NULLOFF; 139 } 140 141 if (from->magic == XFS_ATTR3_LEAF_MAGIC) { 142 hdr3 = (struct xfs_attr3_leaf_hdr *) to; 143 hdr3->firstused = cpu_to_be16(firstused); 144 } else { 145 to->hdr.firstused = cpu_to_be16(firstused); 146 } 147 } 148 149 void 150 xfs_attr3_leaf_hdr_from_disk( 151 struct xfs_da_geometry *geo, 152 struct xfs_attr3_icleaf_hdr *to, 153 struct xfs_attr_leafblock *from) 154 { 155 int i; 156 157 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || 158 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); 159 160 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) { 161 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from; 162 163 to->forw = be32_to_cpu(hdr3->info.hdr.forw); 164 to->back = be32_to_cpu(hdr3->info.hdr.back); 165 to->magic = be16_to_cpu(hdr3->info.hdr.magic); 166 to->count = be16_to_cpu(hdr3->count); 167 to->usedbytes = be16_to_cpu(hdr3->usedbytes); 168 xfs_attr3_leaf_firstused_from_disk(geo, to, from); 169 to->holes = hdr3->holes; 170 171 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 172 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base); 173 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size); 174 } 175 return; 176 } 177 to->forw = be32_to_cpu(from->hdr.info.forw); 178 to->back = be32_to_cpu(from->hdr.info.back); 179 to->magic = be16_to_cpu(from->hdr.info.magic); 180 to->count = be16_to_cpu(from->hdr.count); 181 to->usedbytes = be16_to_cpu(from->hdr.usedbytes); 182 xfs_attr3_leaf_firstused_from_disk(geo, to, from); 183 to->holes = from->hdr.holes; 184 185 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 186 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base); 187 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size); 188 } 189 } 190 191 void 192 xfs_attr3_leaf_hdr_to_disk( 193 struct xfs_da_geometry *geo, 194 struct xfs_attr_leafblock *to, 195 struct xfs_attr3_icleaf_hdr *from) 196 { 197 int i; 198 199 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC || 200 from->magic == XFS_ATTR3_LEAF_MAGIC); 201 202 if (from->magic == XFS_ATTR3_LEAF_MAGIC) { 203 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to; 204 205 hdr3->info.hdr.forw = cpu_to_be32(from->forw); 206 hdr3->info.hdr.back = cpu_to_be32(from->back); 207 hdr3->info.hdr.magic = cpu_to_be16(from->magic); 208 hdr3->count = cpu_to_be16(from->count); 209 hdr3->usedbytes = cpu_to_be16(from->usedbytes); 210 xfs_attr3_leaf_firstused_to_disk(geo, to, from); 211 hdr3->holes = from->holes; 212 hdr3->pad1 = 0; 213 214 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 215 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base); 216 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size); 217 } 218 return; 219 } 220 to->hdr.info.forw = cpu_to_be32(from->forw); 221 to->hdr.info.back = cpu_to_be32(from->back); 222 to->hdr.info.magic = cpu_to_be16(from->magic); 223 to->hdr.count = cpu_to_be16(from->count); 224 to->hdr.usedbytes = cpu_to_be16(from->usedbytes); 225 xfs_attr3_leaf_firstused_to_disk(geo, to, from); 226 to->hdr.holes = from->holes; 227 to->hdr.pad1 = 0; 228 229 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 230 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base); 231 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size); 232 } 233 } 234 235 static xfs_failaddr_t 236 xfs_attr3_leaf_verify_entry( 237 struct xfs_mount *mp, 238 char *buf_end, 239 struct xfs_attr_leafblock *leaf, 240 struct xfs_attr3_icleaf_hdr *leafhdr, 241 struct xfs_attr_leaf_entry *ent, 242 int idx, 243 __u32 *last_hashval) 244 { 245 struct xfs_attr_leaf_name_local *lentry; 246 struct xfs_attr_leaf_name_remote *rentry; 247 char *name_end; 248 unsigned int nameidx; 249 unsigned int namesize; 250 __u32 hashval; 251 252 /* hash order check */ 253 hashval = be32_to_cpu(ent->hashval); 254 if (hashval < *last_hashval) 255 return __this_address; 256 *last_hashval = hashval; 257 258 nameidx = be16_to_cpu(ent->nameidx); 259 if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize) 260 return __this_address; 261 262 /* 263 * Check the name information. The namelen fields are u8 so we can't 264 * possibly exceed the maximum name length of 255 bytes. 265 */ 266 if (ent->flags & XFS_ATTR_LOCAL) { 267 lentry = xfs_attr3_leaf_name_local(leaf, idx); 268 namesize = xfs_attr_leaf_entsize_local(lentry->namelen, 269 be16_to_cpu(lentry->valuelen)); 270 name_end = (char *)lentry + namesize; 271 if (lentry->namelen == 0) 272 return __this_address; 273 } else { 274 rentry = xfs_attr3_leaf_name_remote(leaf, idx); 275 namesize = xfs_attr_leaf_entsize_remote(rentry->namelen); 276 name_end = (char *)rentry + namesize; 277 if (rentry->namelen == 0) 278 return __this_address; 279 if (!(ent->flags & XFS_ATTR_INCOMPLETE) && 280 rentry->valueblk == 0) 281 return __this_address; 282 } 283 284 if (name_end > buf_end) 285 return __this_address; 286 287 return NULL; 288 } 289 290 static xfs_failaddr_t 291 xfs_attr3_leaf_verify( 292 struct xfs_buf *bp) 293 { 294 struct xfs_attr3_icleaf_hdr ichdr; 295 struct xfs_mount *mp = bp->b_mount; 296 struct xfs_attr_leafblock *leaf = bp->b_addr; 297 struct xfs_attr_leaf_entry *entries; 298 struct xfs_attr_leaf_entry *ent; 299 char *buf_end; 300 uint32_t end; /* must be 32bit - see below */ 301 __u32 last_hashval = 0; 302 int i; 303 xfs_failaddr_t fa; 304 305 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf); 306 307 fa = xfs_da3_blkinfo_verify(bp, bp->b_addr); 308 if (fa) 309 return fa; 310 311 /* 312 * firstused is the block offset of the first name info structure. 313 * Make sure it doesn't go off the block or crash into the header. 314 */ 315 if (ichdr.firstused > mp->m_attr_geo->blksize) 316 return __this_address; 317 if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf)) 318 return __this_address; 319 320 /* Make sure the entries array doesn't crash into the name info. */ 321 entries = xfs_attr3_leaf_entryp(bp->b_addr); 322 if ((char *)&entries[ichdr.count] > 323 (char *)bp->b_addr + ichdr.firstused) 324 return __this_address; 325 326 /* 327 * NOTE: This verifier historically failed empty leaf buffers because 328 * we expect the fork to be in another format. Empty attr fork format 329 * conversions are possible during xattr set, however, and format 330 * conversion is not atomic with the xattr set that triggers it. We 331 * cannot assume leaf blocks are non-empty until that is addressed. 332 */ 333 buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize; 334 for (i = 0, ent = entries; i < ichdr.count; ent++, i++) { 335 fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr, 336 ent, i, &last_hashval); 337 if (fa) 338 return fa; 339 } 340 341 /* 342 * Quickly check the freemap information. Attribute data has to be 343 * aligned to 4-byte boundaries, and likewise for the free space. 344 * 345 * Note that for 64k block size filesystems, the freemap entries cannot 346 * overflow as they are only be16 fields. However, when checking end 347 * pointer of the freemap, we have to be careful to detect overflows and 348 * so use uint32_t for those checks. 349 */ 350 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 351 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize) 352 return __this_address; 353 if (ichdr.freemap[i].base & 0x3) 354 return __this_address; 355 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize) 356 return __this_address; 357 if (ichdr.freemap[i].size & 0x3) 358 return __this_address; 359 360 /* be care of 16 bit overflows here */ 361 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size; 362 if (end < ichdr.freemap[i].base) 363 return __this_address; 364 if (end > mp->m_attr_geo->blksize) 365 return __this_address; 366 } 367 368 return NULL; 369 } 370 371 static void 372 xfs_attr3_leaf_write_verify( 373 struct xfs_buf *bp) 374 { 375 struct xfs_mount *mp = bp->b_mount; 376 struct xfs_buf_log_item *bip = bp->b_log_item; 377 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr; 378 xfs_failaddr_t fa; 379 380 fa = xfs_attr3_leaf_verify(bp); 381 if (fa) { 382 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 383 return; 384 } 385 386 if (!xfs_sb_version_hascrc(&mp->m_sb)) 387 return; 388 389 if (bip) 390 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); 391 392 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF); 393 } 394 395 /* 396 * leaf/node format detection on trees is sketchy, so a node read can be done on 397 * leaf level blocks when detection identifies the tree as a node format tree 398 * incorrectly. In this case, we need to swap the verifier to match the correct 399 * format of the block being read. 400 */ 401 static void 402 xfs_attr3_leaf_read_verify( 403 struct xfs_buf *bp) 404 { 405 struct xfs_mount *mp = bp->b_mount; 406 xfs_failaddr_t fa; 407 408 if (xfs_sb_version_hascrc(&mp->m_sb) && 409 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF)) 410 xfs_verifier_error(bp, -EFSBADCRC, __this_address); 411 else { 412 fa = xfs_attr3_leaf_verify(bp); 413 if (fa) 414 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 415 } 416 } 417 418 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = { 419 .name = "xfs_attr3_leaf", 420 .magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC), 421 cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) }, 422 .verify_read = xfs_attr3_leaf_read_verify, 423 .verify_write = xfs_attr3_leaf_write_verify, 424 .verify_struct = xfs_attr3_leaf_verify, 425 }; 426 427 int 428 xfs_attr3_leaf_read( 429 struct xfs_trans *tp, 430 struct xfs_inode *dp, 431 xfs_dablk_t bno, 432 struct xfs_buf **bpp) 433 { 434 int err; 435 436 err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK, 437 &xfs_attr3_leaf_buf_ops); 438 if (!err && tp && *bpp) 439 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF); 440 return err; 441 } 442 443 /*======================================================================== 444 * Namespace helper routines 445 *========================================================================*/ 446 447 static bool 448 xfs_attr_match( 449 struct xfs_da_args *args, 450 uint8_t namelen, 451 unsigned char *name, 452 int flags) 453 { 454 if (args->namelen != namelen) 455 return false; 456 if (memcmp(args->name, name, namelen) != 0) 457 return false; 458 /* 459 * If we are looking for incomplete entries, show only those, else only 460 * show complete entries. 461 */ 462 if (args->attr_filter != 463 (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE))) 464 return false; 465 return true; 466 } 467 468 static int 469 xfs_attr_copy_value( 470 struct xfs_da_args *args, 471 unsigned char *value, 472 int valuelen) 473 { 474 /* 475 * No copy if all we have to do is get the length 476 */ 477 if (!args->valuelen) { 478 args->valuelen = valuelen; 479 return 0; 480 } 481 482 /* 483 * No copy if the length of the existing buffer is too small 484 */ 485 if (args->valuelen < valuelen) { 486 args->valuelen = valuelen; 487 return -ERANGE; 488 } 489 490 if (!args->value) { 491 args->value = kmem_alloc_large(valuelen, KM_NOLOCKDEP); 492 if (!args->value) 493 return -ENOMEM; 494 } 495 args->valuelen = valuelen; 496 497 /* remote block xattr requires IO for copy-in */ 498 if (args->rmtblkno) 499 return xfs_attr_rmtval_get(args); 500 501 /* 502 * This is to prevent a GCC warning because the remote xattr case 503 * doesn't have a value to pass in. In that case, we never reach here, 504 * but GCC can't work that out and so throws a "passing NULL to 505 * memcpy" warning. 506 */ 507 if (!value) 508 return -EINVAL; 509 memcpy(args->value, value, valuelen); 510 return 0; 511 } 512 513 /*======================================================================== 514 * External routines when attribute fork size < XFS_LITINO(mp). 515 *========================================================================*/ 516 517 /* 518 * Query whether the total requested number of attr fork bytes of extended 519 * attribute space will be able to fit inline. 520 * 521 * Returns zero if not, else the 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 /* 539 * Check if the new size could fit at all first: 540 */ 541 if (bytes > XFS_LITINO(mp)) 542 return 0; 543 544 /* rounded down */ 545 offset = (XFS_LITINO(mp) - bytes) >> 3; 546 547 if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) { 548 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; 549 return (offset >= minforkoff) ? minforkoff : 0; 550 } 551 552 /* 553 * If the requested numbers of bytes is smaller or equal to the 554 * current attribute fork size we can always proceed. 555 * 556 * Note that if_bytes in the data fork might actually be larger than 557 * the current data fork size is due to delalloc extents. In that 558 * case either the extent count will go down when they are converted 559 * to real extents, or the delalloc conversion will take care of the 560 * literal area rebalancing. 561 */ 562 if (bytes <= XFS_IFORK_ASIZE(dp)) 563 return dp->i_d.di_forkoff; 564 565 /* 566 * For attr2 we can try to move the forkoff if there is space in the 567 * literal area, but for the old format we are done if there is no 568 * space in the fixed attribute fork. 569 */ 570 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) 571 return 0; 572 573 dsize = dp->i_df.if_bytes; 574 575 switch (dp->i_df.if_format) { 576 case XFS_DINODE_FMT_EXTENTS: 577 /* 578 * If there is no attr fork and the data fork is extents, 579 * determine if creating the default attr fork will result 580 * in the extents form migrating to btree. If so, the 581 * minimum offset only needs to be the space required for 582 * the btree root. 583 */ 584 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes > 585 xfs_default_attroffset(dp)) 586 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS); 587 break; 588 case XFS_DINODE_FMT_BTREE: 589 /* 590 * If we have a data btree then keep forkoff if we have one, 591 * otherwise we are adding a new attr, so then we set 592 * minforkoff to where the btree root can finish so we have 593 * plenty of room for attrs 594 */ 595 if (dp->i_d.di_forkoff) { 596 if (offset < dp->i_d.di_forkoff) 597 return 0; 598 return dp->i_d.di_forkoff; 599 } 600 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot); 601 break; 602 } 603 604 /* 605 * A data fork btree root must have space for at least 606 * MINDBTPTRS key/ptr pairs if the data fork is small or empty. 607 */ 608 minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS)); 609 minforkoff = roundup(minforkoff, 8) >> 3; 610 611 /* attr fork btree root can have at least this many key/ptr pairs */ 612 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS); 613 maxforkoff = maxforkoff >> 3; /* rounded down */ 614 615 if (offset >= maxforkoff) 616 return maxforkoff; 617 if (offset >= minforkoff) 618 return offset; 619 return 0; 620 } 621 622 /* 623 * Switch on the ATTR2 superblock bit (implies also FEATURES2) 624 */ 625 STATIC void 626 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp) 627 { 628 if ((mp->m_flags & XFS_MOUNT_ATTR2) && 629 !(xfs_sb_version_hasattr2(&mp->m_sb))) { 630 spin_lock(&mp->m_sb_lock); 631 if (!xfs_sb_version_hasattr2(&mp->m_sb)) { 632 xfs_sb_version_addattr2(&mp->m_sb); 633 spin_unlock(&mp->m_sb_lock); 634 xfs_log_sb(tp); 635 } else 636 spin_unlock(&mp->m_sb_lock); 637 } 638 } 639 640 /* 641 * Create the initial contents of a shortform attribute list. 642 */ 643 void 644 xfs_attr_shortform_create( 645 struct xfs_da_args *args) 646 { 647 struct xfs_inode *dp = args->dp; 648 struct xfs_ifork *ifp = dp->i_afp; 649 struct xfs_attr_sf_hdr *hdr; 650 651 trace_xfs_attr_sf_create(args); 652 653 ASSERT(ifp->if_bytes == 0); 654 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS) { 655 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */ 656 ifp->if_format = XFS_DINODE_FMT_LOCAL; 657 ifp->if_flags |= XFS_IFINLINE; 658 } else { 659 ASSERT(ifp->if_flags & XFS_IFINLINE); 660 } 661 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK); 662 hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data; 663 memset(hdr, 0, sizeof(*hdr)); 664 hdr->totsize = cpu_to_be16(sizeof(*hdr)); 665 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA); 666 } 667 668 /* 669 * Return -EEXIST if attr is found, or -ENOATTR if not 670 * args: args containing attribute name and namelen 671 * sfep: If not null, pointer will be set to the last attr entry found on 672 -EEXIST. On -ENOATTR pointer is left at the last entry in the list 673 * basep: If not null, pointer is set to the byte offset of the entry in the 674 * list on -EEXIST. On -ENOATTR, pointer is left at the byte offset of 675 * the last entry in the list 676 */ 677 int 678 xfs_attr_sf_findname( 679 struct xfs_da_args *args, 680 struct xfs_attr_sf_entry **sfep, 681 unsigned int *basep) 682 { 683 struct xfs_attr_shortform *sf; 684 struct xfs_attr_sf_entry *sfe; 685 unsigned int base = sizeof(struct xfs_attr_sf_hdr); 686 int size = 0; 687 int end; 688 int i; 689 690 sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data; 691 sfe = &sf->list[0]; 692 end = sf->hdr.count; 693 for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe), 694 base += size, i++) { 695 size = xfs_attr_sf_entsize(sfe); 696 if (!xfs_attr_match(args, sfe->namelen, sfe->nameval, 697 sfe->flags)) 698 continue; 699 break; 700 } 701 702 if (sfep != NULL) 703 *sfep = sfe; 704 705 if (basep != NULL) 706 *basep = base; 707 708 if (i == end) 709 return -ENOATTR; 710 return -EEXIST; 711 } 712 713 /* 714 * Add a name/value pair to the shortform attribute list. 715 * Overflow from the inode has already been checked for. 716 */ 717 void 718 xfs_attr_shortform_add( 719 struct xfs_da_args *args, 720 int forkoff) 721 { 722 struct xfs_attr_shortform *sf; 723 struct xfs_attr_sf_entry *sfe; 724 int offset, size; 725 struct xfs_mount *mp; 726 struct xfs_inode *dp; 727 struct xfs_ifork *ifp; 728 729 trace_xfs_attr_sf_add(args); 730 731 dp = args->dp; 732 mp = dp->i_mount; 733 dp->i_d.di_forkoff = forkoff; 734 735 ifp = dp->i_afp; 736 ASSERT(ifp->if_flags & XFS_IFINLINE); 737 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 738 if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST) 739 ASSERT(0); 740 741 offset = (char *)sfe - (char *)sf; 742 size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen); 743 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); 744 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 745 sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset); 746 747 sfe->namelen = args->namelen; 748 sfe->valuelen = args->valuelen; 749 sfe->flags = args->attr_filter; 750 memcpy(sfe->nameval, args->name, args->namelen); 751 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen); 752 sf->hdr.count++; 753 be16_add_cpu(&sf->hdr.totsize, size); 754 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA); 755 756 xfs_sbversion_add_attr2(mp, args->trans); 757 } 758 759 /* 760 * After the last attribute is removed revert to original inode format, 761 * making all literal area available to the data fork once more. 762 */ 763 void 764 xfs_attr_fork_remove( 765 struct xfs_inode *ip, 766 struct xfs_trans *tp) 767 { 768 ASSERT(ip->i_afp->if_nextents == 0); 769 770 xfs_idestroy_fork(ip->i_afp); 771 kmem_cache_free(xfs_ifork_zone, ip->i_afp); 772 ip->i_afp = NULL; 773 ip->i_d.di_forkoff = 0; 774 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 775 } 776 777 /* 778 * Remove an attribute from the shortform attribute list structure. 779 */ 780 int 781 xfs_attr_shortform_remove( 782 struct xfs_da_args *args) 783 { 784 struct xfs_attr_shortform *sf; 785 struct xfs_attr_sf_entry *sfe; 786 int size = 0, end, totsize; 787 unsigned int base; 788 struct xfs_mount *mp; 789 struct xfs_inode *dp; 790 int error; 791 792 trace_xfs_attr_sf_remove(args); 793 794 dp = args->dp; 795 mp = dp->i_mount; 796 sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data; 797 798 error = xfs_attr_sf_findname(args, &sfe, &base); 799 if (error != -EEXIST) 800 return error; 801 size = xfs_attr_sf_entsize(sfe); 802 803 /* 804 * Fix up the attribute fork data, covering the hole 805 */ 806 end = base + size; 807 totsize = be16_to_cpu(sf->hdr.totsize); 808 if (end != totsize) 809 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end); 810 sf->hdr.count--; 811 be16_add_cpu(&sf->hdr.totsize, -size); 812 813 /* 814 * Fix up the start offset of the attribute fork 815 */ 816 totsize -= size; 817 if (totsize == sizeof(xfs_attr_sf_hdr_t) && 818 (mp->m_flags & XFS_MOUNT_ATTR2) && 819 (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) && 820 !(args->op_flags & XFS_DA_OP_ADDNAME)) { 821 xfs_attr_fork_remove(dp, args->trans); 822 } else { 823 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK); 824 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize); 825 ASSERT(dp->i_d.di_forkoff); 826 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || 827 (args->op_flags & XFS_DA_OP_ADDNAME) || 828 !(mp->m_flags & XFS_MOUNT_ATTR2) || 829 dp->i_df.if_format == XFS_DINODE_FMT_BTREE); 830 xfs_trans_log_inode(args->trans, dp, 831 XFS_ILOG_CORE | XFS_ILOG_ADATA); 832 } 833 834 xfs_sbversion_add_attr2(mp, args->trans); 835 836 return 0; 837 } 838 839 /* 840 * Look up a name in a shortform attribute list structure. 841 */ 842 /*ARGSUSED*/ 843 int 844 xfs_attr_shortform_lookup(xfs_da_args_t *args) 845 { 846 struct xfs_attr_shortform *sf; 847 struct xfs_attr_sf_entry *sfe; 848 int i; 849 struct xfs_ifork *ifp; 850 851 trace_xfs_attr_sf_lookup(args); 852 853 ifp = args->dp->i_afp; 854 ASSERT(ifp->if_flags & XFS_IFINLINE); 855 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 856 sfe = &sf->list[0]; 857 for (i = 0; i < sf->hdr.count; 858 sfe = xfs_attr_sf_nextentry(sfe), i++) { 859 if (xfs_attr_match(args, sfe->namelen, sfe->nameval, 860 sfe->flags)) 861 return -EEXIST; 862 } 863 return -ENOATTR; 864 } 865 866 /* 867 * Retrieve the attribute value and length. 868 * 869 * If args->valuelen is zero, only the length needs to be returned. Unlike a 870 * lookup, we only return an error if the attribute does not exist or we can't 871 * retrieve the value. 872 */ 873 int 874 xfs_attr_shortform_getvalue( 875 struct xfs_da_args *args) 876 { 877 struct xfs_attr_shortform *sf; 878 struct xfs_attr_sf_entry *sfe; 879 int i; 880 881 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE); 882 sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data; 883 sfe = &sf->list[0]; 884 for (i = 0; i < sf->hdr.count; 885 sfe = xfs_attr_sf_nextentry(sfe), i++) { 886 if (xfs_attr_match(args, sfe->namelen, sfe->nameval, 887 sfe->flags)) 888 return xfs_attr_copy_value(args, 889 &sfe->nameval[args->namelen], sfe->valuelen); 890 } 891 return -ENOATTR; 892 } 893 894 /* 895 * Convert from using the shortform to the leaf. On success, return the 896 * buffer so that we can keep it locked until we're totally done with it. 897 */ 898 int 899 xfs_attr_shortform_to_leaf( 900 struct xfs_da_args *args, 901 struct xfs_buf **leaf_bp) 902 { 903 struct xfs_inode *dp; 904 struct xfs_attr_shortform *sf; 905 struct xfs_attr_sf_entry *sfe; 906 struct xfs_da_args nargs; 907 char *tmpbuffer; 908 int error, i, size; 909 xfs_dablk_t blkno; 910 struct xfs_buf *bp; 911 struct xfs_ifork *ifp; 912 913 trace_xfs_attr_sf_to_leaf(args); 914 915 dp = args->dp; 916 ifp = dp->i_afp; 917 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 918 size = be16_to_cpu(sf->hdr.totsize); 919 tmpbuffer = kmem_alloc(size, 0); 920 ASSERT(tmpbuffer != NULL); 921 memcpy(tmpbuffer, ifp->if_u1.if_data, size); 922 sf = (struct xfs_attr_shortform *)tmpbuffer; 923 924 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK); 925 xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK); 926 927 bp = NULL; 928 error = xfs_da_grow_inode(args, &blkno); 929 if (error) 930 goto out; 931 932 ASSERT(blkno == 0); 933 error = xfs_attr3_leaf_create(args, blkno, &bp); 934 if (error) 935 goto out; 936 937 memset((char *)&nargs, 0, sizeof(nargs)); 938 nargs.dp = dp; 939 nargs.geo = args->geo; 940 nargs.total = args->total; 941 nargs.whichfork = XFS_ATTR_FORK; 942 nargs.trans = args->trans; 943 nargs.op_flags = XFS_DA_OP_OKNOENT; 944 945 sfe = &sf->list[0]; 946 for (i = 0; i < sf->hdr.count; i++) { 947 nargs.name = sfe->nameval; 948 nargs.namelen = sfe->namelen; 949 nargs.value = &sfe->nameval[nargs.namelen]; 950 nargs.valuelen = sfe->valuelen; 951 nargs.hashval = xfs_da_hashname(sfe->nameval, 952 sfe->namelen); 953 nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK; 954 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */ 955 ASSERT(error == -ENOATTR); 956 error = xfs_attr3_leaf_add(bp, &nargs); 957 ASSERT(error != -ENOSPC); 958 if (error) 959 goto out; 960 sfe = xfs_attr_sf_nextentry(sfe); 961 } 962 error = 0; 963 *leaf_bp = bp; 964 out: 965 kmem_free(tmpbuffer); 966 return error; 967 } 968 969 /* 970 * Check a leaf attribute block to see if all the entries would fit into 971 * a shortform attribute list. 972 */ 973 int 974 xfs_attr_shortform_allfit( 975 struct xfs_buf *bp, 976 struct xfs_inode *dp) 977 { 978 struct xfs_attr_leafblock *leaf; 979 struct xfs_attr_leaf_entry *entry; 980 xfs_attr_leaf_name_local_t *name_loc; 981 struct xfs_attr3_icleaf_hdr leafhdr; 982 int bytes; 983 int i; 984 struct xfs_mount *mp = bp->b_mount; 985 986 leaf = bp->b_addr; 987 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf); 988 entry = xfs_attr3_leaf_entryp(leaf); 989 990 bytes = sizeof(struct xfs_attr_sf_hdr); 991 for (i = 0; i < leafhdr.count; entry++, i++) { 992 if (entry->flags & XFS_ATTR_INCOMPLETE) 993 continue; /* don't copy partial entries */ 994 if (!(entry->flags & XFS_ATTR_LOCAL)) 995 return 0; 996 name_loc = xfs_attr3_leaf_name_local(leaf, i); 997 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX) 998 return 0; 999 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX) 1000 return 0; 1001 bytes += xfs_attr_sf_entsize_byname(name_loc->namelen, 1002 be16_to_cpu(name_loc->valuelen)); 1003 } 1004 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) && 1005 (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) && 1006 (bytes == sizeof(struct xfs_attr_sf_hdr))) 1007 return -1; 1008 return xfs_attr_shortform_bytesfit(dp, bytes); 1009 } 1010 1011 /* Verify the consistency of an inline attribute fork. */ 1012 xfs_failaddr_t 1013 xfs_attr_shortform_verify( 1014 struct xfs_inode *ip) 1015 { 1016 struct xfs_attr_shortform *sfp; 1017 struct xfs_attr_sf_entry *sfep; 1018 struct xfs_attr_sf_entry *next_sfep; 1019 char *endp; 1020 struct xfs_ifork *ifp; 1021 int i; 1022 int64_t size; 1023 1024 ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL); 1025 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK); 1026 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 1027 size = ifp->if_bytes; 1028 1029 /* 1030 * Give up if the attribute is way too short. 1031 */ 1032 if (size < sizeof(struct xfs_attr_sf_hdr)) 1033 return __this_address; 1034 1035 endp = (char *)sfp + size; 1036 1037 /* Check all reported entries */ 1038 sfep = &sfp->list[0]; 1039 for (i = 0; i < sfp->hdr.count; i++) { 1040 /* 1041 * struct xfs_attr_sf_entry has a variable length. 1042 * Check the fixed-offset parts of the structure are 1043 * within the data buffer. 1044 * xfs_attr_sf_entry is defined with a 1-byte variable 1045 * array at the end, so we must subtract that off. 1046 */ 1047 if (((char *)sfep + sizeof(*sfep)) >= endp) 1048 return __this_address; 1049 1050 /* Don't allow names with known bad length. */ 1051 if (sfep->namelen == 0) 1052 return __this_address; 1053 1054 /* 1055 * Check that the variable-length part of the structure is 1056 * within the data buffer. The next entry starts after the 1057 * name component, so nextentry is an acceptable test. 1058 */ 1059 next_sfep = xfs_attr_sf_nextentry(sfep); 1060 if ((char *)next_sfep > endp) 1061 return __this_address; 1062 1063 /* 1064 * Check for unknown flags. Short form doesn't support 1065 * the incomplete or local bits, so we can use the namespace 1066 * mask here. 1067 */ 1068 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK) 1069 return __this_address; 1070 1071 /* 1072 * Check for invalid namespace combinations. We only allow 1073 * one namespace flag per xattr, so we can just count the 1074 * bits (i.e. hweight) here. 1075 */ 1076 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1) 1077 return __this_address; 1078 1079 sfep = next_sfep; 1080 } 1081 if ((void *)sfep != (void *)endp) 1082 return __this_address; 1083 1084 return NULL; 1085 } 1086 1087 /* 1088 * Convert a leaf attribute list to shortform attribute list 1089 */ 1090 int 1091 xfs_attr3_leaf_to_shortform( 1092 struct xfs_buf *bp, 1093 struct xfs_da_args *args, 1094 int forkoff) 1095 { 1096 struct xfs_attr_leafblock *leaf; 1097 struct xfs_attr3_icleaf_hdr ichdr; 1098 struct xfs_attr_leaf_entry *entry; 1099 struct xfs_attr_leaf_name_local *name_loc; 1100 struct xfs_da_args nargs; 1101 struct xfs_inode *dp = args->dp; 1102 char *tmpbuffer; 1103 int error; 1104 int i; 1105 1106 trace_xfs_attr_leaf_to_sf(args); 1107 1108 tmpbuffer = kmem_alloc(args->geo->blksize, 0); 1109 if (!tmpbuffer) 1110 return -ENOMEM; 1111 1112 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize); 1113 1114 leaf = (xfs_attr_leafblock_t *)tmpbuffer; 1115 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 1116 entry = xfs_attr3_leaf_entryp(leaf); 1117 1118 /* XXX (dgc): buffer is about to be marked stale - why zero it? */ 1119 memset(bp->b_addr, 0, args->geo->blksize); 1120 1121 /* 1122 * Clean out the prior contents of the attribute list. 1123 */ 1124 error = xfs_da_shrink_inode(args, 0, bp); 1125 if (error) 1126 goto out; 1127 1128 if (forkoff == -1) { 1129 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2); 1130 ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE); 1131 xfs_attr_fork_remove(dp, args->trans); 1132 goto out; 1133 } 1134 1135 xfs_attr_shortform_create(args); 1136 1137 /* 1138 * Copy the attributes 1139 */ 1140 memset((char *)&nargs, 0, sizeof(nargs)); 1141 nargs.geo = args->geo; 1142 nargs.dp = dp; 1143 nargs.total = args->total; 1144 nargs.whichfork = XFS_ATTR_FORK; 1145 nargs.trans = args->trans; 1146 nargs.op_flags = XFS_DA_OP_OKNOENT; 1147 1148 for (i = 0; i < ichdr.count; entry++, i++) { 1149 if (entry->flags & XFS_ATTR_INCOMPLETE) 1150 continue; /* don't copy partial entries */ 1151 if (!entry->nameidx) 1152 continue; 1153 ASSERT(entry->flags & XFS_ATTR_LOCAL); 1154 name_loc = xfs_attr3_leaf_name_local(leaf, i); 1155 nargs.name = name_loc->nameval; 1156 nargs.namelen = name_loc->namelen; 1157 nargs.value = &name_loc->nameval[nargs.namelen]; 1158 nargs.valuelen = be16_to_cpu(name_loc->valuelen); 1159 nargs.hashval = be32_to_cpu(entry->hashval); 1160 nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK; 1161 xfs_attr_shortform_add(&nargs, forkoff); 1162 } 1163 error = 0; 1164 1165 out: 1166 kmem_free(tmpbuffer); 1167 return error; 1168 } 1169 1170 /* 1171 * Convert from using a single leaf to a root node and a leaf. 1172 */ 1173 int 1174 xfs_attr3_leaf_to_node( 1175 struct xfs_da_args *args) 1176 { 1177 struct xfs_attr_leafblock *leaf; 1178 struct xfs_attr3_icleaf_hdr icleafhdr; 1179 struct xfs_attr_leaf_entry *entries; 1180 struct xfs_da3_icnode_hdr icnodehdr; 1181 struct xfs_da_intnode *node; 1182 struct xfs_inode *dp = args->dp; 1183 struct xfs_mount *mp = dp->i_mount; 1184 struct xfs_buf *bp1 = NULL; 1185 struct xfs_buf *bp2 = NULL; 1186 xfs_dablk_t blkno; 1187 int error; 1188 1189 trace_xfs_attr_leaf_to_node(args); 1190 1191 error = xfs_da_grow_inode(args, &blkno); 1192 if (error) 1193 goto out; 1194 error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1); 1195 if (error) 1196 goto out; 1197 1198 error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK); 1199 if (error) 1200 goto out; 1201 1202 /* copy leaf to new buffer, update identifiers */ 1203 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF); 1204 bp2->b_ops = bp1->b_ops; 1205 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize); 1206 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1207 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr; 1208 hdr3->blkno = cpu_to_be64(bp2->b_bn); 1209 } 1210 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1); 1211 1212 /* 1213 * Set up the new root node. 1214 */ 1215 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK); 1216 if (error) 1217 goto out; 1218 node = bp1->b_addr; 1219 xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node); 1220 1221 leaf = bp2->b_addr; 1222 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf); 1223 entries = xfs_attr3_leaf_entryp(leaf); 1224 1225 /* both on-disk, don't endian-flip twice */ 1226 icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval; 1227 icnodehdr.btree[0].before = cpu_to_be32(blkno); 1228 icnodehdr.count = 1; 1229 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr); 1230 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1); 1231 error = 0; 1232 out: 1233 return error; 1234 } 1235 1236 /*======================================================================== 1237 * Routines used for growing the Btree. 1238 *========================================================================*/ 1239 1240 /* 1241 * Create the initial contents of a leaf attribute list 1242 * or a leaf in a node attribute list. 1243 */ 1244 STATIC int 1245 xfs_attr3_leaf_create( 1246 struct xfs_da_args *args, 1247 xfs_dablk_t blkno, 1248 struct xfs_buf **bpp) 1249 { 1250 struct xfs_attr_leafblock *leaf; 1251 struct xfs_attr3_icleaf_hdr ichdr; 1252 struct xfs_inode *dp = args->dp; 1253 struct xfs_mount *mp = dp->i_mount; 1254 struct xfs_buf *bp; 1255 int error; 1256 1257 trace_xfs_attr_leaf_create(args); 1258 1259 error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp, 1260 XFS_ATTR_FORK); 1261 if (error) 1262 return error; 1263 bp->b_ops = &xfs_attr3_leaf_buf_ops; 1264 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF); 1265 leaf = bp->b_addr; 1266 memset(leaf, 0, args->geo->blksize); 1267 1268 memset(&ichdr, 0, sizeof(ichdr)); 1269 ichdr.firstused = args->geo->blksize; 1270 1271 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1272 struct xfs_da3_blkinfo *hdr3 = bp->b_addr; 1273 1274 ichdr.magic = XFS_ATTR3_LEAF_MAGIC; 1275 1276 hdr3->blkno = cpu_to_be64(bp->b_bn); 1277 hdr3->owner = cpu_to_be64(dp->i_ino); 1278 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid); 1279 1280 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr); 1281 } else { 1282 ichdr.magic = XFS_ATTR_LEAF_MAGIC; 1283 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr); 1284 } 1285 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base; 1286 1287 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 1288 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1); 1289 1290 *bpp = bp; 1291 return 0; 1292 } 1293 1294 /* 1295 * Split the leaf node, rebalance, then add the new entry. 1296 */ 1297 int 1298 xfs_attr3_leaf_split( 1299 struct xfs_da_state *state, 1300 struct xfs_da_state_blk *oldblk, 1301 struct xfs_da_state_blk *newblk) 1302 { 1303 xfs_dablk_t blkno; 1304 int error; 1305 1306 trace_xfs_attr_leaf_split(state->args); 1307 1308 /* 1309 * Allocate space for a new leaf node. 1310 */ 1311 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC); 1312 error = xfs_da_grow_inode(state->args, &blkno); 1313 if (error) 1314 return error; 1315 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp); 1316 if (error) 1317 return error; 1318 newblk->blkno = blkno; 1319 newblk->magic = XFS_ATTR_LEAF_MAGIC; 1320 1321 /* 1322 * Rebalance the entries across the two leaves. 1323 * NOTE: rebalance() currently depends on the 2nd block being empty. 1324 */ 1325 xfs_attr3_leaf_rebalance(state, oldblk, newblk); 1326 error = xfs_da3_blk_link(state, oldblk, newblk); 1327 if (error) 1328 return error; 1329 1330 /* 1331 * Save info on "old" attribute for "atomic rename" ops, leaf_add() 1332 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the 1333 * "new" attrs info. Will need the "old" info to remove it later. 1334 * 1335 * Insert the "new" entry in the correct block. 1336 */ 1337 if (state->inleaf) { 1338 trace_xfs_attr_leaf_add_old(state->args); 1339 error = xfs_attr3_leaf_add(oldblk->bp, state->args); 1340 } else { 1341 trace_xfs_attr_leaf_add_new(state->args); 1342 error = xfs_attr3_leaf_add(newblk->bp, state->args); 1343 } 1344 1345 /* 1346 * Update last hashval in each block since we added the name. 1347 */ 1348 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL); 1349 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL); 1350 return error; 1351 } 1352 1353 /* 1354 * Add a name to the leaf attribute list structure. 1355 */ 1356 int 1357 xfs_attr3_leaf_add( 1358 struct xfs_buf *bp, 1359 struct xfs_da_args *args) 1360 { 1361 struct xfs_attr_leafblock *leaf; 1362 struct xfs_attr3_icleaf_hdr ichdr; 1363 int tablesize; 1364 int entsize; 1365 int sum; 1366 int tmp; 1367 int i; 1368 1369 trace_xfs_attr_leaf_add(args); 1370 1371 leaf = bp->b_addr; 1372 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 1373 ASSERT(args->index >= 0 && args->index <= ichdr.count); 1374 entsize = xfs_attr_leaf_newentsize(args, NULL); 1375 1376 /* 1377 * Search through freemap for first-fit on new name length. 1378 * (may need to figure in size of entry struct too) 1379 */ 1380 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t) 1381 + xfs_attr3_leaf_hdr_size(leaf); 1382 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) { 1383 if (tablesize > ichdr.firstused) { 1384 sum += ichdr.freemap[i].size; 1385 continue; 1386 } 1387 if (!ichdr.freemap[i].size) 1388 continue; /* no space in this map */ 1389 tmp = entsize; 1390 if (ichdr.freemap[i].base < ichdr.firstused) 1391 tmp += sizeof(xfs_attr_leaf_entry_t); 1392 if (ichdr.freemap[i].size >= tmp) { 1393 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i); 1394 goto out_log_hdr; 1395 } 1396 sum += ichdr.freemap[i].size; 1397 } 1398 1399 /* 1400 * If there are no holes in the address space of the block, 1401 * and we don't have enough freespace, then compaction will do us 1402 * no good and we should just give up. 1403 */ 1404 if (!ichdr.holes && sum < entsize) 1405 return -ENOSPC; 1406 1407 /* 1408 * Compact the entries to coalesce free space. 1409 * This may change the hdr->count via dropping INCOMPLETE entries. 1410 */ 1411 xfs_attr3_leaf_compact(args, &ichdr, bp); 1412 1413 /* 1414 * After compaction, the block is guaranteed to have only one 1415 * free region, in freemap[0]. If it is not big enough, give up. 1416 */ 1417 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) { 1418 tmp = -ENOSPC; 1419 goto out_log_hdr; 1420 } 1421 1422 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0); 1423 1424 out_log_hdr: 1425 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 1426 xfs_trans_log_buf(args->trans, bp, 1427 XFS_DA_LOGRANGE(leaf, &leaf->hdr, 1428 xfs_attr3_leaf_hdr_size(leaf))); 1429 return tmp; 1430 } 1431 1432 /* 1433 * Add a name to a leaf attribute list structure. 1434 */ 1435 STATIC int 1436 xfs_attr3_leaf_add_work( 1437 struct xfs_buf *bp, 1438 struct xfs_attr3_icleaf_hdr *ichdr, 1439 struct xfs_da_args *args, 1440 int mapindex) 1441 { 1442 struct xfs_attr_leafblock *leaf; 1443 struct xfs_attr_leaf_entry *entry; 1444 struct xfs_attr_leaf_name_local *name_loc; 1445 struct xfs_attr_leaf_name_remote *name_rmt; 1446 struct xfs_mount *mp; 1447 int tmp; 1448 int i; 1449 1450 trace_xfs_attr_leaf_add_work(args); 1451 1452 leaf = bp->b_addr; 1453 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE); 1454 ASSERT(args->index >= 0 && args->index <= ichdr->count); 1455 1456 /* 1457 * Force open some space in the entry array and fill it in. 1458 */ 1459 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 1460 if (args->index < ichdr->count) { 1461 tmp = ichdr->count - args->index; 1462 tmp *= sizeof(xfs_attr_leaf_entry_t); 1463 memmove(entry + 1, entry, tmp); 1464 xfs_trans_log_buf(args->trans, bp, 1465 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry))); 1466 } 1467 ichdr->count++; 1468 1469 /* 1470 * Allocate space for the new string (at the end of the run). 1471 */ 1472 mp = args->trans->t_mountp; 1473 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize); 1474 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0); 1475 ASSERT(ichdr->freemap[mapindex].size >= 1476 xfs_attr_leaf_newentsize(args, NULL)); 1477 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize); 1478 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0); 1479 1480 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp); 1481 1482 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base + 1483 ichdr->freemap[mapindex].size); 1484 entry->hashval = cpu_to_be32(args->hashval); 1485 entry->flags = args->attr_filter; 1486 if (tmp) 1487 entry->flags |= XFS_ATTR_LOCAL; 1488 if (args->op_flags & XFS_DA_OP_RENAME) { 1489 entry->flags |= XFS_ATTR_INCOMPLETE; 1490 if ((args->blkno2 == args->blkno) && 1491 (args->index2 <= args->index)) { 1492 args->index2++; 1493 } 1494 } 1495 xfs_trans_log_buf(args->trans, bp, 1496 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 1497 ASSERT((args->index == 0) || 1498 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval))); 1499 ASSERT((args->index == ichdr->count - 1) || 1500 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval))); 1501 1502 /* 1503 * For "remote" attribute values, simply note that we need to 1504 * allocate space for the "remote" value. We can't actually 1505 * allocate the extents in this transaction, and we can't decide 1506 * which blocks they should be as we might allocate more blocks 1507 * as part of this transaction (a split operation for example). 1508 */ 1509 if (entry->flags & XFS_ATTR_LOCAL) { 1510 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 1511 name_loc->namelen = args->namelen; 1512 name_loc->valuelen = cpu_to_be16(args->valuelen); 1513 memcpy((char *)name_loc->nameval, args->name, args->namelen); 1514 memcpy((char *)&name_loc->nameval[args->namelen], args->value, 1515 be16_to_cpu(name_loc->valuelen)); 1516 } else { 1517 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 1518 name_rmt->namelen = args->namelen; 1519 memcpy((char *)name_rmt->name, args->name, args->namelen); 1520 entry->flags |= XFS_ATTR_INCOMPLETE; 1521 /* just in case */ 1522 name_rmt->valuelen = 0; 1523 name_rmt->valueblk = 0; 1524 args->rmtblkno = 1; 1525 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen); 1526 args->rmtvaluelen = args->valuelen; 1527 } 1528 xfs_trans_log_buf(args->trans, bp, 1529 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index), 1530 xfs_attr_leaf_entsize(leaf, args->index))); 1531 1532 /* 1533 * Update the control info for this leaf node 1534 */ 1535 if (be16_to_cpu(entry->nameidx) < ichdr->firstused) 1536 ichdr->firstused = be16_to_cpu(entry->nameidx); 1537 1538 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t) 1539 + xfs_attr3_leaf_hdr_size(leaf)); 1540 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t) 1541 + xfs_attr3_leaf_hdr_size(leaf); 1542 1543 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 1544 if (ichdr->freemap[i].base == tmp) { 1545 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t); 1546 ichdr->freemap[i].size -= 1547 min_t(uint16_t, ichdr->freemap[i].size, 1548 sizeof(xfs_attr_leaf_entry_t)); 1549 } 1550 } 1551 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index); 1552 return 0; 1553 } 1554 1555 /* 1556 * Garbage collect a leaf attribute list block by copying it to a new buffer. 1557 */ 1558 STATIC void 1559 xfs_attr3_leaf_compact( 1560 struct xfs_da_args *args, 1561 struct xfs_attr3_icleaf_hdr *ichdr_dst, 1562 struct xfs_buf *bp) 1563 { 1564 struct xfs_attr_leafblock *leaf_src; 1565 struct xfs_attr_leafblock *leaf_dst; 1566 struct xfs_attr3_icleaf_hdr ichdr_src; 1567 struct xfs_trans *trans = args->trans; 1568 char *tmpbuffer; 1569 1570 trace_xfs_attr_leaf_compact(args); 1571 1572 tmpbuffer = kmem_alloc(args->geo->blksize, 0); 1573 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize); 1574 memset(bp->b_addr, 0, args->geo->blksize); 1575 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer; 1576 leaf_dst = bp->b_addr; 1577 1578 /* 1579 * Copy the on-disk header back into the destination buffer to ensure 1580 * all the information in the header that is not part of the incore 1581 * header structure is preserved. 1582 */ 1583 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src)); 1584 1585 /* Initialise the incore headers */ 1586 ichdr_src = *ichdr_dst; /* struct copy */ 1587 ichdr_dst->firstused = args->geo->blksize; 1588 ichdr_dst->usedbytes = 0; 1589 ichdr_dst->count = 0; 1590 ichdr_dst->holes = 0; 1591 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src); 1592 ichdr_dst->freemap[0].size = ichdr_dst->firstused - 1593 ichdr_dst->freemap[0].base; 1594 1595 /* write the header back to initialise the underlying buffer */ 1596 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst); 1597 1598 /* 1599 * Copy all entry's in the same (sorted) order, 1600 * but allocate name/value pairs packed and in sequence. 1601 */ 1602 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0, 1603 leaf_dst, ichdr_dst, 0, ichdr_src.count); 1604 /* 1605 * this logs the entire buffer, but the caller must write the header 1606 * back to the buffer when it is finished modifying it. 1607 */ 1608 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1); 1609 1610 kmem_free(tmpbuffer); 1611 } 1612 1613 /* 1614 * Compare two leaf blocks "order". 1615 * Return 0 unless leaf2 should go before leaf1. 1616 */ 1617 static int 1618 xfs_attr3_leaf_order( 1619 struct xfs_buf *leaf1_bp, 1620 struct xfs_attr3_icleaf_hdr *leaf1hdr, 1621 struct xfs_buf *leaf2_bp, 1622 struct xfs_attr3_icleaf_hdr *leaf2hdr) 1623 { 1624 struct xfs_attr_leaf_entry *entries1; 1625 struct xfs_attr_leaf_entry *entries2; 1626 1627 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr); 1628 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr); 1629 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 && 1630 ((be32_to_cpu(entries2[0].hashval) < 1631 be32_to_cpu(entries1[0].hashval)) || 1632 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) < 1633 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) { 1634 return 1; 1635 } 1636 return 0; 1637 } 1638 1639 int 1640 xfs_attr_leaf_order( 1641 struct xfs_buf *leaf1_bp, 1642 struct xfs_buf *leaf2_bp) 1643 { 1644 struct xfs_attr3_icleaf_hdr ichdr1; 1645 struct xfs_attr3_icleaf_hdr ichdr2; 1646 struct xfs_mount *mp = leaf1_bp->b_mount; 1647 1648 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr); 1649 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr); 1650 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2); 1651 } 1652 1653 /* 1654 * Redistribute the attribute list entries between two leaf nodes, 1655 * taking into account the size of the new entry. 1656 * 1657 * NOTE: if new block is empty, then it will get the upper half of the 1658 * old block. At present, all (one) callers pass in an empty second block. 1659 * 1660 * This code adjusts the args->index/blkno and args->index2/blkno2 fields 1661 * to match what it is doing in splitting the attribute leaf block. Those 1662 * values are used in "atomic rename" operations on attributes. Note that 1663 * the "new" and "old" values can end up in different blocks. 1664 */ 1665 STATIC void 1666 xfs_attr3_leaf_rebalance( 1667 struct xfs_da_state *state, 1668 struct xfs_da_state_blk *blk1, 1669 struct xfs_da_state_blk *blk2) 1670 { 1671 struct xfs_da_args *args; 1672 struct xfs_attr_leafblock *leaf1; 1673 struct xfs_attr_leafblock *leaf2; 1674 struct xfs_attr3_icleaf_hdr ichdr1; 1675 struct xfs_attr3_icleaf_hdr ichdr2; 1676 struct xfs_attr_leaf_entry *entries1; 1677 struct xfs_attr_leaf_entry *entries2; 1678 int count; 1679 int totallen; 1680 int max; 1681 int space; 1682 int swap; 1683 1684 /* 1685 * Set up environment. 1686 */ 1687 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC); 1688 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC); 1689 leaf1 = blk1->bp->b_addr; 1690 leaf2 = blk2->bp->b_addr; 1691 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1); 1692 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2); 1693 ASSERT(ichdr2.count == 0); 1694 args = state->args; 1695 1696 trace_xfs_attr_leaf_rebalance(args); 1697 1698 /* 1699 * Check ordering of blocks, reverse if it makes things simpler. 1700 * 1701 * NOTE: Given that all (current) callers pass in an empty 1702 * second block, this code should never set "swap". 1703 */ 1704 swap = 0; 1705 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) { 1706 swap(blk1, blk2); 1707 1708 /* swap structures rather than reconverting them */ 1709 swap(ichdr1, ichdr2); 1710 1711 leaf1 = blk1->bp->b_addr; 1712 leaf2 = blk2->bp->b_addr; 1713 swap = 1; 1714 } 1715 1716 /* 1717 * Examine entries until we reduce the absolute difference in 1718 * byte usage between the two blocks to a minimum. Then get 1719 * the direction to copy and the number of elements to move. 1720 * 1721 * "inleaf" is true if the new entry should be inserted into blk1. 1722 * If "swap" is also true, then reverse the sense of "inleaf". 1723 */ 1724 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1, 1725 blk2, &ichdr2, 1726 &count, &totallen); 1727 if (swap) 1728 state->inleaf = !state->inleaf; 1729 1730 /* 1731 * Move any entries required from leaf to leaf: 1732 */ 1733 if (count < ichdr1.count) { 1734 /* 1735 * Figure the total bytes to be added to the destination leaf. 1736 */ 1737 /* number entries being moved */ 1738 count = ichdr1.count - count; 1739 space = ichdr1.usedbytes - totallen; 1740 space += count * sizeof(xfs_attr_leaf_entry_t); 1741 1742 /* 1743 * leaf2 is the destination, compact it if it looks tight. 1744 */ 1745 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1); 1746 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t); 1747 if (space > max) 1748 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp); 1749 1750 /* 1751 * Move high entries from leaf1 to low end of leaf2. 1752 */ 1753 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1, 1754 ichdr1.count - count, leaf2, &ichdr2, 0, count); 1755 1756 } else if (count > ichdr1.count) { 1757 /* 1758 * I assert that since all callers pass in an empty 1759 * second buffer, this code should never execute. 1760 */ 1761 ASSERT(0); 1762 1763 /* 1764 * Figure the total bytes to be added to the destination leaf. 1765 */ 1766 /* number entries being moved */ 1767 count -= ichdr1.count; 1768 space = totallen - ichdr1.usedbytes; 1769 space += count * sizeof(xfs_attr_leaf_entry_t); 1770 1771 /* 1772 * leaf1 is the destination, compact it if it looks tight. 1773 */ 1774 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1); 1775 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t); 1776 if (space > max) 1777 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp); 1778 1779 /* 1780 * Move low entries from leaf2 to high end of leaf1. 1781 */ 1782 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1, 1783 ichdr1.count, count); 1784 } 1785 1786 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1); 1787 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2); 1788 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1); 1789 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1); 1790 1791 /* 1792 * Copy out last hashval in each block for B-tree code. 1793 */ 1794 entries1 = xfs_attr3_leaf_entryp(leaf1); 1795 entries2 = xfs_attr3_leaf_entryp(leaf2); 1796 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval); 1797 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval); 1798 1799 /* 1800 * Adjust the expected index for insertion. 1801 * NOTE: this code depends on the (current) situation that the 1802 * second block was originally empty. 1803 * 1804 * If the insertion point moved to the 2nd block, we must adjust 1805 * the index. We must also track the entry just following the 1806 * new entry for use in an "atomic rename" operation, that entry 1807 * is always the "old" entry and the "new" entry is what we are 1808 * inserting. The index/blkno fields refer to the "old" entry, 1809 * while the index2/blkno2 fields refer to the "new" entry. 1810 */ 1811 if (blk1->index > ichdr1.count) { 1812 ASSERT(state->inleaf == 0); 1813 blk2->index = blk1->index - ichdr1.count; 1814 args->index = args->index2 = blk2->index; 1815 args->blkno = args->blkno2 = blk2->blkno; 1816 } else if (blk1->index == ichdr1.count) { 1817 if (state->inleaf) { 1818 args->index = blk1->index; 1819 args->blkno = blk1->blkno; 1820 args->index2 = 0; 1821 args->blkno2 = blk2->blkno; 1822 } else { 1823 /* 1824 * On a double leaf split, the original attr location 1825 * is already stored in blkno2/index2, so don't 1826 * overwrite it overwise we corrupt the tree. 1827 */ 1828 blk2->index = blk1->index - ichdr1.count; 1829 args->index = blk2->index; 1830 args->blkno = blk2->blkno; 1831 if (!state->extravalid) { 1832 /* 1833 * set the new attr location to match the old 1834 * one and let the higher level split code 1835 * decide where in the leaf to place it. 1836 */ 1837 args->index2 = blk2->index; 1838 args->blkno2 = blk2->blkno; 1839 } 1840 } 1841 } else { 1842 ASSERT(state->inleaf == 1); 1843 args->index = args->index2 = blk1->index; 1844 args->blkno = args->blkno2 = blk1->blkno; 1845 } 1846 } 1847 1848 /* 1849 * Examine entries until we reduce the absolute difference in 1850 * byte usage between the two blocks to a minimum. 1851 * GROT: Is this really necessary? With other than a 512 byte blocksize, 1852 * GROT: there will always be enough room in either block for a new entry. 1853 * GROT: Do a double-split for this case? 1854 */ 1855 STATIC int 1856 xfs_attr3_leaf_figure_balance( 1857 struct xfs_da_state *state, 1858 struct xfs_da_state_blk *blk1, 1859 struct xfs_attr3_icleaf_hdr *ichdr1, 1860 struct xfs_da_state_blk *blk2, 1861 struct xfs_attr3_icleaf_hdr *ichdr2, 1862 int *countarg, 1863 int *usedbytesarg) 1864 { 1865 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr; 1866 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr; 1867 struct xfs_attr_leaf_entry *entry; 1868 int count; 1869 int max; 1870 int index; 1871 int totallen = 0; 1872 int half; 1873 int lastdelta; 1874 int foundit = 0; 1875 int tmp; 1876 1877 /* 1878 * Examine entries until we reduce the absolute difference in 1879 * byte usage between the two blocks to a minimum. 1880 */ 1881 max = ichdr1->count + ichdr2->count; 1882 half = (max + 1) * sizeof(*entry); 1883 half += ichdr1->usedbytes + ichdr2->usedbytes + 1884 xfs_attr_leaf_newentsize(state->args, NULL); 1885 half /= 2; 1886 lastdelta = state->args->geo->blksize; 1887 entry = xfs_attr3_leaf_entryp(leaf1); 1888 for (count = index = 0; count < max; entry++, index++, count++) { 1889 1890 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A)) 1891 /* 1892 * The new entry is in the first block, account for it. 1893 */ 1894 if (count == blk1->index) { 1895 tmp = totallen + sizeof(*entry) + 1896 xfs_attr_leaf_newentsize(state->args, NULL); 1897 if (XFS_ATTR_ABS(half - tmp) > lastdelta) 1898 break; 1899 lastdelta = XFS_ATTR_ABS(half - tmp); 1900 totallen = tmp; 1901 foundit = 1; 1902 } 1903 1904 /* 1905 * Wrap around into the second block if necessary. 1906 */ 1907 if (count == ichdr1->count) { 1908 leaf1 = leaf2; 1909 entry = xfs_attr3_leaf_entryp(leaf1); 1910 index = 0; 1911 } 1912 1913 /* 1914 * Figure out if next leaf entry would be too much. 1915 */ 1916 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1, 1917 index); 1918 if (XFS_ATTR_ABS(half - tmp) > lastdelta) 1919 break; 1920 lastdelta = XFS_ATTR_ABS(half - tmp); 1921 totallen = tmp; 1922 #undef XFS_ATTR_ABS 1923 } 1924 1925 /* 1926 * Calculate the number of usedbytes that will end up in lower block. 1927 * If new entry not in lower block, fix up the count. 1928 */ 1929 totallen -= count * sizeof(*entry); 1930 if (foundit) { 1931 totallen -= sizeof(*entry) + 1932 xfs_attr_leaf_newentsize(state->args, NULL); 1933 } 1934 1935 *countarg = count; 1936 *usedbytesarg = totallen; 1937 return foundit; 1938 } 1939 1940 /*======================================================================== 1941 * Routines used for shrinking the Btree. 1942 *========================================================================*/ 1943 1944 /* 1945 * Check a leaf block and its neighbors to see if the block should be 1946 * collapsed into one or the other neighbor. Always keep the block 1947 * with the smaller block number. 1948 * If the current block is over 50% full, don't try to join it, return 0. 1949 * If the block is empty, fill in the state structure and return 2. 1950 * If it can be collapsed, fill in the state structure and return 1. 1951 * If nothing can be done, return 0. 1952 * 1953 * GROT: allow for INCOMPLETE entries in calculation. 1954 */ 1955 int 1956 xfs_attr3_leaf_toosmall( 1957 struct xfs_da_state *state, 1958 int *action) 1959 { 1960 struct xfs_attr_leafblock *leaf; 1961 struct xfs_da_state_blk *blk; 1962 struct xfs_attr3_icleaf_hdr ichdr; 1963 struct xfs_buf *bp; 1964 xfs_dablk_t blkno; 1965 int bytes; 1966 int forward; 1967 int error; 1968 int retval; 1969 int i; 1970 1971 trace_xfs_attr_leaf_toosmall(state->args); 1972 1973 /* 1974 * Check for the degenerate case of the block being over 50% full. 1975 * If so, it's not worth even looking to see if we might be able 1976 * to coalesce with a sibling. 1977 */ 1978 blk = &state->path.blk[ state->path.active-1 ]; 1979 leaf = blk->bp->b_addr; 1980 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf); 1981 bytes = xfs_attr3_leaf_hdr_size(leaf) + 1982 ichdr.count * sizeof(xfs_attr_leaf_entry_t) + 1983 ichdr.usedbytes; 1984 if (bytes > (state->args->geo->blksize >> 1)) { 1985 *action = 0; /* blk over 50%, don't try to join */ 1986 return 0; 1987 } 1988 1989 /* 1990 * Check for the degenerate case of the block being empty. 1991 * If the block is empty, we'll simply delete it, no need to 1992 * coalesce it with a sibling block. We choose (arbitrarily) 1993 * to merge with the forward block unless it is NULL. 1994 */ 1995 if (ichdr.count == 0) { 1996 /* 1997 * Make altpath point to the block we want to keep and 1998 * path point to the block we want to drop (this one). 1999 */ 2000 forward = (ichdr.forw != 0); 2001 memcpy(&state->altpath, &state->path, sizeof(state->path)); 2002 error = xfs_da3_path_shift(state, &state->altpath, forward, 2003 0, &retval); 2004 if (error) 2005 return error; 2006 if (retval) { 2007 *action = 0; 2008 } else { 2009 *action = 2; 2010 } 2011 return 0; 2012 } 2013 2014 /* 2015 * Examine each sibling block to see if we can coalesce with 2016 * at least 25% free space to spare. We need to figure out 2017 * whether to merge with the forward or the backward block. 2018 * We prefer coalescing with the lower numbered sibling so as 2019 * to shrink an attribute list over time. 2020 */ 2021 /* start with smaller blk num */ 2022 forward = ichdr.forw < ichdr.back; 2023 for (i = 0; i < 2; forward = !forward, i++) { 2024 struct xfs_attr3_icleaf_hdr ichdr2; 2025 if (forward) 2026 blkno = ichdr.forw; 2027 else 2028 blkno = ichdr.back; 2029 if (blkno == 0) 2030 continue; 2031 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp, 2032 blkno, &bp); 2033 if (error) 2034 return error; 2035 2036 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr); 2037 2038 bytes = state->args->geo->blksize - 2039 (state->args->geo->blksize >> 2) - 2040 ichdr.usedbytes - ichdr2.usedbytes - 2041 ((ichdr.count + ichdr2.count) * 2042 sizeof(xfs_attr_leaf_entry_t)) - 2043 xfs_attr3_leaf_hdr_size(leaf); 2044 2045 xfs_trans_brelse(state->args->trans, bp); 2046 if (bytes >= 0) 2047 break; /* fits with at least 25% to spare */ 2048 } 2049 if (i >= 2) { 2050 *action = 0; 2051 return 0; 2052 } 2053 2054 /* 2055 * Make altpath point to the block we want to keep (the lower 2056 * numbered block) and path point to the block we want to drop. 2057 */ 2058 memcpy(&state->altpath, &state->path, sizeof(state->path)); 2059 if (blkno < blk->blkno) { 2060 error = xfs_da3_path_shift(state, &state->altpath, forward, 2061 0, &retval); 2062 } else { 2063 error = xfs_da3_path_shift(state, &state->path, forward, 2064 0, &retval); 2065 } 2066 if (error) 2067 return error; 2068 if (retval) { 2069 *action = 0; 2070 } else { 2071 *action = 1; 2072 } 2073 return 0; 2074 } 2075 2076 /* 2077 * Remove a name from the leaf attribute list structure. 2078 * 2079 * Return 1 if leaf is less than 37% full, 0 if >= 37% full. 2080 * If two leaves are 37% full, when combined they will leave 25% free. 2081 */ 2082 int 2083 xfs_attr3_leaf_remove( 2084 struct xfs_buf *bp, 2085 struct xfs_da_args *args) 2086 { 2087 struct xfs_attr_leafblock *leaf; 2088 struct xfs_attr3_icleaf_hdr ichdr; 2089 struct xfs_attr_leaf_entry *entry; 2090 int before; 2091 int after; 2092 int smallest; 2093 int entsize; 2094 int tablesize; 2095 int tmp; 2096 int i; 2097 2098 trace_xfs_attr_leaf_remove(args); 2099 2100 leaf = bp->b_addr; 2101 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2102 2103 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8); 2104 ASSERT(args->index >= 0 && args->index < ichdr.count); 2105 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) + 2106 xfs_attr3_leaf_hdr_size(leaf)); 2107 2108 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2109 2110 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused); 2111 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize); 2112 2113 /* 2114 * Scan through free region table: 2115 * check for adjacency of free'd entry with an existing one, 2116 * find smallest free region in case we need to replace it, 2117 * adjust any map that borders the entry table, 2118 */ 2119 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t) 2120 + xfs_attr3_leaf_hdr_size(leaf); 2121 tmp = ichdr.freemap[0].size; 2122 before = after = -1; 2123 smallest = XFS_ATTR_LEAF_MAPSIZE - 1; 2124 entsize = xfs_attr_leaf_entsize(leaf, args->index); 2125 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 2126 ASSERT(ichdr.freemap[i].base < args->geo->blksize); 2127 ASSERT(ichdr.freemap[i].size < args->geo->blksize); 2128 if (ichdr.freemap[i].base == tablesize) { 2129 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t); 2130 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t); 2131 } 2132 2133 if (ichdr.freemap[i].base + ichdr.freemap[i].size == 2134 be16_to_cpu(entry->nameidx)) { 2135 before = i; 2136 } else if (ichdr.freemap[i].base == 2137 (be16_to_cpu(entry->nameidx) + entsize)) { 2138 after = i; 2139 } else if (ichdr.freemap[i].size < tmp) { 2140 tmp = ichdr.freemap[i].size; 2141 smallest = i; 2142 } 2143 } 2144 2145 /* 2146 * Coalesce adjacent freemap regions, 2147 * or replace the smallest region. 2148 */ 2149 if ((before >= 0) || (after >= 0)) { 2150 if ((before >= 0) && (after >= 0)) { 2151 ichdr.freemap[before].size += entsize; 2152 ichdr.freemap[before].size += ichdr.freemap[after].size; 2153 ichdr.freemap[after].base = 0; 2154 ichdr.freemap[after].size = 0; 2155 } else if (before >= 0) { 2156 ichdr.freemap[before].size += entsize; 2157 } else { 2158 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx); 2159 ichdr.freemap[after].size += entsize; 2160 } 2161 } else { 2162 /* 2163 * Replace smallest region (if it is smaller than free'd entry) 2164 */ 2165 if (ichdr.freemap[smallest].size < entsize) { 2166 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx); 2167 ichdr.freemap[smallest].size = entsize; 2168 } 2169 } 2170 2171 /* 2172 * Did we remove the first entry? 2173 */ 2174 if (be16_to_cpu(entry->nameidx) == ichdr.firstused) 2175 smallest = 1; 2176 else 2177 smallest = 0; 2178 2179 /* 2180 * Compress the remaining entries and zero out the removed stuff. 2181 */ 2182 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize); 2183 ichdr.usedbytes -= entsize; 2184 xfs_trans_log_buf(args->trans, bp, 2185 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index), 2186 entsize)); 2187 2188 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t); 2189 memmove(entry, entry + 1, tmp); 2190 ichdr.count--; 2191 xfs_trans_log_buf(args->trans, bp, 2192 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t))); 2193 2194 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count]; 2195 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t)); 2196 2197 /* 2198 * If we removed the first entry, re-find the first used byte 2199 * in the name area. Note that if the entry was the "firstused", 2200 * then we don't have a "hole" in our block resulting from 2201 * removing the name. 2202 */ 2203 if (smallest) { 2204 tmp = args->geo->blksize; 2205 entry = xfs_attr3_leaf_entryp(leaf); 2206 for (i = ichdr.count - 1; i >= 0; entry++, i--) { 2207 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused); 2208 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize); 2209 2210 if (be16_to_cpu(entry->nameidx) < tmp) 2211 tmp = be16_to_cpu(entry->nameidx); 2212 } 2213 ichdr.firstused = tmp; 2214 ASSERT(ichdr.firstused != 0); 2215 } else { 2216 ichdr.holes = 1; /* mark as needing compaction */ 2217 } 2218 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 2219 xfs_trans_log_buf(args->trans, bp, 2220 XFS_DA_LOGRANGE(leaf, &leaf->hdr, 2221 xfs_attr3_leaf_hdr_size(leaf))); 2222 2223 /* 2224 * Check if leaf is less than 50% full, caller may want to 2225 * "join" the leaf with a sibling if so. 2226 */ 2227 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) + 2228 ichdr.count * sizeof(xfs_attr_leaf_entry_t); 2229 2230 return tmp < args->geo->magicpct; /* leaf is < 37% full */ 2231 } 2232 2233 /* 2234 * Move all the attribute list entries from drop_leaf into save_leaf. 2235 */ 2236 void 2237 xfs_attr3_leaf_unbalance( 2238 struct xfs_da_state *state, 2239 struct xfs_da_state_blk *drop_blk, 2240 struct xfs_da_state_blk *save_blk) 2241 { 2242 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr; 2243 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr; 2244 struct xfs_attr3_icleaf_hdr drophdr; 2245 struct xfs_attr3_icleaf_hdr savehdr; 2246 struct xfs_attr_leaf_entry *entry; 2247 2248 trace_xfs_attr_leaf_unbalance(state->args); 2249 2250 drop_leaf = drop_blk->bp->b_addr; 2251 save_leaf = save_blk->bp->b_addr; 2252 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf); 2253 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf); 2254 entry = xfs_attr3_leaf_entryp(drop_leaf); 2255 2256 /* 2257 * Save last hashval from dying block for later Btree fixup. 2258 */ 2259 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval); 2260 2261 /* 2262 * Check if we need a temp buffer, or can we do it in place. 2263 * Note that we don't check "leaf" for holes because we will 2264 * always be dropping it, toosmall() decided that for us already. 2265 */ 2266 if (savehdr.holes == 0) { 2267 /* 2268 * dest leaf has no holes, so we add there. May need 2269 * to make some room in the entry array. 2270 */ 2271 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr, 2272 drop_blk->bp, &drophdr)) { 2273 xfs_attr3_leaf_moveents(state->args, 2274 drop_leaf, &drophdr, 0, 2275 save_leaf, &savehdr, 0, 2276 drophdr.count); 2277 } else { 2278 xfs_attr3_leaf_moveents(state->args, 2279 drop_leaf, &drophdr, 0, 2280 save_leaf, &savehdr, 2281 savehdr.count, drophdr.count); 2282 } 2283 } else { 2284 /* 2285 * Destination has holes, so we make a temporary copy 2286 * of the leaf and add them both to that. 2287 */ 2288 struct xfs_attr_leafblock *tmp_leaf; 2289 struct xfs_attr3_icleaf_hdr tmphdr; 2290 2291 tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0); 2292 2293 /* 2294 * Copy the header into the temp leaf so that all the stuff 2295 * not in the incore header is present and gets copied back in 2296 * once we've moved all the entries. 2297 */ 2298 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf)); 2299 2300 memset(&tmphdr, 0, sizeof(tmphdr)); 2301 tmphdr.magic = savehdr.magic; 2302 tmphdr.forw = savehdr.forw; 2303 tmphdr.back = savehdr.back; 2304 tmphdr.firstused = state->args->geo->blksize; 2305 2306 /* write the header to the temp buffer to initialise it */ 2307 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr); 2308 2309 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr, 2310 drop_blk->bp, &drophdr)) { 2311 xfs_attr3_leaf_moveents(state->args, 2312 drop_leaf, &drophdr, 0, 2313 tmp_leaf, &tmphdr, 0, 2314 drophdr.count); 2315 xfs_attr3_leaf_moveents(state->args, 2316 save_leaf, &savehdr, 0, 2317 tmp_leaf, &tmphdr, tmphdr.count, 2318 savehdr.count); 2319 } else { 2320 xfs_attr3_leaf_moveents(state->args, 2321 save_leaf, &savehdr, 0, 2322 tmp_leaf, &tmphdr, 0, 2323 savehdr.count); 2324 xfs_attr3_leaf_moveents(state->args, 2325 drop_leaf, &drophdr, 0, 2326 tmp_leaf, &tmphdr, tmphdr.count, 2327 drophdr.count); 2328 } 2329 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize); 2330 savehdr = tmphdr; /* struct copy */ 2331 kmem_free(tmp_leaf); 2332 } 2333 2334 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr); 2335 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0, 2336 state->args->geo->blksize - 1); 2337 2338 /* 2339 * Copy out last hashval in each block for B-tree code. 2340 */ 2341 entry = xfs_attr3_leaf_entryp(save_leaf); 2342 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval); 2343 } 2344 2345 /*======================================================================== 2346 * Routines used for finding things in the Btree. 2347 *========================================================================*/ 2348 2349 /* 2350 * Look up a name in a leaf attribute list structure. 2351 * This is the internal routine, it uses the caller's buffer. 2352 * 2353 * Note that duplicate keys are allowed, but only check within the 2354 * current leaf node. The Btree code must check in adjacent leaf nodes. 2355 * 2356 * Return in args->index the index into the entry[] array of either 2357 * the found entry, or where the entry should have been (insert before 2358 * that entry). 2359 * 2360 * Don't change the args->value unless we find the attribute. 2361 */ 2362 int 2363 xfs_attr3_leaf_lookup_int( 2364 struct xfs_buf *bp, 2365 struct xfs_da_args *args) 2366 { 2367 struct xfs_attr_leafblock *leaf; 2368 struct xfs_attr3_icleaf_hdr ichdr; 2369 struct xfs_attr_leaf_entry *entry; 2370 struct xfs_attr_leaf_entry *entries; 2371 struct xfs_attr_leaf_name_local *name_loc; 2372 struct xfs_attr_leaf_name_remote *name_rmt; 2373 xfs_dahash_t hashval; 2374 int probe; 2375 int span; 2376 2377 trace_xfs_attr_leaf_lookup(args); 2378 2379 leaf = bp->b_addr; 2380 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2381 entries = xfs_attr3_leaf_entryp(leaf); 2382 if (ichdr.count >= args->geo->blksize / 8) { 2383 xfs_buf_mark_corrupt(bp); 2384 return -EFSCORRUPTED; 2385 } 2386 2387 /* 2388 * Binary search. (note: small blocks will skip this loop) 2389 */ 2390 hashval = args->hashval; 2391 probe = span = ichdr.count / 2; 2392 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) { 2393 span /= 2; 2394 if (be32_to_cpu(entry->hashval) < hashval) 2395 probe += span; 2396 else if (be32_to_cpu(entry->hashval) > hashval) 2397 probe -= span; 2398 else 2399 break; 2400 } 2401 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) { 2402 xfs_buf_mark_corrupt(bp); 2403 return -EFSCORRUPTED; 2404 } 2405 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) { 2406 xfs_buf_mark_corrupt(bp); 2407 return -EFSCORRUPTED; 2408 } 2409 2410 /* 2411 * Since we may have duplicate hashval's, find the first matching 2412 * hashval in the leaf. 2413 */ 2414 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) { 2415 entry--; 2416 probe--; 2417 } 2418 while (probe < ichdr.count && 2419 be32_to_cpu(entry->hashval) < hashval) { 2420 entry++; 2421 probe++; 2422 } 2423 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) { 2424 args->index = probe; 2425 return -ENOATTR; 2426 } 2427 2428 /* 2429 * Duplicate keys may be present, so search all of them for a match. 2430 */ 2431 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval); 2432 entry++, probe++) { 2433 /* 2434 * GROT: Add code to remove incomplete entries. 2435 */ 2436 if (entry->flags & XFS_ATTR_LOCAL) { 2437 name_loc = xfs_attr3_leaf_name_local(leaf, probe); 2438 if (!xfs_attr_match(args, name_loc->namelen, 2439 name_loc->nameval, entry->flags)) 2440 continue; 2441 args->index = probe; 2442 return -EEXIST; 2443 } else { 2444 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe); 2445 if (!xfs_attr_match(args, name_rmt->namelen, 2446 name_rmt->name, entry->flags)) 2447 continue; 2448 args->index = probe; 2449 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen); 2450 args->rmtblkno = be32_to_cpu(name_rmt->valueblk); 2451 args->rmtblkcnt = xfs_attr3_rmt_blocks( 2452 args->dp->i_mount, 2453 args->rmtvaluelen); 2454 return -EEXIST; 2455 } 2456 } 2457 args->index = probe; 2458 return -ENOATTR; 2459 } 2460 2461 /* 2462 * Get the value associated with an attribute name from a leaf attribute 2463 * list structure. 2464 * 2465 * If args->valuelen is zero, only the length needs to be returned. Unlike a 2466 * lookup, we only return an error if the attribute does not exist or we can't 2467 * retrieve the value. 2468 */ 2469 int 2470 xfs_attr3_leaf_getvalue( 2471 struct xfs_buf *bp, 2472 struct xfs_da_args *args) 2473 { 2474 struct xfs_attr_leafblock *leaf; 2475 struct xfs_attr3_icleaf_hdr ichdr; 2476 struct xfs_attr_leaf_entry *entry; 2477 struct xfs_attr_leaf_name_local *name_loc; 2478 struct xfs_attr_leaf_name_remote *name_rmt; 2479 2480 leaf = bp->b_addr; 2481 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2482 ASSERT(ichdr.count < args->geo->blksize / 8); 2483 ASSERT(args->index < ichdr.count); 2484 2485 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2486 if (entry->flags & XFS_ATTR_LOCAL) { 2487 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 2488 ASSERT(name_loc->namelen == args->namelen); 2489 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0); 2490 return xfs_attr_copy_value(args, 2491 &name_loc->nameval[args->namelen], 2492 be16_to_cpu(name_loc->valuelen)); 2493 } 2494 2495 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2496 ASSERT(name_rmt->namelen == args->namelen); 2497 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0); 2498 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen); 2499 args->rmtblkno = be32_to_cpu(name_rmt->valueblk); 2500 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount, 2501 args->rmtvaluelen); 2502 return xfs_attr_copy_value(args, NULL, args->rmtvaluelen); 2503 } 2504 2505 /*======================================================================== 2506 * Utility routines. 2507 *========================================================================*/ 2508 2509 /* 2510 * Move the indicated entries from one leaf to another. 2511 * NOTE: this routine modifies both source and destination leaves. 2512 */ 2513 /*ARGSUSED*/ 2514 STATIC void 2515 xfs_attr3_leaf_moveents( 2516 struct xfs_da_args *args, 2517 struct xfs_attr_leafblock *leaf_s, 2518 struct xfs_attr3_icleaf_hdr *ichdr_s, 2519 int start_s, 2520 struct xfs_attr_leafblock *leaf_d, 2521 struct xfs_attr3_icleaf_hdr *ichdr_d, 2522 int start_d, 2523 int count) 2524 { 2525 struct xfs_attr_leaf_entry *entry_s; 2526 struct xfs_attr_leaf_entry *entry_d; 2527 int desti; 2528 int tmp; 2529 int i; 2530 2531 /* 2532 * Check for nothing to do. 2533 */ 2534 if (count == 0) 2535 return; 2536 2537 /* 2538 * Set up environment. 2539 */ 2540 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC || 2541 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC); 2542 ASSERT(ichdr_s->magic == ichdr_d->magic); 2543 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8); 2544 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s)) 2545 + xfs_attr3_leaf_hdr_size(leaf_s)); 2546 ASSERT(ichdr_d->count < args->geo->blksize / 8); 2547 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d)) 2548 + xfs_attr3_leaf_hdr_size(leaf_d)); 2549 2550 ASSERT(start_s < ichdr_s->count); 2551 ASSERT(start_d <= ichdr_d->count); 2552 ASSERT(count <= ichdr_s->count); 2553 2554 2555 /* 2556 * Move the entries in the destination leaf up to make a hole? 2557 */ 2558 if (start_d < ichdr_d->count) { 2559 tmp = ichdr_d->count - start_d; 2560 tmp *= sizeof(xfs_attr_leaf_entry_t); 2561 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d]; 2562 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count]; 2563 memmove(entry_d, entry_s, tmp); 2564 } 2565 2566 /* 2567 * Copy all entry's in the same (sorted) order, 2568 * but allocate attribute info packed and in sequence. 2569 */ 2570 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2571 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d]; 2572 desti = start_d; 2573 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) { 2574 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused); 2575 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i); 2576 #ifdef GROT 2577 /* 2578 * Code to drop INCOMPLETE entries. Difficult to use as we 2579 * may also need to change the insertion index. Code turned 2580 * off for 6.2, should be revisited later. 2581 */ 2582 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */ 2583 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp); 2584 ichdr_s->usedbytes -= tmp; 2585 ichdr_s->count -= 1; 2586 entry_d--; /* to compensate for ++ in loop hdr */ 2587 desti--; 2588 if ((start_s + i) < offset) 2589 result++; /* insertion index adjustment */ 2590 } else { 2591 #endif /* GROT */ 2592 ichdr_d->firstused -= tmp; 2593 /* both on-disk, don't endian flip twice */ 2594 entry_d->hashval = entry_s->hashval; 2595 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused); 2596 entry_d->flags = entry_s->flags; 2597 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp 2598 <= args->geo->blksize); 2599 memmove(xfs_attr3_leaf_name(leaf_d, desti), 2600 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp); 2601 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp 2602 <= args->geo->blksize); 2603 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp); 2604 ichdr_s->usedbytes -= tmp; 2605 ichdr_d->usedbytes += tmp; 2606 ichdr_s->count -= 1; 2607 ichdr_d->count += 1; 2608 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t) 2609 + xfs_attr3_leaf_hdr_size(leaf_d); 2610 ASSERT(ichdr_d->firstused >= tmp); 2611 #ifdef GROT 2612 } 2613 #endif /* GROT */ 2614 } 2615 2616 /* 2617 * Zero out the entries we just copied. 2618 */ 2619 if (start_s == ichdr_s->count) { 2620 tmp = count * sizeof(xfs_attr_leaf_entry_t); 2621 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2622 ASSERT(((char *)entry_s + tmp) <= 2623 ((char *)leaf_s + args->geo->blksize)); 2624 memset(entry_s, 0, tmp); 2625 } else { 2626 /* 2627 * Move the remaining entries down to fill the hole, 2628 * then zero the entries at the top. 2629 */ 2630 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t); 2631 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count]; 2632 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2633 memmove(entry_d, entry_s, tmp); 2634 2635 tmp = count * sizeof(xfs_attr_leaf_entry_t); 2636 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count]; 2637 ASSERT(((char *)entry_s + tmp) <= 2638 ((char *)leaf_s + args->geo->blksize)); 2639 memset(entry_s, 0, tmp); 2640 } 2641 2642 /* 2643 * Fill in the freemap information 2644 */ 2645 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d); 2646 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t); 2647 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base; 2648 ichdr_d->freemap[1].base = 0; 2649 ichdr_d->freemap[2].base = 0; 2650 ichdr_d->freemap[1].size = 0; 2651 ichdr_d->freemap[2].size = 0; 2652 ichdr_s->holes = 1; /* leaf may not be compact */ 2653 } 2654 2655 /* 2656 * Pick up the last hashvalue from a leaf block. 2657 */ 2658 xfs_dahash_t 2659 xfs_attr_leaf_lasthash( 2660 struct xfs_buf *bp, 2661 int *count) 2662 { 2663 struct xfs_attr3_icleaf_hdr ichdr; 2664 struct xfs_attr_leaf_entry *entries; 2665 struct xfs_mount *mp = bp->b_mount; 2666 2667 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr); 2668 entries = xfs_attr3_leaf_entryp(bp->b_addr); 2669 if (count) 2670 *count = ichdr.count; 2671 if (!ichdr.count) 2672 return 0; 2673 return be32_to_cpu(entries[ichdr.count - 1].hashval); 2674 } 2675 2676 /* 2677 * Calculate the number of bytes used to store the indicated attribute 2678 * (whether local or remote only calculate bytes in this block). 2679 */ 2680 STATIC int 2681 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index) 2682 { 2683 struct xfs_attr_leaf_entry *entries; 2684 xfs_attr_leaf_name_local_t *name_loc; 2685 xfs_attr_leaf_name_remote_t *name_rmt; 2686 int size; 2687 2688 entries = xfs_attr3_leaf_entryp(leaf); 2689 if (entries[index].flags & XFS_ATTR_LOCAL) { 2690 name_loc = xfs_attr3_leaf_name_local(leaf, index); 2691 size = xfs_attr_leaf_entsize_local(name_loc->namelen, 2692 be16_to_cpu(name_loc->valuelen)); 2693 } else { 2694 name_rmt = xfs_attr3_leaf_name_remote(leaf, index); 2695 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen); 2696 } 2697 return size; 2698 } 2699 2700 /* 2701 * Calculate the number of bytes that would be required to store the new 2702 * attribute (whether local or remote only calculate bytes in this block). 2703 * This routine decides as a side effect whether the attribute will be 2704 * a "local" or a "remote" attribute. 2705 */ 2706 int 2707 xfs_attr_leaf_newentsize( 2708 struct xfs_da_args *args, 2709 int *local) 2710 { 2711 int size; 2712 2713 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen); 2714 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) { 2715 if (local) 2716 *local = 1; 2717 return size; 2718 } 2719 if (local) 2720 *local = 0; 2721 return xfs_attr_leaf_entsize_remote(args->namelen); 2722 } 2723 2724 2725 /*======================================================================== 2726 * Manage the INCOMPLETE flag in a leaf entry 2727 *========================================================================*/ 2728 2729 /* 2730 * Clear the INCOMPLETE flag on an entry in a leaf block. 2731 */ 2732 int 2733 xfs_attr3_leaf_clearflag( 2734 struct xfs_da_args *args) 2735 { 2736 struct xfs_attr_leafblock *leaf; 2737 struct xfs_attr_leaf_entry *entry; 2738 struct xfs_attr_leaf_name_remote *name_rmt; 2739 struct xfs_buf *bp; 2740 int error; 2741 #ifdef DEBUG 2742 struct xfs_attr3_icleaf_hdr ichdr; 2743 xfs_attr_leaf_name_local_t *name_loc; 2744 int namelen; 2745 char *name; 2746 #endif /* DEBUG */ 2747 2748 trace_xfs_attr_leaf_clearflag(args); 2749 /* 2750 * Set up the operation. 2751 */ 2752 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp); 2753 if (error) 2754 return error; 2755 2756 leaf = bp->b_addr; 2757 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2758 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE); 2759 2760 #ifdef DEBUG 2761 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2762 ASSERT(args->index < ichdr.count); 2763 ASSERT(args->index >= 0); 2764 2765 if (entry->flags & XFS_ATTR_LOCAL) { 2766 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 2767 namelen = name_loc->namelen; 2768 name = (char *)name_loc->nameval; 2769 } else { 2770 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2771 namelen = name_rmt->namelen; 2772 name = (char *)name_rmt->name; 2773 } 2774 ASSERT(be32_to_cpu(entry->hashval) == args->hashval); 2775 ASSERT(namelen == args->namelen); 2776 ASSERT(memcmp(name, args->name, namelen) == 0); 2777 #endif /* DEBUG */ 2778 2779 entry->flags &= ~XFS_ATTR_INCOMPLETE; 2780 xfs_trans_log_buf(args->trans, bp, 2781 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 2782 2783 if (args->rmtblkno) { 2784 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0); 2785 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2786 name_rmt->valueblk = cpu_to_be32(args->rmtblkno); 2787 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen); 2788 xfs_trans_log_buf(args->trans, bp, 2789 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt))); 2790 } 2791 2792 return 0; 2793 } 2794 2795 /* 2796 * Set the INCOMPLETE flag on an entry in a leaf block. 2797 */ 2798 int 2799 xfs_attr3_leaf_setflag( 2800 struct xfs_da_args *args) 2801 { 2802 struct xfs_attr_leafblock *leaf; 2803 struct xfs_attr_leaf_entry *entry; 2804 struct xfs_attr_leaf_name_remote *name_rmt; 2805 struct xfs_buf *bp; 2806 int error; 2807 #ifdef DEBUG 2808 struct xfs_attr3_icleaf_hdr ichdr; 2809 #endif 2810 2811 trace_xfs_attr_leaf_setflag(args); 2812 2813 /* 2814 * Set up the operation. 2815 */ 2816 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp); 2817 if (error) 2818 return error; 2819 2820 leaf = bp->b_addr; 2821 #ifdef DEBUG 2822 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2823 ASSERT(args->index < ichdr.count); 2824 ASSERT(args->index >= 0); 2825 #endif 2826 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2827 2828 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0); 2829 entry->flags |= XFS_ATTR_INCOMPLETE; 2830 xfs_trans_log_buf(args->trans, bp, 2831 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 2832 if ((entry->flags & XFS_ATTR_LOCAL) == 0) { 2833 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2834 name_rmt->valueblk = 0; 2835 name_rmt->valuelen = 0; 2836 xfs_trans_log_buf(args->trans, bp, 2837 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt))); 2838 } 2839 2840 return 0; 2841 } 2842 2843 /* 2844 * In a single transaction, clear the INCOMPLETE flag on the leaf entry 2845 * given by args->blkno/index and set the INCOMPLETE flag on the leaf 2846 * entry given by args->blkno2/index2. 2847 * 2848 * Note that they could be in different blocks, or in the same block. 2849 */ 2850 int 2851 xfs_attr3_leaf_flipflags( 2852 struct xfs_da_args *args) 2853 { 2854 struct xfs_attr_leafblock *leaf1; 2855 struct xfs_attr_leafblock *leaf2; 2856 struct xfs_attr_leaf_entry *entry1; 2857 struct xfs_attr_leaf_entry *entry2; 2858 struct xfs_attr_leaf_name_remote *name_rmt; 2859 struct xfs_buf *bp1; 2860 struct xfs_buf *bp2; 2861 int error; 2862 #ifdef DEBUG 2863 struct xfs_attr3_icleaf_hdr ichdr1; 2864 struct xfs_attr3_icleaf_hdr ichdr2; 2865 xfs_attr_leaf_name_local_t *name_loc; 2866 int namelen1, namelen2; 2867 char *name1, *name2; 2868 #endif /* DEBUG */ 2869 2870 trace_xfs_attr_leaf_flipflags(args); 2871 2872 /* 2873 * Read the block containing the "old" attr 2874 */ 2875 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1); 2876 if (error) 2877 return error; 2878 2879 /* 2880 * Read the block containing the "new" attr, if it is different 2881 */ 2882 if (args->blkno2 != args->blkno) { 2883 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2, 2884 &bp2); 2885 if (error) 2886 return error; 2887 } else { 2888 bp2 = bp1; 2889 } 2890 2891 leaf1 = bp1->b_addr; 2892 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index]; 2893 2894 leaf2 = bp2->b_addr; 2895 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2]; 2896 2897 #ifdef DEBUG 2898 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1); 2899 ASSERT(args->index < ichdr1.count); 2900 ASSERT(args->index >= 0); 2901 2902 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2); 2903 ASSERT(args->index2 < ichdr2.count); 2904 ASSERT(args->index2 >= 0); 2905 2906 if (entry1->flags & XFS_ATTR_LOCAL) { 2907 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index); 2908 namelen1 = name_loc->namelen; 2909 name1 = (char *)name_loc->nameval; 2910 } else { 2911 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index); 2912 namelen1 = name_rmt->namelen; 2913 name1 = (char *)name_rmt->name; 2914 } 2915 if (entry2->flags & XFS_ATTR_LOCAL) { 2916 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2); 2917 namelen2 = name_loc->namelen; 2918 name2 = (char *)name_loc->nameval; 2919 } else { 2920 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2); 2921 namelen2 = name_rmt->namelen; 2922 name2 = (char *)name_rmt->name; 2923 } 2924 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval)); 2925 ASSERT(namelen1 == namelen2); 2926 ASSERT(memcmp(name1, name2, namelen1) == 0); 2927 #endif /* DEBUG */ 2928 2929 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE); 2930 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0); 2931 2932 entry1->flags &= ~XFS_ATTR_INCOMPLETE; 2933 xfs_trans_log_buf(args->trans, bp1, 2934 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1))); 2935 if (args->rmtblkno) { 2936 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0); 2937 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index); 2938 name_rmt->valueblk = cpu_to_be32(args->rmtblkno); 2939 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen); 2940 xfs_trans_log_buf(args->trans, bp1, 2941 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt))); 2942 } 2943 2944 entry2->flags |= XFS_ATTR_INCOMPLETE; 2945 xfs_trans_log_buf(args->trans, bp2, 2946 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2))); 2947 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) { 2948 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2); 2949 name_rmt->valueblk = 0; 2950 name_rmt->valuelen = 0; 2951 xfs_trans_log_buf(args->trans, bp2, 2952 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt))); 2953 } 2954 2955 return 0; 2956 } 2957