1 /* 2 * Copyright (c) 2001 Ian Dowse. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 * 25 * $FreeBSD: src/sys/ufs/ufs/ufs_dirhash.c,v 1.3.2.6 2002/04/10 21:41:14 dwmalone Exp $ 26 */ 27 /* 28 * This implements a hash-based lookup scheme for UFS directories. 29 */ 30 31 #include "opt_ufs.h" 32 33 #ifdef UFS_DIRHASH 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/malloc.h> 39 #include <sys/fnv_hash.h> 40 #include <sys/proc.h> 41 #include <sys/buf.h> 42 #include <sys/vnode.h> 43 #include <sys/mount.h> 44 #include <sys/sysctl.h> 45 #include <vm/vm_zone.h> 46 47 #include <ufs/ufs/quota.h> 48 #include <ufs/ufs/inode.h> 49 #include <ufs/ufs/dir.h> 50 #include <ufs/ufs/dirhash.h> 51 #include <ufs/ufs/ufsmount.h> 52 #include <ufs/ufs/ufs_extern.h> 53 54 #define WRAPINCR(val, limit) (((val) + 1 == (limit)) ? 0 : ((val) + 1)) 55 #define WRAPDECR(val, limit) (((val) == 0) ? ((limit) - 1) : ((val) - 1)) 56 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 57 #define BLKFREE2IDX(n) ((n) > DH_NFSTATS ? DH_NFSTATS : (n)) 58 59 static MALLOC_DEFINE(M_DIRHASH, "UFS dirhash", "UFS directory hash tables"); 60 61 SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD, 0, "UFS filesystem"); 62 63 static int ufs_mindirhashsize = DIRBLKSIZ * 5; 64 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_minsize, CTLFLAG_RW, 65 &ufs_mindirhashsize, 66 0, "minimum directory size in bytes for which to use hashed lookup"); 67 static int ufs_dirhashmaxmem = 2 * 1024 * 1024; 68 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_maxmem, CTLFLAG_RW, &ufs_dirhashmaxmem, 69 0, "maximum allowed dirhash memory usage"); 70 static int ufs_dirhashmem; 71 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_mem, CTLFLAG_RD, &ufs_dirhashmem, 72 0, "current dirhash memory usage"); 73 static int ufs_dirhashcheck = 0; 74 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_docheck, CTLFLAG_RW, &ufs_dirhashcheck, 75 0, "enable extra sanity tests"); 76 77 78 static int ufsdirhash_hash(struct dirhash *dh, char *name, int namelen); 79 static void ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff); 80 static void ufsdirhash_delslot(struct dirhash *dh, int slot); 81 static int ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, 82 doff_t offset); 83 static doff_t ufsdirhash_getprev(struct direct *dp, doff_t offset); 84 static void ufsdirhash_init(void); 85 static int ufsdirhash_recycle(int wanted); 86 87 static vm_zone_t ufsdirhash_zone; 88 89 /* Dirhash list; recently-used entries are near the tail. */ 90 static TAILQ_HEAD(, dirhash) ufsdirhash_list; 91 92 /* 93 * Attempt to build up a hash table for the directory contents in 94 * inode 'ip'. Returns 0 on success, or -1 of the operation failed. 95 */ 96 int 97 ufsdirhash_build(struct inode *ip) 98 { 99 struct dirhash *dh; 100 struct buf *bp = NULL; 101 struct direct *ep; 102 struct vnode *vp; 103 doff_t bmask, pos; 104 int dirblocks, i, j, memreqd, nblocks, narrays, nslots, slot; 105 106 /* Check if we can/should use dirhash. */ 107 if (ip->i_dirhash == NULL) { 108 if (ip->i_size < ufs_mindirhashsize || OFSFMT(ip->i_vnode)) 109 return (-1); 110 } else { 111 /* Hash exists, but sysctls could have changed. */ 112 if (ip->i_size < ufs_mindirhashsize || 113 ufs_dirhashmem > ufs_dirhashmaxmem) { 114 ufsdirhash_free(ip); 115 return (-1); 116 } 117 /* Check if hash exists and is intact (note: unlocked read). */ 118 if (ip->i_dirhash->dh_hash != NULL) 119 return (0); 120 /* Free the old, recycled hash and build a new one. */ 121 ufsdirhash_free(ip); 122 } 123 124 /* Don't hash removed directories. */ 125 if (ip->i_effnlink == 0) 126 return (-1); 127 128 vp = ip->i_vnode; 129 /* Allocate 50% more entries than this dir size could ever need. */ 130 KASSERT(ip->i_size >= DIRBLKSIZ, ("ufsdirhash_build size")); 131 nslots = ip->i_size / DIRECTSIZ(1); 132 nslots = (nslots * 3 + 1) / 2; 133 narrays = howmany(nslots, DH_NBLKOFF); 134 nslots = narrays * DH_NBLKOFF; 135 dirblocks = howmany(ip->i_size, DIRBLKSIZ); 136 nblocks = (dirblocks * 3 + 1) / 2; 137 138 memreqd = sizeof(*dh) + narrays * sizeof(*dh->dh_hash) + 139 narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 140 nblocks * sizeof(*dh->dh_blkfree); 141 if (memreqd + ufs_dirhashmem > ufs_dirhashmaxmem) { 142 if (memreqd > ufs_dirhashmaxmem / 2) 143 return (-1); 144 145 /* Try to free some space. */ 146 if (ufsdirhash_recycle(memreqd) != 0) 147 return (-1); 148 /* Enough was freed. */ 149 } 150 ufs_dirhashmem += memreqd; 151 152 /* 153 * Use non-blocking mallocs so that we will revert to a linear 154 * lookup on failure rather than potentially blocking forever. 155 */ 156 MALLOC(dh, struct dirhash *, sizeof *dh, M_DIRHASH, M_NOWAIT | M_ZERO); 157 if (dh == NULL) 158 return (-1); 159 MALLOC(dh->dh_hash, doff_t **, narrays * sizeof(dh->dh_hash[0]), 160 M_DIRHASH, M_NOWAIT | M_ZERO); 161 MALLOC(dh->dh_blkfree, u_int8_t *, nblocks * sizeof(dh->dh_blkfree[0]), 162 M_DIRHASH, M_NOWAIT); 163 if (dh->dh_hash == NULL || dh->dh_blkfree == NULL) 164 goto fail; 165 for (i = 0; i < narrays; i++) { 166 if ((dh->dh_hash[i] = zalloc(ufsdirhash_zone)) == NULL) 167 goto fail; 168 for (j = 0; j < DH_NBLKOFF; j++) 169 dh->dh_hash[i][j] = DIRHASH_EMPTY; 170 } 171 172 /* Initialise the hash table and block statistics. */ 173 dh->dh_narrays = narrays; 174 dh->dh_hlen = nslots; 175 dh->dh_nblk = nblocks; 176 dh->dh_dirblks = dirblocks; 177 for (i = 0; i < dirblocks; i++) 178 dh->dh_blkfree[i] = DIRBLKSIZ / DIRALIGN; 179 for (i = 0; i < DH_NFSTATS; i++) 180 dh->dh_firstfree[i] = -1; 181 dh->dh_firstfree[DH_NFSTATS] = 0; 182 dh->dh_seqopt = 0; 183 dh->dh_seqoff = 0; 184 dh->dh_score = DH_SCOREINIT; 185 ip->i_dirhash = dh; 186 187 bmask = VFSTOUFS(vp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 188 pos = 0; 189 while (pos < ip->i_size) { 190 /* If necessary, get the next directory block. */ 191 if ((pos & bmask) == 0) { 192 if (bp != NULL) 193 brelse(bp); 194 if (UFS_BLKATOFF(vp, (off_t)pos, NULL, &bp) != 0) 195 goto fail; 196 } 197 198 /* Add this entry to the hash. */ 199 ep = (struct direct *)((char *)bp->b_data + (pos & bmask)); 200 if (ep->d_reclen == 0 || ep->d_reclen > 201 DIRBLKSIZ - (pos & (DIRBLKSIZ - 1))) { 202 /* Corrupted directory. */ 203 brelse(bp); 204 goto fail; 205 } 206 if (ep->d_ino != 0) { 207 /* Add the entry (simplified ufsdirhash_add). */ 208 slot = ufsdirhash_hash(dh, ep->d_name, ep->d_namlen); 209 while (DH_ENTRY(dh, slot) != DIRHASH_EMPTY) 210 slot = WRAPINCR(slot, dh->dh_hlen); 211 dh->dh_hused++; 212 DH_ENTRY(dh, slot) = pos; 213 ufsdirhash_adjfree(dh, pos, -DIRSIZ(0, ep)); 214 } 215 pos += ep->d_reclen; 216 } 217 218 if (bp != NULL) 219 brelse(bp); 220 TAILQ_INSERT_TAIL(&ufsdirhash_list, dh, dh_list); 221 dh->dh_onlist = 1; 222 return (0); 223 224 fail: 225 if (dh->dh_hash != NULL) { 226 for (i = 0; i < narrays; i++) 227 if (dh->dh_hash[i] != NULL) 228 zfree(ufsdirhash_zone, dh->dh_hash[i]); 229 FREE(dh->dh_hash, M_DIRHASH); 230 } 231 if (dh->dh_blkfree != NULL) 232 FREE(dh->dh_blkfree, M_DIRHASH); 233 FREE(dh, M_DIRHASH); 234 ip->i_dirhash = NULL; 235 ufs_dirhashmem -= memreqd; 236 return (-1); 237 } 238 239 /* 240 * Free any hash table associated with inode 'ip'. 241 */ 242 void 243 ufsdirhash_free(struct inode *ip) 244 { 245 struct dirhash *dh; 246 int i, mem; 247 248 if ((dh = ip->i_dirhash) == NULL) 249 return; 250 if (dh->dh_onlist) 251 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 252 253 /* The dirhash pointed to by 'dh' is exclusively ours now. */ 254 255 mem = sizeof(*dh); 256 if (dh->dh_hash != NULL) { 257 for (i = 0; i < dh->dh_narrays; i++) 258 zfree(ufsdirhash_zone, dh->dh_hash[i]); 259 FREE(dh->dh_hash, M_DIRHASH); 260 FREE(dh->dh_blkfree, M_DIRHASH); 261 mem += dh->dh_narrays * sizeof(*dh->dh_hash) + 262 dh->dh_narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 263 dh->dh_nblk * sizeof(*dh->dh_blkfree); 264 } 265 FREE(dh, M_DIRHASH); 266 ip->i_dirhash = NULL; 267 268 ufs_dirhashmem -= mem; 269 } 270 271 /* 272 * Find the offset of the specified name within the given inode. 273 * Returns 0 on success, ENOENT if the entry does not exist, or 274 * EJUSTRETURN if the caller should revert to a linear search. 275 * 276 * If successful, the directory offset is stored in *offp, and a 277 * pointer to a struct buf containing the entry is stored in *bpp. If 278 * prevoffp is non-NULL, the offset of the previous entry within 279 * the DIRBLKSIZ-sized block is stored in *prevoffp (if the entry 280 * is the first in a block, the start of the block is used). 281 */ 282 int 283 ufsdirhash_lookup(struct inode *ip, char *name, int namelen, doff_t *offp, 284 struct buf **bpp, doff_t *prevoffp) 285 { 286 struct dirhash *dh, *dh_next; 287 struct direct *dp; 288 struct vnode *vp; 289 struct buf *bp; 290 doff_t blkoff, bmask, offset, prevoff; 291 int i, slot; 292 293 if ((dh = ip->i_dirhash) == NULL) 294 return (EJUSTRETURN); 295 /* 296 * Move this dirhash towards the end of the list if it has a 297 * score higher than the next entry. 298 * Optimise the case where it's already the last by performing 299 * an unlocked read of the TAILQ_NEXT pointer. 300 */ 301 if (TAILQ_NEXT(dh, dh_list) != NULL) { 302 /* 303 * If the new score will be greater than that of the next 304 * entry, then move this entry past it. With both mutexes 305 * held, dh_next won't go away, but its dh_score could 306 * change; that's not important since it is just a hint. 307 */ 308 if (dh->dh_hash != NULL && 309 (dh_next = TAILQ_NEXT(dh, dh_list)) != NULL && 310 dh->dh_score >= dh_next->dh_score) { 311 KASSERT(dh->dh_onlist, ("dirhash: not on list")); 312 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 313 TAILQ_INSERT_AFTER(&ufsdirhash_list, dh_next, dh, 314 dh_list); 315 } 316 } 317 if (dh->dh_hash == NULL) { 318 ufsdirhash_free(ip); 319 return (EJUSTRETURN); 320 } 321 322 /* Update the score. */ 323 if (dh->dh_score < DH_SCOREMAX) 324 dh->dh_score++; 325 326 vp = ip->i_vnode; 327 bmask = VFSTOUFS(vp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 328 blkoff = -1; 329 bp = NULL; 330 restart: 331 slot = ufsdirhash_hash(dh, name, namelen); 332 333 if (dh->dh_seqopt) { 334 /* 335 * Sequential access optimisation. dh_seqoff contains the 336 * offset of the directory entry immediately following 337 * the last entry that was looked up. Check if this offset 338 * appears in the hash chain for the name we are looking for. 339 */ 340 for (i = slot; (offset = DH_ENTRY(dh, i)) != DIRHASH_EMPTY; 341 i = WRAPINCR(i, dh->dh_hlen)) 342 if (offset == dh->dh_seqoff) 343 break; 344 if (offset == dh->dh_seqoff) { 345 /* 346 * We found an entry with the expected offset. This 347 * is probably the entry we want, but if not, the 348 * code below will turn off seqoff and retry. 349 */ 350 slot = i; 351 } else 352 dh->dh_seqopt = 0; 353 } 354 355 for (; (offset = DH_ENTRY(dh, slot)) != DIRHASH_EMPTY; 356 slot = WRAPINCR(slot, dh->dh_hlen)) { 357 if (offset == DIRHASH_DEL) 358 continue; 359 360 if (offset < 0 || offset >= ip->i_size) 361 panic("ufsdirhash_lookup: bad offset in hash array"); 362 if ((offset & ~bmask) != blkoff) { 363 if (bp != NULL) 364 brelse(bp); 365 blkoff = offset & ~bmask; 366 if (UFS_BLKATOFF(vp, (off_t)blkoff, NULL, &bp) != 0) 367 return (EJUSTRETURN); 368 } 369 dp = (struct direct *)(bp->b_data + (offset & bmask)); 370 if (dp->d_reclen == 0 || dp->d_reclen > 371 DIRBLKSIZ - (offset & (DIRBLKSIZ - 1))) { 372 /* Corrupted directory. */ 373 brelse(bp); 374 return (EJUSTRETURN); 375 } 376 if (dp->d_namlen == namelen && 377 bcmp(dp->d_name, name, namelen) == 0) { 378 /* Found. Get the prev offset if needed. */ 379 if (prevoffp != NULL) { 380 if (offset & (DIRBLKSIZ - 1)) { 381 prevoff = ufsdirhash_getprev(dp, 382 offset); 383 if (prevoff == -1) { 384 brelse(bp); 385 return (EJUSTRETURN); 386 } 387 } else 388 prevoff = offset; 389 *prevoffp = prevoff; 390 } 391 392 /* Check for sequential access, and update offset. */ 393 if (dh->dh_seqopt == 0 && dh->dh_seqoff == offset) 394 dh->dh_seqopt = 1; 395 dh->dh_seqoff = offset + DIRSIZ(0, dp); 396 397 *bpp = bp; 398 *offp = offset; 399 return (0); 400 } 401 402 if (dh->dh_hash == NULL) { 403 if (bp != NULL) 404 brelse(bp); 405 ufsdirhash_free(ip); 406 return (EJUSTRETURN); 407 } 408 /* 409 * When the name doesn't match in the seqopt case, go back 410 * and search normally. 411 */ 412 if (dh->dh_seqopt) { 413 dh->dh_seqopt = 0; 414 goto restart; 415 } 416 } 417 if (bp != NULL) 418 brelse(bp); 419 return (ENOENT); 420 } 421 422 /* 423 * Find a directory block with room for 'slotneeded' bytes. Returns 424 * the offset of the directory entry that begins the free space. 425 * This will either be the offset of an existing entry that has free 426 * space at the end, or the offset of an entry with d_ino == 0 at 427 * the start of a DIRBLKSIZ block. 428 * 429 * To use the space, the caller may need to compact existing entries in 430 * the directory. The total number of bytes in all of the entries involved 431 * in the compaction is stored in *slotsize. In other words, all of 432 * the entries that must be compacted are exactly contained in the 433 * region beginning at the returned offset and spanning *slotsize bytes. 434 * 435 * Returns -1 if no space was found, indicating that the directory 436 * must be extended. 437 */ 438 doff_t 439 ufsdirhash_findfree(struct inode *ip, int slotneeded, int *slotsize) 440 { 441 struct direct *dp; 442 struct dirhash *dh; 443 struct buf *bp; 444 doff_t pos, slotstart; 445 int dirblock, error, freebytes, i; 446 447 if ((dh = ip->i_dirhash) == NULL) 448 return (-1); 449 if (dh->dh_hash == NULL) { 450 ufsdirhash_free(ip); 451 return (-1); 452 } 453 454 /* Find a directory block with the desired free space. */ 455 dirblock = -1; 456 for (i = howmany(slotneeded, DIRALIGN); i <= DH_NFSTATS; i++) 457 if ((dirblock = dh->dh_firstfree[i]) != -1) 458 break; 459 if (dirblock == -1) { 460 return (-1); 461 } 462 463 KASSERT(dirblock < dh->dh_nblk && 464 dh->dh_blkfree[dirblock] >= howmany(slotneeded, DIRALIGN), 465 ("ufsdirhash_findfree: bad stats")); 466 pos = dirblock * DIRBLKSIZ; 467 error = UFS_BLKATOFF(ip->i_vnode, (off_t)pos, (char **)&dp, &bp); 468 if (error) 469 return (-1); 470 471 /* Find the first entry with free space. */ 472 for (i = 0; i < DIRBLKSIZ; ) { 473 if (dp->d_reclen == 0) { 474 brelse(bp); 475 return (-1); 476 } 477 if (dp->d_ino == 0 || dp->d_reclen > DIRSIZ(0, dp)) 478 break; 479 i += dp->d_reclen; 480 dp = (struct direct *)((char *)dp + dp->d_reclen); 481 } 482 if (i > DIRBLKSIZ) { 483 brelse(bp); 484 return (-1); 485 } 486 slotstart = pos + i; 487 488 /* Find the range of entries needed to get enough space */ 489 freebytes = 0; 490 while (i < DIRBLKSIZ && freebytes < slotneeded) { 491 freebytes += dp->d_reclen; 492 if (dp->d_ino != 0) 493 freebytes -= DIRSIZ(0, dp); 494 if (dp->d_reclen == 0) { 495 brelse(bp); 496 return (-1); 497 } 498 i += dp->d_reclen; 499 dp = (struct direct *)((char *)dp + dp->d_reclen); 500 } 501 if (i > DIRBLKSIZ) { 502 brelse(bp); 503 return (-1); 504 } 505 if (freebytes < slotneeded) 506 panic("ufsdirhash_findfree: free mismatch"); 507 brelse(bp); 508 *slotsize = pos + i - slotstart; 509 return (slotstart); 510 } 511 512 /* 513 * Return the start of the unused space at the end of a directory, or 514 * -1 if there are no trailing unused blocks. 515 */ 516 doff_t 517 ufsdirhash_enduseful(struct inode *ip) 518 { 519 520 struct dirhash *dh; 521 int i; 522 523 if ((dh = ip->i_dirhash) == NULL) 524 return (-1); 525 if (dh->dh_hash == NULL) { 526 ufsdirhash_free(ip); 527 return (-1); 528 } 529 530 if (dh->dh_blkfree[dh->dh_dirblks - 1] != DIRBLKSIZ / DIRALIGN) { 531 return (-1); 532 } 533 534 for (i = dh->dh_dirblks - 1; i >= 0; i--) 535 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN) 536 break; 537 return ((doff_t)(i + 1) * DIRBLKSIZ); 538 } 539 540 /* 541 * Insert information into the hash about a new directory entry. dirp 542 * points to a struct direct containing the entry, and offset specifies 543 * the offset of this entry. 544 */ 545 void 546 ufsdirhash_add(struct inode *ip, struct direct *dirp, doff_t offset) 547 { 548 struct dirhash *dh; 549 int slot; 550 551 if ((dh = ip->i_dirhash) == NULL) 552 return; 553 if (dh->dh_hash == NULL) { 554 ufsdirhash_free(ip); 555 return; 556 } 557 558 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ, 559 ("ufsdirhash_add: bad offset")); 560 /* 561 * Normal hash usage is < 66%. If the usage gets too high then 562 * remove the hash entirely and let it be rebuilt later. 563 */ 564 if (dh->dh_hused >= (dh->dh_hlen * 3) / 4) { 565 ufsdirhash_free(ip); 566 return; 567 } 568 569 /* Find a free hash slot (empty or deleted), and add the entry. */ 570 slot = ufsdirhash_hash(dh, dirp->d_name, dirp->d_namlen); 571 while (DH_ENTRY(dh, slot) >= 0) 572 slot = WRAPINCR(slot, dh->dh_hlen); 573 if (DH_ENTRY(dh, slot) == DIRHASH_EMPTY) 574 dh->dh_hused++; 575 DH_ENTRY(dh, slot) = offset; 576 577 /* Update the per-block summary info. */ 578 ufsdirhash_adjfree(dh, offset, -DIRSIZ(0, dirp)); 579 } 580 581 /* 582 * Remove the specified directory entry from the hash. The entry to remove 583 * is defined by the name in `dirp', which must exist at the specified 584 * `offset' within the directory. 585 */ 586 void 587 ufsdirhash_remove(struct inode *ip, struct direct *dirp, doff_t offset) 588 { 589 struct dirhash *dh; 590 int slot; 591 592 if ((dh = ip->i_dirhash) == NULL) 593 return; 594 if (dh->dh_hash == NULL) { 595 ufsdirhash_free(ip); 596 return; 597 } 598 599 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ, 600 ("ufsdirhash_remove: bad offset")); 601 /* Find the entry */ 602 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, offset); 603 604 /* Remove the hash entry. */ 605 ufsdirhash_delslot(dh, slot); 606 607 /* Update the per-block summary info. */ 608 ufsdirhash_adjfree(dh, offset, DIRSIZ(0, dirp)); 609 } 610 611 /* 612 * Change the offset associated with a directory entry in the hash. Used 613 * when compacting directory blocks. 614 */ 615 void 616 ufsdirhash_move(struct inode *ip, struct direct *dirp, doff_t oldoff, 617 doff_t newoff) 618 { 619 struct dirhash *dh; 620 int slot; 621 622 if ((dh = ip->i_dirhash) == NULL) 623 return; 624 if (dh->dh_hash == NULL) { 625 ufsdirhash_free(ip); 626 return; 627 } 628 629 KASSERT(oldoff < dh->dh_dirblks * DIRBLKSIZ && 630 newoff < dh->dh_dirblks * DIRBLKSIZ, 631 ("ufsdirhash_move: bad offset")); 632 /* Find the entry, and update the offset. */ 633 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, oldoff); 634 DH_ENTRY(dh, slot) = newoff; 635 } 636 637 /* 638 * Inform dirhash that the directory has grown by one block that 639 * begins at offset (i.e. the new length is offset + DIRBLKSIZ). 640 */ 641 void 642 ufsdirhash_newblk(struct inode *ip, doff_t offset) 643 { 644 struct dirhash *dh; 645 int block; 646 647 if ((dh = ip->i_dirhash) == NULL) 648 return; 649 if (dh->dh_hash == NULL) { 650 ufsdirhash_free(ip); 651 return; 652 } 653 654 KASSERT(offset == dh->dh_dirblks * DIRBLKSIZ, 655 ("ufsdirhash_newblk: bad offset")); 656 block = offset / DIRBLKSIZ; 657 if (block >= dh->dh_nblk) { 658 /* Out of space; must rebuild. */ 659 ufsdirhash_free(ip); 660 return; 661 } 662 dh->dh_dirblks = block + 1; 663 664 /* Account for the new free block. */ 665 dh->dh_blkfree[block] = DIRBLKSIZ / DIRALIGN; 666 if (dh->dh_firstfree[DH_NFSTATS] == -1) 667 dh->dh_firstfree[DH_NFSTATS] = block; 668 } 669 670 /* 671 * Inform dirhash that the directory is being truncated. 672 */ 673 void 674 ufsdirhash_dirtrunc(struct inode *ip, doff_t offset) 675 { 676 struct dirhash *dh; 677 int block, i; 678 679 if ((dh = ip->i_dirhash) == NULL) 680 return; 681 if (dh->dh_hash == NULL) { 682 ufsdirhash_free(ip); 683 return; 684 } 685 686 KASSERT(offset <= dh->dh_dirblks * DIRBLKSIZ, 687 ("ufsdirhash_dirtrunc: bad offset")); 688 block = howmany(offset, DIRBLKSIZ); 689 /* 690 * If the directory shrinks to less than 1/8 of dh_nblk blocks 691 * (about 20% of its original size due to the 50% extra added in 692 * ufsdirhash_build) then free it, and let the caller rebuild 693 * if necessary. 694 */ 695 if (block < dh->dh_nblk / 8 && dh->dh_narrays > 1) { 696 ufsdirhash_free(ip); 697 return; 698 } 699 700 /* 701 * Remove any `first free' information pertaining to the 702 * truncated blocks. All blocks we're removing should be 703 * completely unused. 704 */ 705 if (dh->dh_firstfree[DH_NFSTATS] >= block) 706 dh->dh_firstfree[DH_NFSTATS] = -1; 707 for (i = block; i < dh->dh_dirblks; i++) 708 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN) 709 panic("ufsdirhash_dirtrunc: blocks in use"); 710 for (i = 0; i < DH_NFSTATS; i++) 711 if (dh->dh_firstfree[i] >= block) 712 panic("ufsdirhash_dirtrunc: first free corrupt"); 713 dh->dh_dirblks = block; 714 } 715 716 /* 717 * Debugging function to check that the dirhash information about 718 * a directory block matches its actual contents. Panics if a mismatch 719 * is detected. 720 * 721 * On entry, `buf' should point to the start of an in-core 722 * DIRBLKSIZ-sized directory block, and `offset' should contain the 723 * offset from the start of the directory of that block. 724 */ 725 void 726 ufsdirhash_checkblock(struct inode *ip, char *buf, doff_t offset) 727 { 728 struct dirhash *dh; 729 struct direct *dp; 730 int block, ffslot, i, nfree; 731 732 if (!ufs_dirhashcheck) 733 return; 734 if ((dh = ip->i_dirhash) == NULL) 735 return; 736 if (dh->dh_hash == NULL) { 737 ufsdirhash_free(ip); 738 return; 739 } 740 741 block = offset / DIRBLKSIZ; 742 if ((offset & (DIRBLKSIZ - 1)) != 0 || block >= dh->dh_dirblks) 743 panic("ufsdirhash_checkblock: bad offset"); 744 745 nfree = 0; 746 for (i = 0; i < DIRBLKSIZ; i += dp->d_reclen) { 747 dp = (struct direct *)(buf + i); 748 if (dp->d_reclen == 0 || i + dp->d_reclen > DIRBLKSIZ) 749 panic("ufsdirhash_checkblock: bad dir"); 750 751 if (dp->d_ino == 0) { 752 #if 0 753 /* 754 * XXX entries with d_ino == 0 should only occur 755 * at the start of a DIRBLKSIZ block. However the 756 * ufs code is tolerant of such entries at other 757 * offsets, and fsck does not fix them. 758 */ 759 if (i != 0) 760 panic("ufsdirhash_checkblock: bad dir inode"); 761 #endif 762 nfree += dp->d_reclen; 763 continue; 764 } 765 766 /* Check that the entry exists (will panic if it doesn't). */ 767 ufsdirhash_findslot(dh, dp->d_name, dp->d_namlen, offset + i); 768 769 nfree += dp->d_reclen - DIRSIZ(0, dp); 770 } 771 if (i != DIRBLKSIZ) 772 panic("ufsdirhash_checkblock: bad dir end"); 773 774 if (dh->dh_blkfree[block] * DIRALIGN != nfree) 775 panic("ufsdirhash_checkblock: bad free count"); 776 777 ffslot = BLKFREE2IDX(nfree / DIRALIGN); 778 for (i = 0; i <= DH_NFSTATS; i++) 779 if (dh->dh_firstfree[i] == block && i != ffslot) 780 panic("ufsdirhash_checkblock: bad first-free"); 781 if (dh->dh_firstfree[ffslot] == -1) 782 panic("ufsdirhash_checkblock: missing first-free entry"); 783 } 784 785 /* 786 * Hash the specified filename into a dirhash slot. 787 */ 788 static int 789 ufsdirhash_hash(struct dirhash *dh, char *name, int namelen) 790 { 791 u_int32_t hash; 792 793 /* 794 * We hash the name and then some ofther bit of data which is 795 * invarient over the dirhash's lifetime. Otherwise names 796 * differing only in the last byte are placed close to one 797 * another in the table, which is bad for linear probing. 798 */ 799 hash = fnv_32_buf(name, namelen, FNV1_32_INIT); 800 hash = fnv_32_buf(dh, sizeof(dh), hash); 801 return (hash % dh->dh_hlen); 802 } 803 804 /* 805 * Adjust the number of free bytes in the block containing `offset' 806 * by the value specified by `diff'. 807 * 808 * The caller must ensure we have exclusive access to `dh'. 809 */ 810 static void 811 ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff) 812 { 813 int block, i, nfidx, ofidx; 814 815 /* Update the per-block summary info. */ 816 block = offset / DIRBLKSIZ; 817 KASSERT(block < dh->dh_nblk && block < dh->dh_dirblks, 818 ("dirhash bad offset")); 819 ofidx = BLKFREE2IDX(dh->dh_blkfree[block]); 820 dh->dh_blkfree[block] = (int)dh->dh_blkfree[block] + (diff / DIRALIGN); 821 nfidx = BLKFREE2IDX(dh->dh_blkfree[block]); 822 823 /* Update the `first free' list if necessary. */ 824 if (ofidx != nfidx) { 825 /* If removing, scan forward for the next block. */ 826 if (dh->dh_firstfree[ofidx] == block) { 827 for (i = block + 1; i < dh->dh_dirblks; i++) 828 if (BLKFREE2IDX(dh->dh_blkfree[i]) == ofidx) 829 break; 830 dh->dh_firstfree[ofidx] = (i < dh->dh_dirblks) ? i : -1; 831 } 832 833 /* Make this the new `first free' if necessary */ 834 if (dh->dh_firstfree[nfidx] > block || 835 dh->dh_firstfree[nfidx] == -1) 836 dh->dh_firstfree[nfidx] = block; 837 } 838 } 839 840 /* 841 * Find the specified name which should have the specified offset. 842 * Returns a slot number, and panics on failure. 843 * 844 * `dh' must be locked on entry and remains so on return. 845 */ 846 static int 847 ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, doff_t offset) 848 { 849 int slot; 850 851 /* Find the entry. */ 852 KASSERT(dh->dh_hused < dh->dh_hlen, ("dirhash find full")); 853 slot = ufsdirhash_hash(dh, name, namelen); 854 while (DH_ENTRY(dh, slot) != offset && 855 DH_ENTRY(dh, slot) != DIRHASH_EMPTY) 856 slot = WRAPINCR(slot, dh->dh_hlen); 857 if (DH_ENTRY(dh, slot) != offset) 858 panic("ufsdirhash_findslot: '%.*s' not found", namelen, name); 859 860 return (slot); 861 } 862 863 /* 864 * Remove the entry corresponding to the specified slot from the hash array. 865 * 866 * `dh' must be locked on entry and remains so on return. 867 */ 868 static void 869 ufsdirhash_delslot(struct dirhash *dh, int slot) 870 { 871 int i; 872 873 /* Mark the entry as deleted. */ 874 DH_ENTRY(dh, slot) = DIRHASH_DEL; 875 876 /* If this is the end of a chain of DIRHASH_DEL slots, remove them. */ 877 for (i = slot; DH_ENTRY(dh, i) == DIRHASH_DEL; ) 878 i = WRAPINCR(i, dh->dh_hlen); 879 if (DH_ENTRY(dh, i) == DIRHASH_EMPTY) { 880 i = WRAPDECR(i, dh->dh_hlen); 881 while (DH_ENTRY(dh, i) == DIRHASH_DEL) { 882 DH_ENTRY(dh, i) = DIRHASH_EMPTY; 883 dh->dh_hused--; 884 i = WRAPDECR(i, dh->dh_hlen); 885 } 886 KASSERT(dh->dh_hused >= 0, ("ufsdirhash_delslot neg hlen")); 887 } 888 } 889 890 /* 891 * Given a directory entry and its offset, find the offset of the 892 * previous entry in the same DIRBLKSIZ-sized block. Returns an 893 * offset, or -1 if there is no previous entry in the block or some 894 * other problem occurred. 895 */ 896 static doff_t 897 ufsdirhash_getprev(struct direct *dirp, doff_t offset) 898 { 899 struct direct *dp; 900 char *blkbuf; 901 doff_t blkoff, prevoff; 902 int entrypos, i; 903 904 blkoff = offset & ~(DIRBLKSIZ - 1); /* offset of start of block */ 905 entrypos = offset & (DIRBLKSIZ - 1); /* entry relative to block */ 906 blkbuf = (char *)dirp - entrypos; 907 prevoff = blkoff; 908 909 /* If `offset' is the start of a block, there is no previous entry. */ 910 if (entrypos == 0) 911 return (-1); 912 913 /* Scan from the start of the block until we get to the entry. */ 914 for (i = 0; i < entrypos; i += dp->d_reclen) { 915 dp = (struct direct *)(blkbuf + i); 916 if (dp->d_reclen == 0 || i + dp->d_reclen > entrypos) 917 return (-1); /* Corrupted directory. */ 918 prevoff = blkoff + i; 919 } 920 return (prevoff); 921 } 922 923 /* 924 * Try to free up `wanted' bytes by stealing memory from existing 925 * dirhashes. Returns zero if successful. 926 */ 927 static int 928 ufsdirhash_recycle(int wanted) 929 { 930 struct dirhash *dh; 931 doff_t **hash; 932 u_int8_t *blkfree; 933 int i, mem, narrays; 934 935 while (wanted + ufs_dirhashmem > ufs_dirhashmaxmem) { 936 /* Find a dirhash, and lock it. */ 937 if ((dh = TAILQ_FIRST(&ufsdirhash_list)) == NULL) { 938 return (-1); 939 } 940 KASSERT(dh->dh_hash != NULL, ("dirhash: NULL hash on list")); 941 942 /* Decrement the score; only recycle if it becomes zero. */ 943 if (--dh->dh_score > 0) { 944 return (-1); 945 } 946 947 /* Remove it from the list and detach its memory. */ 948 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 949 dh->dh_onlist = 0; 950 hash = dh->dh_hash; 951 dh->dh_hash = NULL; 952 blkfree = dh->dh_blkfree; 953 dh->dh_blkfree = NULL; 954 narrays = dh->dh_narrays; 955 mem = narrays * sizeof(*dh->dh_hash) + 956 narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 957 dh->dh_nblk * sizeof(*dh->dh_blkfree); 958 959 /* Free the detached memory. */ 960 for (i = 0; i < narrays; i++) 961 zfree(ufsdirhash_zone, hash[i]); 962 FREE(hash, M_DIRHASH); 963 FREE(blkfree, M_DIRHASH); 964 965 /* Account for the returned memory, and repeat if necessary. */ 966 ufs_dirhashmem -= mem; 967 } 968 /* Success. */ 969 return (0); 970 } 971 972 973 static void 974 ufsdirhash_init() 975 { 976 ufsdirhash_zone = zinit("DIRHASH", DH_NBLKOFF * sizeof(daddr_t), 0, 977 0, 1); 978 TAILQ_INIT(&ufsdirhash_list); 979 } 980 SYSINIT(ufsdirhash, SI_SUB_PSEUDO, SI_ORDER_ANY, ufsdirhash_init, NULL) 981 982 983 #endif /* UFS_DIRHASH */ 984