1 /* 2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 3 * Copyright (c) 2013 Red Hat, Inc. 4 * All Rights Reserved. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License as 8 * published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it would be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 18 */ 19 #include "xfs.h" 20 #include "xfs_fs.h" 21 #include "xfs_shared.h" 22 #include "xfs_format.h" 23 #include "xfs_log_format.h" 24 #include "xfs_trans_resv.h" 25 #include "xfs_bit.h" 26 #include "xfs_sb.h" 27 #include "xfs_mount.h" 28 #include "xfs_da_format.h" 29 #include "xfs_da_btree.h" 30 #include "xfs_inode.h" 31 #include "xfs_trans.h" 32 #include "xfs_inode_item.h" 33 #include "xfs_bmap_btree.h" 34 #include "xfs_bmap.h" 35 #include "xfs_attr_sf.h" 36 #include "xfs_attr_remote.h" 37 #include "xfs_attr.h" 38 #include "xfs_attr_leaf.h" 39 #include "xfs_error.h" 40 #include "xfs_trace.h" 41 #include "xfs_buf_item.h" 42 #include "xfs_cksum.h" 43 #include "xfs_dir2.h" 44 #include "xfs_log.h" 45 46 47 /* 48 * xfs_attr_leaf.c 49 * 50 * Routines to implement leaf blocks of attributes as Btrees of hashed names. 51 */ 52 53 /*======================================================================== 54 * Function prototypes for the kernel. 55 *========================================================================*/ 56 57 /* 58 * Routines used for growing the Btree. 59 */ 60 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args, 61 xfs_dablk_t which_block, struct xfs_buf **bpp); 62 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer, 63 struct xfs_attr3_icleaf_hdr *ichdr, 64 struct xfs_da_args *args, int freemap_index); 65 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args, 66 struct xfs_attr3_icleaf_hdr *ichdr, 67 struct xfs_buf *leaf_buffer); 68 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state, 69 xfs_da_state_blk_t *blk1, 70 xfs_da_state_blk_t *blk2); 71 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state, 72 xfs_da_state_blk_t *leaf_blk_1, 73 struct xfs_attr3_icleaf_hdr *ichdr1, 74 xfs_da_state_blk_t *leaf_blk_2, 75 struct xfs_attr3_icleaf_hdr *ichdr2, 76 int *number_entries_in_blk1, 77 int *number_usedbytes_in_blk1); 78 79 /* 80 * Utility routines. 81 */ 82 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args, 83 struct xfs_attr_leafblock *src_leaf, 84 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start, 85 struct xfs_attr_leafblock *dst_leaf, 86 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start, 87 int move_count); 88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index); 89 90 /* 91 * attr3 block 'firstused' conversion helpers. 92 * 93 * firstused refers to the offset of the first used byte of the nameval region 94 * of an attr leaf block. The region starts at the tail of the block and expands 95 * backwards towards the middle. As such, firstused is initialized to the block 96 * size for an empty leaf block and is reduced from there. 97 * 98 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k. 99 * The in-core firstused field is 32-bit and thus supports the maximum fsb size. 100 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this 101 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent 102 * the attr block size. The following helpers manage the conversion between the 103 * in-core and on-disk formats. 104 */ 105 106 static void 107 xfs_attr3_leaf_firstused_from_disk( 108 struct xfs_da_geometry *geo, 109 struct xfs_attr3_icleaf_hdr *to, 110 struct xfs_attr_leafblock *from) 111 { 112 struct xfs_attr3_leaf_hdr *hdr3; 113 114 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) { 115 hdr3 = (struct xfs_attr3_leaf_hdr *) from; 116 to->firstused = be16_to_cpu(hdr3->firstused); 117 } else { 118 to->firstused = be16_to_cpu(from->hdr.firstused); 119 } 120 121 /* 122 * Convert from the magic fsb size value to actual blocksize. This 123 * should only occur for empty blocks when the block size overflows 124 * 16-bits. 125 */ 126 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) { 127 ASSERT(!to->count && !to->usedbytes); 128 ASSERT(geo->blksize > USHRT_MAX); 129 to->firstused = geo->blksize; 130 } 131 } 132 133 static void 134 xfs_attr3_leaf_firstused_to_disk( 135 struct xfs_da_geometry *geo, 136 struct xfs_attr_leafblock *to, 137 struct xfs_attr3_icleaf_hdr *from) 138 { 139 struct xfs_attr3_leaf_hdr *hdr3; 140 uint32_t firstused; 141 142 /* magic value should only be seen on disk */ 143 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF); 144 145 /* 146 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk 147 * value. This only overflows at the max supported value of 64k. Use the 148 * magic on-disk value to represent block size in this case. 149 */ 150 firstused = from->firstused; 151 if (firstused > USHRT_MAX) { 152 ASSERT(from->firstused == geo->blksize); 153 firstused = XFS_ATTR3_LEAF_NULLOFF; 154 } 155 156 if (from->magic == XFS_ATTR3_LEAF_MAGIC) { 157 hdr3 = (struct xfs_attr3_leaf_hdr *) to; 158 hdr3->firstused = cpu_to_be16(firstused); 159 } else { 160 to->hdr.firstused = cpu_to_be16(firstused); 161 } 162 } 163 164 void 165 xfs_attr3_leaf_hdr_from_disk( 166 struct xfs_da_geometry *geo, 167 struct xfs_attr3_icleaf_hdr *to, 168 struct xfs_attr_leafblock *from) 169 { 170 int i; 171 172 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || 173 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); 174 175 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) { 176 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from; 177 178 to->forw = be32_to_cpu(hdr3->info.hdr.forw); 179 to->back = be32_to_cpu(hdr3->info.hdr.back); 180 to->magic = be16_to_cpu(hdr3->info.hdr.magic); 181 to->count = be16_to_cpu(hdr3->count); 182 to->usedbytes = be16_to_cpu(hdr3->usedbytes); 183 xfs_attr3_leaf_firstused_from_disk(geo, to, from); 184 to->holes = hdr3->holes; 185 186 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 187 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base); 188 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size); 189 } 190 return; 191 } 192 to->forw = be32_to_cpu(from->hdr.info.forw); 193 to->back = be32_to_cpu(from->hdr.info.back); 194 to->magic = be16_to_cpu(from->hdr.info.magic); 195 to->count = be16_to_cpu(from->hdr.count); 196 to->usedbytes = be16_to_cpu(from->hdr.usedbytes); 197 xfs_attr3_leaf_firstused_from_disk(geo, to, from); 198 to->holes = from->hdr.holes; 199 200 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 201 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base); 202 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size); 203 } 204 } 205 206 void 207 xfs_attr3_leaf_hdr_to_disk( 208 struct xfs_da_geometry *geo, 209 struct xfs_attr_leafblock *to, 210 struct xfs_attr3_icleaf_hdr *from) 211 { 212 int i; 213 214 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC || 215 from->magic == XFS_ATTR3_LEAF_MAGIC); 216 217 if (from->magic == XFS_ATTR3_LEAF_MAGIC) { 218 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to; 219 220 hdr3->info.hdr.forw = cpu_to_be32(from->forw); 221 hdr3->info.hdr.back = cpu_to_be32(from->back); 222 hdr3->info.hdr.magic = cpu_to_be16(from->magic); 223 hdr3->count = cpu_to_be16(from->count); 224 hdr3->usedbytes = cpu_to_be16(from->usedbytes); 225 xfs_attr3_leaf_firstused_to_disk(geo, to, from); 226 hdr3->holes = from->holes; 227 hdr3->pad1 = 0; 228 229 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 230 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base); 231 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size); 232 } 233 return; 234 } 235 to->hdr.info.forw = cpu_to_be32(from->forw); 236 to->hdr.info.back = cpu_to_be32(from->back); 237 to->hdr.info.magic = cpu_to_be16(from->magic); 238 to->hdr.count = cpu_to_be16(from->count); 239 to->hdr.usedbytes = cpu_to_be16(from->usedbytes); 240 xfs_attr3_leaf_firstused_to_disk(geo, to, from); 241 to->hdr.holes = from->holes; 242 to->hdr.pad1 = 0; 243 244 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 245 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base); 246 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size); 247 } 248 } 249 250 static xfs_failaddr_t 251 xfs_attr3_leaf_verify( 252 struct xfs_buf *bp) 253 { 254 struct xfs_attr3_icleaf_hdr ichdr; 255 struct xfs_mount *mp = bp->b_target->bt_mount; 256 struct xfs_attr_leafblock *leaf = bp->b_addr; 257 struct xfs_perag *pag = bp->b_pag; 258 struct xfs_attr_leaf_entry *entries; 259 260 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf); 261 262 if (xfs_sb_version_hascrc(&mp->m_sb)) { 263 struct xfs_da3_node_hdr *hdr3 = bp->b_addr; 264 265 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC) 266 return __this_address; 267 268 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid)) 269 return __this_address; 270 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn) 271 return __this_address; 272 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn))) 273 return __this_address; 274 } else { 275 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC) 276 return __this_address; 277 } 278 /* 279 * In recovery there is a transient state where count == 0 is valid 280 * because we may have transitioned an empty shortform attr to a leaf 281 * if the attr didn't fit in shortform. 282 */ 283 if (pag && pag->pagf_init && ichdr.count == 0) 284 return __this_address; 285 286 /* 287 * firstused is the block offset of the first name info structure. 288 * Make sure it doesn't go off the block or crash into the header. 289 */ 290 if (ichdr.firstused > mp->m_attr_geo->blksize) 291 return __this_address; 292 if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf)) 293 return __this_address; 294 295 /* Make sure the entries array doesn't crash into the name info. */ 296 entries = xfs_attr3_leaf_entryp(bp->b_addr); 297 if ((char *)&entries[ichdr.count] > 298 (char *)bp->b_addr + ichdr.firstused) 299 return __this_address; 300 301 /* XXX: need to range check rest of attr header values */ 302 /* XXX: hash order check? */ 303 304 return NULL; 305 } 306 307 static void 308 xfs_attr3_leaf_write_verify( 309 struct xfs_buf *bp) 310 { 311 struct xfs_mount *mp = bp->b_target->bt_mount; 312 struct xfs_buf_log_item *bip = bp->b_log_item; 313 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr; 314 xfs_failaddr_t fa; 315 316 fa = xfs_attr3_leaf_verify(bp); 317 if (fa) { 318 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 319 return; 320 } 321 322 if (!xfs_sb_version_hascrc(&mp->m_sb)) 323 return; 324 325 if (bip) 326 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); 327 328 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF); 329 } 330 331 /* 332 * leaf/node format detection on trees is sketchy, so a node read can be done on 333 * leaf level blocks when detection identifies the tree as a node format tree 334 * incorrectly. In this case, we need to swap the verifier to match the correct 335 * format of the block being read. 336 */ 337 static void 338 xfs_attr3_leaf_read_verify( 339 struct xfs_buf *bp) 340 { 341 struct xfs_mount *mp = bp->b_target->bt_mount; 342 xfs_failaddr_t fa; 343 344 if (xfs_sb_version_hascrc(&mp->m_sb) && 345 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF)) 346 xfs_verifier_error(bp, -EFSBADCRC, __this_address); 347 else { 348 fa = xfs_attr3_leaf_verify(bp); 349 if (fa) 350 xfs_verifier_error(bp, -EFSCORRUPTED, fa); 351 } 352 } 353 354 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = { 355 .name = "xfs_attr3_leaf", 356 .verify_read = xfs_attr3_leaf_read_verify, 357 .verify_write = xfs_attr3_leaf_write_verify, 358 .verify_struct = xfs_attr3_leaf_verify, 359 }; 360 361 int 362 xfs_attr3_leaf_read( 363 struct xfs_trans *tp, 364 struct xfs_inode *dp, 365 xfs_dablk_t bno, 366 xfs_daddr_t mappedbno, 367 struct xfs_buf **bpp) 368 { 369 int err; 370 371 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp, 372 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops); 373 if (!err && tp && *bpp) 374 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF); 375 return err; 376 } 377 378 /*======================================================================== 379 * Namespace helper routines 380 *========================================================================*/ 381 382 /* 383 * If namespace bits don't match return 0. 384 * If all match then return 1. 385 */ 386 STATIC int 387 xfs_attr_namesp_match(int arg_flags, int ondisk_flags) 388 { 389 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags); 390 } 391 392 393 /*======================================================================== 394 * External routines when attribute fork size < XFS_LITINO(mp). 395 *========================================================================*/ 396 397 /* 398 * Query whether the requested number of additional bytes of extended 399 * attribute space will be able to fit inline. 400 * 401 * Returns zero if not, else the di_forkoff fork offset to be used in the 402 * literal area for attribute data once the new bytes have been added. 403 * 404 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value; 405 * special case for dev/uuid inodes, they have fixed size data forks. 406 */ 407 int 408 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes) 409 { 410 int offset; 411 int minforkoff; /* lower limit on valid forkoff locations */ 412 int maxforkoff; /* upper limit on valid forkoff locations */ 413 int dsize; 414 xfs_mount_t *mp = dp->i_mount; 415 416 /* rounded down */ 417 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3; 418 419 if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) { 420 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; 421 return (offset >= minforkoff) ? minforkoff : 0; 422 } 423 424 /* 425 * If the requested numbers of bytes is smaller or equal to the 426 * current attribute fork size we can always proceed. 427 * 428 * Note that if_bytes in the data fork might actually be larger than 429 * the current data fork size is due to delalloc extents. In that 430 * case either the extent count will go down when they are converted 431 * to real extents, or the delalloc conversion will take care of the 432 * literal area rebalancing. 433 */ 434 if (bytes <= XFS_IFORK_ASIZE(dp)) 435 return dp->i_d.di_forkoff; 436 437 /* 438 * For attr2 we can try to move the forkoff if there is space in the 439 * literal area, but for the old format we are done if there is no 440 * space in the fixed attribute fork. 441 */ 442 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) 443 return 0; 444 445 dsize = dp->i_df.if_bytes; 446 447 switch (dp->i_d.di_format) { 448 case XFS_DINODE_FMT_EXTENTS: 449 /* 450 * If there is no attr fork and the data fork is extents, 451 * determine if creating the default attr fork will result 452 * in the extents form migrating to btree. If so, the 453 * minimum offset only needs to be the space required for 454 * the btree root. 455 */ 456 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes > 457 xfs_default_attroffset(dp)) 458 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS); 459 break; 460 case XFS_DINODE_FMT_BTREE: 461 /* 462 * If we have a data btree then keep forkoff if we have one, 463 * otherwise we are adding a new attr, so then we set 464 * minforkoff to where the btree root can finish so we have 465 * plenty of room for attrs 466 */ 467 if (dp->i_d.di_forkoff) { 468 if (offset < dp->i_d.di_forkoff) 469 return 0; 470 return dp->i_d.di_forkoff; 471 } 472 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot); 473 break; 474 } 475 476 /* 477 * A data fork btree root must have space for at least 478 * MINDBTPTRS key/ptr pairs if the data fork is small or empty. 479 */ 480 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS)); 481 minforkoff = roundup(minforkoff, 8) >> 3; 482 483 /* attr fork btree root can have at least this many key/ptr pairs */ 484 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) - 485 XFS_BMDR_SPACE_CALC(MINABTPTRS); 486 maxforkoff = maxforkoff >> 3; /* rounded down */ 487 488 if (offset >= maxforkoff) 489 return maxforkoff; 490 if (offset >= minforkoff) 491 return offset; 492 return 0; 493 } 494 495 /* 496 * Switch on the ATTR2 superblock bit (implies also FEATURES2) 497 */ 498 STATIC void 499 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp) 500 { 501 if ((mp->m_flags & XFS_MOUNT_ATTR2) && 502 !(xfs_sb_version_hasattr2(&mp->m_sb))) { 503 spin_lock(&mp->m_sb_lock); 504 if (!xfs_sb_version_hasattr2(&mp->m_sb)) { 505 xfs_sb_version_addattr2(&mp->m_sb); 506 spin_unlock(&mp->m_sb_lock); 507 xfs_log_sb(tp); 508 } else 509 spin_unlock(&mp->m_sb_lock); 510 } 511 } 512 513 /* 514 * Create the initial contents of a shortform attribute list. 515 */ 516 void 517 xfs_attr_shortform_create(xfs_da_args_t *args) 518 { 519 xfs_attr_sf_hdr_t *hdr; 520 xfs_inode_t *dp; 521 xfs_ifork_t *ifp; 522 523 trace_xfs_attr_sf_create(args); 524 525 dp = args->dp; 526 ASSERT(dp != NULL); 527 ifp = dp->i_afp; 528 ASSERT(ifp != NULL); 529 ASSERT(ifp->if_bytes == 0); 530 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) { 531 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */ 532 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL; 533 ifp->if_flags |= XFS_IFINLINE; 534 } else { 535 ASSERT(ifp->if_flags & XFS_IFINLINE); 536 } 537 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK); 538 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data; 539 hdr->count = 0; 540 hdr->totsize = cpu_to_be16(sizeof(*hdr)); 541 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA); 542 } 543 544 /* 545 * Add a name/value pair to the shortform attribute list. 546 * Overflow from the inode has already been checked for. 547 */ 548 void 549 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff) 550 { 551 xfs_attr_shortform_t *sf; 552 xfs_attr_sf_entry_t *sfe; 553 int i, offset, size; 554 xfs_mount_t *mp; 555 xfs_inode_t *dp; 556 xfs_ifork_t *ifp; 557 558 trace_xfs_attr_sf_add(args); 559 560 dp = args->dp; 561 mp = dp->i_mount; 562 dp->i_d.di_forkoff = forkoff; 563 564 ifp = dp->i_afp; 565 ASSERT(ifp->if_flags & XFS_IFINLINE); 566 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data; 567 sfe = &sf->list[0]; 568 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) { 569 #ifdef DEBUG 570 if (sfe->namelen != args->namelen) 571 continue; 572 if (memcmp(args->name, sfe->nameval, args->namelen) != 0) 573 continue; 574 if (!xfs_attr_namesp_match(args->flags, sfe->flags)) 575 continue; 576 ASSERT(0); 577 #endif 578 } 579 580 offset = (char *)sfe - (char *)sf; 581 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen); 582 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); 583 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data; 584 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset); 585 586 sfe->namelen = args->namelen; 587 sfe->valuelen = args->valuelen; 588 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags); 589 memcpy(sfe->nameval, args->name, args->namelen); 590 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen); 591 sf->hdr.count++; 592 be16_add_cpu(&sf->hdr.totsize, size); 593 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA); 594 595 xfs_sbversion_add_attr2(mp, args->trans); 596 } 597 598 /* 599 * After the last attribute is removed revert to original inode format, 600 * making all literal area available to the data fork once more. 601 */ 602 void 603 xfs_attr_fork_remove( 604 struct xfs_inode *ip, 605 struct xfs_trans *tp) 606 { 607 xfs_idestroy_fork(ip, XFS_ATTR_FORK); 608 ip->i_d.di_forkoff = 0; 609 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 610 611 ASSERT(ip->i_d.di_anextents == 0); 612 ASSERT(ip->i_afp == NULL); 613 614 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 615 } 616 617 /* 618 * Remove an attribute from the shortform attribute list structure. 619 */ 620 int 621 xfs_attr_shortform_remove(xfs_da_args_t *args) 622 { 623 xfs_attr_shortform_t *sf; 624 xfs_attr_sf_entry_t *sfe; 625 int base, size=0, end, totsize, i; 626 xfs_mount_t *mp; 627 xfs_inode_t *dp; 628 629 trace_xfs_attr_sf_remove(args); 630 631 dp = args->dp; 632 mp = dp->i_mount; 633 base = sizeof(xfs_attr_sf_hdr_t); 634 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data; 635 sfe = &sf->list[0]; 636 end = sf->hdr.count; 637 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), 638 base += size, i++) { 639 size = XFS_ATTR_SF_ENTSIZE(sfe); 640 if (sfe->namelen != args->namelen) 641 continue; 642 if (memcmp(sfe->nameval, args->name, args->namelen) != 0) 643 continue; 644 if (!xfs_attr_namesp_match(args->flags, sfe->flags)) 645 continue; 646 break; 647 } 648 if (i == end) 649 return -ENOATTR; 650 651 /* 652 * Fix up the attribute fork data, covering the hole 653 */ 654 end = base + size; 655 totsize = be16_to_cpu(sf->hdr.totsize); 656 if (end != totsize) 657 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end); 658 sf->hdr.count--; 659 be16_add_cpu(&sf->hdr.totsize, -size); 660 661 /* 662 * Fix up the start offset of the attribute fork 663 */ 664 totsize -= size; 665 if (totsize == sizeof(xfs_attr_sf_hdr_t) && 666 (mp->m_flags & XFS_MOUNT_ATTR2) && 667 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) && 668 !(args->op_flags & XFS_DA_OP_ADDNAME)) { 669 xfs_attr_fork_remove(dp, args->trans); 670 } else { 671 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK); 672 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize); 673 ASSERT(dp->i_d.di_forkoff); 674 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || 675 (args->op_flags & XFS_DA_OP_ADDNAME) || 676 !(mp->m_flags & XFS_MOUNT_ATTR2) || 677 dp->i_d.di_format == XFS_DINODE_FMT_BTREE); 678 xfs_trans_log_inode(args->trans, dp, 679 XFS_ILOG_CORE | XFS_ILOG_ADATA); 680 } 681 682 xfs_sbversion_add_attr2(mp, args->trans); 683 684 return 0; 685 } 686 687 /* 688 * Look up a name in a shortform attribute list structure. 689 */ 690 /*ARGSUSED*/ 691 int 692 xfs_attr_shortform_lookup(xfs_da_args_t *args) 693 { 694 xfs_attr_shortform_t *sf; 695 xfs_attr_sf_entry_t *sfe; 696 int i; 697 xfs_ifork_t *ifp; 698 699 trace_xfs_attr_sf_lookup(args); 700 701 ifp = args->dp->i_afp; 702 ASSERT(ifp->if_flags & XFS_IFINLINE); 703 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data; 704 sfe = &sf->list[0]; 705 for (i = 0; i < sf->hdr.count; 706 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) { 707 if (sfe->namelen != args->namelen) 708 continue; 709 if (memcmp(args->name, sfe->nameval, args->namelen) != 0) 710 continue; 711 if (!xfs_attr_namesp_match(args->flags, sfe->flags)) 712 continue; 713 return -EEXIST; 714 } 715 return -ENOATTR; 716 } 717 718 /* 719 * Look up a name in a shortform attribute list structure. 720 */ 721 /*ARGSUSED*/ 722 int 723 xfs_attr_shortform_getvalue(xfs_da_args_t *args) 724 { 725 xfs_attr_shortform_t *sf; 726 xfs_attr_sf_entry_t *sfe; 727 int i; 728 729 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE); 730 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data; 731 sfe = &sf->list[0]; 732 for (i = 0; i < sf->hdr.count; 733 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) { 734 if (sfe->namelen != args->namelen) 735 continue; 736 if (memcmp(args->name, sfe->nameval, args->namelen) != 0) 737 continue; 738 if (!xfs_attr_namesp_match(args->flags, sfe->flags)) 739 continue; 740 if (args->flags & ATTR_KERNOVAL) { 741 args->valuelen = sfe->valuelen; 742 return -EEXIST; 743 } 744 if (args->valuelen < sfe->valuelen) { 745 args->valuelen = sfe->valuelen; 746 return -ERANGE; 747 } 748 args->valuelen = sfe->valuelen; 749 memcpy(args->value, &sfe->nameval[args->namelen], 750 args->valuelen); 751 return -EEXIST; 752 } 753 return -ENOATTR; 754 } 755 756 /* 757 * Convert from using the shortform to the leaf. On success, return the 758 * buffer so that we can keep it locked until we're totally done with it. 759 */ 760 int 761 xfs_attr_shortform_to_leaf( 762 struct xfs_da_args *args, 763 struct xfs_buf **leaf_bp) 764 { 765 xfs_inode_t *dp; 766 xfs_attr_shortform_t *sf; 767 xfs_attr_sf_entry_t *sfe; 768 xfs_da_args_t nargs; 769 char *tmpbuffer; 770 int error, i, size; 771 xfs_dablk_t blkno; 772 struct xfs_buf *bp; 773 xfs_ifork_t *ifp; 774 775 trace_xfs_attr_sf_to_leaf(args); 776 777 dp = args->dp; 778 ifp = dp->i_afp; 779 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data; 780 size = be16_to_cpu(sf->hdr.totsize); 781 tmpbuffer = kmem_alloc(size, KM_SLEEP); 782 ASSERT(tmpbuffer != NULL); 783 memcpy(tmpbuffer, ifp->if_u1.if_data, size); 784 sf = (xfs_attr_shortform_t *)tmpbuffer; 785 786 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK); 787 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK); 788 789 bp = NULL; 790 error = xfs_da_grow_inode(args, &blkno); 791 if (error) { 792 /* 793 * If we hit an IO error middle of the transaction inside 794 * grow_inode(), we may have inconsistent data. Bail out. 795 */ 796 if (error == -EIO) 797 goto out; 798 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */ 799 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */ 800 goto out; 801 } 802 803 ASSERT(blkno == 0); 804 error = xfs_attr3_leaf_create(args, blkno, &bp); 805 if (error) { 806 error = xfs_da_shrink_inode(args, 0, bp); 807 bp = NULL; 808 if (error) 809 goto out; 810 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */ 811 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */ 812 goto out; 813 } 814 815 memset((char *)&nargs, 0, sizeof(nargs)); 816 nargs.dp = dp; 817 nargs.geo = args->geo; 818 nargs.firstblock = args->firstblock; 819 nargs.dfops = args->dfops; 820 nargs.total = args->total; 821 nargs.whichfork = XFS_ATTR_FORK; 822 nargs.trans = args->trans; 823 nargs.op_flags = XFS_DA_OP_OKNOENT; 824 825 sfe = &sf->list[0]; 826 for (i = 0; i < sf->hdr.count; i++) { 827 nargs.name = sfe->nameval; 828 nargs.namelen = sfe->namelen; 829 nargs.value = &sfe->nameval[nargs.namelen]; 830 nargs.valuelen = sfe->valuelen; 831 nargs.hashval = xfs_da_hashname(sfe->nameval, 832 sfe->namelen); 833 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags); 834 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */ 835 ASSERT(error == -ENOATTR); 836 error = xfs_attr3_leaf_add(bp, &nargs); 837 ASSERT(error != -ENOSPC); 838 if (error) 839 goto out; 840 sfe = XFS_ATTR_SF_NEXTENTRY(sfe); 841 } 842 error = 0; 843 *leaf_bp = bp; 844 out: 845 kmem_free(tmpbuffer); 846 return error; 847 } 848 849 /* 850 * Check a leaf attribute block to see if all the entries would fit into 851 * a shortform attribute list. 852 */ 853 int 854 xfs_attr_shortform_allfit( 855 struct xfs_buf *bp, 856 struct xfs_inode *dp) 857 { 858 struct xfs_attr_leafblock *leaf; 859 struct xfs_attr_leaf_entry *entry; 860 xfs_attr_leaf_name_local_t *name_loc; 861 struct xfs_attr3_icleaf_hdr leafhdr; 862 int bytes; 863 int i; 864 struct xfs_mount *mp = bp->b_target->bt_mount; 865 866 leaf = bp->b_addr; 867 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf); 868 entry = xfs_attr3_leaf_entryp(leaf); 869 870 bytes = sizeof(struct xfs_attr_sf_hdr); 871 for (i = 0; i < leafhdr.count; entry++, i++) { 872 if (entry->flags & XFS_ATTR_INCOMPLETE) 873 continue; /* don't copy partial entries */ 874 if (!(entry->flags & XFS_ATTR_LOCAL)) 875 return 0; 876 name_loc = xfs_attr3_leaf_name_local(leaf, i); 877 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX) 878 return 0; 879 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX) 880 return 0; 881 bytes += sizeof(struct xfs_attr_sf_entry) - 1 882 + name_loc->namelen 883 + be16_to_cpu(name_loc->valuelen); 884 } 885 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) && 886 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) && 887 (bytes == sizeof(struct xfs_attr_sf_hdr))) 888 return -1; 889 return xfs_attr_shortform_bytesfit(dp, bytes); 890 } 891 892 /* Verify the consistency of an inline attribute fork. */ 893 xfs_failaddr_t 894 xfs_attr_shortform_verify( 895 struct xfs_inode *ip) 896 { 897 struct xfs_attr_shortform *sfp; 898 struct xfs_attr_sf_entry *sfep; 899 struct xfs_attr_sf_entry *next_sfep; 900 char *endp; 901 struct xfs_ifork *ifp; 902 int i; 903 int size; 904 905 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL); 906 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK); 907 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data; 908 size = ifp->if_bytes; 909 910 /* 911 * Give up if the attribute is way too short. 912 */ 913 if (size < sizeof(struct xfs_attr_sf_hdr)) 914 return __this_address; 915 916 endp = (char *)sfp + size; 917 918 /* Check all reported entries */ 919 sfep = &sfp->list[0]; 920 for (i = 0; i < sfp->hdr.count; i++) { 921 /* 922 * struct xfs_attr_sf_entry has a variable length. 923 * Check the fixed-offset parts of the structure are 924 * within the data buffer. 925 */ 926 if (((char *)sfep + sizeof(*sfep)) >= endp) 927 return __this_address; 928 929 /* Don't allow names with known bad length. */ 930 if (sfep->namelen == 0) 931 return __this_address; 932 933 /* 934 * Check that the variable-length part of the structure is 935 * within the data buffer. The next entry starts after the 936 * name component, so nextentry is an acceptable test. 937 */ 938 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep); 939 if ((char *)next_sfep > endp) 940 return __this_address; 941 942 /* 943 * Check for unknown flags. Short form doesn't support 944 * the incomplete or local bits, so we can use the namespace 945 * mask here. 946 */ 947 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK) 948 return __this_address; 949 950 /* 951 * Check for invalid namespace combinations. We only allow 952 * one namespace flag per xattr, so we can just count the 953 * bits (i.e. hweight) here. 954 */ 955 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1) 956 return __this_address; 957 958 sfep = next_sfep; 959 } 960 if ((void *)sfep != (void *)endp) 961 return __this_address; 962 963 return NULL; 964 } 965 966 /* 967 * Convert a leaf attribute list to shortform attribute list 968 */ 969 int 970 xfs_attr3_leaf_to_shortform( 971 struct xfs_buf *bp, 972 struct xfs_da_args *args, 973 int forkoff) 974 { 975 struct xfs_attr_leafblock *leaf; 976 struct xfs_attr3_icleaf_hdr ichdr; 977 struct xfs_attr_leaf_entry *entry; 978 struct xfs_attr_leaf_name_local *name_loc; 979 struct xfs_da_args nargs; 980 struct xfs_inode *dp = args->dp; 981 char *tmpbuffer; 982 int error; 983 int i; 984 985 trace_xfs_attr_leaf_to_sf(args); 986 987 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP); 988 if (!tmpbuffer) 989 return -ENOMEM; 990 991 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize); 992 993 leaf = (xfs_attr_leafblock_t *)tmpbuffer; 994 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 995 entry = xfs_attr3_leaf_entryp(leaf); 996 997 /* XXX (dgc): buffer is about to be marked stale - why zero it? */ 998 memset(bp->b_addr, 0, args->geo->blksize); 999 1000 /* 1001 * Clean out the prior contents of the attribute list. 1002 */ 1003 error = xfs_da_shrink_inode(args, 0, bp); 1004 if (error) 1005 goto out; 1006 1007 if (forkoff == -1) { 1008 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2); 1009 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE); 1010 xfs_attr_fork_remove(dp, args->trans); 1011 goto out; 1012 } 1013 1014 xfs_attr_shortform_create(args); 1015 1016 /* 1017 * Copy the attributes 1018 */ 1019 memset((char *)&nargs, 0, sizeof(nargs)); 1020 nargs.geo = args->geo; 1021 nargs.dp = dp; 1022 nargs.firstblock = args->firstblock; 1023 nargs.dfops = args->dfops; 1024 nargs.total = args->total; 1025 nargs.whichfork = XFS_ATTR_FORK; 1026 nargs.trans = args->trans; 1027 nargs.op_flags = XFS_DA_OP_OKNOENT; 1028 1029 for (i = 0; i < ichdr.count; entry++, i++) { 1030 if (entry->flags & XFS_ATTR_INCOMPLETE) 1031 continue; /* don't copy partial entries */ 1032 if (!entry->nameidx) 1033 continue; 1034 ASSERT(entry->flags & XFS_ATTR_LOCAL); 1035 name_loc = xfs_attr3_leaf_name_local(leaf, i); 1036 nargs.name = name_loc->nameval; 1037 nargs.namelen = name_loc->namelen; 1038 nargs.value = &name_loc->nameval[nargs.namelen]; 1039 nargs.valuelen = be16_to_cpu(name_loc->valuelen); 1040 nargs.hashval = be32_to_cpu(entry->hashval); 1041 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags); 1042 xfs_attr_shortform_add(&nargs, forkoff); 1043 } 1044 error = 0; 1045 1046 out: 1047 kmem_free(tmpbuffer); 1048 return error; 1049 } 1050 1051 /* 1052 * Convert from using a single leaf to a root node and a leaf. 1053 */ 1054 int 1055 xfs_attr3_leaf_to_node( 1056 struct xfs_da_args *args) 1057 { 1058 struct xfs_attr_leafblock *leaf; 1059 struct xfs_attr3_icleaf_hdr icleafhdr; 1060 struct xfs_attr_leaf_entry *entries; 1061 struct xfs_da_node_entry *btree; 1062 struct xfs_da3_icnode_hdr icnodehdr; 1063 struct xfs_da_intnode *node; 1064 struct xfs_inode *dp = args->dp; 1065 struct xfs_mount *mp = dp->i_mount; 1066 struct xfs_buf *bp1 = NULL; 1067 struct xfs_buf *bp2 = NULL; 1068 xfs_dablk_t blkno; 1069 int error; 1070 1071 trace_xfs_attr_leaf_to_node(args); 1072 1073 error = xfs_da_grow_inode(args, &blkno); 1074 if (error) 1075 goto out; 1076 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1); 1077 if (error) 1078 goto out; 1079 1080 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK); 1081 if (error) 1082 goto out; 1083 1084 /* copy leaf to new buffer, update identifiers */ 1085 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF); 1086 bp2->b_ops = bp1->b_ops; 1087 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize); 1088 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1089 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr; 1090 hdr3->blkno = cpu_to_be64(bp2->b_bn); 1091 } 1092 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1); 1093 1094 /* 1095 * Set up the new root node. 1096 */ 1097 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK); 1098 if (error) 1099 goto out; 1100 node = bp1->b_addr; 1101 dp->d_ops->node_hdr_from_disk(&icnodehdr, node); 1102 btree = dp->d_ops->node_tree_p(node); 1103 1104 leaf = bp2->b_addr; 1105 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf); 1106 entries = xfs_attr3_leaf_entryp(leaf); 1107 1108 /* both on-disk, don't endian-flip twice */ 1109 btree[0].hashval = entries[icleafhdr.count - 1].hashval; 1110 btree[0].before = cpu_to_be32(blkno); 1111 icnodehdr.count = 1; 1112 dp->d_ops->node_hdr_to_disk(node, &icnodehdr); 1113 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1); 1114 error = 0; 1115 out: 1116 return error; 1117 } 1118 1119 /*======================================================================== 1120 * Routines used for growing the Btree. 1121 *========================================================================*/ 1122 1123 /* 1124 * Create the initial contents of a leaf attribute list 1125 * or a leaf in a node attribute list. 1126 */ 1127 STATIC int 1128 xfs_attr3_leaf_create( 1129 struct xfs_da_args *args, 1130 xfs_dablk_t blkno, 1131 struct xfs_buf **bpp) 1132 { 1133 struct xfs_attr_leafblock *leaf; 1134 struct xfs_attr3_icleaf_hdr ichdr; 1135 struct xfs_inode *dp = args->dp; 1136 struct xfs_mount *mp = dp->i_mount; 1137 struct xfs_buf *bp; 1138 int error; 1139 1140 trace_xfs_attr_leaf_create(args); 1141 1142 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp, 1143 XFS_ATTR_FORK); 1144 if (error) 1145 return error; 1146 bp->b_ops = &xfs_attr3_leaf_buf_ops; 1147 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF); 1148 leaf = bp->b_addr; 1149 memset(leaf, 0, args->geo->blksize); 1150 1151 memset(&ichdr, 0, sizeof(ichdr)); 1152 ichdr.firstused = args->geo->blksize; 1153 1154 if (xfs_sb_version_hascrc(&mp->m_sb)) { 1155 struct xfs_da3_blkinfo *hdr3 = bp->b_addr; 1156 1157 ichdr.magic = XFS_ATTR3_LEAF_MAGIC; 1158 1159 hdr3->blkno = cpu_to_be64(bp->b_bn); 1160 hdr3->owner = cpu_to_be64(dp->i_ino); 1161 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid); 1162 1163 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr); 1164 } else { 1165 ichdr.magic = XFS_ATTR_LEAF_MAGIC; 1166 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr); 1167 } 1168 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base; 1169 1170 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 1171 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1); 1172 1173 *bpp = bp; 1174 return 0; 1175 } 1176 1177 /* 1178 * Split the leaf node, rebalance, then add the new entry. 1179 */ 1180 int 1181 xfs_attr3_leaf_split( 1182 struct xfs_da_state *state, 1183 struct xfs_da_state_blk *oldblk, 1184 struct xfs_da_state_blk *newblk) 1185 { 1186 xfs_dablk_t blkno; 1187 int error; 1188 1189 trace_xfs_attr_leaf_split(state->args); 1190 1191 /* 1192 * Allocate space for a new leaf node. 1193 */ 1194 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC); 1195 error = xfs_da_grow_inode(state->args, &blkno); 1196 if (error) 1197 return error; 1198 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp); 1199 if (error) 1200 return error; 1201 newblk->blkno = blkno; 1202 newblk->magic = XFS_ATTR_LEAF_MAGIC; 1203 1204 /* 1205 * Rebalance the entries across the two leaves. 1206 * NOTE: rebalance() currently depends on the 2nd block being empty. 1207 */ 1208 xfs_attr3_leaf_rebalance(state, oldblk, newblk); 1209 error = xfs_da3_blk_link(state, oldblk, newblk); 1210 if (error) 1211 return error; 1212 1213 /* 1214 * Save info on "old" attribute for "atomic rename" ops, leaf_add() 1215 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the 1216 * "new" attrs info. Will need the "old" info to remove it later. 1217 * 1218 * Insert the "new" entry in the correct block. 1219 */ 1220 if (state->inleaf) { 1221 trace_xfs_attr_leaf_add_old(state->args); 1222 error = xfs_attr3_leaf_add(oldblk->bp, state->args); 1223 } else { 1224 trace_xfs_attr_leaf_add_new(state->args); 1225 error = xfs_attr3_leaf_add(newblk->bp, state->args); 1226 } 1227 1228 /* 1229 * Update last hashval in each block since we added the name. 1230 */ 1231 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL); 1232 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL); 1233 return error; 1234 } 1235 1236 /* 1237 * Add a name to the leaf attribute list structure. 1238 */ 1239 int 1240 xfs_attr3_leaf_add( 1241 struct xfs_buf *bp, 1242 struct xfs_da_args *args) 1243 { 1244 struct xfs_attr_leafblock *leaf; 1245 struct xfs_attr3_icleaf_hdr ichdr; 1246 int tablesize; 1247 int entsize; 1248 int sum; 1249 int tmp; 1250 int i; 1251 1252 trace_xfs_attr_leaf_add(args); 1253 1254 leaf = bp->b_addr; 1255 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 1256 ASSERT(args->index >= 0 && args->index <= ichdr.count); 1257 entsize = xfs_attr_leaf_newentsize(args, NULL); 1258 1259 /* 1260 * Search through freemap for first-fit on new name length. 1261 * (may need to figure in size of entry struct too) 1262 */ 1263 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t) 1264 + xfs_attr3_leaf_hdr_size(leaf); 1265 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) { 1266 if (tablesize > ichdr.firstused) { 1267 sum += ichdr.freemap[i].size; 1268 continue; 1269 } 1270 if (!ichdr.freemap[i].size) 1271 continue; /* no space in this map */ 1272 tmp = entsize; 1273 if (ichdr.freemap[i].base < ichdr.firstused) 1274 tmp += sizeof(xfs_attr_leaf_entry_t); 1275 if (ichdr.freemap[i].size >= tmp) { 1276 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i); 1277 goto out_log_hdr; 1278 } 1279 sum += ichdr.freemap[i].size; 1280 } 1281 1282 /* 1283 * If there are no holes in the address space of the block, 1284 * and we don't have enough freespace, then compaction will do us 1285 * no good and we should just give up. 1286 */ 1287 if (!ichdr.holes && sum < entsize) 1288 return -ENOSPC; 1289 1290 /* 1291 * Compact the entries to coalesce free space. 1292 * This may change the hdr->count via dropping INCOMPLETE entries. 1293 */ 1294 xfs_attr3_leaf_compact(args, &ichdr, bp); 1295 1296 /* 1297 * After compaction, the block is guaranteed to have only one 1298 * free region, in freemap[0]. If it is not big enough, give up. 1299 */ 1300 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) { 1301 tmp = -ENOSPC; 1302 goto out_log_hdr; 1303 } 1304 1305 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0); 1306 1307 out_log_hdr: 1308 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 1309 xfs_trans_log_buf(args->trans, bp, 1310 XFS_DA_LOGRANGE(leaf, &leaf->hdr, 1311 xfs_attr3_leaf_hdr_size(leaf))); 1312 return tmp; 1313 } 1314 1315 /* 1316 * Add a name to a leaf attribute list structure. 1317 */ 1318 STATIC int 1319 xfs_attr3_leaf_add_work( 1320 struct xfs_buf *bp, 1321 struct xfs_attr3_icleaf_hdr *ichdr, 1322 struct xfs_da_args *args, 1323 int mapindex) 1324 { 1325 struct xfs_attr_leafblock *leaf; 1326 struct xfs_attr_leaf_entry *entry; 1327 struct xfs_attr_leaf_name_local *name_loc; 1328 struct xfs_attr_leaf_name_remote *name_rmt; 1329 struct xfs_mount *mp; 1330 int tmp; 1331 int i; 1332 1333 trace_xfs_attr_leaf_add_work(args); 1334 1335 leaf = bp->b_addr; 1336 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE); 1337 ASSERT(args->index >= 0 && args->index <= ichdr->count); 1338 1339 /* 1340 * Force open some space in the entry array and fill it in. 1341 */ 1342 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 1343 if (args->index < ichdr->count) { 1344 tmp = ichdr->count - args->index; 1345 tmp *= sizeof(xfs_attr_leaf_entry_t); 1346 memmove(entry + 1, entry, tmp); 1347 xfs_trans_log_buf(args->trans, bp, 1348 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry))); 1349 } 1350 ichdr->count++; 1351 1352 /* 1353 * Allocate space for the new string (at the end of the run). 1354 */ 1355 mp = args->trans->t_mountp; 1356 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize); 1357 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0); 1358 ASSERT(ichdr->freemap[mapindex].size >= 1359 xfs_attr_leaf_newentsize(args, NULL)); 1360 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize); 1361 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0); 1362 1363 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp); 1364 1365 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base + 1366 ichdr->freemap[mapindex].size); 1367 entry->hashval = cpu_to_be32(args->hashval); 1368 entry->flags = tmp ? XFS_ATTR_LOCAL : 0; 1369 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags); 1370 if (args->op_flags & XFS_DA_OP_RENAME) { 1371 entry->flags |= XFS_ATTR_INCOMPLETE; 1372 if ((args->blkno2 == args->blkno) && 1373 (args->index2 <= args->index)) { 1374 args->index2++; 1375 } 1376 } 1377 xfs_trans_log_buf(args->trans, bp, 1378 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 1379 ASSERT((args->index == 0) || 1380 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval))); 1381 ASSERT((args->index == ichdr->count - 1) || 1382 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval))); 1383 1384 /* 1385 * For "remote" attribute values, simply note that we need to 1386 * allocate space for the "remote" value. We can't actually 1387 * allocate the extents in this transaction, and we can't decide 1388 * which blocks they should be as we might allocate more blocks 1389 * as part of this transaction (a split operation for example). 1390 */ 1391 if (entry->flags & XFS_ATTR_LOCAL) { 1392 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 1393 name_loc->namelen = args->namelen; 1394 name_loc->valuelen = cpu_to_be16(args->valuelen); 1395 memcpy((char *)name_loc->nameval, args->name, args->namelen); 1396 memcpy((char *)&name_loc->nameval[args->namelen], args->value, 1397 be16_to_cpu(name_loc->valuelen)); 1398 } else { 1399 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 1400 name_rmt->namelen = args->namelen; 1401 memcpy((char *)name_rmt->name, args->name, args->namelen); 1402 entry->flags |= XFS_ATTR_INCOMPLETE; 1403 /* just in case */ 1404 name_rmt->valuelen = 0; 1405 name_rmt->valueblk = 0; 1406 args->rmtblkno = 1; 1407 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen); 1408 args->rmtvaluelen = args->valuelen; 1409 } 1410 xfs_trans_log_buf(args->trans, bp, 1411 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index), 1412 xfs_attr_leaf_entsize(leaf, args->index))); 1413 1414 /* 1415 * Update the control info for this leaf node 1416 */ 1417 if (be16_to_cpu(entry->nameidx) < ichdr->firstused) 1418 ichdr->firstused = be16_to_cpu(entry->nameidx); 1419 1420 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t) 1421 + xfs_attr3_leaf_hdr_size(leaf)); 1422 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t) 1423 + xfs_attr3_leaf_hdr_size(leaf); 1424 1425 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 1426 if (ichdr->freemap[i].base == tmp) { 1427 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t); 1428 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t); 1429 } 1430 } 1431 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index); 1432 return 0; 1433 } 1434 1435 /* 1436 * Garbage collect a leaf attribute list block by copying it to a new buffer. 1437 */ 1438 STATIC void 1439 xfs_attr3_leaf_compact( 1440 struct xfs_da_args *args, 1441 struct xfs_attr3_icleaf_hdr *ichdr_dst, 1442 struct xfs_buf *bp) 1443 { 1444 struct xfs_attr_leafblock *leaf_src; 1445 struct xfs_attr_leafblock *leaf_dst; 1446 struct xfs_attr3_icleaf_hdr ichdr_src; 1447 struct xfs_trans *trans = args->trans; 1448 char *tmpbuffer; 1449 1450 trace_xfs_attr_leaf_compact(args); 1451 1452 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP); 1453 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize); 1454 memset(bp->b_addr, 0, args->geo->blksize); 1455 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer; 1456 leaf_dst = bp->b_addr; 1457 1458 /* 1459 * Copy the on-disk header back into the destination buffer to ensure 1460 * all the information in the header that is not part of the incore 1461 * header structure is preserved. 1462 */ 1463 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src)); 1464 1465 /* Initialise the incore headers */ 1466 ichdr_src = *ichdr_dst; /* struct copy */ 1467 ichdr_dst->firstused = args->geo->blksize; 1468 ichdr_dst->usedbytes = 0; 1469 ichdr_dst->count = 0; 1470 ichdr_dst->holes = 0; 1471 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src); 1472 ichdr_dst->freemap[0].size = ichdr_dst->firstused - 1473 ichdr_dst->freemap[0].base; 1474 1475 /* write the header back to initialise the underlying buffer */ 1476 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst); 1477 1478 /* 1479 * Copy all entry's in the same (sorted) order, 1480 * but allocate name/value pairs packed and in sequence. 1481 */ 1482 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0, 1483 leaf_dst, ichdr_dst, 0, ichdr_src.count); 1484 /* 1485 * this logs the entire buffer, but the caller must write the header 1486 * back to the buffer when it is finished modifying it. 1487 */ 1488 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1); 1489 1490 kmem_free(tmpbuffer); 1491 } 1492 1493 /* 1494 * Compare two leaf blocks "order". 1495 * Return 0 unless leaf2 should go before leaf1. 1496 */ 1497 static int 1498 xfs_attr3_leaf_order( 1499 struct xfs_buf *leaf1_bp, 1500 struct xfs_attr3_icleaf_hdr *leaf1hdr, 1501 struct xfs_buf *leaf2_bp, 1502 struct xfs_attr3_icleaf_hdr *leaf2hdr) 1503 { 1504 struct xfs_attr_leaf_entry *entries1; 1505 struct xfs_attr_leaf_entry *entries2; 1506 1507 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr); 1508 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr); 1509 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 && 1510 ((be32_to_cpu(entries2[0].hashval) < 1511 be32_to_cpu(entries1[0].hashval)) || 1512 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) < 1513 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) { 1514 return 1; 1515 } 1516 return 0; 1517 } 1518 1519 int 1520 xfs_attr_leaf_order( 1521 struct xfs_buf *leaf1_bp, 1522 struct xfs_buf *leaf2_bp) 1523 { 1524 struct xfs_attr3_icleaf_hdr ichdr1; 1525 struct xfs_attr3_icleaf_hdr ichdr2; 1526 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount; 1527 1528 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr); 1529 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr); 1530 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2); 1531 } 1532 1533 /* 1534 * Redistribute the attribute list entries between two leaf nodes, 1535 * taking into account the size of the new entry. 1536 * 1537 * NOTE: if new block is empty, then it will get the upper half of the 1538 * old block. At present, all (one) callers pass in an empty second block. 1539 * 1540 * This code adjusts the args->index/blkno and args->index2/blkno2 fields 1541 * to match what it is doing in splitting the attribute leaf block. Those 1542 * values are used in "atomic rename" operations on attributes. Note that 1543 * the "new" and "old" values can end up in different blocks. 1544 */ 1545 STATIC void 1546 xfs_attr3_leaf_rebalance( 1547 struct xfs_da_state *state, 1548 struct xfs_da_state_blk *blk1, 1549 struct xfs_da_state_blk *blk2) 1550 { 1551 struct xfs_da_args *args; 1552 struct xfs_attr_leafblock *leaf1; 1553 struct xfs_attr_leafblock *leaf2; 1554 struct xfs_attr3_icleaf_hdr ichdr1; 1555 struct xfs_attr3_icleaf_hdr ichdr2; 1556 struct xfs_attr_leaf_entry *entries1; 1557 struct xfs_attr_leaf_entry *entries2; 1558 int count; 1559 int totallen; 1560 int max; 1561 int space; 1562 int swap; 1563 1564 /* 1565 * Set up environment. 1566 */ 1567 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC); 1568 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC); 1569 leaf1 = blk1->bp->b_addr; 1570 leaf2 = blk2->bp->b_addr; 1571 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1); 1572 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2); 1573 ASSERT(ichdr2.count == 0); 1574 args = state->args; 1575 1576 trace_xfs_attr_leaf_rebalance(args); 1577 1578 /* 1579 * Check ordering of blocks, reverse if it makes things simpler. 1580 * 1581 * NOTE: Given that all (current) callers pass in an empty 1582 * second block, this code should never set "swap". 1583 */ 1584 swap = 0; 1585 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) { 1586 struct xfs_da_state_blk *tmp_blk; 1587 struct xfs_attr3_icleaf_hdr tmp_ichdr; 1588 1589 tmp_blk = blk1; 1590 blk1 = blk2; 1591 blk2 = tmp_blk; 1592 1593 /* struct copies to swap them rather than reconverting */ 1594 tmp_ichdr = ichdr1; 1595 ichdr1 = ichdr2; 1596 ichdr2 = tmp_ichdr; 1597 1598 leaf1 = blk1->bp->b_addr; 1599 leaf2 = blk2->bp->b_addr; 1600 swap = 1; 1601 } 1602 1603 /* 1604 * Examine entries until we reduce the absolute difference in 1605 * byte usage between the two blocks to a minimum. Then get 1606 * the direction to copy and the number of elements to move. 1607 * 1608 * "inleaf" is true if the new entry should be inserted into blk1. 1609 * If "swap" is also true, then reverse the sense of "inleaf". 1610 */ 1611 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1, 1612 blk2, &ichdr2, 1613 &count, &totallen); 1614 if (swap) 1615 state->inleaf = !state->inleaf; 1616 1617 /* 1618 * Move any entries required from leaf to leaf: 1619 */ 1620 if (count < ichdr1.count) { 1621 /* 1622 * Figure the total bytes to be added to the destination leaf. 1623 */ 1624 /* number entries being moved */ 1625 count = ichdr1.count - count; 1626 space = ichdr1.usedbytes - totallen; 1627 space += count * sizeof(xfs_attr_leaf_entry_t); 1628 1629 /* 1630 * leaf2 is the destination, compact it if it looks tight. 1631 */ 1632 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1); 1633 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t); 1634 if (space > max) 1635 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp); 1636 1637 /* 1638 * Move high entries from leaf1 to low end of leaf2. 1639 */ 1640 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1, 1641 ichdr1.count - count, leaf2, &ichdr2, 0, count); 1642 1643 } else if (count > ichdr1.count) { 1644 /* 1645 * I assert that since all callers pass in an empty 1646 * second buffer, this code should never execute. 1647 */ 1648 ASSERT(0); 1649 1650 /* 1651 * Figure the total bytes to be added to the destination leaf. 1652 */ 1653 /* number entries being moved */ 1654 count -= ichdr1.count; 1655 space = totallen - ichdr1.usedbytes; 1656 space += count * sizeof(xfs_attr_leaf_entry_t); 1657 1658 /* 1659 * leaf1 is the destination, compact it if it looks tight. 1660 */ 1661 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1); 1662 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t); 1663 if (space > max) 1664 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp); 1665 1666 /* 1667 * Move low entries from leaf2 to high end of leaf1. 1668 */ 1669 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1, 1670 ichdr1.count, count); 1671 } 1672 1673 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1); 1674 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2); 1675 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1); 1676 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1); 1677 1678 /* 1679 * Copy out last hashval in each block for B-tree code. 1680 */ 1681 entries1 = xfs_attr3_leaf_entryp(leaf1); 1682 entries2 = xfs_attr3_leaf_entryp(leaf2); 1683 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval); 1684 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval); 1685 1686 /* 1687 * Adjust the expected index for insertion. 1688 * NOTE: this code depends on the (current) situation that the 1689 * second block was originally empty. 1690 * 1691 * If the insertion point moved to the 2nd block, we must adjust 1692 * the index. We must also track the entry just following the 1693 * new entry for use in an "atomic rename" operation, that entry 1694 * is always the "old" entry and the "new" entry is what we are 1695 * inserting. The index/blkno fields refer to the "old" entry, 1696 * while the index2/blkno2 fields refer to the "new" entry. 1697 */ 1698 if (blk1->index > ichdr1.count) { 1699 ASSERT(state->inleaf == 0); 1700 blk2->index = blk1->index - ichdr1.count; 1701 args->index = args->index2 = blk2->index; 1702 args->blkno = args->blkno2 = blk2->blkno; 1703 } else if (blk1->index == ichdr1.count) { 1704 if (state->inleaf) { 1705 args->index = blk1->index; 1706 args->blkno = blk1->blkno; 1707 args->index2 = 0; 1708 args->blkno2 = blk2->blkno; 1709 } else { 1710 /* 1711 * On a double leaf split, the original attr location 1712 * is already stored in blkno2/index2, so don't 1713 * overwrite it overwise we corrupt the tree. 1714 */ 1715 blk2->index = blk1->index - ichdr1.count; 1716 args->index = blk2->index; 1717 args->blkno = blk2->blkno; 1718 if (!state->extravalid) { 1719 /* 1720 * set the new attr location to match the old 1721 * one and let the higher level split code 1722 * decide where in the leaf to place it. 1723 */ 1724 args->index2 = blk2->index; 1725 args->blkno2 = blk2->blkno; 1726 } 1727 } 1728 } else { 1729 ASSERT(state->inleaf == 1); 1730 args->index = args->index2 = blk1->index; 1731 args->blkno = args->blkno2 = blk1->blkno; 1732 } 1733 } 1734 1735 /* 1736 * Examine entries until we reduce the absolute difference in 1737 * byte usage between the two blocks to a minimum. 1738 * GROT: Is this really necessary? With other than a 512 byte blocksize, 1739 * GROT: there will always be enough room in either block for a new entry. 1740 * GROT: Do a double-split for this case? 1741 */ 1742 STATIC int 1743 xfs_attr3_leaf_figure_balance( 1744 struct xfs_da_state *state, 1745 struct xfs_da_state_blk *blk1, 1746 struct xfs_attr3_icleaf_hdr *ichdr1, 1747 struct xfs_da_state_blk *blk2, 1748 struct xfs_attr3_icleaf_hdr *ichdr2, 1749 int *countarg, 1750 int *usedbytesarg) 1751 { 1752 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr; 1753 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr; 1754 struct xfs_attr_leaf_entry *entry; 1755 int count; 1756 int max; 1757 int index; 1758 int totallen = 0; 1759 int half; 1760 int lastdelta; 1761 int foundit = 0; 1762 int tmp; 1763 1764 /* 1765 * Examine entries until we reduce the absolute difference in 1766 * byte usage between the two blocks to a minimum. 1767 */ 1768 max = ichdr1->count + ichdr2->count; 1769 half = (max + 1) * sizeof(*entry); 1770 half += ichdr1->usedbytes + ichdr2->usedbytes + 1771 xfs_attr_leaf_newentsize(state->args, NULL); 1772 half /= 2; 1773 lastdelta = state->args->geo->blksize; 1774 entry = xfs_attr3_leaf_entryp(leaf1); 1775 for (count = index = 0; count < max; entry++, index++, count++) { 1776 1777 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A)) 1778 /* 1779 * The new entry is in the first block, account for it. 1780 */ 1781 if (count == blk1->index) { 1782 tmp = totallen + sizeof(*entry) + 1783 xfs_attr_leaf_newentsize(state->args, NULL); 1784 if (XFS_ATTR_ABS(half - tmp) > lastdelta) 1785 break; 1786 lastdelta = XFS_ATTR_ABS(half - tmp); 1787 totallen = tmp; 1788 foundit = 1; 1789 } 1790 1791 /* 1792 * Wrap around into the second block if necessary. 1793 */ 1794 if (count == ichdr1->count) { 1795 leaf1 = leaf2; 1796 entry = xfs_attr3_leaf_entryp(leaf1); 1797 index = 0; 1798 } 1799 1800 /* 1801 * Figure out if next leaf entry would be too much. 1802 */ 1803 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1, 1804 index); 1805 if (XFS_ATTR_ABS(half - tmp) > lastdelta) 1806 break; 1807 lastdelta = XFS_ATTR_ABS(half - tmp); 1808 totallen = tmp; 1809 #undef XFS_ATTR_ABS 1810 } 1811 1812 /* 1813 * Calculate the number of usedbytes that will end up in lower block. 1814 * If new entry not in lower block, fix up the count. 1815 */ 1816 totallen -= count * sizeof(*entry); 1817 if (foundit) { 1818 totallen -= sizeof(*entry) + 1819 xfs_attr_leaf_newentsize(state->args, NULL); 1820 } 1821 1822 *countarg = count; 1823 *usedbytesarg = totallen; 1824 return foundit; 1825 } 1826 1827 /*======================================================================== 1828 * Routines used for shrinking the Btree. 1829 *========================================================================*/ 1830 1831 /* 1832 * Check a leaf block and its neighbors to see if the block should be 1833 * collapsed into one or the other neighbor. Always keep the block 1834 * with the smaller block number. 1835 * If the current block is over 50% full, don't try to join it, return 0. 1836 * If the block is empty, fill in the state structure and return 2. 1837 * If it can be collapsed, fill in the state structure and return 1. 1838 * If nothing can be done, return 0. 1839 * 1840 * GROT: allow for INCOMPLETE entries in calculation. 1841 */ 1842 int 1843 xfs_attr3_leaf_toosmall( 1844 struct xfs_da_state *state, 1845 int *action) 1846 { 1847 struct xfs_attr_leafblock *leaf; 1848 struct xfs_da_state_blk *blk; 1849 struct xfs_attr3_icleaf_hdr ichdr; 1850 struct xfs_buf *bp; 1851 xfs_dablk_t blkno; 1852 int bytes; 1853 int forward; 1854 int error; 1855 int retval; 1856 int i; 1857 1858 trace_xfs_attr_leaf_toosmall(state->args); 1859 1860 /* 1861 * Check for the degenerate case of the block being over 50% full. 1862 * If so, it's not worth even looking to see if we might be able 1863 * to coalesce with a sibling. 1864 */ 1865 blk = &state->path.blk[ state->path.active-1 ]; 1866 leaf = blk->bp->b_addr; 1867 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf); 1868 bytes = xfs_attr3_leaf_hdr_size(leaf) + 1869 ichdr.count * sizeof(xfs_attr_leaf_entry_t) + 1870 ichdr.usedbytes; 1871 if (bytes > (state->args->geo->blksize >> 1)) { 1872 *action = 0; /* blk over 50%, don't try to join */ 1873 return 0; 1874 } 1875 1876 /* 1877 * Check for the degenerate case of the block being empty. 1878 * If the block is empty, we'll simply delete it, no need to 1879 * coalesce it with a sibling block. We choose (arbitrarily) 1880 * to merge with the forward block unless it is NULL. 1881 */ 1882 if (ichdr.count == 0) { 1883 /* 1884 * Make altpath point to the block we want to keep and 1885 * path point to the block we want to drop (this one). 1886 */ 1887 forward = (ichdr.forw != 0); 1888 memcpy(&state->altpath, &state->path, sizeof(state->path)); 1889 error = xfs_da3_path_shift(state, &state->altpath, forward, 1890 0, &retval); 1891 if (error) 1892 return error; 1893 if (retval) { 1894 *action = 0; 1895 } else { 1896 *action = 2; 1897 } 1898 return 0; 1899 } 1900 1901 /* 1902 * Examine each sibling block to see if we can coalesce with 1903 * at least 25% free space to spare. We need to figure out 1904 * whether to merge with the forward or the backward block. 1905 * We prefer coalescing with the lower numbered sibling so as 1906 * to shrink an attribute list over time. 1907 */ 1908 /* start with smaller blk num */ 1909 forward = ichdr.forw < ichdr.back; 1910 for (i = 0; i < 2; forward = !forward, i++) { 1911 struct xfs_attr3_icleaf_hdr ichdr2; 1912 if (forward) 1913 blkno = ichdr.forw; 1914 else 1915 blkno = ichdr.back; 1916 if (blkno == 0) 1917 continue; 1918 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp, 1919 blkno, -1, &bp); 1920 if (error) 1921 return error; 1922 1923 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr); 1924 1925 bytes = state->args->geo->blksize - 1926 (state->args->geo->blksize >> 2) - 1927 ichdr.usedbytes - ichdr2.usedbytes - 1928 ((ichdr.count + ichdr2.count) * 1929 sizeof(xfs_attr_leaf_entry_t)) - 1930 xfs_attr3_leaf_hdr_size(leaf); 1931 1932 xfs_trans_brelse(state->args->trans, bp); 1933 if (bytes >= 0) 1934 break; /* fits with at least 25% to spare */ 1935 } 1936 if (i >= 2) { 1937 *action = 0; 1938 return 0; 1939 } 1940 1941 /* 1942 * Make altpath point to the block we want to keep (the lower 1943 * numbered block) and path point to the block we want to drop. 1944 */ 1945 memcpy(&state->altpath, &state->path, sizeof(state->path)); 1946 if (blkno < blk->blkno) { 1947 error = xfs_da3_path_shift(state, &state->altpath, forward, 1948 0, &retval); 1949 } else { 1950 error = xfs_da3_path_shift(state, &state->path, forward, 1951 0, &retval); 1952 } 1953 if (error) 1954 return error; 1955 if (retval) { 1956 *action = 0; 1957 } else { 1958 *action = 1; 1959 } 1960 return 0; 1961 } 1962 1963 /* 1964 * Remove a name from the leaf attribute list structure. 1965 * 1966 * Return 1 if leaf is less than 37% full, 0 if >= 37% full. 1967 * If two leaves are 37% full, when combined they will leave 25% free. 1968 */ 1969 int 1970 xfs_attr3_leaf_remove( 1971 struct xfs_buf *bp, 1972 struct xfs_da_args *args) 1973 { 1974 struct xfs_attr_leafblock *leaf; 1975 struct xfs_attr3_icleaf_hdr ichdr; 1976 struct xfs_attr_leaf_entry *entry; 1977 int before; 1978 int after; 1979 int smallest; 1980 int entsize; 1981 int tablesize; 1982 int tmp; 1983 int i; 1984 1985 trace_xfs_attr_leaf_remove(args); 1986 1987 leaf = bp->b_addr; 1988 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 1989 1990 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8); 1991 ASSERT(args->index >= 0 && args->index < ichdr.count); 1992 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) + 1993 xfs_attr3_leaf_hdr_size(leaf)); 1994 1995 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 1996 1997 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused); 1998 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize); 1999 2000 /* 2001 * Scan through free region table: 2002 * check for adjacency of free'd entry with an existing one, 2003 * find smallest free region in case we need to replace it, 2004 * adjust any map that borders the entry table, 2005 */ 2006 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t) 2007 + xfs_attr3_leaf_hdr_size(leaf); 2008 tmp = ichdr.freemap[0].size; 2009 before = after = -1; 2010 smallest = XFS_ATTR_LEAF_MAPSIZE - 1; 2011 entsize = xfs_attr_leaf_entsize(leaf, args->index); 2012 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) { 2013 ASSERT(ichdr.freemap[i].base < args->geo->blksize); 2014 ASSERT(ichdr.freemap[i].size < args->geo->blksize); 2015 if (ichdr.freemap[i].base == tablesize) { 2016 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t); 2017 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t); 2018 } 2019 2020 if (ichdr.freemap[i].base + ichdr.freemap[i].size == 2021 be16_to_cpu(entry->nameidx)) { 2022 before = i; 2023 } else if (ichdr.freemap[i].base == 2024 (be16_to_cpu(entry->nameidx) + entsize)) { 2025 after = i; 2026 } else if (ichdr.freemap[i].size < tmp) { 2027 tmp = ichdr.freemap[i].size; 2028 smallest = i; 2029 } 2030 } 2031 2032 /* 2033 * Coalesce adjacent freemap regions, 2034 * or replace the smallest region. 2035 */ 2036 if ((before >= 0) || (after >= 0)) { 2037 if ((before >= 0) && (after >= 0)) { 2038 ichdr.freemap[before].size += entsize; 2039 ichdr.freemap[before].size += ichdr.freemap[after].size; 2040 ichdr.freemap[after].base = 0; 2041 ichdr.freemap[after].size = 0; 2042 } else if (before >= 0) { 2043 ichdr.freemap[before].size += entsize; 2044 } else { 2045 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx); 2046 ichdr.freemap[after].size += entsize; 2047 } 2048 } else { 2049 /* 2050 * Replace smallest region (if it is smaller than free'd entry) 2051 */ 2052 if (ichdr.freemap[smallest].size < entsize) { 2053 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx); 2054 ichdr.freemap[smallest].size = entsize; 2055 } 2056 } 2057 2058 /* 2059 * Did we remove the first entry? 2060 */ 2061 if (be16_to_cpu(entry->nameidx) == ichdr.firstused) 2062 smallest = 1; 2063 else 2064 smallest = 0; 2065 2066 /* 2067 * Compress the remaining entries and zero out the removed stuff. 2068 */ 2069 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize); 2070 ichdr.usedbytes -= entsize; 2071 xfs_trans_log_buf(args->trans, bp, 2072 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index), 2073 entsize)); 2074 2075 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t); 2076 memmove(entry, entry + 1, tmp); 2077 ichdr.count--; 2078 xfs_trans_log_buf(args->trans, bp, 2079 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t))); 2080 2081 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count]; 2082 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t)); 2083 2084 /* 2085 * If we removed the first entry, re-find the first used byte 2086 * in the name area. Note that if the entry was the "firstused", 2087 * then we don't have a "hole" in our block resulting from 2088 * removing the name. 2089 */ 2090 if (smallest) { 2091 tmp = args->geo->blksize; 2092 entry = xfs_attr3_leaf_entryp(leaf); 2093 for (i = ichdr.count - 1; i >= 0; entry++, i--) { 2094 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused); 2095 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize); 2096 2097 if (be16_to_cpu(entry->nameidx) < tmp) 2098 tmp = be16_to_cpu(entry->nameidx); 2099 } 2100 ichdr.firstused = tmp; 2101 ASSERT(ichdr.firstused != 0); 2102 } else { 2103 ichdr.holes = 1; /* mark as needing compaction */ 2104 } 2105 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr); 2106 xfs_trans_log_buf(args->trans, bp, 2107 XFS_DA_LOGRANGE(leaf, &leaf->hdr, 2108 xfs_attr3_leaf_hdr_size(leaf))); 2109 2110 /* 2111 * Check if leaf is less than 50% full, caller may want to 2112 * "join" the leaf with a sibling if so. 2113 */ 2114 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) + 2115 ichdr.count * sizeof(xfs_attr_leaf_entry_t); 2116 2117 return tmp < args->geo->magicpct; /* leaf is < 37% full */ 2118 } 2119 2120 /* 2121 * Move all the attribute list entries from drop_leaf into save_leaf. 2122 */ 2123 void 2124 xfs_attr3_leaf_unbalance( 2125 struct xfs_da_state *state, 2126 struct xfs_da_state_blk *drop_blk, 2127 struct xfs_da_state_blk *save_blk) 2128 { 2129 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr; 2130 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr; 2131 struct xfs_attr3_icleaf_hdr drophdr; 2132 struct xfs_attr3_icleaf_hdr savehdr; 2133 struct xfs_attr_leaf_entry *entry; 2134 2135 trace_xfs_attr_leaf_unbalance(state->args); 2136 2137 drop_leaf = drop_blk->bp->b_addr; 2138 save_leaf = save_blk->bp->b_addr; 2139 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf); 2140 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf); 2141 entry = xfs_attr3_leaf_entryp(drop_leaf); 2142 2143 /* 2144 * Save last hashval from dying block for later Btree fixup. 2145 */ 2146 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval); 2147 2148 /* 2149 * Check if we need a temp buffer, or can we do it in place. 2150 * Note that we don't check "leaf" for holes because we will 2151 * always be dropping it, toosmall() decided that for us already. 2152 */ 2153 if (savehdr.holes == 0) { 2154 /* 2155 * dest leaf has no holes, so we add there. May need 2156 * to make some room in the entry array. 2157 */ 2158 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr, 2159 drop_blk->bp, &drophdr)) { 2160 xfs_attr3_leaf_moveents(state->args, 2161 drop_leaf, &drophdr, 0, 2162 save_leaf, &savehdr, 0, 2163 drophdr.count); 2164 } else { 2165 xfs_attr3_leaf_moveents(state->args, 2166 drop_leaf, &drophdr, 0, 2167 save_leaf, &savehdr, 2168 savehdr.count, drophdr.count); 2169 } 2170 } else { 2171 /* 2172 * Destination has holes, so we make a temporary copy 2173 * of the leaf and add them both to that. 2174 */ 2175 struct xfs_attr_leafblock *tmp_leaf; 2176 struct xfs_attr3_icleaf_hdr tmphdr; 2177 2178 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP); 2179 2180 /* 2181 * Copy the header into the temp leaf so that all the stuff 2182 * not in the incore header is present and gets copied back in 2183 * once we've moved all the entries. 2184 */ 2185 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf)); 2186 2187 memset(&tmphdr, 0, sizeof(tmphdr)); 2188 tmphdr.magic = savehdr.magic; 2189 tmphdr.forw = savehdr.forw; 2190 tmphdr.back = savehdr.back; 2191 tmphdr.firstused = state->args->geo->blksize; 2192 2193 /* write the header to the temp buffer to initialise it */ 2194 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr); 2195 2196 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr, 2197 drop_blk->bp, &drophdr)) { 2198 xfs_attr3_leaf_moveents(state->args, 2199 drop_leaf, &drophdr, 0, 2200 tmp_leaf, &tmphdr, 0, 2201 drophdr.count); 2202 xfs_attr3_leaf_moveents(state->args, 2203 save_leaf, &savehdr, 0, 2204 tmp_leaf, &tmphdr, tmphdr.count, 2205 savehdr.count); 2206 } else { 2207 xfs_attr3_leaf_moveents(state->args, 2208 save_leaf, &savehdr, 0, 2209 tmp_leaf, &tmphdr, 0, 2210 savehdr.count); 2211 xfs_attr3_leaf_moveents(state->args, 2212 drop_leaf, &drophdr, 0, 2213 tmp_leaf, &tmphdr, tmphdr.count, 2214 drophdr.count); 2215 } 2216 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize); 2217 savehdr = tmphdr; /* struct copy */ 2218 kmem_free(tmp_leaf); 2219 } 2220 2221 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr); 2222 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0, 2223 state->args->geo->blksize - 1); 2224 2225 /* 2226 * Copy out last hashval in each block for B-tree code. 2227 */ 2228 entry = xfs_attr3_leaf_entryp(save_leaf); 2229 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval); 2230 } 2231 2232 /*======================================================================== 2233 * Routines used for finding things in the Btree. 2234 *========================================================================*/ 2235 2236 /* 2237 * Look up a name in a leaf attribute list structure. 2238 * This is the internal routine, it uses the caller's buffer. 2239 * 2240 * Note that duplicate keys are allowed, but only check within the 2241 * current leaf node. The Btree code must check in adjacent leaf nodes. 2242 * 2243 * Return in args->index the index into the entry[] array of either 2244 * the found entry, or where the entry should have been (insert before 2245 * that entry). 2246 * 2247 * Don't change the args->value unless we find the attribute. 2248 */ 2249 int 2250 xfs_attr3_leaf_lookup_int( 2251 struct xfs_buf *bp, 2252 struct xfs_da_args *args) 2253 { 2254 struct xfs_attr_leafblock *leaf; 2255 struct xfs_attr3_icleaf_hdr ichdr; 2256 struct xfs_attr_leaf_entry *entry; 2257 struct xfs_attr_leaf_entry *entries; 2258 struct xfs_attr_leaf_name_local *name_loc; 2259 struct xfs_attr_leaf_name_remote *name_rmt; 2260 xfs_dahash_t hashval; 2261 int probe; 2262 int span; 2263 2264 trace_xfs_attr_leaf_lookup(args); 2265 2266 leaf = bp->b_addr; 2267 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2268 entries = xfs_attr3_leaf_entryp(leaf); 2269 if (ichdr.count >= args->geo->blksize / 8) 2270 return -EFSCORRUPTED; 2271 2272 /* 2273 * Binary search. (note: small blocks will skip this loop) 2274 */ 2275 hashval = args->hashval; 2276 probe = span = ichdr.count / 2; 2277 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) { 2278 span /= 2; 2279 if (be32_to_cpu(entry->hashval) < hashval) 2280 probe += span; 2281 else if (be32_to_cpu(entry->hashval) > hashval) 2282 probe -= span; 2283 else 2284 break; 2285 } 2286 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) 2287 return -EFSCORRUPTED; 2288 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) 2289 return -EFSCORRUPTED; 2290 2291 /* 2292 * Since we may have duplicate hashval's, find the first matching 2293 * hashval in the leaf. 2294 */ 2295 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) { 2296 entry--; 2297 probe--; 2298 } 2299 while (probe < ichdr.count && 2300 be32_to_cpu(entry->hashval) < hashval) { 2301 entry++; 2302 probe++; 2303 } 2304 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) { 2305 args->index = probe; 2306 return -ENOATTR; 2307 } 2308 2309 /* 2310 * Duplicate keys may be present, so search all of them for a match. 2311 */ 2312 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval); 2313 entry++, probe++) { 2314 /* 2315 * GROT: Add code to remove incomplete entries. 2316 */ 2317 /* 2318 * If we are looking for INCOMPLETE entries, show only those. 2319 * If we are looking for complete entries, show only those. 2320 */ 2321 if ((args->flags & XFS_ATTR_INCOMPLETE) != 2322 (entry->flags & XFS_ATTR_INCOMPLETE)) { 2323 continue; 2324 } 2325 if (entry->flags & XFS_ATTR_LOCAL) { 2326 name_loc = xfs_attr3_leaf_name_local(leaf, probe); 2327 if (name_loc->namelen != args->namelen) 2328 continue; 2329 if (memcmp(args->name, name_loc->nameval, 2330 args->namelen) != 0) 2331 continue; 2332 if (!xfs_attr_namesp_match(args->flags, entry->flags)) 2333 continue; 2334 args->index = probe; 2335 return -EEXIST; 2336 } else { 2337 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe); 2338 if (name_rmt->namelen != args->namelen) 2339 continue; 2340 if (memcmp(args->name, name_rmt->name, 2341 args->namelen) != 0) 2342 continue; 2343 if (!xfs_attr_namesp_match(args->flags, entry->flags)) 2344 continue; 2345 args->index = probe; 2346 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen); 2347 args->rmtblkno = be32_to_cpu(name_rmt->valueblk); 2348 args->rmtblkcnt = xfs_attr3_rmt_blocks( 2349 args->dp->i_mount, 2350 args->rmtvaluelen); 2351 return -EEXIST; 2352 } 2353 } 2354 args->index = probe; 2355 return -ENOATTR; 2356 } 2357 2358 /* 2359 * Get the value associated with an attribute name from a leaf attribute 2360 * list structure. 2361 */ 2362 int 2363 xfs_attr3_leaf_getvalue( 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_name_local *name_loc; 2371 struct xfs_attr_leaf_name_remote *name_rmt; 2372 int valuelen; 2373 2374 leaf = bp->b_addr; 2375 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2376 ASSERT(ichdr.count < args->geo->blksize / 8); 2377 ASSERT(args->index < ichdr.count); 2378 2379 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2380 if (entry->flags & XFS_ATTR_LOCAL) { 2381 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 2382 ASSERT(name_loc->namelen == args->namelen); 2383 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0); 2384 valuelen = be16_to_cpu(name_loc->valuelen); 2385 if (args->flags & ATTR_KERNOVAL) { 2386 args->valuelen = valuelen; 2387 return 0; 2388 } 2389 if (args->valuelen < valuelen) { 2390 args->valuelen = valuelen; 2391 return -ERANGE; 2392 } 2393 args->valuelen = valuelen; 2394 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen); 2395 } else { 2396 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2397 ASSERT(name_rmt->namelen == args->namelen); 2398 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0); 2399 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen); 2400 args->rmtblkno = be32_to_cpu(name_rmt->valueblk); 2401 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount, 2402 args->rmtvaluelen); 2403 if (args->flags & ATTR_KERNOVAL) { 2404 args->valuelen = args->rmtvaluelen; 2405 return 0; 2406 } 2407 if (args->valuelen < args->rmtvaluelen) { 2408 args->valuelen = args->rmtvaluelen; 2409 return -ERANGE; 2410 } 2411 args->valuelen = args->rmtvaluelen; 2412 } 2413 return 0; 2414 } 2415 2416 /*======================================================================== 2417 * Utility routines. 2418 *========================================================================*/ 2419 2420 /* 2421 * Move the indicated entries from one leaf to another. 2422 * NOTE: this routine modifies both source and destination leaves. 2423 */ 2424 /*ARGSUSED*/ 2425 STATIC void 2426 xfs_attr3_leaf_moveents( 2427 struct xfs_da_args *args, 2428 struct xfs_attr_leafblock *leaf_s, 2429 struct xfs_attr3_icleaf_hdr *ichdr_s, 2430 int start_s, 2431 struct xfs_attr_leafblock *leaf_d, 2432 struct xfs_attr3_icleaf_hdr *ichdr_d, 2433 int start_d, 2434 int count) 2435 { 2436 struct xfs_attr_leaf_entry *entry_s; 2437 struct xfs_attr_leaf_entry *entry_d; 2438 int desti; 2439 int tmp; 2440 int i; 2441 2442 /* 2443 * Check for nothing to do. 2444 */ 2445 if (count == 0) 2446 return; 2447 2448 /* 2449 * Set up environment. 2450 */ 2451 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC || 2452 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC); 2453 ASSERT(ichdr_s->magic == ichdr_d->magic); 2454 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8); 2455 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s)) 2456 + xfs_attr3_leaf_hdr_size(leaf_s)); 2457 ASSERT(ichdr_d->count < args->geo->blksize / 8); 2458 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d)) 2459 + xfs_attr3_leaf_hdr_size(leaf_d)); 2460 2461 ASSERT(start_s < ichdr_s->count); 2462 ASSERT(start_d <= ichdr_d->count); 2463 ASSERT(count <= ichdr_s->count); 2464 2465 2466 /* 2467 * Move the entries in the destination leaf up to make a hole? 2468 */ 2469 if (start_d < ichdr_d->count) { 2470 tmp = ichdr_d->count - start_d; 2471 tmp *= sizeof(xfs_attr_leaf_entry_t); 2472 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d]; 2473 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count]; 2474 memmove(entry_d, entry_s, tmp); 2475 } 2476 2477 /* 2478 * Copy all entry's in the same (sorted) order, 2479 * but allocate attribute info packed and in sequence. 2480 */ 2481 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2482 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d]; 2483 desti = start_d; 2484 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) { 2485 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused); 2486 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i); 2487 #ifdef GROT 2488 /* 2489 * Code to drop INCOMPLETE entries. Difficult to use as we 2490 * may also need to change the insertion index. Code turned 2491 * off for 6.2, should be revisited later. 2492 */ 2493 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */ 2494 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp); 2495 ichdr_s->usedbytes -= tmp; 2496 ichdr_s->count -= 1; 2497 entry_d--; /* to compensate for ++ in loop hdr */ 2498 desti--; 2499 if ((start_s + i) < offset) 2500 result++; /* insertion index adjustment */ 2501 } else { 2502 #endif /* GROT */ 2503 ichdr_d->firstused -= tmp; 2504 /* both on-disk, don't endian flip twice */ 2505 entry_d->hashval = entry_s->hashval; 2506 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused); 2507 entry_d->flags = entry_s->flags; 2508 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp 2509 <= args->geo->blksize); 2510 memmove(xfs_attr3_leaf_name(leaf_d, desti), 2511 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp); 2512 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp 2513 <= args->geo->blksize); 2514 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp); 2515 ichdr_s->usedbytes -= tmp; 2516 ichdr_d->usedbytes += tmp; 2517 ichdr_s->count -= 1; 2518 ichdr_d->count += 1; 2519 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t) 2520 + xfs_attr3_leaf_hdr_size(leaf_d); 2521 ASSERT(ichdr_d->firstused >= tmp); 2522 #ifdef GROT 2523 } 2524 #endif /* GROT */ 2525 } 2526 2527 /* 2528 * Zero out the entries we just copied. 2529 */ 2530 if (start_s == ichdr_s->count) { 2531 tmp = count * sizeof(xfs_attr_leaf_entry_t); 2532 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2533 ASSERT(((char *)entry_s + tmp) <= 2534 ((char *)leaf_s + args->geo->blksize)); 2535 memset(entry_s, 0, tmp); 2536 } else { 2537 /* 2538 * Move the remaining entries down to fill the hole, 2539 * then zero the entries at the top. 2540 */ 2541 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t); 2542 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count]; 2543 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s]; 2544 memmove(entry_d, entry_s, tmp); 2545 2546 tmp = count * sizeof(xfs_attr_leaf_entry_t); 2547 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count]; 2548 ASSERT(((char *)entry_s + tmp) <= 2549 ((char *)leaf_s + args->geo->blksize)); 2550 memset(entry_s, 0, tmp); 2551 } 2552 2553 /* 2554 * Fill in the freemap information 2555 */ 2556 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d); 2557 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t); 2558 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base; 2559 ichdr_d->freemap[1].base = 0; 2560 ichdr_d->freemap[2].base = 0; 2561 ichdr_d->freemap[1].size = 0; 2562 ichdr_d->freemap[2].size = 0; 2563 ichdr_s->holes = 1; /* leaf may not be compact */ 2564 } 2565 2566 /* 2567 * Pick up the last hashvalue from a leaf block. 2568 */ 2569 xfs_dahash_t 2570 xfs_attr_leaf_lasthash( 2571 struct xfs_buf *bp, 2572 int *count) 2573 { 2574 struct xfs_attr3_icleaf_hdr ichdr; 2575 struct xfs_attr_leaf_entry *entries; 2576 struct xfs_mount *mp = bp->b_target->bt_mount; 2577 2578 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr); 2579 entries = xfs_attr3_leaf_entryp(bp->b_addr); 2580 if (count) 2581 *count = ichdr.count; 2582 if (!ichdr.count) 2583 return 0; 2584 return be32_to_cpu(entries[ichdr.count - 1].hashval); 2585 } 2586 2587 /* 2588 * Calculate the number of bytes used to store the indicated attribute 2589 * (whether local or remote only calculate bytes in this block). 2590 */ 2591 STATIC int 2592 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index) 2593 { 2594 struct xfs_attr_leaf_entry *entries; 2595 xfs_attr_leaf_name_local_t *name_loc; 2596 xfs_attr_leaf_name_remote_t *name_rmt; 2597 int size; 2598 2599 entries = xfs_attr3_leaf_entryp(leaf); 2600 if (entries[index].flags & XFS_ATTR_LOCAL) { 2601 name_loc = xfs_attr3_leaf_name_local(leaf, index); 2602 size = xfs_attr_leaf_entsize_local(name_loc->namelen, 2603 be16_to_cpu(name_loc->valuelen)); 2604 } else { 2605 name_rmt = xfs_attr3_leaf_name_remote(leaf, index); 2606 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen); 2607 } 2608 return size; 2609 } 2610 2611 /* 2612 * Calculate the number of bytes that would be required to store the new 2613 * attribute (whether local or remote only calculate bytes in this block). 2614 * This routine decides as a side effect whether the attribute will be 2615 * a "local" or a "remote" attribute. 2616 */ 2617 int 2618 xfs_attr_leaf_newentsize( 2619 struct xfs_da_args *args, 2620 int *local) 2621 { 2622 int size; 2623 2624 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen); 2625 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) { 2626 if (local) 2627 *local = 1; 2628 return size; 2629 } 2630 if (local) 2631 *local = 0; 2632 return xfs_attr_leaf_entsize_remote(args->namelen); 2633 } 2634 2635 2636 /*======================================================================== 2637 * Manage the INCOMPLETE flag in a leaf entry 2638 *========================================================================*/ 2639 2640 /* 2641 * Clear the INCOMPLETE flag on an entry in a leaf block. 2642 */ 2643 int 2644 xfs_attr3_leaf_clearflag( 2645 struct xfs_da_args *args) 2646 { 2647 struct xfs_attr_leafblock *leaf; 2648 struct xfs_attr_leaf_entry *entry; 2649 struct xfs_attr_leaf_name_remote *name_rmt; 2650 struct xfs_buf *bp; 2651 int error; 2652 #ifdef DEBUG 2653 struct xfs_attr3_icleaf_hdr ichdr; 2654 xfs_attr_leaf_name_local_t *name_loc; 2655 int namelen; 2656 char *name; 2657 #endif /* DEBUG */ 2658 2659 trace_xfs_attr_leaf_clearflag(args); 2660 /* 2661 * Set up the operation. 2662 */ 2663 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp); 2664 if (error) 2665 return error; 2666 2667 leaf = bp->b_addr; 2668 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2669 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE); 2670 2671 #ifdef DEBUG 2672 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2673 ASSERT(args->index < ichdr.count); 2674 ASSERT(args->index >= 0); 2675 2676 if (entry->flags & XFS_ATTR_LOCAL) { 2677 name_loc = xfs_attr3_leaf_name_local(leaf, args->index); 2678 namelen = name_loc->namelen; 2679 name = (char *)name_loc->nameval; 2680 } else { 2681 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2682 namelen = name_rmt->namelen; 2683 name = (char *)name_rmt->name; 2684 } 2685 ASSERT(be32_to_cpu(entry->hashval) == args->hashval); 2686 ASSERT(namelen == args->namelen); 2687 ASSERT(memcmp(name, args->name, namelen) == 0); 2688 #endif /* DEBUG */ 2689 2690 entry->flags &= ~XFS_ATTR_INCOMPLETE; 2691 xfs_trans_log_buf(args->trans, bp, 2692 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 2693 2694 if (args->rmtblkno) { 2695 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0); 2696 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2697 name_rmt->valueblk = cpu_to_be32(args->rmtblkno); 2698 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen); 2699 xfs_trans_log_buf(args->trans, bp, 2700 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt))); 2701 } 2702 2703 /* 2704 * Commit the flag value change and start the next trans in series. 2705 */ 2706 return xfs_trans_roll_inode(&args->trans, args->dp); 2707 } 2708 2709 /* 2710 * Set the INCOMPLETE flag on an entry in a leaf block. 2711 */ 2712 int 2713 xfs_attr3_leaf_setflag( 2714 struct xfs_da_args *args) 2715 { 2716 struct xfs_attr_leafblock *leaf; 2717 struct xfs_attr_leaf_entry *entry; 2718 struct xfs_attr_leaf_name_remote *name_rmt; 2719 struct xfs_buf *bp; 2720 int error; 2721 #ifdef DEBUG 2722 struct xfs_attr3_icleaf_hdr ichdr; 2723 #endif 2724 2725 trace_xfs_attr_leaf_setflag(args); 2726 2727 /* 2728 * Set up the operation. 2729 */ 2730 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp); 2731 if (error) 2732 return error; 2733 2734 leaf = bp->b_addr; 2735 #ifdef DEBUG 2736 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf); 2737 ASSERT(args->index < ichdr.count); 2738 ASSERT(args->index >= 0); 2739 #endif 2740 entry = &xfs_attr3_leaf_entryp(leaf)[args->index]; 2741 2742 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0); 2743 entry->flags |= XFS_ATTR_INCOMPLETE; 2744 xfs_trans_log_buf(args->trans, bp, 2745 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry))); 2746 if ((entry->flags & XFS_ATTR_LOCAL) == 0) { 2747 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index); 2748 name_rmt->valueblk = 0; 2749 name_rmt->valuelen = 0; 2750 xfs_trans_log_buf(args->trans, bp, 2751 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt))); 2752 } 2753 2754 /* 2755 * Commit the flag value change and start the next trans in series. 2756 */ 2757 return xfs_trans_roll_inode(&args->trans, args->dp); 2758 } 2759 2760 /* 2761 * In a single transaction, clear the INCOMPLETE flag on the leaf entry 2762 * given by args->blkno/index and set the INCOMPLETE flag on the leaf 2763 * entry given by args->blkno2/index2. 2764 * 2765 * Note that they could be in different blocks, or in the same block. 2766 */ 2767 int 2768 xfs_attr3_leaf_flipflags( 2769 struct xfs_da_args *args) 2770 { 2771 struct xfs_attr_leafblock *leaf1; 2772 struct xfs_attr_leafblock *leaf2; 2773 struct xfs_attr_leaf_entry *entry1; 2774 struct xfs_attr_leaf_entry *entry2; 2775 struct xfs_attr_leaf_name_remote *name_rmt; 2776 struct xfs_buf *bp1; 2777 struct xfs_buf *bp2; 2778 int error; 2779 #ifdef DEBUG 2780 struct xfs_attr3_icleaf_hdr ichdr1; 2781 struct xfs_attr3_icleaf_hdr ichdr2; 2782 xfs_attr_leaf_name_local_t *name_loc; 2783 int namelen1, namelen2; 2784 char *name1, *name2; 2785 #endif /* DEBUG */ 2786 2787 trace_xfs_attr_leaf_flipflags(args); 2788 2789 /* 2790 * Read the block containing the "old" attr 2791 */ 2792 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1); 2793 if (error) 2794 return error; 2795 2796 /* 2797 * Read the block containing the "new" attr, if it is different 2798 */ 2799 if (args->blkno2 != args->blkno) { 2800 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2, 2801 -1, &bp2); 2802 if (error) 2803 return error; 2804 } else { 2805 bp2 = bp1; 2806 } 2807 2808 leaf1 = bp1->b_addr; 2809 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index]; 2810 2811 leaf2 = bp2->b_addr; 2812 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2]; 2813 2814 #ifdef DEBUG 2815 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1); 2816 ASSERT(args->index < ichdr1.count); 2817 ASSERT(args->index >= 0); 2818 2819 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2); 2820 ASSERT(args->index2 < ichdr2.count); 2821 ASSERT(args->index2 >= 0); 2822 2823 if (entry1->flags & XFS_ATTR_LOCAL) { 2824 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index); 2825 namelen1 = name_loc->namelen; 2826 name1 = (char *)name_loc->nameval; 2827 } else { 2828 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index); 2829 namelen1 = name_rmt->namelen; 2830 name1 = (char *)name_rmt->name; 2831 } 2832 if (entry2->flags & XFS_ATTR_LOCAL) { 2833 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2); 2834 namelen2 = name_loc->namelen; 2835 name2 = (char *)name_loc->nameval; 2836 } else { 2837 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2); 2838 namelen2 = name_rmt->namelen; 2839 name2 = (char *)name_rmt->name; 2840 } 2841 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval)); 2842 ASSERT(namelen1 == namelen2); 2843 ASSERT(memcmp(name1, name2, namelen1) == 0); 2844 #endif /* DEBUG */ 2845 2846 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE); 2847 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0); 2848 2849 entry1->flags &= ~XFS_ATTR_INCOMPLETE; 2850 xfs_trans_log_buf(args->trans, bp1, 2851 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1))); 2852 if (args->rmtblkno) { 2853 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0); 2854 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index); 2855 name_rmt->valueblk = cpu_to_be32(args->rmtblkno); 2856 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen); 2857 xfs_trans_log_buf(args->trans, bp1, 2858 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt))); 2859 } 2860 2861 entry2->flags |= XFS_ATTR_INCOMPLETE; 2862 xfs_trans_log_buf(args->trans, bp2, 2863 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2))); 2864 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) { 2865 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2); 2866 name_rmt->valueblk = 0; 2867 name_rmt->valuelen = 0; 2868 xfs_trans_log_buf(args->trans, bp2, 2869 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt))); 2870 } 2871 2872 /* 2873 * Commit the flag value change and start the next trans in series. 2874 */ 2875 error = xfs_trans_roll_inode(&args->trans, args->dp); 2876 2877 return error; 2878 } 2879