ufs_dirhash.c revision 107868
1/* 2 * Copyright (c) 2001, 2002 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: head/sys/ufs/ufs/ufs_dirhash.c 107868 2002-12-14 17:16:16Z iedowse $ 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/lock.h> 39#include <sys/mutex.h> 40#include <sys/malloc.h> 41#include <sys/fnv_hash.h> 42#include <sys/proc.h> 43#include <sys/bio.h> 44#include <sys/buf.h> 45#include <sys/vnode.h> 46#include <sys/mount.h> 47#include <sys/sysctl.h> 48#include <vm/uma.h> 49 50#include <ufs/ufs/quota.h> 51#include <ufs/ufs/inode.h> 52#include <ufs/ufs/dir.h> 53#include <ufs/ufs/dirhash.h> 54#include <ufs/ufs/extattr.h> 55#include <ufs/ufs/ufsmount.h> 56#include <ufs/ufs/ufs_extern.h> 57 58#define WRAPINCR(val, limit) (((val) + 1 == (limit)) ? 0 : ((val) + 1)) 59#define WRAPDECR(val, limit) (((val) == 0) ? ((limit) - 1) : ((val) - 1)) 60#define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0) 61#define BLKFREE2IDX(n) ((n) > DH_NFSTATS ? DH_NFSTATS : (n)) 62 63static MALLOC_DEFINE(M_DIRHASH, "UFS dirhash", "UFS directory hash tables"); 64 65SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD, 0, "UFS filesystem"); 66 67static int ufs_mindirhashsize = DIRBLKSIZ * 5; 68SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_minsize, CTLFLAG_RW, 69 &ufs_mindirhashsize, 70 0, "minimum directory size in bytes for which to use hashed lookup"); 71static int ufs_dirhashmaxmem = 2 * 1024 * 1024; 72SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_maxmem, CTLFLAG_RW, &ufs_dirhashmaxmem, 73 0, "maximum allowed dirhash memory usage"); 74static int ufs_dirhashmem; 75SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_mem, CTLFLAG_RD, &ufs_dirhashmem, 76 0, "current dirhash memory usage"); 77static int ufs_dirhashcheck = 0; 78SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_docheck, CTLFLAG_RW, &ufs_dirhashcheck, 79 0, "enable extra sanity tests"); 80 81 82static int ufsdirhash_hash(struct dirhash *dh, char *name, int namelen); 83static void ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff); 84static void ufsdirhash_delslot(struct dirhash *dh, int slot); 85static int ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, 86 doff_t offset); 87static doff_t ufsdirhash_getprev(struct direct *dp, doff_t offset); 88static int ufsdirhash_recycle(int wanted); 89 90static uma_zone_t ufsdirhash_zone; 91 92/* Dirhash list; recently-used entries are near the tail. */ 93static TAILQ_HEAD(, dirhash) ufsdirhash_list; 94 95/* Protects: ufsdirhash_list, `dh_list' field, ufs_dirhashmem. */ 96static struct mtx ufsdirhash_mtx; 97 98/* 99 * Locking order: 100 * ufsdirhash_mtx 101 * dh_mtx 102 * 103 * The dh_mtx mutex should be acquired either via the inode lock, or via 104 * ufsdirhash_mtx. Only the owner of the inode may free the associated 105 * dirhash, but anything can steal its memory and set dh_hash to NULL. 106 */ 107 108/* 109 * Attempt to build up a hash table for the directory contents in 110 * inode 'ip'. Returns 0 on success, or -1 of the operation failed. 111 */ 112int 113ufsdirhash_build(struct inode *ip) 114{ 115 struct dirhash *dh; 116 struct buf *bp = NULL; 117 struct direct *ep; 118 struct vnode *vp; 119 doff_t bmask, pos; 120 int dirblocks, i, j, memreqd, nblocks, narrays, nslots, slot; 121 122 /* Check if we can/should use dirhash. */ 123 if (ip->i_dirhash == NULL) { 124 if (ip->i_size < ufs_mindirhashsize || OFSFMT(ip->i_vnode)) 125 return (-1); 126 } else { 127 /* Hash exists, but sysctls could have changed. */ 128 if (ip->i_size < ufs_mindirhashsize || 129 ufs_dirhashmem > ufs_dirhashmaxmem) { 130 ufsdirhash_free(ip); 131 return (-1); 132 } 133 /* Check if hash exists and is intact (note: unlocked read). */ 134 if (ip->i_dirhash->dh_hash != NULL) 135 return (0); 136 /* Free the old, recycled hash and build a new one. */ 137 ufsdirhash_free(ip); 138 } 139 140 /* Don't hash removed directories. */ 141 if (ip->i_effnlink == 0) 142 return (-1); 143 144 vp = ip->i_vnode; 145 /* Allocate 50% more entries than this dir size could ever need. */ 146 KASSERT(ip->i_size >= DIRBLKSIZ, ("ufsdirhash_build size")); 147 nslots = ip->i_size / DIRECTSIZ(1); 148 nslots = (nslots * 3 + 1) / 2; 149 narrays = howmany(nslots, DH_NBLKOFF); 150 nslots = narrays * DH_NBLKOFF; 151 dirblocks = howmany(ip->i_size, DIRBLKSIZ); 152 nblocks = (dirblocks * 3 + 1) / 2; 153 154 memreqd = sizeof(*dh) + narrays * sizeof(*dh->dh_hash) + 155 narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 156 nblocks * sizeof(*dh->dh_blkfree); 157 mtx_lock(&ufsdirhash_mtx); 158 if (memreqd + ufs_dirhashmem > ufs_dirhashmaxmem) { 159 mtx_unlock(&ufsdirhash_mtx); 160 if (memreqd > ufs_dirhashmaxmem / 2) 161 return (-1); 162 163 /* Try to free some space. */ 164 if (ufsdirhash_recycle(memreqd) != 0) 165 return (-1); 166 /* Enough was freed, and ufsdirhash_mtx has been locked. */ 167 } 168 ufs_dirhashmem += memreqd; 169 mtx_unlock(&ufsdirhash_mtx); 170 171 /* 172 * Use non-blocking mallocs so that we will revert to a linear 173 * lookup on failure rather than potentially blocking forever. 174 */ 175 MALLOC(dh, struct dirhash *, sizeof *dh, M_DIRHASH, M_NOWAIT | M_ZERO); 176 if (dh == NULL) { 177 mtx_lock(&ufsdirhash_mtx); 178 ufs_dirhashmem -= memreqd; 179 mtx_unlock(&ufsdirhash_mtx); 180 return (-1); 181 } 182 MALLOC(dh->dh_hash, doff_t **, narrays * sizeof(dh->dh_hash[0]), 183 M_DIRHASH, M_NOWAIT | M_ZERO); 184 MALLOC(dh->dh_blkfree, u_int8_t *, nblocks * sizeof(dh->dh_blkfree[0]), 185 M_DIRHASH, M_NOWAIT); 186 if (dh->dh_hash == NULL || dh->dh_blkfree == NULL) 187 goto fail; 188 for (i = 0; i < narrays; i++) { 189 if ((dh->dh_hash[i] = uma_zalloc(ufsdirhash_zone, 190 M_WAITOK)) == NULL) 191 goto fail; 192 for (j = 0; j < DH_NBLKOFF; j++) 193 dh->dh_hash[i][j] = DIRHASH_EMPTY; 194 } 195 196 /* Initialise the hash table and block statistics. */ 197 mtx_init(&dh->dh_mtx, "dirhash", NULL, MTX_DEF); 198 dh->dh_narrays = narrays; 199 dh->dh_hlen = nslots; 200 dh->dh_nblk = nblocks; 201 dh->dh_dirblks = dirblocks; 202 for (i = 0; i < dirblocks; i++) 203 dh->dh_blkfree[i] = DIRBLKSIZ / DIRALIGN; 204 for (i = 0; i < DH_NFSTATS; i++) 205 dh->dh_firstfree[i] = -1; 206 dh->dh_firstfree[DH_NFSTATS] = 0; 207 dh->dh_seqopt = 0; 208 dh->dh_seqoff = 0; 209 dh->dh_score = DH_SCOREINIT; 210 ip->i_dirhash = dh; 211 212 bmask = VFSTOUFS(vp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 213 pos = 0; 214 while (pos < ip->i_size) { 215 /* If necessary, get the next directory block. */ 216 if ((pos & bmask) == 0) { 217 if (bp != NULL) 218 brelse(bp); 219 if (UFS_BLKATOFF(vp, (off_t)pos, NULL, &bp) != 0) 220 goto fail; 221 } 222 223 /* Add this entry to the hash. */ 224 ep = (struct direct *)((char *)bp->b_data + (pos & bmask)); 225 if (ep->d_reclen == 0 || ep->d_reclen > 226 DIRBLKSIZ - (pos & (DIRBLKSIZ - 1))) { 227 /* Corrupted directory. */ 228 brelse(bp); 229 goto fail; 230 } 231 if (ep->d_ino != 0) { 232 /* Add the entry (simplified ufsdirhash_add). */ 233 slot = ufsdirhash_hash(dh, ep->d_name, ep->d_namlen); 234 while (DH_ENTRY(dh, slot) != DIRHASH_EMPTY) 235 slot = WRAPINCR(slot, dh->dh_hlen); 236 dh->dh_hused++; 237 DH_ENTRY(dh, slot) = pos; 238 ufsdirhash_adjfree(dh, pos, -DIRSIZ(0, ep)); 239 } 240 pos += ep->d_reclen; 241 } 242 243 if (bp != NULL) 244 brelse(bp); 245 mtx_lock(&ufsdirhash_mtx); 246 TAILQ_INSERT_TAIL(&ufsdirhash_list, dh, dh_list); 247 dh->dh_onlist = 1; 248 mtx_unlock(&ufsdirhash_mtx); 249 return (0); 250 251fail: 252 if (dh->dh_hash != NULL) { 253 for (i = 0; i < narrays; i++) 254 if (dh->dh_hash[i] != NULL) 255 uma_zfree(ufsdirhash_zone, dh->dh_hash[i]); 256 FREE(dh->dh_hash, M_DIRHASH); 257 } 258 if (dh->dh_blkfree != NULL) 259 FREE(dh->dh_blkfree, M_DIRHASH); 260 FREE(dh, M_DIRHASH); 261 ip->i_dirhash = NULL; 262 mtx_lock(&ufsdirhash_mtx); 263 ufs_dirhashmem -= memreqd; 264 mtx_unlock(&ufsdirhash_mtx); 265 return (-1); 266} 267 268/* 269 * Free any hash table associated with inode 'ip'. 270 */ 271void 272ufsdirhash_free(struct inode *ip) 273{ 274 struct dirhash *dh; 275 int i, mem; 276 277 if ((dh = ip->i_dirhash) == NULL) 278 return; 279 mtx_lock(&ufsdirhash_mtx); 280 mtx_lock(&dh->dh_mtx); 281 if (dh->dh_onlist) 282 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 283 mtx_unlock(&dh->dh_mtx); 284 mtx_unlock(&ufsdirhash_mtx); 285 286 /* The dirhash pointed to by 'dh' is exclusively ours now. */ 287 288 mem = sizeof(*dh); 289 if (dh->dh_hash != NULL) { 290 for (i = 0; i < dh->dh_narrays; i++) 291 uma_zfree(ufsdirhash_zone, dh->dh_hash[i]); 292 FREE(dh->dh_hash, M_DIRHASH); 293 FREE(dh->dh_blkfree, M_DIRHASH); 294 mem += dh->dh_narrays * sizeof(*dh->dh_hash) + 295 dh->dh_narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 296 dh->dh_nblk * sizeof(*dh->dh_blkfree); 297 } 298 mtx_destroy(&dh->dh_mtx); 299 FREE(dh, M_DIRHASH); 300 ip->i_dirhash = NULL; 301 302 mtx_lock(&ufsdirhash_mtx); 303 ufs_dirhashmem -= mem; 304 mtx_unlock(&ufsdirhash_mtx); 305} 306 307/* 308 * Find the offset of the specified name within the given inode. 309 * Returns 0 on success, ENOENT if the entry does not exist, or 310 * EJUSTRETURN if the caller should revert to a linear search. 311 * 312 * If successful, the directory offset is stored in *offp, and a 313 * pointer to a struct buf containing the entry is stored in *bpp. If 314 * prevoffp is non-NULL, the offset of the previous entry within 315 * the DIRBLKSIZ-sized block is stored in *prevoffp (if the entry 316 * is the first in a block, the start of the block is used). 317 */ 318int 319ufsdirhash_lookup(struct inode *ip, char *name, int namelen, doff_t *offp, 320 struct buf **bpp, doff_t *prevoffp) 321{ 322 struct dirhash *dh, *dh_next; 323 struct direct *dp; 324 struct vnode *vp; 325 struct buf *bp; 326 doff_t blkoff, bmask, offset, prevoff; 327 int i, slot; 328 329 if ((dh = ip->i_dirhash) == NULL) 330 return (EJUSTRETURN); 331 /* 332 * Move this dirhash towards the end of the list if it has a 333 * score higher than the next entry, and acquire the dh_mtx. 334 * Optimise the case where it's already the last by performing 335 * an unlocked read of the TAILQ_NEXT pointer. 336 * 337 * In both cases, end up holding just dh_mtx. 338 */ 339 if (TAILQ_NEXT(dh, dh_list) != NULL) { 340 mtx_lock(&ufsdirhash_mtx); 341 mtx_lock(&dh->dh_mtx); 342 /* 343 * If the new score will be greater than that of the next 344 * entry, then move this entry past it. With both mutexes 345 * held, dh_next won't go away, but its dh_score could 346 * change; that's not important since it is just a hint. 347 */ 348 if (dh->dh_hash != NULL && 349 (dh_next = TAILQ_NEXT(dh, dh_list)) != NULL && 350 dh->dh_score >= dh_next->dh_score) { 351 KASSERT(dh->dh_onlist, ("dirhash: not on list")); 352 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 353 TAILQ_INSERT_AFTER(&ufsdirhash_list, dh_next, dh, 354 dh_list); 355 } 356 mtx_unlock(&ufsdirhash_mtx); 357 } else { 358 /* Already the last, though that could change as we wait. */ 359 mtx_lock(&dh->dh_mtx); 360 } 361 if (dh->dh_hash == NULL) { 362 mtx_unlock(&dh->dh_mtx); 363 ufsdirhash_free(ip); 364 return (EJUSTRETURN); 365 } 366 367 /* Update the score. */ 368 if (dh->dh_score < DH_SCOREMAX) 369 dh->dh_score++; 370 371 vp = ip->i_vnode; 372 bmask = VFSTOUFS(vp->v_mount)->um_mountp->mnt_stat.f_iosize - 1; 373 blkoff = -1; 374 bp = NULL; 375restart: 376 slot = ufsdirhash_hash(dh, name, namelen); 377 378 if (dh->dh_seqopt) { 379 /* 380 * Sequential access optimisation. dh_seqoff contains the 381 * offset of the directory entry immediately following 382 * the last entry that was looked up. Check if this offset 383 * appears in the hash chain for the name we are looking for. 384 */ 385 for (i = slot; (offset = DH_ENTRY(dh, i)) != DIRHASH_EMPTY; 386 i = WRAPINCR(i, dh->dh_hlen)) 387 if (offset == dh->dh_seqoff) 388 break; 389 if (offset == dh->dh_seqoff) { 390 /* 391 * We found an entry with the expected offset. This 392 * is probably the entry we want, but if not, the 393 * code below will turn off seqoff and retry. 394 */ 395 slot = i; 396 } else 397 dh->dh_seqopt = 0; 398 } 399 400 for (; (offset = DH_ENTRY(dh, slot)) != DIRHASH_EMPTY; 401 slot = WRAPINCR(slot, dh->dh_hlen)) { 402 if (offset == DIRHASH_DEL) 403 continue; 404 mtx_unlock(&dh->dh_mtx); 405 406 if (offset < 0 || offset >= ip->i_size) 407 panic("ufsdirhash_lookup: bad offset in hash array"); 408 if ((offset & ~bmask) != blkoff) { 409 if (bp != NULL) 410 brelse(bp); 411 blkoff = offset & ~bmask; 412 if (UFS_BLKATOFF(vp, (off_t)blkoff, NULL, &bp) != 0) 413 return (EJUSTRETURN); 414 } 415 dp = (struct direct *)(bp->b_data + (offset & bmask)); 416 if (dp->d_reclen == 0 || dp->d_reclen > 417 DIRBLKSIZ - (offset & (DIRBLKSIZ - 1))) { 418 /* Corrupted directory. */ 419 brelse(bp); 420 return (EJUSTRETURN); 421 } 422 if (dp->d_namlen == namelen && 423 bcmp(dp->d_name, name, namelen) == 0) { 424 /* Found. Get the prev offset if needed. */ 425 if (prevoffp != NULL) { 426 if (offset & (DIRBLKSIZ - 1)) { 427 prevoff = ufsdirhash_getprev(dp, 428 offset); 429 if (prevoff == -1) { 430 brelse(bp); 431 return (EJUSTRETURN); 432 } 433 } else 434 prevoff = offset; 435 *prevoffp = prevoff; 436 } 437 438 /* Check for sequential access, and update offset. */ 439 if (dh->dh_seqopt == 0 && dh->dh_seqoff == offset) 440 dh->dh_seqopt = 1; 441 dh->dh_seqoff = offset + DIRSIZ(0, dp); 442 443 *bpp = bp; 444 *offp = offset; 445 return (0); 446 } 447 448 mtx_lock(&dh->dh_mtx); 449 if (dh->dh_hash == NULL) { 450 mtx_unlock(&dh->dh_mtx); 451 if (bp != NULL) 452 brelse(bp); 453 ufsdirhash_free(ip); 454 return (EJUSTRETURN); 455 } 456 /* 457 * When the name doesn't match in the seqopt case, go back 458 * and search normally. 459 */ 460 if (dh->dh_seqopt) { 461 dh->dh_seqopt = 0; 462 goto restart; 463 } 464 } 465 mtx_unlock(&dh->dh_mtx); 466 if (bp != NULL) 467 brelse(bp); 468 return (ENOENT); 469} 470 471/* 472 * Find a directory block with room for 'slotneeded' bytes. Returns 473 * the offset of the directory entry that begins the free space. 474 * This will either be the offset of an existing entry that has free 475 * space at the end, or the offset of an entry with d_ino == 0 at 476 * the start of a DIRBLKSIZ block. 477 * 478 * To use the space, the caller may need to compact existing entries in 479 * the directory. The total number of bytes in all of the entries involved 480 * in the compaction is stored in *slotsize. In other words, all of 481 * the entries that must be compacted are exactly contained in the 482 * region beginning at the returned offset and spanning *slotsize bytes. 483 * 484 * Returns -1 if no space was found, indicating that the directory 485 * must be extended. 486 */ 487doff_t 488ufsdirhash_findfree(struct inode *ip, int slotneeded, int *slotsize) 489{ 490 struct direct *dp; 491 struct dirhash *dh; 492 struct buf *bp; 493 doff_t pos, slotstart; 494 int dirblock, error, freebytes, i; 495 496 if ((dh = ip->i_dirhash) == NULL) 497 return (-1); 498 mtx_lock(&dh->dh_mtx); 499 if (dh->dh_hash == NULL) { 500 mtx_unlock(&dh->dh_mtx); 501 ufsdirhash_free(ip); 502 return (-1); 503 } 504 505 /* Find a directory block with the desired free space. */ 506 dirblock = -1; 507 for (i = howmany(slotneeded, DIRALIGN); i <= DH_NFSTATS; i++) 508 if ((dirblock = dh->dh_firstfree[i]) != -1) 509 break; 510 if (dirblock == -1) { 511 mtx_unlock(&dh->dh_mtx); 512 return (-1); 513 } 514 515 KASSERT(dirblock < dh->dh_nblk && 516 dh->dh_blkfree[dirblock] >= howmany(slotneeded, DIRALIGN), 517 ("ufsdirhash_findfree: bad stats")); 518 mtx_unlock(&dh->dh_mtx); 519 pos = dirblock * DIRBLKSIZ; 520 error = UFS_BLKATOFF(ip->i_vnode, (off_t)pos, (char **)&dp, &bp); 521 if (error) 522 return (-1); 523 524 /* Find the first entry with free space. */ 525 for (i = 0; i < DIRBLKSIZ; ) { 526 if (dp->d_reclen == 0) { 527 brelse(bp); 528 return (-1); 529 } 530 if (dp->d_ino == 0 || dp->d_reclen > DIRSIZ(0, dp)) 531 break; 532 i += dp->d_reclen; 533 dp = (struct direct *)((char *)dp + dp->d_reclen); 534 } 535 if (i > DIRBLKSIZ) { 536 brelse(bp); 537 return (-1); 538 } 539 slotstart = pos + i; 540 541 /* Find the range of entries needed to get enough space */ 542 freebytes = 0; 543 while (i < DIRBLKSIZ && freebytes < slotneeded) { 544 freebytes += dp->d_reclen; 545 if (dp->d_ino != 0) 546 freebytes -= DIRSIZ(0, dp); 547 if (dp->d_reclen == 0) { 548 brelse(bp); 549 return (-1); 550 } 551 i += dp->d_reclen; 552 dp = (struct direct *)((char *)dp + dp->d_reclen); 553 } 554 if (i > DIRBLKSIZ) { 555 brelse(bp); 556 return (-1); 557 } 558 if (freebytes < slotneeded) 559 panic("ufsdirhash_findfree: free mismatch"); 560 brelse(bp); 561 *slotsize = pos + i - slotstart; 562 return (slotstart); 563} 564 565/* 566 * Return the start of the unused space at the end of a directory, or 567 * -1 if there are no trailing unused blocks. 568 */ 569doff_t 570ufsdirhash_enduseful(struct inode *ip) 571{ 572 573 struct dirhash *dh; 574 int i; 575 576 if ((dh = ip->i_dirhash) == NULL) 577 return (-1); 578 mtx_lock(&dh->dh_mtx); 579 if (dh->dh_hash == NULL) { 580 mtx_unlock(&dh->dh_mtx); 581 ufsdirhash_free(ip); 582 return (-1); 583 } 584 585 if (dh->dh_blkfree[dh->dh_dirblks - 1] != DIRBLKSIZ / DIRALIGN) { 586 mtx_unlock(&dh->dh_mtx); 587 return (-1); 588 } 589 590 for (i = dh->dh_dirblks - 1; i >= 0; i--) 591 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN) 592 break; 593 mtx_unlock(&dh->dh_mtx); 594 return ((doff_t)(i + 1) * DIRBLKSIZ); 595} 596 597/* 598 * Insert information into the hash about a new directory entry. dirp 599 * points to a struct direct containing the entry, and offset specifies 600 * the offset of this entry. 601 */ 602void 603ufsdirhash_add(struct inode *ip, struct direct *dirp, doff_t offset) 604{ 605 struct dirhash *dh; 606 int slot; 607 608 if ((dh = ip->i_dirhash) == NULL) 609 return; 610 mtx_lock(&dh->dh_mtx); 611 if (dh->dh_hash == NULL) { 612 mtx_unlock(&dh->dh_mtx); 613 ufsdirhash_free(ip); 614 return; 615 } 616 617 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ, 618 ("ufsdirhash_add: bad offset")); 619 /* 620 * Normal hash usage is < 66%. If the usage gets too high then 621 * remove the hash entirely and let it be rebuilt later. 622 */ 623 if (dh->dh_hused >= (dh->dh_hlen * 3) / 4) { 624 mtx_unlock(&dh->dh_mtx); 625 ufsdirhash_free(ip); 626 return; 627 } 628 629 /* Find a free hash slot (empty or deleted), and add the entry. */ 630 slot = ufsdirhash_hash(dh, dirp->d_name, dirp->d_namlen); 631 while (DH_ENTRY(dh, slot) >= 0) 632 slot = WRAPINCR(slot, dh->dh_hlen); 633 if (DH_ENTRY(dh, slot) == DIRHASH_EMPTY) 634 dh->dh_hused++; 635 DH_ENTRY(dh, slot) = offset; 636 637 /* Update the per-block summary info. */ 638 ufsdirhash_adjfree(dh, offset, -DIRSIZ(0, dirp)); 639 mtx_unlock(&dh->dh_mtx); 640} 641 642/* 643 * Remove the specified directory entry from the hash. The entry to remove 644 * is defined by the name in `dirp', which must exist at the specified 645 * `offset' within the directory. 646 */ 647void 648ufsdirhash_remove(struct inode *ip, struct direct *dirp, doff_t offset) 649{ 650 struct dirhash *dh; 651 int slot; 652 653 if ((dh = ip->i_dirhash) == NULL) 654 return; 655 mtx_lock(&dh->dh_mtx); 656 if (dh->dh_hash == NULL) { 657 mtx_unlock(&dh->dh_mtx); 658 ufsdirhash_free(ip); 659 return; 660 } 661 662 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ, 663 ("ufsdirhash_remove: bad offset")); 664 /* Find the entry */ 665 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, offset); 666 667 /* Remove the hash entry. */ 668 ufsdirhash_delslot(dh, slot); 669 670 /* Update the per-block summary info. */ 671 ufsdirhash_adjfree(dh, offset, DIRSIZ(0, dirp)); 672 mtx_unlock(&dh->dh_mtx); 673} 674 675/* 676 * Change the offset associated with a directory entry in the hash. Used 677 * when compacting directory blocks. 678 */ 679void 680ufsdirhash_move(struct inode *ip, struct direct *dirp, doff_t oldoff, 681 doff_t newoff) 682{ 683 struct dirhash *dh; 684 int slot; 685 686 if ((dh = ip->i_dirhash) == NULL) 687 return; 688 mtx_lock(&dh->dh_mtx); 689 if (dh->dh_hash == NULL) { 690 mtx_unlock(&dh->dh_mtx); 691 ufsdirhash_free(ip); 692 return; 693 } 694 695 KASSERT(oldoff < dh->dh_dirblks * DIRBLKSIZ && 696 newoff < dh->dh_dirblks * DIRBLKSIZ, 697 ("ufsdirhash_move: bad offset")); 698 /* Find the entry, and update the offset. */ 699 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, oldoff); 700 DH_ENTRY(dh, slot) = newoff; 701 mtx_unlock(&dh->dh_mtx); 702} 703 704/* 705 * Inform dirhash that the directory has grown by one block that 706 * begins at offset (i.e. the new length is offset + DIRBLKSIZ). 707 */ 708void 709ufsdirhash_newblk(struct inode *ip, doff_t offset) 710{ 711 struct dirhash *dh; 712 int block; 713 714 if ((dh = ip->i_dirhash) == NULL) 715 return; 716 mtx_lock(&dh->dh_mtx); 717 if (dh->dh_hash == NULL) { 718 mtx_unlock(&dh->dh_mtx); 719 ufsdirhash_free(ip); 720 return; 721 } 722 723 KASSERT(offset == dh->dh_dirblks * DIRBLKSIZ, 724 ("ufsdirhash_newblk: bad offset")); 725 block = offset / DIRBLKSIZ; 726 if (block >= dh->dh_nblk) { 727 /* Out of space; must rebuild. */ 728 mtx_unlock(&dh->dh_mtx); 729 ufsdirhash_free(ip); 730 return; 731 } 732 dh->dh_dirblks = block + 1; 733 734 /* Account for the new free block. */ 735 dh->dh_blkfree[block] = DIRBLKSIZ / DIRALIGN; 736 if (dh->dh_firstfree[DH_NFSTATS] == -1) 737 dh->dh_firstfree[DH_NFSTATS] = block; 738 mtx_unlock(&dh->dh_mtx); 739} 740 741/* 742 * Inform dirhash that the directory is being truncated. 743 */ 744void 745ufsdirhash_dirtrunc(struct inode *ip, doff_t offset) 746{ 747 struct dirhash *dh; 748 int block, i; 749 750 if ((dh = ip->i_dirhash) == NULL) 751 return; 752 mtx_lock(&dh->dh_mtx); 753 if (dh->dh_hash == NULL) { 754 mtx_unlock(&dh->dh_mtx); 755 ufsdirhash_free(ip); 756 return; 757 } 758 759 KASSERT(offset <= dh->dh_dirblks * DIRBLKSIZ, 760 ("ufsdirhash_dirtrunc: bad offset")); 761 block = howmany(offset, DIRBLKSIZ); 762 /* 763 * If the directory shrinks to less than 1/8 of dh_nblk blocks 764 * (about 20% of its original size due to the 50% extra added in 765 * ufsdirhash_build) then free it, and let the caller rebuild 766 * if necessary. 767 */ 768 if (block < dh->dh_nblk / 8 && dh->dh_narrays > 1) { 769 mtx_unlock(&dh->dh_mtx); 770 ufsdirhash_free(ip); 771 return; 772 } 773 774 /* 775 * Remove any `first free' information pertaining to the 776 * truncated blocks. All blocks we're removing should be 777 * completely unused. 778 */ 779 if (dh->dh_firstfree[DH_NFSTATS] >= block) 780 dh->dh_firstfree[DH_NFSTATS] = -1; 781 for (i = block; i < dh->dh_dirblks; i++) 782 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN) 783 panic("ufsdirhash_dirtrunc: blocks in use"); 784 for (i = 0; i < DH_NFSTATS; i++) 785 if (dh->dh_firstfree[i] >= block) 786 panic("ufsdirhash_dirtrunc: first free corrupt"); 787 dh->dh_dirblks = block; 788 mtx_unlock(&dh->dh_mtx); 789} 790 791/* 792 * Debugging function to check that the dirhash information about 793 * a directory block matches its actual contents. Panics if a mismatch 794 * is detected. 795 * 796 * On entry, `buf' should point to the start of an in-core 797 * DIRBLKSIZ-sized directory block, and `offset' should contain the 798 * offset from the start of the directory of that block. 799 */ 800void 801ufsdirhash_checkblock(struct inode *ip, char *buf, doff_t offset) 802{ 803 struct dirhash *dh; 804 struct direct *dp; 805 int block, ffslot, i, nfree; 806 807 if (!ufs_dirhashcheck) 808 return; 809 if ((dh = ip->i_dirhash) == NULL) 810 return; 811 mtx_lock(&dh->dh_mtx); 812 if (dh->dh_hash == NULL) { 813 mtx_unlock(&dh->dh_mtx); 814 ufsdirhash_free(ip); 815 return; 816 } 817 818 block = offset / DIRBLKSIZ; 819 if ((offset & (DIRBLKSIZ - 1)) != 0 || block >= dh->dh_dirblks) 820 panic("ufsdirhash_checkblock: bad offset"); 821 822 nfree = 0; 823 for (i = 0; i < DIRBLKSIZ; i += dp->d_reclen) { 824 dp = (struct direct *)(buf + i); 825 if (dp->d_reclen == 0 || i + dp->d_reclen > DIRBLKSIZ) 826 panic("ufsdirhash_checkblock: bad dir"); 827 828 if (dp->d_ino == 0) { 829#if 0 830 /* 831 * XXX entries with d_ino == 0 should only occur 832 * at the start of a DIRBLKSIZ block. However the 833 * ufs code is tolerant of such entries at other 834 * offsets, and fsck does not fix them. 835 */ 836 if (i != 0) 837 panic("ufsdirhash_checkblock: bad dir inode"); 838#endif 839 nfree += dp->d_reclen; 840 continue; 841 } 842 843 /* Check that the entry exists (will panic if it doesn't). */ 844 ufsdirhash_findslot(dh, dp->d_name, dp->d_namlen, offset + i); 845 846 nfree += dp->d_reclen - DIRSIZ(0, dp); 847 } 848 if (i != DIRBLKSIZ) 849 panic("ufsdirhash_checkblock: bad dir end"); 850 851 if (dh->dh_blkfree[block] * DIRALIGN != nfree) 852 panic("ufsdirhash_checkblock: bad free count"); 853 854 ffslot = BLKFREE2IDX(nfree / DIRALIGN); 855 for (i = 0; i <= DH_NFSTATS; i++) 856 if (dh->dh_firstfree[i] == block && i != ffslot) 857 panic("ufsdirhash_checkblock: bad first-free"); 858 if (dh->dh_firstfree[ffslot] == -1) 859 panic("ufsdirhash_checkblock: missing first-free entry"); 860 mtx_unlock(&dh->dh_mtx); 861} 862 863/* 864 * Hash the specified filename into a dirhash slot. 865 */ 866static int 867ufsdirhash_hash(struct dirhash *dh, char *name, int namelen) 868{ 869 u_int32_t hash; 870 871 /* 872 * We hash the name and then some other bit of data that is 873 * invariant over the dirhash's lifetime. Otherwise names 874 * differing only in the last byte are placed close to one 875 * another in the table, which is bad for linear probing. 876 */ 877 hash = fnv_32_buf(name, namelen, FNV1_32_INIT); 878 hash = fnv_32_buf(dh, sizeof(dh), hash); 879 return (hash % dh->dh_hlen); 880} 881 882/* 883 * Adjust the number of free bytes in the block containing `offset' 884 * by the value specified by `diff'. 885 * 886 * The caller must ensure we have exclusive access to `dh'; normally 887 * that means that dh_mtx should be held, but this is also called 888 * from ufsdirhash_build() where exclusive access can be assumed. 889 */ 890static void 891ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff) 892{ 893 int block, i, nfidx, ofidx; 894 895 /* Update the per-block summary info. */ 896 block = offset / DIRBLKSIZ; 897 KASSERT(block < dh->dh_nblk && block < dh->dh_dirblks, 898 ("dirhash bad offset")); 899 ofidx = BLKFREE2IDX(dh->dh_blkfree[block]); 900 dh->dh_blkfree[block] = (int)dh->dh_blkfree[block] + (diff / DIRALIGN); 901 nfidx = BLKFREE2IDX(dh->dh_blkfree[block]); 902 903 /* Update the `first free' list if necessary. */ 904 if (ofidx != nfidx) { 905 /* If removing, scan forward for the next block. */ 906 if (dh->dh_firstfree[ofidx] == block) { 907 for (i = block + 1; i < dh->dh_dirblks; i++) 908 if (BLKFREE2IDX(dh->dh_blkfree[i]) == ofidx) 909 break; 910 dh->dh_firstfree[ofidx] = (i < dh->dh_dirblks) ? i : -1; 911 } 912 913 /* Make this the new `first free' if necessary */ 914 if (dh->dh_firstfree[nfidx] > block || 915 dh->dh_firstfree[nfidx] == -1) 916 dh->dh_firstfree[nfidx] = block; 917 } 918} 919 920/* 921 * Find the specified name which should have the specified offset. 922 * Returns a slot number, and panics on failure. 923 * 924 * `dh' must be locked on entry and remains so on return. 925 */ 926static int 927ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, doff_t offset) 928{ 929 int slot; 930 931 mtx_assert(&dh->dh_mtx, MA_OWNED); 932 933 /* Find the entry. */ 934 KASSERT(dh->dh_hused < dh->dh_hlen, ("dirhash find full")); 935 slot = ufsdirhash_hash(dh, name, namelen); 936 while (DH_ENTRY(dh, slot) != offset && 937 DH_ENTRY(dh, slot) != DIRHASH_EMPTY) 938 slot = WRAPINCR(slot, dh->dh_hlen); 939 if (DH_ENTRY(dh, slot) != offset) 940 panic("ufsdirhash_findslot: '%.*s' not found", namelen, name); 941 942 return (slot); 943} 944 945/* 946 * Remove the entry corresponding to the specified slot from the hash array. 947 * 948 * `dh' must be locked on entry and remains so on return. 949 */ 950static void 951ufsdirhash_delslot(struct dirhash *dh, int slot) 952{ 953 int i; 954 955 mtx_assert(&dh->dh_mtx, MA_OWNED); 956 957 /* Mark the entry as deleted. */ 958 DH_ENTRY(dh, slot) = DIRHASH_DEL; 959 960 /* If this is the end of a chain of DIRHASH_DEL slots, remove them. */ 961 for (i = slot; DH_ENTRY(dh, i) == DIRHASH_DEL; ) 962 i = WRAPINCR(i, dh->dh_hlen); 963 if (DH_ENTRY(dh, i) == DIRHASH_EMPTY) { 964 i = WRAPDECR(i, dh->dh_hlen); 965 while (DH_ENTRY(dh, i) == DIRHASH_DEL) { 966 DH_ENTRY(dh, i) = DIRHASH_EMPTY; 967 dh->dh_hused--; 968 i = WRAPDECR(i, dh->dh_hlen); 969 } 970 KASSERT(dh->dh_hused >= 0, ("ufsdirhash_delslot neg hlen")); 971 } 972} 973 974/* 975 * Given a directory entry and its offset, find the offset of the 976 * previous entry in the same DIRBLKSIZ-sized block. Returns an 977 * offset, or -1 if there is no previous entry in the block or some 978 * other problem occurred. 979 */ 980static doff_t 981ufsdirhash_getprev(struct direct *dirp, doff_t offset) 982{ 983 struct direct *dp; 984 char *blkbuf; 985 doff_t blkoff, prevoff; 986 int entrypos, i; 987 988 blkoff = offset & ~(DIRBLKSIZ - 1); /* offset of start of block */ 989 entrypos = offset & (DIRBLKSIZ - 1); /* entry relative to block */ 990 blkbuf = (char *)dirp - entrypos; 991 prevoff = blkoff; 992 993 /* If `offset' is the start of a block, there is no previous entry. */ 994 if (entrypos == 0) 995 return (-1); 996 997 /* Scan from the start of the block until we get to the entry. */ 998 for (i = 0; i < entrypos; i += dp->d_reclen) { 999 dp = (struct direct *)(blkbuf + i); 1000 if (dp->d_reclen == 0 || i + dp->d_reclen > entrypos) 1001 return (-1); /* Corrupted directory. */ 1002 prevoff = blkoff + i; 1003 } 1004 return (prevoff); 1005} 1006 1007/* 1008 * Try to free up `wanted' bytes by stealing memory from existing 1009 * dirhashes. Returns zero with ufsdirhash_mtx locked if successful. 1010 */ 1011static int 1012ufsdirhash_recycle(int wanted) 1013{ 1014 struct dirhash *dh; 1015 doff_t **hash; 1016 u_int8_t *blkfree; 1017 int i, mem, narrays; 1018 1019 mtx_lock(&ufsdirhash_mtx); 1020 while (wanted + ufs_dirhashmem > ufs_dirhashmaxmem) { 1021 /* Find a dirhash, and lock it. */ 1022 if ((dh = TAILQ_FIRST(&ufsdirhash_list)) == NULL) { 1023 mtx_unlock(&ufsdirhash_mtx); 1024 return (-1); 1025 } 1026 mtx_lock(&dh->dh_mtx); 1027 KASSERT(dh->dh_hash != NULL, ("dirhash: NULL hash on list")); 1028 1029 /* Decrement the score; only recycle if it becomes zero. */ 1030 if (--dh->dh_score > 0) { 1031 mtx_unlock(&dh->dh_mtx); 1032 mtx_unlock(&ufsdirhash_mtx); 1033 return (-1); 1034 } 1035 1036 /* Remove it from the list and detach its memory. */ 1037 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 1038 dh->dh_onlist = 0; 1039 hash = dh->dh_hash; 1040 dh->dh_hash = NULL; 1041 blkfree = dh->dh_blkfree; 1042 dh->dh_blkfree = NULL; 1043 narrays = dh->dh_narrays; 1044 mem = narrays * sizeof(*dh->dh_hash) + 1045 narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 1046 dh->dh_nblk * sizeof(*dh->dh_blkfree); 1047 1048 /* Unlock everything, free the detached memory. */ 1049 mtx_unlock(&dh->dh_mtx); 1050 mtx_unlock(&ufsdirhash_mtx); 1051 for (i = 0; i < narrays; i++) 1052 uma_zfree(ufsdirhash_zone, hash[i]); 1053 FREE(hash, M_DIRHASH); 1054 FREE(blkfree, M_DIRHASH); 1055 1056 /* Account for the returned memory, and repeat if necessary. */ 1057 mtx_lock(&ufsdirhash_mtx); 1058 ufs_dirhashmem -= mem; 1059 } 1060 /* Success; return with ufsdirhash_mtx locked. */ 1061 return (0); 1062} 1063 1064 1065void 1066ufsdirhash_init() 1067{ 1068 ufsdirhash_zone = uma_zcreate("DIRHASH", DH_NBLKOFF * sizeof(doff_t), 1069 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1070 mtx_init(&ufsdirhash_mtx, "dirhash list", NULL, MTX_DEF); 1071 TAILQ_INIT(&ufsdirhash_list); 1072} 1073 1074void 1075ufsdirhash_uninit() 1076{ 1077 KASSERT(TAILQ_EMPTY(&ufsdirhash_list), ("ufsdirhash_uninit")); 1078 uma_zdestroy(ufsdirhash_zone); 1079 mtx_destroy(&ufsdirhash_mtx); 1080} 1081 1082#endif /* UFS_DIRHASH */ 1083