2 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/kern/vfs_lock.c,v 1.30 2008/06/30 03:57:41 dillon Exp $
38 * External virtual filesystem routines
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
46 #include <sys/mount.h>
48 #include <sys/vnode.h>
50 #include <sys/sysctl.h>
52 #include <machine/limits.h>
55 #include <vm/vm_object.h>
58 #include <sys/thread2.h>
59 #include <sys/sysref2.h>
61 static void vnode_terminate(struct vnode *vp);
62 static boolean_t vnode_ctor(void *obj, void *private, int ocflags);
63 static void vnode_dtor(void *obj, void *private);
65 static MALLOC_DEFINE(M_VNODE, "vnodes", "vnode structures");
66 static struct sysref_class vnode_sysref_class = {
69 .proto = SYSREF_PROTO_VNODE,
70 .offset = offsetof(struct vnode, v_sysref),
71 .objsize = sizeof(struct vnode),
73 .flags = SRC_MANAGEDINIT,
77 .terminate = (sysref_terminate_func_t)vnode_terminate,
78 .lock = (sysref_terminate_func_t)vx_lock,
79 .unlock = (sysref_terminate_func_t)vx_unlock
84 * The vnode free list hold inactive vnodes. Aged inactive vnodes
85 * are inserted prior to the mid point, and otherwise inserted
88 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
89 static struct vnode vnode_free_mid1;
90 static struct vnode vnode_free_mid2;
91 static struct vnode vnode_free_rover;
92 static struct spinlock vfs_spin = SPINLOCK_INITIALIZER(vfs_spin);
93 static enum { ROVER_MID1, ROVER_MID2 } rover_state = ROVER_MID2;
96 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD,
97 &freevnodes, 0, "Number of free nodes");
98 static int wantfreevnodes = 25;
99 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW,
100 &wantfreevnodes, 0, "Desired number of free vnodes");
102 static ulong trackvnode;
103 SYSCTL_ULONG(_debug, OID_AUTO, trackvnode, CTLFLAG_RW,
108 * Called from vfsinit()
113 TAILQ_INIT(&vnode_free_list);
114 TAILQ_INSERT_TAIL(&vnode_free_list, &vnode_free_mid1, v_freelist);
115 TAILQ_INSERT_TAIL(&vnode_free_list, &vnode_free_mid2, v_freelist);
116 TAILQ_INSERT_TAIL(&vnode_free_list, &vnode_free_rover, v_freelist);
117 spin_init(&vfs_spin);
118 kmalloc_raise_limit(M_VNODE, 0); /* unlimited */
126 _vsetflags(struct vnode *vp, int flags)
128 atomic_set_int(&vp->v_flag, flags);
133 _vclrflags(struct vnode *vp, int flags)
135 atomic_clear_int(&vp->v_flag, flags);
139 vsetflags(struct vnode *vp, int flags)
141 _vsetflags(vp, flags);
145 vclrflags(struct vnode *vp, int flags)
147 _vclrflags(vp, flags);
151 * Inline helper functions.
153 * WARNING: vbusy() may only be called while the vnode lock or VX lock
154 * is held. The vnode spinlock need not be held.
160 __vbusy_interlocked(struct vnode *vp)
162 KKASSERT(vp->v_flag & VFREE);
163 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
165 _vclrflags(vp, VFREE);
170 __vbusy(struct vnode *vp)
173 if ((ulong)vp == trackvnode)
174 kprintf("__vbusy %p %08x\n", vp, vp->v_flag);
176 spin_lock(&vfs_spin);
177 __vbusy_interlocked(vp);
178 spin_unlock(&vfs_spin);
182 * Put a vnode on the free list. The caller has cleared VCACHED or owns the
183 * implied sysref related to having removed the vnode from the freelist
184 * (and VCACHED is already clear in that case).
190 __vfree(struct vnode *vp)
193 if ((ulong)vp == trackvnode) {
194 kprintf("__vfree %p %08x\n", vp, vp->v_flag);
198 spin_lock(&vfs_spin);
199 KKASSERT((vp->v_flag & VFREE) == 0);
202 * Distinguish between basically dead vnodes, vnodes with cached
203 * data, and vnodes without cached data. A rover will shift the
204 * vnodes around as their cache status is lost.
206 if (vp->v_flag & VRECLAIMED) {
207 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
208 } else if (vp->v_object && vp->v_object->resident_page_count) {
209 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
210 } else if (vp->v_object && vp->v_object->swblock_count) {
211 TAILQ_INSERT_BEFORE(&vnode_free_mid2, vp, v_freelist);
213 TAILQ_INSERT_BEFORE(&vnode_free_mid1, vp, v_freelist);
216 _vsetflags(vp, VFREE);
217 spin_unlock(&vfs_spin);
221 * Put a vnode on the free list. The caller has cleared VCACHED or owns the
222 * implied sysref related to having removed the vnode from the freelist
223 * (and VCACHED is already clear in that case).
229 __vfreetail(struct vnode *vp)
232 if ((ulong)vp == trackvnode)
233 kprintf("__vfreetail %p %08x\n", vp, vp->v_flag);
235 spin_lock(&vfs_spin);
236 KKASSERT((vp->v_flag & VFREE) == 0);
237 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
239 _vsetflags(vp, VFREE);
240 spin_unlock(&vfs_spin);
244 * Return a C boolean if we should put the vnode on the freelist (VFREE),
245 * or leave it / mark it as VCACHED.
247 * This routine is only valid if the vnode is already either VFREE or
248 * VCACHED, or if it can become VFREE or VCACHED via vnode_terminate().
250 * WARNING! This functions is typically called with v_spin held.
254 static __inline boolean_t
255 vshouldfree(struct vnode *vp)
257 return (vp->v_auxrefs == 0 &&
258 (vp->v_object == NULL || vp->v_object->resident_page_count == 0));
262 * Add a ref to an active vnode. This function should never be called
263 * with an inactive vnode (use vget() instead).
268 vref(struct vnode *vp)
270 KKASSERT(vp->v_sysref.refcnt > 0 &&
271 (vp->v_flag & (VFREE|VINACTIVE)) == 0);
272 sysref_get(&vp->v_sysref);
276 * Release a ref on an active or inactive vnode. The sysref termination
277 * function will be called when the active last active reference is released,
278 * and the vnode is returned to the objcache when the last inactive
279 * reference is released.
282 vrele(struct vnode *vp)
284 sysref_put(&vp->v_sysref);
288 * Add an auxiliary data structure reference to the vnode. Auxiliary
289 * references do not change the state of the vnode or prevent them
290 * from being deactivated, reclaimed, or placed on or removed from
293 * An auxiliary reference DOES prevent the vnode from being destroyed,
294 * allowing you to vx_lock() it, test state, etc.
296 * An auxiliary reference DOES NOT move a vnode out of the VFREE state
297 * once it has entered it.
299 * WARNING! vhold() and vhold_interlocked() must not acquire v_spin.
300 * The spinlock may or may not already be held by the caller.
301 * vdrop() will clean up the free list state.
306 vhold(struct vnode *vp)
308 KKASSERT(vp->v_sysref.refcnt != 0);
309 atomic_add_int(&vp->v_auxrefs, 1);
313 vhold_interlocked(struct vnode *vp)
315 atomic_add_int(&vp->v_auxrefs, 1);
319 * Remove an auxiliary reference from the vnode.
321 * vdrop needs to check for a VCACHE->VFREE transition to catch cases
322 * where a vnode is held past its reclamation. We use v_spin to
323 * interlock VCACHED -> !VCACHED transitions.
328 vdrop(struct vnode *vp)
330 KKASSERT(vp->v_sysref.refcnt != 0 && vp->v_auxrefs > 0);
331 spin_lock(&vp->v_spin);
332 atomic_subtract_int(&vp->v_auxrefs, 1);
333 if ((vp->v_flag & VCACHED) && vshouldfree(vp)) {
334 _vclrflags(vp, VCACHED);
337 spin_unlock(&vp->v_spin);
341 * This function is called when the last active reference on the vnode
342 * is released, typically via vrele(). SYSREF will VX lock the vnode
343 * and then give the vnode a negative ref count, indicating that it is
344 * undergoing termination or is being set aside for the cache, and one
345 * final sysref_put() is required to actually return it to the memory
348 * Additional inactive sysrefs may race us but that's ok. Reactivations
349 * cannot race us because the sysref code interlocked with the VX lock
350 * (which is held on call).
355 vnode_terminate(struct vnode *vp)
358 * We own the VX lock, it should not be possible for someone else
359 * to have reactivated the vp.
361 KKASSERT(sysref_isinactive(&vp->v_sysref));
364 * Deactivate the vnode by marking it VFREE or VCACHED.
365 * The vnode can be reactivated from either state until
366 * reclaimed. These states inherit the 'last' sysref on the
369 * NOTE: There may be additional inactive references from
370 * other entities blocking on the VX lock while we hold it,
371 * but this does not prevent us from changing the vnode's
374 * NOTE: The vnode could already be marked inactive. XXX
377 * NOTE: v_mount may be NULL due to assignment to
380 * NOTE: The vnode may be marked inactive with dirty buffers
381 * or dirty pages in its cached VM object still present.
383 * NOTE: VCACHED should not be set on entry. We lose control
384 * of the sysref the instant the vnode is placed on the
385 * free list or when VCACHED is set.
387 * The VX lock is required when transitioning to
388 * +VCACHED but is not sufficient for the vshouldfree()
389 * interlocked test or when transitioning to -VCACHED.
391 if ((vp->v_flag & VINACTIVE) == 0) {
392 _vsetflags(vp, VINACTIVE);
396 spin_lock(&vp->v_spin);
397 KKASSERT((vp->v_flag & (VFREE|VCACHED)) == 0);
401 _vsetflags(vp, VCACHED); /* inactive but not yet free*/
402 spin_unlock(&vp->v_spin);
407 * Physical vnode constructor / destructor. These are only executed on
408 * the backend of the objcache. They are NOT executed on every vnode
409 * allocation or deallocation.
414 vnode_ctor(void *obj, void *private, int ocflags)
416 struct vnode *vp = obj;
418 lwkt_token_init(&vp->v_token, "vnode");
419 lockinit(&vp->v_lock, "vnode", 0, 0);
420 ccms_dataspace_init(&vp->v_ccms);
421 TAILQ_INIT(&vp->v_namecache);
422 RB_INIT(&vp->v_rbclean_tree);
423 RB_INIT(&vp->v_rbdirty_tree);
424 RB_INIT(&vp->v_rbhash_tree);
425 spin_init(&vp->v_spin);
433 vnode_dtor(void *obj, void *private)
435 struct vnode *vp = obj;
437 KKASSERT((vp->v_flag & (VCACHED|VFREE)) == 0);
438 ccms_dataspace_destroy(&vp->v_ccms);
441 /****************************************************************
442 * VX LOCKING FUNCTIONS *
443 ****************************************************************
445 * These functions lock vnodes for reclamation and deactivation related
446 * activities. The caller must already be holding some sort of reference
452 vx_lock(struct vnode *vp)
454 lockmgr(&vp->v_lock, LK_EXCLUSIVE);
458 * The non-blocking version also uses a slightly different mechanic.
459 * This function will explicitly fail not only if it cannot acquire
460 * the lock normally, but also if the caller already holds a lock.
462 * The adjusted mechanic is used to close a loophole where complex
463 * VOP_RECLAIM code can circle around recursively and allocate the
464 * same vnode it is trying to destroy from the freelist.
466 * Any filesystem (aka UFS) which puts LK_CANRECURSE in lk_flags can
467 * cause the incorrect behavior to occur. If not for that lockmgr()
468 * would do the right thing.
471 vx_lock_nonblock(struct vnode *vp)
473 if (lockcountnb(&vp->v_lock))
475 return(lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT));
479 vx_unlock(struct vnode *vp)
481 lockmgr(&vp->v_lock, LK_RELEASE);
484 /****************************************************************
485 * VNODE ACQUISITION FUNCTIONS *
486 ****************************************************************
488 * These functions must be used when accessing a vnode via an auxiliary
489 * reference such as the namecache or free list, or when you wish to
490 * do a combo ref+lock sequence.
492 * These functions are MANDATORY for any code chain accessing a vnode
493 * whos activation state is not known.
495 * vget() can be called with LK_NOWAIT and will return EBUSY if the
496 * lock cannot be immediately acquired.
498 * vget()/vput() are used when reactivation is desired.
500 * vx_get() and vx_put() are used when reactivation is not desired.
503 vget(struct vnode *vp, int flags)
508 * A lock type must be passed
510 if ((flags & LK_TYPE_MASK) == 0) {
511 panic("vget() called with no lock specified!");
516 * Reference the structure and then acquire the lock. 0->1
517 * transitions and refs during termination are allowed here so
518 * call sysref directly.
520 * NOTE: The requested lock might be a shared lock and does
521 * not protect our access to the refcnt or other fields.
523 sysref_get(&vp->v_sysref);
524 if ((error = vn_lock(vp, flags)) != 0) {
526 * The lock failed, undo and return an error.
528 sysref_put(&vp->v_sysref);
529 } else if (vp->v_flag & VRECLAIMED) {
531 * The node is being reclaimed and cannot be reactivated
532 * any more, undo and return ENOENT.
539 * If the vnode is marked VFREE or VCACHED it needs to be
540 * reactivated, otherwise it had better already be active.
541 * VINACTIVE must also be cleared.
543 * In the VFREE/VCACHED case we have to throw away the
544 * sysref that was earmarking those cases and preventing
545 * the vnode from being destroyed. Our sysref is still held.
547 * We are allowed to reactivate the vnode while we hold
548 * the VX lock, assuming it can be reactivated.
550 spin_lock(&vp->v_spin);
551 if (vp->v_flag & VFREE) {
553 sysref_activate(&vp->v_sysref);
554 spin_unlock(&vp->v_spin);
555 sysref_put(&vp->v_sysref);
556 } else if (vp->v_flag & VCACHED) {
557 _vclrflags(vp, VCACHED);
558 sysref_activate(&vp->v_sysref);
559 spin_unlock(&vp->v_spin);
560 sysref_put(&vp->v_sysref);
562 if (sysref_isinactive(&vp->v_sysref)) {
563 sysref_activate(&vp->v_sysref);
564 kprintf("Warning vp %p reactivation race\n",
567 spin_unlock(&vp->v_spin);
569 _vclrflags(vp, VINACTIVE);
578 debug_vput(struct vnode *vp, const char *filename, int line)
580 kprintf("vput(%p) %s:%d\n", vp, filename, line);
591 vput(struct vnode *vp)
600 * XXX The vx_*() locks should use auxrefs, not the main reference counter.
605 vx_get(struct vnode *vp)
607 sysref_get(&vp->v_sysref);
608 lockmgr(&vp->v_lock, LK_EXCLUSIVE);
615 vx_get_nonblock(struct vnode *vp)
619 sysref_get(&vp->v_sysref);
620 error = lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT);
622 sysref_put(&vp->v_sysref);
627 * Relase a VX lock that also held a ref on the vnode.
629 * vx_put needs to check for a VCACHED->VFREE transition to catch the
630 * case where e.g. vnlru issues a vgone*().
635 vx_put(struct vnode *vp)
637 spin_lock(&vp->v_spin);
638 if ((vp->v_flag & VCACHED) && vshouldfree(vp)) {
639 _vclrflags(vp, VCACHED);
642 spin_unlock(&vp->v_spin);
643 lockmgr(&vp->v_lock, LK_RELEASE);
644 sysref_put(&vp->v_sysref);
648 * The rover looks for vnodes past the midline with no cached data and
649 * moves them to before the midline. If we do not do this the midline
650 * can wind up in a degenerate state.
654 vnode_rover_locked(void)
659 * Get the vnode after the rover. The rover roves between mid1 and
660 * the end so the only special vnode it can encounter is mid2.
662 vp = TAILQ_NEXT(&vnode_free_rover, v_freelist);
663 if (vp == &vnode_free_mid2) {
664 vp = TAILQ_NEXT(vp, v_freelist);
665 rover_state = ROVER_MID2;
667 KKASSERT(vp != &vnode_free_mid1);
670 * Start over if we finished the scan.
672 TAILQ_REMOVE(&vnode_free_list, &vnode_free_rover, v_freelist);
674 TAILQ_INSERT_AFTER(&vnode_free_list, &vnode_free_mid1,
675 &vnode_free_rover, v_freelist);
676 rover_state = ROVER_MID1;
679 TAILQ_INSERT_AFTER(&vnode_free_list, vp, &vnode_free_rover, v_freelist);
682 * Shift vp if appropriate.
684 if (vp->v_object && vp->v_object->resident_page_count) {
686 * Promote vnode with resident pages to section 3.
687 * (This case shouldn't happen).
689 if (rover_state == ROVER_MID1) {
690 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
691 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
693 } else if (vp->v_object && vp->v_object->swblock_count) {
695 * Demote vnode with only swap pages to section 2
697 if (rover_state == ROVER_MID2) {
698 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
699 TAILQ_INSERT_BEFORE(&vnode_free_mid2, vp, v_freelist);
703 * Demote vnode with no cached data to section 1
705 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
706 TAILQ_INSERT_BEFORE(&vnode_free_mid1, vp, v_freelist);
711 * Try to reuse a vnode from the free list.
713 * NOTE: The returned vnode is not completely initialized.
715 * WARNING: The freevnodes count can race, NULL can be returned even if
727 for (count = 0; count < freevnodes; count++) {
729 * Try to lock the first vnode on the free list.
732 * We use a bad hack in vx_lock_nonblock() which avoids
733 * the lock order reversal between vfs_spin and v_spin.
734 * This is very fragile code and I don't want to use
737 spin_lock(&vfs_spin);
738 vnode_rover_locked();
739 vnode_rover_locked();
740 vp = TAILQ_FIRST(&vnode_free_list);
741 while (vp == &vnode_free_mid1 || vp == &vnode_free_mid2 ||
742 vp == &vnode_free_rover) {
743 vp = TAILQ_NEXT(vp, v_freelist);
747 if (vx_lock_nonblock(vp)) {
748 KKASSERT(vp->v_flag & VFREE);
749 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
750 TAILQ_INSERT_TAIL(&vnode_free_list,
752 spin_unlock(&vfs_spin);
757 * We inherit the sysref associated the vnode on the free
758 * list. Because VCACHED is clear the vnode will not
759 * be placed back on the free list. We own the sysref
760 * free and clear and thus control the disposition of
763 __vbusy_interlocked(vp);
764 spin_unlock(&vfs_spin);
766 if ((ulong)vp == trackvnode)
767 kprintf("allocfreevnode %p %08x\n", vp, vp->v_flag);
770 * Do not reclaim/reuse a vnode while auxillary refs exists.
771 * This includes namecache refs due to a related ncp being
772 * locked or having children.
774 * We will make this test several times as auxrefs can
775 * get incremented on us without any spinlocks being held
776 * until we have removed all namecache and inode references
779 * Because VCACHED is already in the correct state (cleared)
780 * we cannot race other vdrop()s occuring at the same time
781 * and can safely place vp on the free list.
783 * The free list association reinherits the sysref.
792 * We inherit the reference that was previously associated
793 * with the vnode being on the free list. VCACHED had better
794 * not be set because the reference and VX lock prevents
795 * the sysref from transitioning to an active state.
797 KKASSERT((vp->v_flag & (VINACTIVE|VCACHED)) == VINACTIVE);
798 KKASSERT(sysref_isinactive(&vp->v_sysref));
801 * Holding the VX lock on an inactive vnode prevents it
802 * from being reactivated or reused. New namecache
803 * associations can only be made using active vnodes.
805 * Another thread may be blocked on our vnode lock while
806 * holding a namecache lock. We can only reuse this vnode
807 * if we can clear all namecache associations without
810 * Because VCACHED is already in the correct state (cleared)
811 * we cannot race other vdrop()s occuring at the same time
812 * and can safely place vp on the free list.
814 if ((vp->v_flag & VRECLAIMED) == 0) {
815 if (cache_inval_vp_nonblock(vp)) {
821 /* vnode is still VX locked */
825 * We can reuse the vnode if no primary or auxiliary
826 * references remain other then ours, else put it
827 * back on the free list and keep looking.
829 * Either the free list inherits the last reference
830 * or we fall through and sysref_activate() the last
833 * Since the vnode is in a VRECLAIMED state, no new
834 * namecache associations could have been made.
836 KKASSERT(TAILQ_EMPTY(&vp->v_namecache));
838 !sysref_islastdeactivation(&vp->v_sysref)) {
845 * Return a VX locked vnode suitable for reuse. The caller
846 * inherits the sysref.
854 * Obtain a new vnode from the freelist, allocating more if necessary.
855 * The returned vnode is VX locked & vrefd.
857 * All new vnodes set the VAGE flags. An open() of the vnode will
858 * decrement the (2-bit) flags. Vnodes which are opened several times
859 * are thus retained in the cache over vnodes which are merely stat()d.
864 allocvnode(int lktimeout, int lkflags)
869 * Try to reuse vnodes if we hit the max. This situation only
870 * occurs in certain large-memory (2G+) situations. We cannot
871 * attempt to directly reclaim vnodes due to nasty recursion
874 while (numvnodes - freevnodes > desiredvnodes)
878 * Try to build up as many vnodes as we can before reallocating
879 * from the free list. A vnode on the free list simply means
880 * that it is inactive with no resident pages. It may or may not
881 * have been reclaimed and could have valuable information associated
882 * with it that we shouldn't throw away unless we really need to.
884 * HAMMER NOTE: Re-establishing a vnode is a fairly expensive
885 * operation for HAMMER but this should benefit UFS as well.
887 if (freevnodes >= wantfreevnodes && numvnodes >= desiredvnodes)
888 vp = allocfreevnode();
892 vp = sysref_alloc(&vnode_sysref_class);
893 KKASSERT((vp->v_flag & (VCACHED|VFREE)) == 0);
894 lockmgr(&vp->v_lock, LK_EXCLUSIVE);
899 * We are using a managed sysref class, vnode fields are only
900 * zerod on initial allocation from the backing store, not
901 * on reallocation. Thus we have to clear these fields for both
902 * reallocation and reuse.
906 panic("cleaned vnode isn't");
907 if (bio_track_active(&vp->v_track_read) ||
908 bio_track_active(&vp->v_track_write)) {
909 panic("Clean vnode has pending I/O's");
911 if (vp->v_flag & VONWORKLST)
912 panic("Clean vnode still pending on syncer worklist!");
913 if (!RB_EMPTY(&vp->v_rbdirty_tree))
914 panic("Clean vnode still has dirty buffers!");
915 if (!RB_EMPTY(&vp->v_rbclean_tree))
916 panic("Clean vnode still has clean buffers!");
917 if (!RB_EMPTY(&vp->v_rbhash_tree))
918 panic("Clean vnode still on hash tree!");
919 KKASSERT(vp->v_mount == NULL);
921 vp->v_flag = VAGE0 | VAGE1;
928 vp->v_writecount = 0; /* XXX */
931 * lktimeout only applies when LK_TIMELOCK is used, and only
932 * the pageout daemon uses it. The timeout may not be zero
933 * or the pageout daemon can deadlock in low-VM situations.
937 lockreinit(&vp->v_lock, "vnode", lktimeout, lkflags);
938 KKASSERT(TAILQ_EMPTY(&vp->v_namecache));
939 /* exclusive lock still held */
942 * Note: sysref needs to be activated to convert -0x40000000 to +1.
943 * The -0x40000000 comes from the last ref on reuse, and from
944 * sysref_init() on allocate.
946 sysref_activate(&vp->v_sysref);
947 vp->v_filesize = NOOFFSET;
953 KKASSERT(vp->v_mount == NULL);
962 freesomevnodes(int n)
969 if ((vp = allocfreevnode()) == NULL)