1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2001, 2002 Ian Dowse. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 /* 29 * This implements a hash-based lookup scheme for UFS directories. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_ufs.h" 36 37 #ifdef UFS_DIRHASH 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/mutex.h> 44 #include <sys/malloc.h> 45 #include <sys/fnv_hash.h> 46 #include <sys/proc.h> 47 #include <sys/bio.h> 48 #include <sys/buf.h> 49 #include <sys/vnode.h> 50 #include <sys/mount.h> 51 #include <sys/refcount.h> 52 #include <sys/sysctl.h> 53 #include <sys/sx.h> 54 #include <sys/eventhandler.h> 55 #include <sys/time.h> 56 #include <vm/uma.h> 57 58 #include <ufs/ufs/quota.h> 59 #include <ufs/ufs/inode.h> 60 #include <ufs/ufs/dir.h> 61 #include <ufs/ufs/dirhash.h> 62 #include <ufs/ufs/extattr.h> 63 #include <ufs/ufs/ufsmount.h> 64 #include <ufs/ufs/ufs_extern.h> 65 66 #define WRAPINCR(val, limit) (((val) + 1 == (limit)) ? 0 : ((val) + 1)) 67 #define WRAPDECR(val, limit) (((val) == 0) ? ((limit) - 1) : ((val) - 1)) 68 #define BLKFREE2IDX(n) ((n) > DH_NFSTATS ? DH_NFSTATS : (n)) 69 70 static MALLOC_DEFINE(M_DIRHASH, "ufs_dirhash", "UFS directory hash tables"); 71 72 static int ufs_mindirhashsize = DIRBLKSIZ * 5; 73 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_minsize, CTLFLAG_RW, 74 &ufs_mindirhashsize, 75 0, "minimum directory size in bytes for which to use hashed lookup"); 76 static int ufs_dirhashmaxmem = 2 * 1024 * 1024; /* NOTE: initial value. It is 77 tuned in ufsdirhash_init() */ 78 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_maxmem, CTLFLAG_RW, &ufs_dirhashmaxmem, 79 0, "maximum allowed dirhash memory usage"); 80 static int ufs_dirhashmem; 81 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_mem, CTLFLAG_RD, &ufs_dirhashmem, 82 0, "current dirhash memory usage"); 83 static int ufs_dirhashcheck = 0; 84 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_docheck, CTLFLAG_RW, &ufs_dirhashcheck, 85 0, "enable extra sanity tests"); 86 static int ufs_dirhashlowmemcount = 0; 87 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_lowmemcount, CTLFLAG_RD, 88 &ufs_dirhashlowmemcount, 0, "number of times low memory hook called"); 89 static int ufs_dirhashreclaimpercent = 10; 90 static int ufsdirhash_set_reclaimpercent(SYSCTL_HANDLER_ARGS); 91 SYSCTL_PROC(_vfs_ufs, OID_AUTO, dirhash_reclaimpercent, 92 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 93 0, 0, ufsdirhash_set_reclaimpercent, "I", 94 "set percentage of dirhash cache to be removed in low VM events"); 95 96 static int ufsdirhash_hash(struct dirhash *dh, char *name, int namelen); 97 static void ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff); 98 static void ufsdirhash_delslot(struct dirhash *dh, int slot); 99 static int ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, 100 doff_t offset); 101 static doff_t ufsdirhash_getprev(struct direct *dp, doff_t offset); 102 static int ufsdirhash_recycle(int wanted); 103 static void ufsdirhash_lowmem(void); 104 static void ufsdirhash_free_locked(struct inode *ip); 105 106 static uma_zone_t ufsdirhash_zone; 107 108 #define DIRHASHLIST_LOCK() mtx_lock(&ufsdirhash_mtx) 109 #define DIRHASHLIST_UNLOCK() mtx_unlock(&ufsdirhash_mtx) 110 #define DIRHASH_BLKALLOC() uma_zalloc(ufsdirhash_zone, M_NOWAIT) 111 #define DIRHASH_BLKFREE(ptr) uma_zfree(ufsdirhash_zone, (ptr)) 112 #define DIRHASH_ASSERT_LOCKED(dh) \ 113 sx_assert(&(dh)->dh_lock, SA_LOCKED) 114 115 /* Dirhash list; recently-used entries are near the tail. */ 116 static TAILQ_HEAD(, dirhash) ufsdirhash_list; 117 118 /* Protects: ufsdirhash_list, `dh_list' field, ufs_dirhashmem. */ 119 static struct mtx ufsdirhash_mtx; 120 121 /* 122 * Locking: 123 * 124 * The relationship between inode and dirhash is protected either by an 125 * exclusive vnode lock or the vnode interlock where a shared vnode lock 126 * may be used. The dirhash_mtx is acquired after the dirhash lock. To 127 * handle teardown races, code wishing to lock the dirhash for an inode 128 * when using a shared vnode lock must obtain a private reference on the 129 * dirhash while holding the vnode interlock. They can drop it once they 130 * have obtained the dirhash lock and verified that the dirhash wasn't 131 * recycled while they waited for the dirhash lock. 132 * 133 * ufsdirhash_build() acquires a shared lock on the dirhash when it is 134 * successful. This lock is released after a call to ufsdirhash_lookup(). 135 * 136 * Functions requiring exclusive access use ufsdirhash_acquire() which may 137 * free a dirhash structure that was recycled by ufsdirhash_recycle(). 138 * 139 * The dirhash lock may be held across io operations. 140 * 141 * WITNESS reports a lock order reversal between the "bufwait" lock 142 * and the "dirhash" lock. However, this specific reversal will not 143 * cause a deadlock. To get a deadlock, one would have to lock a 144 * buffer followed by the dirhash while a second thread locked a 145 * buffer while holding the dirhash lock. The second order can happen 146 * under a shared or exclusive vnode lock for the associated directory 147 * in lookup(). The first order, however, can only happen under an 148 * exclusive vnode lock (e.g. unlink(), rename(), etc.). Thus, for 149 * a thread to be doing a "bufwait" -> "dirhash" order, it has to hold 150 * an exclusive vnode lock. That exclusive vnode lock will prevent 151 * any other threads from doing a "dirhash" -> "bufwait" order. 152 */ 153 154 static void 155 ufsdirhash_hold(struct dirhash *dh) 156 { 157 158 refcount_acquire(&dh->dh_refcount); 159 } 160 161 static void 162 ufsdirhash_drop(struct dirhash *dh) 163 { 164 165 if (refcount_release(&dh->dh_refcount)) { 166 sx_destroy(&dh->dh_lock); 167 free(dh, M_DIRHASH); 168 } 169 } 170 171 /* 172 * Release the lock on a dirhash. 173 */ 174 static void 175 ufsdirhash_release(struct dirhash *dh) 176 { 177 178 sx_unlock(&dh->dh_lock); 179 } 180 181 /* 182 * Either acquire an existing hash locked shared or create a new hash and 183 * return it exclusively locked. May return NULL if the allocation fails. 184 * 185 * The vnode interlock is used to protect the i_dirhash pointer from 186 * simultaneous access while only a shared vnode lock is held. 187 */ 188 static struct dirhash * 189 ufsdirhash_create(struct inode *ip) 190 { 191 struct dirhash *ndh; 192 struct dirhash *dh; 193 struct vnode *vp; 194 bool excl; 195 196 ndh = dh = NULL; 197 vp = ip->i_vnode; 198 excl = false; 199 for (;;) { 200 /* Racy check for i_dirhash to prefetch a dirhash structure. */ 201 if (ip->i_dirhash == NULL && ndh == NULL) { 202 ndh = malloc(sizeof *dh, M_DIRHASH, 203 M_NOWAIT | M_ZERO); 204 if (ndh == NULL) 205 return (NULL); 206 refcount_init(&ndh->dh_refcount, 1); 207 208 /* 209 * The DUPOK is to prevent warnings from the 210 * sx_slock() a few lines down which is safe 211 * since the duplicate lock in that case is 212 * the one for this dirhash we are creating 213 * now which has no external references until 214 * after this function returns. 215 */ 216 sx_init_flags(&ndh->dh_lock, "dirhash", SX_DUPOK); 217 sx_xlock(&ndh->dh_lock); 218 } 219 /* 220 * Check i_dirhash. If it's NULL just try to use a 221 * preallocated structure. If none exists loop and try again. 222 */ 223 VI_LOCK(vp); 224 dh = ip->i_dirhash; 225 if (dh == NULL) { 226 ip->i_dirhash = ndh; 227 VI_UNLOCK(vp); 228 if (ndh == NULL) 229 continue; 230 return (ndh); 231 } 232 ufsdirhash_hold(dh); 233 VI_UNLOCK(vp); 234 235 /* Acquire a lock on existing hashes. */ 236 if (excl) 237 sx_xlock(&dh->dh_lock); 238 else 239 sx_slock(&dh->dh_lock); 240 241 /* The hash could've been recycled while we were waiting. */ 242 VI_LOCK(vp); 243 if (ip->i_dirhash != dh) { 244 VI_UNLOCK(vp); 245 ufsdirhash_release(dh); 246 ufsdirhash_drop(dh); 247 continue; 248 } 249 VI_UNLOCK(vp); 250 ufsdirhash_drop(dh); 251 252 /* If the hash is still valid we've succeeded. */ 253 if (dh->dh_hash != NULL) 254 break; 255 /* 256 * If the hash is NULL it has been recycled. Try to upgrade 257 * so we can recreate it. If we fail the upgrade, drop our 258 * lock and try again. 259 */ 260 if (excl || sx_try_upgrade(&dh->dh_lock)) 261 break; 262 sx_sunlock(&dh->dh_lock); 263 excl = true; 264 } 265 /* Free the preallocated structure if it was not necessary. */ 266 if (ndh) { 267 ufsdirhash_release(ndh); 268 ufsdirhash_drop(ndh); 269 } 270 return (dh); 271 } 272 273 /* 274 * Acquire an exclusive lock on an existing hash. Requires an exclusive 275 * vnode lock to protect the i_dirhash pointer. hashes that have been 276 * recycled are reclaimed here and NULL is returned. 277 */ 278 static struct dirhash * 279 ufsdirhash_acquire(struct inode *ip) 280 { 281 struct dirhash *dh; 282 283 ASSERT_VOP_ELOCKED(ip->i_vnode, __FUNCTION__); 284 285 dh = ip->i_dirhash; 286 if (dh == NULL) 287 return (NULL); 288 sx_xlock(&dh->dh_lock); 289 if (dh->dh_hash != NULL) 290 return (dh); 291 ufsdirhash_free_locked(ip); 292 return (NULL); 293 } 294 295 /* 296 * Acquire exclusively and free the hash pointed to by ip. Works with a 297 * shared or exclusive vnode lock. 298 */ 299 void 300 ufsdirhash_free(struct inode *ip) 301 { 302 struct dirhash *dh; 303 struct vnode *vp; 304 305 vp = ip->i_vnode; 306 for (;;) { 307 /* Grab a reference on this inode's dirhash if it has one. */ 308 VI_LOCK(vp); 309 dh = ip->i_dirhash; 310 if (dh == NULL) { 311 VI_UNLOCK(vp); 312 return; 313 } 314 ufsdirhash_hold(dh); 315 VI_UNLOCK(vp); 316 317 /* Exclusively lock the dirhash. */ 318 sx_xlock(&dh->dh_lock); 319 320 /* If this dirhash still belongs to this inode, then free it. */ 321 VI_LOCK(vp); 322 if (ip->i_dirhash == dh) { 323 VI_UNLOCK(vp); 324 ufsdirhash_drop(dh); 325 break; 326 } 327 VI_UNLOCK(vp); 328 329 /* 330 * This inode's dirhash has changed while we were 331 * waiting for the dirhash lock, so try again. 332 */ 333 ufsdirhash_release(dh); 334 ufsdirhash_drop(dh); 335 } 336 ufsdirhash_free_locked(ip); 337 } 338 339 /* 340 * Attempt to build up a hash table for the directory contents in 341 * inode 'ip'. Returns 0 on success, or -1 of the operation failed. 342 */ 343 int 344 ufsdirhash_build(struct inode *ip) 345 { 346 struct dirhash *dh; 347 struct buf *bp = NULL; 348 struct direct *ep; 349 struct vnode *vp; 350 doff_t bmask, pos; 351 u_int dirblocks, i, narrays, nblocks, nslots; 352 int j, memreqd, slot; 353 354 /* Take care of a decreased sysctl value. */ 355 while (ufs_dirhashmem > ufs_dirhashmaxmem) { 356 if (ufsdirhash_recycle(0) != 0) 357 return (-1); 358 /* Recycled enough memory, so unlock the list. */ 359 DIRHASHLIST_UNLOCK(); 360 } 361 362 /* Check if we can/should use dirhash. */ 363 if (ip->i_size < ufs_mindirhashsize || OFSFMT(ip->i_vnode) || 364 ip->i_effnlink == 0) { 365 if (ip->i_dirhash) 366 ufsdirhash_free(ip); 367 return (-1); 368 } 369 dh = ufsdirhash_create(ip); 370 if (dh == NULL) 371 return (-1); 372 if (dh->dh_hash != NULL) 373 return (0); 374 375 vp = ip->i_vnode; 376 /* Allocate 50% more entries than this dir size could ever need. */ 377 KASSERT(ip->i_size >= DIRBLKSIZ, ("ufsdirhash_build size")); 378 nslots = ip->i_size / DIRECTSIZ(1); 379 nslots = (nslots * 3 + 1) / 2; 380 narrays = howmany(nslots, DH_NBLKOFF); 381 nslots = narrays * DH_NBLKOFF; 382 dirblocks = howmany(ip->i_size, DIRBLKSIZ); 383 nblocks = (dirblocks * 3 + 1) / 2; 384 memreqd = sizeof(*dh) + narrays * sizeof(*dh->dh_hash) + 385 narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) + 386 nblocks * sizeof(*dh->dh_blkfree); 387 DIRHASHLIST_LOCK(); 388 if (memreqd + ufs_dirhashmem > ufs_dirhashmaxmem) { 389 DIRHASHLIST_UNLOCK(); 390 if (memreqd > ufs_dirhashmaxmem / 2) 391 goto fail; 392 /* Try to free some space. */ 393 if (ufsdirhash_recycle(memreqd) != 0) 394 goto fail; 395 /* Enough was freed, and list has been locked. */ 396 } 397 ufs_dirhashmem += memreqd; 398 DIRHASHLIST_UNLOCK(); 399 400 /* Initialise the hash table and block statistics. */ 401 dh->dh_memreq = memreqd; 402 dh->dh_narrays = narrays; 403 dh->dh_hlen = nslots; 404 dh->dh_nblk = nblocks; 405 dh->dh_dirblks = dirblocks; 406 for (i = 0; i < DH_NFSTATS; i++) 407 dh->dh_firstfree[i] = -1; 408 dh->dh_firstfree[DH_NFSTATS] = 0; 409 dh->dh_hused = 0; 410 dh->dh_seqoff = -1; 411 dh->dh_score = DH_SCOREINIT; 412 dh->dh_lastused = time_second; 413 414 /* 415 * Use non-blocking mallocs so that we will revert to a linear 416 * lookup on failure rather than potentially blocking forever. 417 */ 418 dh->dh_hash = malloc(narrays * sizeof(dh->dh_hash[0]), 419 M_DIRHASH, M_NOWAIT | M_ZERO); 420 if (dh->dh_hash == NULL) 421 goto fail; 422 dh->dh_blkfree = malloc(nblocks * sizeof(dh->dh_blkfree[0]), 423 M_DIRHASH, M_NOWAIT); 424 if (dh->dh_blkfree == NULL) 425 goto fail; 426 for (i = 0; i < narrays; i++) { 427 if ((dh->dh_hash[i] = DIRHASH_BLKALLOC()) == NULL) 428 goto fail; 429 for (j = 0; j < DH_NBLKOFF; j++) 430 dh->dh_hash[i][j] = DIRHASH_EMPTY; 431 } 432 for (i = 0; i < dirblocks; i++) 433 dh->dh_blkfree[i] = DIRBLKSIZ / DIRALIGN; 434 bmask = vp->v_mount->mnt_stat.f_iosize - 1; 435 pos = 0; 436 while (pos < ip->i_size) { 437 /* If necessary, get the next directory block. */ 438 if ((pos & bmask) == 0) { 439 if (bp != NULL) 440 brelse(bp); 441 if (UFS_BLKATOFF(vp, (off_t)pos, NULL, &bp) != 0) 442 goto fail; 443 } 444 445 /* Add this entry to the hash. */ 446 ep = (struct direct *)((char *)bp->b_data + (pos & bmask)); 447 if (ep->d_reclen == 0 || ep->d_reclen > 448 DIRBLKSIZ - (pos & (DIRBLKSIZ - 1))) { 449 /* Corrupted directory. */ 450 brelse(bp); 451 goto fail; 452 } 453 if (ep->d_ino != 0) { 454 /* Add the entry (simplified ufsdirhash_add). */ 455 slot = ufsdirhash_hash(dh, ep->d_name, ep->d_namlen); 456 while (DH_ENTRY(dh, slot) != DIRHASH_EMPTY) 457 slot = WRAPINCR(slot, dh->dh_hlen); 458 dh->dh_hused++; 459 DH_ENTRY(dh, slot) = pos; 460 ufsdirhash_adjfree(dh, pos, -DIRSIZ(0, ep)); 461 } 462 pos += ep->d_reclen; 463 } 464 465 if (bp != NULL) 466 brelse(bp); 467 DIRHASHLIST_LOCK(); 468 TAILQ_INSERT_TAIL(&ufsdirhash_list, dh, dh_list); 469 dh->dh_onlist = 1; 470 DIRHASHLIST_UNLOCK(); 471 sx_downgrade(&dh->dh_lock); 472 return (0); 473 474 fail: 475 ufsdirhash_free_locked(ip); 476 return (-1); 477 } 478 479 /* 480 * Free any hash table associated with inode 'ip'. 481 */ 482 static void 483 ufsdirhash_free_locked(struct inode *ip) 484 { 485 struct dirhash *dh; 486 struct vnode *vp; 487 int i; 488 489 DIRHASH_ASSERT_LOCKED(ip->i_dirhash); 490 491 /* 492 * Clear the pointer in the inode to prevent new threads from 493 * finding the dead structure. 494 */ 495 vp = ip->i_vnode; 496 VI_LOCK(vp); 497 dh = ip->i_dirhash; 498 ip->i_dirhash = NULL; 499 VI_UNLOCK(vp); 500 501 /* 502 * Remove the hash from the list since we are going to free its 503 * memory. 504 */ 505 DIRHASHLIST_LOCK(); 506 if (dh->dh_onlist) 507 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 508 ufs_dirhashmem -= dh->dh_memreq; 509 DIRHASHLIST_UNLOCK(); 510 511 /* 512 * At this point, any waiters for the lock should hold their 513 * own reference on the dirhash structure. They will drop 514 * that reference once they grab the vnode interlock and see 515 * that ip->i_dirhash is NULL. 516 */ 517 sx_xunlock(&dh->dh_lock); 518 519 /* 520 * Handle partially recycled as well as fully constructed hashes. 521 */ 522 if (dh->dh_hash != NULL) { 523 for (i = 0; i < dh->dh_narrays; i++) 524 if (dh->dh_hash[i] != NULL) 525 DIRHASH_BLKFREE(dh->dh_hash[i]); 526 free(dh->dh_hash, M_DIRHASH); 527 if (dh->dh_blkfree != NULL) 528 free(dh->dh_blkfree, M_DIRHASH); 529 } 530 531 /* 532 * Drop the inode's reference to the data structure. 533 */ 534 ufsdirhash_drop(dh); 535 } 536 537 /* 538 * Find the offset of the specified name within the given inode. 539 * Returns 0 on success, ENOENT if the entry does not exist, or 540 * EJUSTRETURN if the caller should revert to a linear search. 541 * 542 * If successful, the directory offset is stored in *offp, and a 543 * pointer to a struct buf containing the entry is stored in *bpp. If 544 * prevoffp is non-NULL, the offset of the previous entry within 545 * the DIRBLKSIZ-sized block is stored in *prevoffp (if the entry 546 * is the first in a block, the start of the block is used). 547 * 548 * Must be called with the hash locked. Returns with the hash unlocked. 549 */ 550 int 551 ufsdirhash_lookup(struct inode *ip, char *name, int namelen, doff_t *offp, 552 struct buf **bpp, doff_t *prevoffp) 553 { 554 struct dirhash *dh, *dh_next; 555 struct direct *dp; 556 struct vnode *vp; 557 struct buf *bp; 558 doff_t blkoff, bmask, offset, prevoff, seqoff; 559 int i, slot; 560 int error; 561 562 dh = ip->i_dirhash; 563 KASSERT(dh != NULL && dh->dh_hash != NULL, 564 ("ufsdirhash_lookup: Invalid dirhash %p\n", dh)); 565 DIRHASH_ASSERT_LOCKED(dh); 566 /* 567 * Move this dirhash towards the end of the list if it has a 568 * score higher than the next entry, and acquire the dh_lock. 569 */ 570 DIRHASHLIST_LOCK(); 571 if (TAILQ_NEXT(dh, dh_list) != NULL) { 572 /* 573 * If the new score will be greater than that of the next 574 * entry, then move this entry past it. With both mutexes 575 * held, dh_next won't go away, but its dh_score could 576 * change; that's not important since it is just a hint. 577 */ 578 if ((dh_next = TAILQ_NEXT(dh, dh_list)) != NULL && 579 dh->dh_score >= dh_next->dh_score) { 580 KASSERT(dh->dh_onlist, ("dirhash: not on list")); 581 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 582 TAILQ_INSERT_AFTER(&ufsdirhash_list, dh_next, dh, 583 dh_list); 584 } 585 } 586 /* Update the score. */ 587 if (dh->dh_score < DH_SCOREMAX) 588 dh->dh_score++; 589 590 /* Update last used time. */ 591 dh->dh_lastused = time_second; 592 DIRHASHLIST_UNLOCK(); 593 594 vp = ip->i_vnode; 595 bmask = vp->v_mount->mnt_stat.f_iosize - 1; 596 blkoff = -1; 597 bp = NULL; 598 seqoff = dh->dh_seqoff; 599 restart: 600 slot = ufsdirhash_hash(dh, name, namelen); 601 602 if (seqoff != -1) { 603 /* 604 * Sequential access optimisation. seqoff contains the 605 * offset of the directory entry immediately following 606 * the last entry that was looked up. Check if this offset 607 * appears in the hash chain for the name we are looking for. 608 */ 609 for (i = slot; (offset = DH_ENTRY(dh, i)) != DIRHASH_EMPTY; 610 i = WRAPINCR(i, dh->dh_hlen)) 611 if (offset == seqoff) 612 break; 613 if (offset == seqoff) { 614 /* 615 * We found an entry with the expected offset. This 616 * is probably the entry we want, but if not, the 617 * code below will retry. 618 */ 619 slot = i; 620 } else 621 seqoff = -1; 622 } 623 624 for (; (offset = DH_ENTRY(dh, slot)) != DIRHASH_EMPTY; 625 slot = WRAPINCR(slot, dh->dh_hlen)) { 626 if (offset == DIRHASH_DEL) 627 continue; 628 if (offset < 0 || offset >= ip->i_size) 629 panic("ufsdirhash_lookup: bad offset in hash array"); 630 if ((offset & ~bmask) != blkoff) { 631 if (bp != NULL) 632 brelse(bp); 633 blkoff = offset & ~bmask; 634 if (UFS_BLKATOFF(vp, (off_t)blkoff, NULL, &bp) != 0) { 635 error = EJUSTRETURN; 636 goto fail; 637 } 638 } 639 KASSERT(bp != NULL, ("no buffer allocated")); 640 dp = (struct direct *)(bp->b_data + (offset & bmask)); 641 if (dp->d_reclen == 0 || dp->d_reclen > 642 DIRBLKSIZ - (offset & (DIRBLKSIZ - 1))) { 643 /* Corrupted directory. */ 644 error = EJUSTRETURN; 645 goto fail; 646 } 647 if (dp->d_namlen == namelen && 648 bcmp(dp->d_name, name, namelen) == 0) { 649 /* Found. Get the prev offset if needed. */ 650 if (prevoffp != NULL) { 651 if (offset & (DIRBLKSIZ - 1)) { 652 prevoff = ufsdirhash_getprev(dp, 653 offset); 654 if (prevoff == -1) { 655 error = EJUSTRETURN; 656 goto fail; 657 } 658 } else 659 prevoff = offset; 660 *prevoffp = prevoff; 661 } 662 663 /* Update offset. */ 664 dh->dh_seqoff = offset + DIRSIZ(0, dp); 665 *bpp = bp; 666 *offp = offset; 667 ufsdirhash_release(dh); 668 return (0); 669 } 670 671 /* 672 * When the name doesn't match in the sequential 673 * optimization case, go back and search normally. 674 */ 675 if (seqoff != -1) { 676 seqoff = -1; 677 goto restart; 678 } 679 } 680 error = ENOENT; 681 fail: 682 ufsdirhash_release(dh); 683 if (bp != NULL) 684 brelse(bp); 685 return (error); 686 } 687 688 /* 689 * Find a directory block with room for 'slotneeded' bytes. Returns 690 * the offset of the directory entry that begins the free space. 691 * This will either be the offset of an existing entry that has free 692 * space at the end, or the offset of an entry with d_ino == 0 at 693 * the start of a DIRBLKSIZ block. 694 * 695 * To use the space, the caller may need to compact existing entries in 696 * the directory. The total number of bytes in all of the entries involved 697 * in the compaction is stored in *slotsize. In other words, all of 698 * the entries that must be compacted are exactly contained in the 699 * region beginning at the returned offset and spanning *slotsize bytes. 700 * 701 * Returns -1 if no space was found, indicating that the directory 702 * must be extended. 703 */ 704 doff_t 705 ufsdirhash_findfree(struct inode *ip, int slotneeded, int *slotsize) 706 { 707 struct direct *dp; 708 struct dirhash *dh; 709 struct buf *bp; 710 doff_t pos, slotstart; 711 int dirblock, error, freebytes, i; 712 713 dh = ip->i_dirhash; 714 KASSERT(dh != NULL && dh->dh_hash != NULL, 715 ("ufsdirhash_findfree: Invalid dirhash %p\n", dh)); 716 DIRHASH_ASSERT_LOCKED(dh); 717 718 /* Find a directory block with the desired free space. */ 719 dirblock = -1; 720 for (i = howmany(slotneeded, DIRALIGN); i <= DH_NFSTATS; i++) 721 if ((dirblock = dh->dh_firstfree[i]) != -1) 722 break; 723 if (dirblock == -1) 724 return (-1); 725 726 KASSERT(dirblock < dh->dh_nblk && 727 dh->dh_blkfree[dirblock] >= howmany(slotneeded, DIRALIGN), 728 ("ufsdirhash_findfree: bad stats")); 729 pos = dirblock * DIRBLKSIZ; 730 error = UFS_BLKATOFF(ip->i_vnode, (off_t)pos, (char **)&dp, &bp); 731 if (error) 732 return (-1); 733 734 /* Find the first entry with free space. */ 735 for (i = 0; i < DIRBLKSIZ; ) { 736 if (dp->d_reclen == 0) { 737 brelse(bp); 738 return (-1); 739 } 740 if (dp->d_ino == 0 || dp->d_reclen > DIRSIZ(0, dp)) 741 break; 742 i += dp->d_reclen; 743 dp = (struct direct *)((char *)dp + dp->d_reclen); 744 } 745 if (i > DIRBLKSIZ) { 746 brelse(bp); 747 return (-1); 748 } 749 slotstart = pos + i; 750 751 /* Find the range of entries needed to get enough space */ 752 freebytes = 0; 753 while (i < DIRBLKSIZ && freebytes < slotneeded) { 754 freebytes += dp->d_reclen; 755 if (dp->d_ino != 0) 756 freebytes -= DIRSIZ(0, dp); 757 if (dp->d_reclen == 0) { 758 brelse(bp); 759 return (-1); 760 } 761 i += dp->d_reclen; 762 dp = (struct direct *)((char *)dp + dp->d_reclen); 763 } 764 if (i > DIRBLKSIZ) { 765 brelse(bp); 766 return (-1); 767 } 768 if (freebytes < slotneeded) 769 panic("ufsdirhash_findfree: free mismatch"); 770 brelse(bp); 771 *slotsize = pos + i - slotstart; 772 return (slotstart); 773 } 774 775 /* 776 * Return the start of the unused space at the end of a directory, or 777 * -1 if there are no trailing unused blocks. 778 */ 779 doff_t 780 ufsdirhash_enduseful(struct inode *ip) 781 { 782 783 struct dirhash *dh; 784 int i; 785 786 dh = ip->i_dirhash; 787 DIRHASH_ASSERT_LOCKED(dh); 788 KASSERT(dh != NULL && dh->dh_hash != NULL, 789 ("ufsdirhash_enduseful: Invalid dirhash %p\n", dh)); 790 791 if (dh->dh_blkfree[dh->dh_dirblks - 1] != DIRBLKSIZ / DIRALIGN) 792 return (-1); 793 794 for (i = dh->dh_dirblks - 1; i >= 0; i--) 795 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN) 796 break; 797 798 return ((doff_t)(i + 1) * DIRBLKSIZ); 799 } 800 801 /* 802 * Insert information into the hash about a new directory entry. dirp 803 * points to a struct direct containing the entry, and offset specifies 804 * the offset of this entry. 805 */ 806 void 807 ufsdirhash_add(struct inode *ip, struct direct *dirp, doff_t offset) 808 { 809 struct dirhash *dh; 810 int slot; 811 812 if ((dh = ufsdirhash_acquire(ip)) == NULL) 813 return; 814 815 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ, 816 ("ufsdirhash_add: bad offset")); 817 /* 818 * Normal hash usage is < 66%. If the usage gets too high then 819 * remove the hash entirely and let it be rebuilt later. 820 */ 821 if (dh->dh_hused >= (dh->dh_hlen * 3) / 4) { 822 ufsdirhash_free_locked(ip); 823 return; 824 } 825 826 /* Find a free hash slot (empty or deleted), and add the entry. */ 827 slot = ufsdirhash_hash(dh, dirp->d_name, dirp->d_namlen); 828 while (DH_ENTRY(dh, slot) >= 0) 829 slot = WRAPINCR(slot, dh->dh_hlen); 830 if (DH_ENTRY(dh, slot) == DIRHASH_EMPTY) 831 dh->dh_hused++; 832 DH_ENTRY(dh, slot) = offset; 833 834 /* Update last used time. */ 835 dh->dh_lastused = time_second; 836 837 /* Update the per-block summary info. */ 838 ufsdirhash_adjfree(dh, offset, -DIRSIZ(0, dirp)); 839 ufsdirhash_release(dh); 840 } 841 842 /* 843 * Remove the specified directory entry from the hash. The entry to remove 844 * is defined by the name in `dirp', which must exist at the specified 845 * `offset' within the directory. 846 */ 847 void 848 ufsdirhash_remove(struct inode *ip, struct direct *dirp, doff_t offset) 849 { 850 struct dirhash *dh; 851 int slot; 852 853 if ((dh = ufsdirhash_acquire(ip)) == NULL) 854 return; 855 856 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ, 857 ("ufsdirhash_remove: bad offset")); 858 /* Find the entry */ 859 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, offset); 860 861 /* Remove the hash entry. */ 862 ufsdirhash_delslot(dh, slot); 863 864 /* Update the per-block summary info. */ 865 ufsdirhash_adjfree(dh, offset, DIRSIZ(0, dirp)); 866 ufsdirhash_release(dh); 867 } 868 869 /* 870 * Change the offset associated with a directory entry in the hash. Used 871 * when compacting directory blocks. 872 */ 873 void 874 ufsdirhash_move(struct inode *ip, struct direct *dirp, doff_t oldoff, 875 doff_t newoff) 876 { 877 struct dirhash *dh; 878 int slot; 879 880 if ((dh = ufsdirhash_acquire(ip)) == NULL) 881 return; 882 883 KASSERT(oldoff < dh->dh_dirblks * DIRBLKSIZ && 884 newoff < dh->dh_dirblks * DIRBLKSIZ, 885 ("ufsdirhash_move: bad offset")); 886 /* Find the entry, and update the offset. */ 887 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, oldoff); 888 DH_ENTRY(dh, slot) = newoff; 889 ufsdirhash_release(dh); 890 } 891 892 /* 893 * Inform dirhash that the directory has grown by one block that 894 * begins at offset (i.e. the new length is offset + DIRBLKSIZ). 895 */ 896 void 897 ufsdirhash_newblk(struct inode *ip, doff_t offset) 898 { 899 struct dirhash *dh; 900 int block; 901 902 if ((dh = ufsdirhash_acquire(ip)) == NULL) 903 return; 904 905 KASSERT(offset == dh->dh_dirblks * DIRBLKSIZ, 906 ("ufsdirhash_newblk: bad offset")); 907 block = offset / DIRBLKSIZ; 908 if (block >= dh->dh_nblk) { 909 /* Out of space; must rebuild. */ 910 ufsdirhash_free_locked(ip); 911 return; 912 } 913 dh->dh_dirblks = block + 1; 914 915 /* Account for the new free block. */ 916 dh->dh_blkfree[block] = DIRBLKSIZ / DIRALIGN; 917 if (dh->dh_firstfree[DH_NFSTATS] == -1) 918 dh->dh_firstfree[DH_NFSTATS] = block; 919 ufsdirhash_release(dh); 920 } 921 922 /* 923 * Inform dirhash that the directory is being truncated. 924 */ 925 void 926 ufsdirhash_dirtrunc(struct inode *ip, doff_t offset) 927 { 928 struct dirhash *dh; 929 int block, i; 930 931 if ((dh = ufsdirhash_acquire(ip)) == NULL) 932 return; 933 934 KASSERT(offset <= dh->dh_dirblks * DIRBLKSIZ, 935 ("ufsdirhash_dirtrunc: bad offset")); 936 block = howmany(offset, DIRBLKSIZ); 937 /* 938 * If the directory shrinks to less than 1/8 of dh_nblk blocks 939 * (about 20% of its original size due to the 50% extra added in 940 * ufsdirhash_build) then free it, and let the caller rebuild 941 * if necessary. 942 */ 943 if (block < dh->dh_nblk / 8 && dh->dh_narrays > 1) { 944 ufsdirhash_free_locked(ip); 945 return; 946 } 947 948 /* 949 * Remove any `first free' information pertaining to the 950 * truncated blocks. All blocks we're removing should be 951 * completely unused. 952 */ 953 if (dh->dh_firstfree[DH_NFSTATS] >= block) 954 dh->dh_firstfree[DH_NFSTATS] = -1; 955 for (i = block; i < dh->dh_dirblks; i++) 956 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN) 957 panic("ufsdirhash_dirtrunc: blocks in use"); 958 for (i = 0; i < DH_NFSTATS; i++) 959 if (dh->dh_firstfree[i] >= block) 960 panic("ufsdirhash_dirtrunc: first free corrupt"); 961 dh->dh_dirblks = block; 962 ufsdirhash_release(dh); 963 } 964 965 /* 966 * Debugging function to check that the dirhash information about 967 * a directory block matches its actual contents. Panics if a mismatch 968 * is detected. 969 * 970 * On entry, `buf' should point to the start of an in-core 971 * DIRBLKSIZ-sized directory block, and `offset' should contain the 972 * offset from the start of the directory of that block. 973 */ 974 void 975 ufsdirhash_checkblock(struct inode *ip, char *buf, doff_t offset) 976 { 977 struct dirhash *dh; 978 struct direct *dp; 979 int block, ffslot, i, nfree; 980 981 if (!ufs_dirhashcheck) 982 return; 983 if ((dh = ufsdirhash_acquire(ip)) == NULL) 984 return; 985 986 block = offset / DIRBLKSIZ; 987 if ((offset & (DIRBLKSIZ - 1)) != 0 || block >= dh->dh_dirblks) 988 panic("ufsdirhash_checkblock: bad offset"); 989 990 nfree = 0; 991 for (i = 0; i < DIRBLKSIZ; i += dp->d_reclen) { 992 dp = (struct direct *)(buf + i); 993 if (dp->d_reclen == 0 || i + dp->d_reclen > DIRBLKSIZ) 994 panic("ufsdirhash_checkblock: bad dir"); 995 996 if (dp->d_ino == 0) { 997 #if 0 998 /* 999 * XXX entries with d_ino == 0 should only occur 1000 * at the start of a DIRBLKSIZ block. However the 1001 * ufs code is tolerant of such entries at other 1002 * offsets, and fsck does not fix them. 1003 */ 1004 if (i != 0) 1005 panic("ufsdirhash_checkblock: bad dir inode"); 1006 #endif 1007 nfree += dp->d_reclen; 1008 continue; 1009 } 1010 1011 /* Check that the entry exists (will panic if it doesn't). */ 1012 ufsdirhash_findslot(dh, dp->d_name, dp->d_namlen, offset + i); 1013 1014 nfree += dp->d_reclen - DIRSIZ(0, dp); 1015 } 1016 if (i != DIRBLKSIZ) 1017 panic("ufsdirhash_checkblock: bad dir end"); 1018 1019 if (dh->dh_blkfree[block] * DIRALIGN != nfree) 1020 panic("ufsdirhash_checkblock: bad free count"); 1021 1022 ffslot = BLKFREE2IDX(nfree / DIRALIGN); 1023 for (i = 0; i <= DH_NFSTATS; i++) 1024 if (dh->dh_firstfree[i] == block && i != ffslot) 1025 panic("ufsdirhash_checkblock: bad first-free"); 1026 if (dh->dh_firstfree[ffslot] == -1) 1027 panic("ufsdirhash_checkblock: missing first-free entry"); 1028 ufsdirhash_release(dh); 1029 } 1030 1031 /* 1032 * Hash the specified filename into a dirhash slot. 1033 */ 1034 static int 1035 ufsdirhash_hash(struct dirhash *dh, char *name, int namelen) 1036 { 1037 u_int32_t hash; 1038 1039 /* 1040 * We hash the name and then some other bit of data that is 1041 * invariant over the dirhash's lifetime. Otherwise names 1042 * differing only in the last byte are placed close to one 1043 * another in the table, which is bad for linear probing. 1044 */ 1045 hash = fnv_32_buf(name, namelen, FNV1_32_INIT); 1046 hash = fnv_32_buf(&dh, sizeof(dh), hash); 1047 return (hash % dh->dh_hlen); 1048 } 1049 1050 /* 1051 * Adjust the number of free bytes in the block containing `offset' 1052 * by the value specified by `diff'. 1053 * 1054 * The caller must ensure we have exclusive access to `dh'; normally 1055 * that means that dh_lock should be held, but this is also called 1056 * from ufsdirhash_build() where exclusive access can be assumed. 1057 */ 1058 static void 1059 ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff) 1060 { 1061 int block, i, nfidx, ofidx; 1062 1063 /* Update the per-block summary info. */ 1064 block = offset / DIRBLKSIZ; 1065 KASSERT(block < dh->dh_nblk && block < dh->dh_dirblks, 1066 ("dirhash bad offset")); 1067 ofidx = BLKFREE2IDX(dh->dh_blkfree[block]); 1068 dh->dh_blkfree[block] = (int)dh->dh_blkfree[block] + (diff / DIRALIGN); 1069 nfidx = BLKFREE2IDX(dh->dh_blkfree[block]); 1070 1071 /* Update the `first free' list if necessary. */ 1072 if (ofidx != nfidx) { 1073 /* If removing, scan forward for the next block. */ 1074 if (dh->dh_firstfree[ofidx] == block) { 1075 for (i = block + 1; i < dh->dh_dirblks; i++) 1076 if (BLKFREE2IDX(dh->dh_blkfree[i]) == ofidx) 1077 break; 1078 dh->dh_firstfree[ofidx] = (i < dh->dh_dirblks) ? i : -1; 1079 } 1080 1081 /* Make this the new `first free' if necessary */ 1082 if (dh->dh_firstfree[nfidx] > block || 1083 dh->dh_firstfree[nfidx] == -1) 1084 dh->dh_firstfree[nfidx] = block; 1085 } 1086 } 1087 1088 /* 1089 * Find the specified name which should have the specified offset. 1090 * Returns a slot number, and panics on failure. 1091 * 1092 * `dh' must be locked on entry and remains so on return. 1093 */ 1094 static int 1095 ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, doff_t offset) 1096 { 1097 int slot; 1098 1099 DIRHASH_ASSERT_LOCKED(dh); 1100 1101 /* Find the entry. */ 1102 KASSERT(dh->dh_hused < dh->dh_hlen, ("dirhash find full")); 1103 slot = ufsdirhash_hash(dh, name, namelen); 1104 while (DH_ENTRY(dh, slot) != offset && 1105 DH_ENTRY(dh, slot) != DIRHASH_EMPTY) 1106 slot = WRAPINCR(slot, dh->dh_hlen); 1107 if (DH_ENTRY(dh, slot) != offset) 1108 panic("ufsdirhash_findslot: '%.*s' not found", namelen, name); 1109 1110 return (slot); 1111 } 1112 1113 /* 1114 * Remove the entry corresponding to the specified slot from the hash array. 1115 * 1116 * `dh' must be locked on entry and remains so on return. 1117 */ 1118 static void 1119 ufsdirhash_delslot(struct dirhash *dh, int slot) 1120 { 1121 int i; 1122 1123 DIRHASH_ASSERT_LOCKED(dh); 1124 1125 /* Mark the entry as deleted. */ 1126 DH_ENTRY(dh, slot) = DIRHASH_DEL; 1127 1128 /* If this is the end of a chain of DIRHASH_DEL slots, remove them. */ 1129 for (i = slot; DH_ENTRY(dh, i) == DIRHASH_DEL; ) 1130 i = WRAPINCR(i, dh->dh_hlen); 1131 if (DH_ENTRY(dh, i) == DIRHASH_EMPTY) { 1132 i = WRAPDECR(i, dh->dh_hlen); 1133 while (DH_ENTRY(dh, i) == DIRHASH_DEL) { 1134 DH_ENTRY(dh, i) = DIRHASH_EMPTY; 1135 dh->dh_hused--; 1136 i = WRAPDECR(i, dh->dh_hlen); 1137 } 1138 KASSERT(dh->dh_hused >= 0, ("ufsdirhash_delslot neg hlen")); 1139 } 1140 } 1141 1142 /* 1143 * Given a directory entry and its offset, find the offset of the 1144 * previous entry in the same DIRBLKSIZ-sized block. Returns an 1145 * offset, or -1 if there is no previous entry in the block or some 1146 * other problem occurred. 1147 */ 1148 static doff_t 1149 ufsdirhash_getprev(struct direct *dirp, doff_t offset) 1150 { 1151 struct direct *dp; 1152 char *blkbuf; 1153 doff_t blkoff, prevoff; 1154 int entrypos, i; 1155 1156 blkoff = rounddown2(offset, DIRBLKSIZ); /* offset of start of block */ 1157 entrypos = offset & (DIRBLKSIZ - 1); /* entry relative to block */ 1158 blkbuf = (char *)dirp - entrypos; 1159 prevoff = blkoff; 1160 1161 /* If `offset' is the start of a block, there is no previous entry. */ 1162 if (entrypos == 0) 1163 return (-1); 1164 1165 /* Scan from the start of the block until we get to the entry. */ 1166 for (i = 0; i < entrypos; i += dp->d_reclen) { 1167 dp = (struct direct *)(blkbuf + i); 1168 if (dp->d_reclen == 0 || i + dp->d_reclen > entrypos) 1169 return (-1); /* Corrupted directory. */ 1170 prevoff = blkoff + i; 1171 } 1172 return (prevoff); 1173 } 1174 1175 /* 1176 * Delete the given dirhash and reclaim its memory. Assumes that 1177 * ufsdirhash_list is locked, and leaves it locked. Also assumes 1178 * that dh is locked. Returns the amount of memory freed. 1179 */ 1180 static int 1181 ufsdirhash_destroy(struct dirhash *dh) 1182 { 1183 doff_t **hash; 1184 u_int8_t *blkfree; 1185 int i, mem, narrays; 1186 1187 KASSERT(dh->dh_hash != NULL, ("dirhash: NULL hash on list")); 1188 1189 /* Remove it from the list and detach its memory. */ 1190 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list); 1191 dh->dh_onlist = 0; 1192 hash = dh->dh_hash; 1193 dh->dh_hash = NULL; 1194 blkfree = dh->dh_blkfree; 1195 dh->dh_blkfree = NULL; 1196 narrays = dh->dh_narrays; 1197 mem = dh->dh_memreq; 1198 dh->dh_memreq = 0; 1199 1200 /* Unlock dirhash and free the detached memory. */ 1201 ufsdirhash_release(dh); 1202 for (i = 0; i < narrays; i++) 1203 DIRHASH_BLKFREE(hash[i]); 1204 free(hash, M_DIRHASH); 1205 free(blkfree, M_DIRHASH); 1206 1207 /* Account for the returned memory. */ 1208 ufs_dirhashmem -= mem; 1209 1210 return (mem); 1211 } 1212 1213 /* 1214 * Try to free up `wanted' bytes by stealing memory from existing 1215 * dirhashes. Returns zero with list locked if successful. 1216 */ 1217 static int 1218 ufsdirhash_recycle(int wanted) 1219 { 1220 struct dirhash *dh; 1221 1222 DIRHASHLIST_LOCK(); 1223 dh = TAILQ_FIRST(&ufsdirhash_list); 1224 while (wanted + ufs_dirhashmem > ufs_dirhashmaxmem) { 1225 /* Decrement the score; only recycle if it becomes zero. */ 1226 if (dh == NULL || --dh->dh_score > 0) { 1227 DIRHASHLIST_UNLOCK(); 1228 return (-1); 1229 } 1230 /* 1231 * If we can't lock it it's in use and we don't want to 1232 * recycle it anyway. 1233 */ 1234 if (!sx_try_xlock(&dh->dh_lock)) { 1235 dh = TAILQ_NEXT(dh, dh_list); 1236 continue; 1237 } 1238 1239 ufsdirhash_destroy(dh); 1240 1241 /* Repeat if necessary. */ 1242 dh = TAILQ_FIRST(&ufsdirhash_list); 1243 } 1244 /* Success; return with list locked. */ 1245 return (0); 1246 } 1247 1248 /* 1249 * Callback that frees some dirhashes when the system is low on virtual memory. 1250 */ 1251 static void 1252 ufsdirhash_lowmem(void) 1253 { 1254 struct dirhash *dh, *dh_temp; 1255 int memfreed, memwanted; 1256 1257 ufs_dirhashlowmemcount++; 1258 memfreed = 0; 1259 memwanted = ufs_dirhashmem * ufs_dirhashreclaimpercent / 100; 1260 1261 DIRHASHLIST_LOCK(); 1262 1263 /* 1264 * Reclaim up to memwanted from the oldest dirhashes. This will allow 1265 * us to make some progress when the system is running out of memory 1266 * without compromising the dinamicity of maximum age. If the situation 1267 * does not improve lowmem will be eventually retriggered and free some 1268 * other entry in the cache. The entries on the head of the list should 1269 * be the oldest. If during list traversal we can't get a lock on the 1270 * dirhash, it will be skipped. 1271 */ 1272 TAILQ_FOREACH_SAFE(dh, &ufsdirhash_list, dh_list, dh_temp) { 1273 if (sx_try_xlock(&dh->dh_lock)) 1274 memfreed += ufsdirhash_destroy(dh); 1275 if (memfreed >= memwanted) 1276 break; 1277 } 1278 DIRHASHLIST_UNLOCK(); 1279 } 1280 1281 static int 1282 ufsdirhash_set_reclaimpercent(SYSCTL_HANDLER_ARGS) 1283 { 1284 int error, v; 1285 1286 v = ufs_dirhashreclaimpercent; 1287 error = sysctl_handle_int(oidp, &v, v, req); 1288 if (error) 1289 return (error); 1290 if (req->newptr == NULL) 1291 return (error); 1292 if (v == ufs_dirhashreclaimpercent) 1293 return (0); 1294 1295 /* Refuse invalid percentages */ 1296 if (v < 0 || v > 100) 1297 return (EINVAL); 1298 ufs_dirhashreclaimpercent = v; 1299 return (0); 1300 } 1301 1302 void 1303 ufsdirhash_init(void) 1304 { 1305 ufs_dirhashmaxmem = lmax(roundup(hibufspace / 64, PAGE_SIZE), 1306 2 * 1024 * 1024); 1307 1308 ufsdirhash_zone = uma_zcreate("DIRHASH", DH_NBLKOFF * sizeof(doff_t), 1309 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1310 mtx_init(&ufsdirhash_mtx, "dirhash list", NULL, MTX_DEF); 1311 TAILQ_INIT(&ufsdirhash_list); 1312 1313 /* Register a callback function to handle low memory signals */ 1314 EVENTHANDLER_REGISTER(vm_lowmem, ufsdirhash_lowmem, NULL, 1315 EVENTHANDLER_PRI_FIRST); 1316 } 1317 1318 void 1319 ufsdirhash_uninit(void) 1320 { 1321 KASSERT(TAILQ_EMPTY(&ufsdirhash_list), ("ufsdirhash_uninit")); 1322 uma_zdestroy(ufsdirhash_zone); 1323 mtx_destroy(&ufsdirhash_mtx); 1324 } 1325 1326 #endif /* UFS_DIRHASH */ 1327