2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2001, 2002 Ian Dowse. All rights reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
29 * This implements a hash-based lookup scheme for UFS directories.
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
43 #include <sys/mutex.h>
44 #include <sys/malloc.h>
45 #include <sys/fnv_hash.h>
49 #include <sys/vnode.h>
50 #include <sys/mount.h>
51 #include <sys/refcount.h>
52 #include <sys/sysctl.h>
54 #include <sys/eventhandler.h>
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>
66 #define WRAPINCR(val, limit) (((val) + 1 == (limit)) ? 0 : ((val) + 1))
67 #define WRAPDECR(val, limit) (((val) == 0) ? ((limit) - 1) : ((val) - 1))
68 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
69 #define BLKFREE2IDX(n) ((n) > DH_NFSTATS ? DH_NFSTATS : (n))
71 static MALLOC_DEFINE(M_DIRHASH, "ufs_dirhash", "UFS directory hash tables");
73 static int ufs_mindirhashsize = DIRBLKSIZ * 5;
74 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_minsize, CTLFLAG_RW,
76 0, "minimum directory size in bytes for which to use hashed lookup");
77 static int ufs_dirhashmaxmem = 2 * 1024 * 1024; /* NOTE: initial value. It is
78 tuned in ufsdirhash_init() */
79 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_maxmem, CTLFLAG_RW, &ufs_dirhashmaxmem,
80 0, "maximum allowed dirhash memory usage");
81 static int ufs_dirhashmem;
82 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_mem, CTLFLAG_RD, &ufs_dirhashmem,
83 0, "current dirhash memory usage");
84 static int ufs_dirhashcheck = 0;
85 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_docheck, CTLFLAG_RW, &ufs_dirhashcheck,
86 0, "enable extra sanity tests");
87 static int ufs_dirhashlowmemcount = 0;
88 SYSCTL_INT(_vfs_ufs, OID_AUTO, dirhash_lowmemcount, CTLFLAG_RD,
89 &ufs_dirhashlowmemcount, 0, "number of times low memory hook called");
90 static int ufs_dirhashreclaimpercent = 10;
91 static int ufsdirhash_set_reclaimpercent(SYSCTL_HANDLER_ARGS);
92 SYSCTL_PROC(_vfs_ufs, OID_AUTO, dirhash_reclaimpercent,
93 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
94 0, 0, ufsdirhash_set_reclaimpercent, "I",
95 "set percentage of dirhash cache to be removed in low VM events");
98 static int ufsdirhash_hash(struct dirhash *dh, char *name, int namelen);
99 static void ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff);
100 static void ufsdirhash_delslot(struct dirhash *dh, int slot);
101 static int ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen,
103 static doff_t ufsdirhash_getprev(struct direct *dp, doff_t offset);
104 static int ufsdirhash_recycle(int wanted);
105 static void ufsdirhash_lowmem(void);
106 static void ufsdirhash_free_locked(struct inode *ip);
108 static uma_zone_t ufsdirhash_zone;
110 #define DIRHASHLIST_LOCK() mtx_lock(&ufsdirhash_mtx)
111 #define DIRHASHLIST_UNLOCK() mtx_unlock(&ufsdirhash_mtx)
112 #define DIRHASH_BLKALLOC_WAITOK() uma_zalloc(ufsdirhash_zone, M_WAITOK)
113 #define DIRHASH_BLKFREE(ptr) uma_zfree(ufsdirhash_zone, (ptr))
114 #define DIRHASH_ASSERT_LOCKED(dh) \
115 sx_assert(&(dh)->dh_lock, SA_LOCKED)
117 /* Dirhash list; recently-used entries are near the tail. */
118 static TAILQ_HEAD(, dirhash) ufsdirhash_list;
120 /* Protects: ufsdirhash_list, `dh_list' field, ufs_dirhashmem. */
121 static struct mtx ufsdirhash_mtx;
126 * The relationship between inode and dirhash is protected either by an
127 * exclusive vnode lock or the vnode interlock where a shared vnode lock
128 * may be used. The dirhash_mtx is acquired after the dirhash lock. To
129 * handle teardown races, code wishing to lock the dirhash for an inode
130 * when using a shared vnode lock must obtain a private reference on the
131 * dirhash while holding the vnode interlock. They can drop it once they
132 * have obtained the dirhash lock and verified that the dirhash wasn't
133 * recycled while they waited for the dirhash lock.
135 * ufsdirhash_build() acquires a shared lock on the dirhash when it is
136 * successful. This lock is released after a call to ufsdirhash_lookup().
138 * Functions requiring exclusive access use ufsdirhash_acquire() which may
139 * free a dirhash structure that was recycled by ufsdirhash_recycle().
141 * The dirhash lock may be held across io operations.
143 * WITNESS reports a lock order reversal between the "bufwait" lock
144 * and the "dirhash" lock. However, this specific reversal will not
145 * cause a deadlock. To get a deadlock, one would have to lock a
146 * buffer followed by the dirhash while a second thread locked a
147 * buffer while holding the dirhash lock. The second order can happen
148 * under a shared or exclusive vnode lock for the associated directory
149 * in lookup(). The first order, however, can only happen under an
150 * exclusive vnode lock (e.g. unlink(), rename(), etc.). Thus, for
151 * a thread to be doing a "bufwait" -> "dirhash" order, it has to hold
152 * an exclusive vnode lock. That exclusive vnode lock will prevent
153 * any other threads from doing a "dirhash" -> "bufwait" order.
157 ufsdirhash_hold(struct dirhash *dh)
160 refcount_acquire(&dh->dh_refcount);
164 ufsdirhash_drop(struct dirhash *dh)
167 if (refcount_release(&dh->dh_refcount)) {
168 sx_destroy(&dh->dh_lock);
174 * Release the lock on a dirhash.
177 ufsdirhash_release(struct dirhash *dh)
180 sx_unlock(&dh->dh_lock);
184 * Either acquire an existing hash locked shared or create a new hash and
185 * return it exclusively locked. May return NULL if the allocation fails.
187 * The vnode interlock is used to protect the i_dirhash pointer from
188 * simultaneous access while only a shared vnode lock is held.
190 static struct dirhash *
191 ufsdirhash_create(struct inode *ip)
202 /* Racy check for i_dirhash to prefetch a dirhash structure. */
203 if (ip->i_dirhash == NULL && ndh == NULL) {
204 ndh = malloc(sizeof *dh, M_DIRHASH,
208 refcount_init(&ndh->dh_refcount, 1);
211 * The DUPOK is to prevent warnings from the
212 * sx_slock() a few lines down which is safe
213 * since the duplicate lock in that case is
214 * the one for this dirhash we are creating
215 * now which has no external references until
216 * after this function returns.
218 sx_init_flags(&ndh->dh_lock, "dirhash", SX_DUPOK);
219 sx_xlock(&ndh->dh_lock);
222 * Check i_dirhash. If it's NULL just try to use a
223 * preallocated structure. If none exists loop and try again.
237 /* Acquire a lock on existing hashes. */
239 sx_xlock(&dh->dh_lock);
241 sx_slock(&dh->dh_lock);
243 /* The hash could've been recycled while we were waiting. */
245 if (ip->i_dirhash != dh) {
247 ufsdirhash_release(dh);
254 /* If the hash is still valid we've succeeded. */
255 if (dh->dh_hash != NULL)
258 * If the hash is NULL it has been recycled. Try to upgrade
259 * so we can recreate it. If we fail the upgrade, drop our
260 * lock and try again.
262 if (excl || sx_try_upgrade(&dh->dh_lock))
264 sx_sunlock(&dh->dh_lock);
267 /* Free the preallocated structure if it was not necessary. */
269 ufsdirhash_release(ndh);
270 ufsdirhash_drop(ndh);
276 * Acquire an exclusive lock on an existing hash. Requires an exclusive
277 * vnode lock to protect the i_dirhash pointer. hashes that have been
278 * recycled are reclaimed here and NULL is returned.
280 static struct dirhash *
281 ufsdirhash_acquire(struct inode *ip)
285 ASSERT_VOP_ELOCKED(ip->i_vnode, __FUNCTION__);
290 sx_xlock(&dh->dh_lock);
291 if (dh->dh_hash != NULL)
293 ufsdirhash_free_locked(ip);
298 * Acquire exclusively and free the hash pointed to by ip. Works with a
299 * shared or exclusive vnode lock.
302 ufsdirhash_free(struct inode *ip)
309 /* Grab a reference on this inode's dirhash if it has one. */
319 /* Exclusively lock the dirhash. */
320 sx_xlock(&dh->dh_lock);
322 /* If this dirhash still belongs to this inode, then free it. */
324 if (ip->i_dirhash == dh) {
332 * This inode's dirhash has changed while we were
333 * waiting for the dirhash lock, so try again.
335 ufsdirhash_release(dh);
338 ufsdirhash_free_locked(ip);
342 * Attempt to build up a hash table for the directory contents in
343 * inode 'ip'. Returns 0 on success, or -1 of the operation failed.
346 ufsdirhash_build(struct inode *ip)
349 struct buf *bp = NULL;
353 u_int dirblocks, i, narrays, nblocks, nslots;
354 int j, memreqd, slot;
356 /* Take care of a decreased sysctl value. */
357 while (ufs_dirhashmem > ufs_dirhashmaxmem) {
358 if (ufsdirhash_recycle(0) != 0)
360 /* Recycled enough memory, so unlock the list. */
361 DIRHASHLIST_UNLOCK();
364 /* Check if we can/should use dirhash. */
365 if (ip->i_size < ufs_mindirhashsize || OFSFMT(ip->i_vnode) ||
366 ip->i_effnlink == 0) {
371 dh = ufsdirhash_create(ip);
374 if (dh->dh_hash != NULL)
378 /* Allocate 50% more entries than this dir size could ever need. */
379 KASSERT(ip->i_size >= DIRBLKSIZ, ("ufsdirhash_build size"));
380 nslots = ip->i_size / DIRECTSIZ(1);
381 nslots = (nslots * 3 + 1) / 2;
382 narrays = howmany(nslots, DH_NBLKOFF);
383 nslots = narrays * DH_NBLKOFF;
384 dirblocks = howmany(ip->i_size, DIRBLKSIZ);
385 nblocks = (dirblocks * 3 + 1) / 2;
386 memreqd = sizeof(*dh) + narrays * sizeof(*dh->dh_hash) +
387 narrays * DH_NBLKOFF * sizeof(**dh->dh_hash) +
388 nblocks * sizeof(*dh->dh_blkfree);
390 if (memreqd + ufs_dirhashmem > ufs_dirhashmaxmem) {
391 DIRHASHLIST_UNLOCK();
392 if (memreqd > ufs_dirhashmaxmem / 2)
394 /* Try to free some space. */
395 if (ufsdirhash_recycle(memreqd) != 0)
397 /* Enough was freed, and list has been locked. */
399 ufs_dirhashmem += memreqd;
400 DIRHASHLIST_UNLOCK();
402 /* Initialise the hash table and block statistics. */
403 dh->dh_memreq = memreqd;
404 dh->dh_narrays = narrays;
405 dh->dh_hlen = nslots;
406 dh->dh_nblk = nblocks;
407 dh->dh_dirblks = dirblocks;
408 for (i = 0; i < DH_NFSTATS; i++)
409 dh->dh_firstfree[i] = -1;
410 dh->dh_firstfree[DH_NFSTATS] = 0;
413 dh->dh_score = DH_SCOREINIT;
414 dh->dh_lastused = time_second;
417 * Use non-blocking mallocs so that we will revert to a linear
418 * lookup on failure rather than potentially blocking forever.
420 dh->dh_hash = malloc(narrays * sizeof(dh->dh_hash[0]),
421 M_DIRHASH, M_NOWAIT | M_ZERO);
422 if (dh->dh_hash == NULL)
424 dh->dh_blkfree = malloc(nblocks * sizeof(dh->dh_blkfree[0]),
425 M_DIRHASH, M_NOWAIT);
426 if (dh->dh_blkfree == NULL)
428 for (i = 0; i < narrays; i++) {
429 if ((dh->dh_hash[i] = DIRHASH_BLKALLOC_WAITOK()) == NULL)
431 for (j = 0; j < DH_NBLKOFF; j++)
432 dh->dh_hash[i][j] = DIRHASH_EMPTY;
434 for (i = 0; i < dirblocks; i++)
435 dh->dh_blkfree[i] = DIRBLKSIZ / DIRALIGN;
436 bmask = vp->v_mount->mnt_stat.f_iosize - 1;
438 while (pos < ip->i_size) {
439 /* If necessary, get the next directory block. */
440 if ((pos & bmask) == 0) {
443 if (UFS_BLKATOFF(vp, (off_t)pos, NULL, &bp) != 0)
447 /* Add this entry to the hash. */
448 ep = (struct direct *)((char *)bp->b_data + (pos & bmask));
449 if (ep->d_reclen == 0 || ep->d_reclen >
450 DIRBLKSIZ - (pos & (DIRBLKSIZ - 1))) {
451 /* Corrupted directory. */
455 if (ep->d_ino != 0) {
456 /* Add the entry (simplified ufsdirhash_add). */
457 slot = ufsdirhash_hash(dh, ep->d_name, ep->d_namlen);
458 while (DH_ENTRY(dh, slot) != DIRHASH_EMPTY)
459 slot = WRAPINCR(slot, dh->dh_hlen);
461 DH_ENTRY(dh, slot) = pos;
462 ufsdirhash_adjfree(dh, pos, -DIRSIZ(0, ep));
470 TAILQ_INSERT_TAIL(&ufsdirhash_list, dh, dh_list);
472 DIRHASHLIST_UNLOCK();
473 sx_downgrade(&dh->dh_lock);
477 ufsdirhash_free_locked(ip);
482 * Free any hash table associated with inode 'ip'.
485 ufsdirhash_free_locked(struct inode *ip)
491 DIRHASH_ASSERT_LOCKED(ip->i_dirhash);
494 * Clear the pointer in the inode to prevent new threads from
495 * finding the dead structure.
500 ip->i_dirhash = NULL;
504 * Remove the hash from the list since we are going to free its
509 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list);
510 ufs_dirhashmem -= dh->dh_memreq;
511 DIRHASHLIST_UNLOCK();
514 * At this point, any waiters for the lock should hold their
515 * own reference on the dirhash structure. They will drop
516 * that reference once they grab the vnode interlock and see
517 * that ip->i_dirhash is NULL.
519 sx_xunlock(&dh->dh_lock);
522 * Handle partially recycled as well as fully constructed hashes.
524 if (dh->dh_hash != NULL) {
525 for (i = 0; i < dh->dh_narrays; i++)
526 if (dh->dh_hash[i] != NULL)
527 DIRHASH_BLKFREE(dh->dh_hash[i]);
528 free(dh->dh_hash, M_DIRHASH);
529 if (dh->dh_blkfree != NULL)
530 free(dh->dh_blkfree, M_DIRHASH);
534 * Drop the inode's reference to the data structure.
540 * Find the offset of the specified name within the given inode.
541 * Returns 0 on success, ENOENT if the entry does not exist, or
542 * EJUSTRETURN if the caller should revert to a linear search.
544 * If successful, the directory offset is stored in *offp, and a
545 * pointer to a struct buf containing the entry is stored in *bpp. If
546 * prevoffp is non-NULL, the offset of the previous entry within
547 * the DIRBLKSIZ-sized block is stored in *prevoffp (if the entry
548 * is the first in a block, the start of the block is used).
550 * Must be called with the hash locked. Returns with the hash unlocked.
553 ufsdirhash_lookup(struct inode *ip, char *name, int namelen, doff_t *offp,
554 struct buf **bpp, doff_t *prevoffp)
556 struct dirhash *dh, *dh_next;
560 doff_t blkoff, bmask, offset, prevoff, seqoff;
565 KASSERT(dh != NULL && dh->dh_hash != NULL,
566 ("ufsdirhash_lookup: Invalid dirhash %p\n", dh));
567 DIRHASH_ASSERT_LOCKED(dh);
569 * Move this dirhash towards the end of the list if it has a
570 * score higher than the next entry, and acquire the dh_lock.
573 if (TAILQ_NEXT(dh, dh_list) != NULL) {
575 * If the new score will be greater than that of the next
576 * entry, then move this entry past it. With both mutexes
577 * held, dh_next won't go away, but its dh_score could
578 * change; that's not important since it is just a hint.
580 if ((dh_next = TAILQ_NEXT(dh, dh_list)) != NULL &&
581 dh->dh_score >= dh_next->dh_score) {
582 KASSERT(dh->dh_onlist, ("dirhash: not on list"));
583 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list);
584 TAILQ_INSERT_AFTER(&ufsdirhash_list, dh_next, dh,
588 /* Update the score. */
589 if (dh->dh_score < DH_SCOREMAX)
592 /* Update last used time. */
593 dh->dh_lastused = time_second;
594 DIRHASHLIST_UNLOCK();
597 bmask = vp->v_mount->mnt_stat.f_iosize - 1;
600 seqoff = dh->dh_seqoff;
602 slot = ufsdirhash_hash(dh, name, namelen);
606 * Sequential access optimisation. seqoff contains the
607 * offset of the directory entry immediately following
608 * the last entry that was looked up. Check if this offset
609 * appears in the hash chain for the name we are looking for.
611 for (i = slot; (offset = DH_ENTRY(dh, i)) != DIRHASH_EMPTY;
612 i = WRAPINCR(i, dh->dh_hlen))
613 if (offset == seqoff)
615 if (offset == seqoff) {
617 * We found an entry with the expected offset. This
618 * is probably the entry we want, but if not, the
619 * code below will retry.
626 for (; (offset = DH_ENTRY(dh, slot)) != DIRHASH_EMPTY;
627 slot = WRAPINCR(slot, dh->dh_hlen)) {
628 if (offset == DIRHASH_DEL)
630 if (offset < 0 || offset >= ip->i_size)
631 panic("ufsdirhash_lookup: bad offset in hash array");
632 if ((offset & ~bmask) != blkoff) {
635 blkoff = offset & ~bmask;
636 if (UFS_BLKATOFF(vp, (off_t)blkoff, NULL, &bp) != 0) {
641 KASSERT(bp != NULL, ("no buffer allocated"));
642 dp = (struct direct *)(bp->b_data + (offset & bmask));
643 if (dp->d_reclen == 0 || dp->d_reclen >
644 DIRBLKSIZ - (offset & (DIRBLKSIZ - 1))) {
645 /* Corrupted directory. */
649 if (dp->d_namlen == namelen &&
650 bcmp(dp->d_name, name, namelen) == 0) {
651 /* Found. Get the prev offset if needed. */
652 if (prevoffp != NULL) {
653 if (offset & (DIRBLKSIZ - 1)) {
654 prevoff = ufsdirhash_getprev(dp,
666 dh->dh_seqoff = offset + DIRSIZ(0, dp);
669 ufsdirhash_release(dh);
674 * When the name doesn't match in the sequential
675 * optimization case, go back and search normally.
684 ufsdirhash_release(dh);
691 * Find a directory block with room for 'slotneeded' bytes. Returns
692 * the offset of the directory entry that begins the free space.
693 * This will either be the offset of an existing entry that has free
694 * space at the end, or the offset of an entry with d_ino == 0 at
695 * the start of a DIRBLKSIZ block.
697 * To use the space, the caller may need to compact existing entries in
698 * the directory. The total number of bytes in all of the entries involved
699 * in the compaction is stored in *slotsize. In other words, all of
700 * the entries that must be compacted are exactly contained in the
701 * region beginning at the returned offset and spanning *slotsize bytes.
703 * Returns -1 if no space was found, indicating that the directory
707 ufsdirhash_findfree(struct inode *ip, int slotneeded, int *slotsize)
712 doff_t pos, slotstart;
713 int dirblock, error, freebytes, i;
716 KASSERT(dh != NULL && dh->dh_hash != NULL,
717 ("ufsdirhash_findfree: Invalid dirhash %p\n", dh));
718 DIRHASH_ASSERT_LOCKED(dh);
720 /* Find a directory block with the desired free space. */
722 for (i = howmany(slotneeded, DIRALIGN); i <= DH_NFSTATS; i++)
723 if ((dirblock = dh->dh_firstfree[i]) != -1)
728 KASSERT(dirblock < dh->dh_nblk &&
729 dh->dh_blkfree[dirblock] >= howmany(slotneeded, DIRALIGN),
730 ("ufsdirhash_findfree: bad stats"));
731 pos = dirblock * DIRBLKSIZ;
732 error = UFS_BLKATOFF(ip->i_vnode, (off_t)pos, (char **)&dp, &bp);
736 /* Find the first entry with free space. */
737 for (i = 0; i < DIRBLKSIZ; ) {
738 if (dp->d_reclen == 0) {
742 if (dp->d_ino == 0 || dp->d_reclen > DIRSIZ(0, dp))
745 dp = (struct direct *)((char *)dp + dp->d_reclen);
753 /* Find the range of entries needed to get enough space */
755 while (i < DIRBLKSIZ && freebytes < slotneeded) {
756 freebytes += dp->d_reclen;
758 freebytes -= DIRSIZ(0, dp);
759 if (dp->d_reclen == 0) {
764 dp = (struct direct *)((char *)dp + dp->d_reclen);
770 if (freebytes < slotneeded)
771 panic("ufsdirhash_findfree: free mismatch");
773 *slotsize = pos + i - slotstart;
778 * Return the start of the unused space at the end of a directory, or
779 * -1 if there are no trailing unused blocks.
782 ufsdirhash_enduseful(struct inode *ip)
789 DIRHASH_ASSERT_LOCKED(dh);
790 KASSERT(dh != NULL && dh->dh_hash != NULL,
791 ("ufsdirhash_enduseful: Invalid dirhash %p\n", dh));
793 if (dh->dh_blkfree[dh->dh_dirblks - 1] != DIRBLKSIZ / DIRALIGN)
796 for (i = dh->dh_dirblks - 1; i >= 0; i--)
797 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN)
800 return ((doff_t)(i + 1) * DIRBLKSIZ);
804 * Insert information into the hash about a new directory entry. dirp
805 * points to a struct direct containing the entry, and offset specifies
806 * the offset of this entry.
809 ufsdirhash_add(struct inode *ip, struct direct *dirp, doff_t offset)
814 if ((dh = ufsdirhash_acquire(ip)) == NULL)
817 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ,
818 ("ufsdirhash_add: bad offset"));
820 * Normal hash usage is < 66%. If the usage gets too high then
821 * remove the hash entirely and let it be rebuilt later.
823 if (dh->dh_hused >= (dh->dh_hlen * 3) / 4) {
824 ufsdirhash_free_locked(ip);
828 /* Find a free hash slot (empty or deleted), and add the entry. */
829 slot = ufsdirhash_hash(dh, dirp->d_name, dirp->d_namlen);
830 while (DH_ENTRY(dh, slot) >= 0)
831 slot = WRAPINCR(slot, dh->dh_hlen);
832 if (DH_ENTRY(dh, slot) == DIRHASH_EMPTY)
834 DH_ENTRY(dh, slot) = offset;
836 /* Update last used time. */
837 dh->dh_lastused = time_second;
839 /* Update the per-block summary info. */
840 ufsdirhash_adjfree(dh, offset, -DIRSIZ(0, dirp));
841 ufsdirhash_release(dh);
845 * Remove the specified directory entry from the hash. The entry to remove
846 * is defined by the name in `dirp', which must exist at the specified
847 * `offset' within the directory.
850 ufsdirhash_remove(struct inode *ip, struct direct *dirp, doff_t offset)
855 if ((dh = ufsdirhash_acquire(ip)) == NULL)
858 KASSERT(offset < dh->dh_dirblks * DIRBLKSIZ,
859 ("ufsdirhash_remove: bad offset"));
861 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, offset);
863 /* Remove the hash entry. */
864 ufsdirhash_delslot(dh, slot);
866 /* Update the per-block summary info. */
867 ufsdirhash_adjfree(dh, offset, DIRSIZ(0, dirp));
868 ufsdirhash_release(dh);
872 * Change the offset associated with a directory entry in the hash. Used
873 * when compacting directory blocks.
876 ufsdirhash_move(struct inode *ip, struct direct *dirp, doff_t oldoff,
882 if ((dh = ufsdirhash_acquire(ip)) == NULL)
885 KASSERT(oldoff < dh->dh_dirblks * DIRBLKSIZ &&
886 newoff < dh->dh_dirblks * DIRBLKSIZ,
887 ("ufsdirhash_move: bad offset"));
888 /* Find the entry, and update the offset. */
889 slot = ufsdirhash_findslot(dh, dirp->d_name, dirp->d_namlen, oldoff);
890 DH_ENTRY(dh, slot) = newoff;
891 ufsdirhash_release(dh);
895 * Inform dirhash that the directory has grown by one block that
896 * begins at offset (i.e. the new length is offset + DIRBLKSIZ).
899 ufsdirhash_newblk(struct inode *ip, doff_t offset)
904 if ((dh = ufsdirhash_acquire(ip)) == NULL)
907 KASSERT(offset == dh->dh_dirblks * DIRBLKSIZ,
908 ("ufsdirhash_newblk: bad offset"));
909 block = offset / DIRBLKSIZ;
910 if (block >= dh->dh_nblk) {
911 /* Out of space; must rebuild. */
912 ufsdirhash_free_locked(ip);
915 dh->dh_dirblks = block + 1;
917 /* Account for the new free block. */
918 dh->dh_blkfree[block] = DIRBLKSIZ / DIRALIGN;
919 if (dh->dh_firstfree[DH_NFSTATS] == -1)
920 dh->dh_firstfree[DH_NFSTATS] = block;
921 ufsdirhash_release(dh);
925 * Inform dirhash that the directory is being truncated.
928 ufsdirhash_dirtrunc(struct inode *ip, doff_t offset)
933 if ((dh = ufsdirhash_acquire(ip)) == NULL)
936 KASSERT(offset <= dh->dh_dirblks * DIRBLKSIZ,
937 ("ufsdirhash_dirtrunc: bad offset"));
938 block = howmany(offset, DIRBLKSIZ);
940 * If the directory shrinks to less than 1/8 of dh_nblk blocks
941 * (about 20% of its original size due to the 50% extra added in
942 * ufsdirhash_build) then free it, and let the caller rebuild
945 if (block < dh->dh_nblk / 8 && dh->dh_narrays > 1) {
946 ufsdirhash_free_locked(ip);
951 * Remove any `first free' information pertaining to the
952 * truncated blocks. All blocks we're removing should be
955 if (dh->dh_firstfree[DH_NFSTATS] >= block)
956 dh->dh_firstfree[DH_NFSTATS] = -1;
957 for (i = block; i < dh->dh_dirblks; i++)
958 if (dh->dh_blkfree[i] != DIRBLKSIZ / DIRALIGN)
959 panic("ufsdirhash_dirtrunc: blocks in use");
960 for (i = 0; i < DH_NFSTATS; i++)
961 if (dh->dh_firstfree[i] >= block)
962 panic("ufsdirhash_dirtrunc: first free corrupt");
963 dh->dh_dirblks = block;
964 ufsdirhash_release(dh);
968 * Debugging function to check that the dirhash information about
969 * a directory block matches its actual contents. Panics if a mismatch
972 * On entry, `buf' should point to the start of an in-core
973 * DIRBLKSIZ-sized directory block, and `offset' should contain the
974 * offset from the start of the directory of that block.
977 ufsdirhash_checkblock(struct inode *ip, char *buf, doff_t offset)
981 int block, ffslot, i, nfree;
983 if (!ufs_dirhashcheck)
985 if ((dh = ufsdirhash_acquire(ip)) == NULL)
988 block = offset / DIRBLKSIZ;
989 if ((offset & (DIRBLKSIZ - 1)) != 0 || block >= dh->dh_dirblks)
990 panic("ufsdirhash_checkblock: bad offset");
993 for (i = 0; i < DIRBLKSIZ; i += dp->d_reclen) {
994 dp = (struct direct *)(buf + i);
995 if (dp->d_reclen == 0 || i + dp->d_reclen > DIRBLKSIZ)
996 panic("ufsdirhash_checkblock: bad dir");
998 if (dp->d_ino == 0) {
1001 * XXX entries with d_ino == 0 should only occur
1002 * at the start of a DIRBLKSIZ block. However the
1003 * ufs code is tolerant of such entries at other
1004 * offsets, and fsck does not fix them.
1007 panic("ufsdirhash_checkblock: bad dir inode");
1009 nfree += dp->d_reclen;
1013 /* Check that the entry exists (will panic if it doesn't). */
1014 ufsdirhash_findslot(dh, dp->d_name, dp->d_namlen, offset + i);
1016 nfree += dp->d_reclen - DIRSIZ(0, dp);
1019 panic("ufsdirhash_checkblock: bad dir end");
1021 if (dh->dh_blkfree[block] * DIRALIGN != nfree)
1022 panic("ufsdirhash_checkblock: bad free count");
1024 ffslot = BLKFREE2IDX(nfree / DIRALIGN);
1025 for (i = 0; i <= DH_NFSTATS; i++)
1026 if (dh->dh_firstfree[i] == block && i != ffslot)
1027 panic("ufsdirhash_checkblock: bad first-free");
1028 if (dh->dh_firstfree[ffslot] == -1)
1029 panic("ufsdirhash_checkblock: missing first-free entry");
1030 ufsdirhash_release(dh);
1034 * Hash the specified filename into a dirhash slot.
1037 ufsdirhash_hash(struct dirhash *dh, char *name, int namelen)
1042 * We hash the name and then some other bit of data that is
1043 * invariant over the dirhash's lifetime. Otherwise names
1044 * differing only in the last byte are placed close to one
1045 * another in the table, which is bad for linear probing.
1047 hash = fnv_32_buf(name, namelen, FNV1_32_INIT);
1048 hash = fnv_32_buf(&dh, sizeof(dh), hash);
1049 return (hash % dh->dh_hlen);
1053 * Adjust the number of free bytes in the block containing `offset'
1054 * by the value specified by `diff'.
1056 * The caller must ensure we have exclusive access to `dh'; normally
1057 * that means that dh_lock should be held, but this is also called
1058 * from ufsdirhash_build() where exclusive access can be assumed.
1061 ufsdirhash_adjfree(struct dirhash *dh, doff_t offset, int diff)
1063 int block, i, nfidx, ofidx;
1065 /* Update the per-block summary info. */
1066 block = offset / DIRBLKSIZ;
1067 KASSERT(block < dh->dh_nblk && block < dh->dh_dirblks,
1068 ("dirhash bad offset"));
1069 ofidx = BLKFREE2IDX(dh->dh_blkfree[block]);
1070 dh->dh_blkfree[block] = (int)dh->dh_blkfree[block] + (diff / DIRALIGN);
1071 nfidx = BLKFREE2IDX(dh->dh_blkfree[block]);
1073 /* Update the `first free' list if necessary. */
1074 if (ofidx != nfidx) {
1075 /* If removing, scan forward for the next block. */
1076 if (dh->dh_firstfree[ofidx] == block) {
1077 for (i = block + 1; i < dh->dh_dirblks; i++)
1078 if (BLKFREE2IDX(dh->dh_blkfree[i]) == ofidx)
1080 dh->dh_firstfree[ofidx] = (i < dh->dh_dirblks) ? i : -1;
1083 /* Make this the new `first free' if necessary */
1084 if (dh->dh_firstfree[nfidx] > block ||
1085 dh->dh_firstfree[nfidx] == -1)
1086 dh->dh_firstfree[nfidx] = block;
1091 * Find the specified name which should have the specified offset.
1092 * Returns a slot number, and panics on failure.
1094 * `dh' must be locked on entry and remains so on return.
1097 ufsdirhash_findslot(struct dirhash *dh, char *name, int namelen, doff_t offset)
1101 DIRHASH_ASSERT_LOCKED(dh);
1103 /* Find the entry. */
1104 KASSERT(dh->dh_hused < dh->dh_hlen, ("dirhash find full"));
1105 slot = ufsdirhash_hash(dh, name, namelen);
1106 while (DH_ENTRY(dh, slot) != offset &&
1107 DH_ENTRY(dh, slot) != DIRHASH_EMPTY)
1108 slot = WRAPINCR(slot, dh->dh_hlen);
1109 if (DH_ENTRY(dh, slot) != offset)
1110 panic("ufsdirhash_findslot: '%.*s' not found", namelen, name);
1116 * Remove the entry corresponding to the specified slot from the hash array.
1118 * `dh' must be locked on entry and remains so on return.
1121 ufsdirhash_delslot(struct dirhash *dh, int slot)
1125 DIRHASH_ASSERT_LOCKED(dh);
1127 /* Mark the entry as deleted. */
1128 DH_ENTRY(dh, slot) = DIRHASH_DEL;
1130 /* If this is the end of a chain of DIRHASH_DEL slots, remove them. */
1131 for (i = slot; DH_ENTRY(dh, i) == DIRHASH_DEL; )
1132 i = WRAPINCR(i, dh->dh_hlen);
1133 if (DH_ENTRY(dh, i) == DIRHASH_EMPTY) {
1134 i = WRAPDECR(i, dh->dh_hlen);
1135 while (DH_ENTRY(dh, i) == DIRHASH_DEL) {
1136 DH_ENTRY(dh, i) = DIRHASH_EMPTY;
1138 i = WRAPDECR(i, dh->dh_hlen);
1140 KASSERT(dh->dh_hused >= 0, ("ufsdirhash_delslot neg hlen"));
1145 * Given a directory entry and its offset, find the offset of the
1146 * previous entry in the same DIRBLKSIZ-sized block. Returns an
1147 * offset, or -1 if there is no previous entry in the block or some
1148 * other problem occurred.
1151 ufsdirhash_getprev(struct direct *dirp, doff_t offset)
1155 doff_t blkoff, prevoff;
1158 blkoff = rounddown2(offset, DIRBLKSIZ); /* offset of start of block */
1159 entrypos = offset & (DIRBLKSIZ - 1); /* entry relative to block */
1160 blkbuf = (char *)dirp - entrypos;
1163 /* If `offset' is the start of a block, there is no previous entry. */
1167 /* Scan from the start of the block until we get to the entry. */
1168 for (i = 0; i < entrypos; i += dp->d_reclen) {
1169 dp = (struct direct *)(blkbuf + i);
1170 if (dp->d_reclen == 0 || i + dp->d_reclen > entrypos)
1171 return (-1); /* Corrupted directory. */
1172 prevoff = blkoff + i;
1178 * Delete the given dirhash and reclaim its memory. Assumes that
1179 * ufsdirhash_list is locked, and leaves it locked. Also assumes
1180 * that dh is locked. Returns the amount of memory freed.
1183 ufsdirhash_destroy(struct dirhash *dh)
1187 int i, mem, narrays;
1189 KASSERT(dh->dh_hash != NULL, ("dirhash: NULL hash on list"));
1191 /* Remove it from the list and detach its memory. */
1192 TAILQ_REMOVE(&ufsdirhash_list, dh, dh_list);
1196 blkfree = dh->dh_blkfree;
1197 dh->dh_blkfree = NULL;
1198 narrays = dh->dh_narrays;
1199 mem = dh->dh_memreq;
1202 /* Unlock dirhash and free the detached memory. */
1203 ufsdirhash_release(dh);
1204 for (i = 0; i < narrays; i++)
1205 DIRHASH_BLKFREE(hash[i]);
1206 free(hash, M_DIRHASH);
1207 free(blkfree, M_DIRHASH);
1209 /* Account for the returned memory. */
1210 ufs_dirhashmem -= mem;
1216 * Try to free up `wanted' bytes by stealing memory from existing
1217 * dirhashes. Returns zero with list locked if successful.
1220 ufsdirhash_recycle(int wanted)
1225 dh = TAILQ_FIRST(&ufsdirhash_list);
1226 while (wanted + ufs_dirhashmem > ufs_dirhashmaxmem) {
1227 /* Decrement the score; only recycle if it becomes zero. */
1228 if (dh == NULL || --dh->dh_score > 0) {
1229 DIRHASHLIST_UNLOCK();
1233 * If we can't lock it it's in use and we don't want to
1234 * recycle it anyway.
1236 if (!sx_try_xlock(&dh->dh_lock)) {
1237 dh = TAILQ_NEXT(dh, dh_list);
1241 ufsdirhash_destroy(dh);
1243 /* Repeat if necessary. */
1244 dh = TAILQ_FIRST(&ufsdirhash_list);
1246 /* Success; return with list locked. */
1251 * Callback that frees some dirhashes when the system is low on virtual memory.
1256 struct dirhash *dh, *dh_temp;
1257 int memfreed, memwanted;
1259 ufs_dirhashlowmemcount++;
1261 memwanted = ufs_dirhashmem * ufs_dirhashreclaimpercent / 100;
1266 * Reclaim up to memwanted from the oldest dirhashes. This will allow
1267 * us to make some progress when the system is running out of memory
1268 * without compromising the dinamicity of maximum age. If the situation
1269 * does not improve lowmem will be eventually retriggered and free some
1270 * other entry in the cache. The entries on the head of the list should
1271 * be the oldest. If during list traversal we can't get a lock on the
1272 * dirhash, it will be skipped.
1274 TAILQ_FOREACH_SAFE(dh, &ufsdirhash_list, dh_list, dh_temp) {
1275 if (sx_try_xlock(&dh->dh_lock))
1276 memfreed += ufsdirhash_destroy(dh);
1277 if (memfreed >= memwanted)
1280 DIRHASHLIST_UNLOCK();
1284 ufsdirhash_set_reclaimpercent(SYSCTL_HANDLER_ARGS)
1288 v = ufs_dirhashreclaimpercent;
1289 error = sysctl_handle_int(oidp, &v, v, req);
1292 if (req->newptr == NULL)
1294 if (v == ufs_dirhashreclaimpercent)
1297 /* Refuse invalid percentages */
1298 if (v < 0 || v > 100)
1300 ufs_dirhashreclaimpercent = v;
1307 ufs_dirhashmaxmem = lmax(roundup(hibufspace / 64, PAGE_SIZE),
1310 ufsdirhash_zone = uma_zcreate("DIRHASH", DH_NBLKOFF * sizeof(doff_t),
1311 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1312 mtx_init(&ufsdirhash_mtx, "dirhash list", NULL, MTX_DEF);
1313 TAILQ_INIT(&ufsdirhash_list);
1315 /* Register a callback function to handle low memory signals */
1316 EVENTHANDLER_REGISTER(vm_lowmem, ufsdirhash_lowmem, NULL,
1317 EVENTHANDLER_PRI_FIRST);
1323 KASSERT(TAILQ_EMPTY(&ufsdirhash_list), ("ufsdirhash_uninit"));
1324 uma_zdestroy(ufsdirhash_zone);
1325 mtx_destroy(&ufsdirhash_mtx);
1328 #endif /* UFS_DIRHASH */