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