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 1.20 2006/08/09 22:47:32 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>
61 static MALLOC_DEFINE(M_VNODE, "vnodes", "vnode structures");
63 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
66 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD,
68 static int wantfreevnodes = 25;
69 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW,
70 &wantfreevnodes, 0, "");
72 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
73 &minvnodes, 0, "Minimum number of vnodes");
76 * Called from vfsinit()
81 minvnodes = desiredvnodes / 4;
83 TAILQ_INIT(&vnode_free_list);
87 * Inline helper functions. vbusy() and vfree() must be called while in a
90 * Warning: must be callable if the caller holds a read spinlock to something
91 * else, meaning we can't use read spinlocks here.
95 __vbusy(struct vnode *vp)
97 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
99 vp->v_flag &= ~(VFREE|VAGE);
104 __vfree(struct vnode *vp)
106 if (vp->v_flag & (VAGE|VRECLAIMED))
107 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
109 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
116 * Return 1 if we can immediately place the vnode on the freelist.
119 vshouldfree(struct vnode *vp, int usecount)
121 if (vp->v_flag & VFREE)
122 return (0); /* already free */
123 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount)
124 return (0); /* other holderse */
126 (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
133 * Reference a vnode or release the reference on a vnode. The vnode will
134 * be taken off the freelist if it is on it and cannot be recycled or
135 * deactivated while refd. The last release of a vnode will deactivate the
136 * vnode via VOP_INACTIVE().
138 * Special cases: refing a vnode does not clear VINACTIVE, you have to vget()
139 * the vnode shared or exclusive to do that.
141 * Warning: must be callable if the caller holds a read spinlock to something
142 * else, meaning we can't use read spinlocks here.
146 __vref(struct vnode *vp)
149 if (vp->v_flag & VFREE)
154 * Add another ref to a vnode. The vnode must already have at least one
158 vref(struct vnode *vp)
160 KKASSERT(vp->v_usecount > 0 && (vp->v_flag & (VFREE|VINACTIVE)) == 0);
161 atomic_add_int(&vp->v_usecount, 1);
165 * Add a ref to a vnode which may not have any refs. This routine is called
166 * from the namecache and vx_get(). The vnode is explicitly removed from
167 * the freelist if necessary. If requested, the vnode will be reactivated.
169 * vget(), cache_vget(), and cache_vref() reactives vnodes. vx_get() does
173 vref_initial(struct vnode *vp, int reactivate)
178 vp->v_flag &= ~VINACTIVE;
183 vrele(struct vnode *vp)
186 if (vp->v_usecount == 1) {
187 KASSERT(lockcountnb(&vp->v_lock) == 0,
188 ("last vrele vp %p still locked", vp));
191 * Deactivation requires an exclusive v_lock (vx_lock()), and
192 * only occurs if the usecount is still 1 after locking.
194 if ((vp->v_flag & VINACTIVE) == 0) {
195 if (vx_lock(vp) == 0) {
196 if ((vp->v_flag & VINACTIVE) == 0 &&
197 vp->v_usecount == 1) {
198 vp->v_flag |= VINACTIVE;
204 if (vshouldfree(vp, 1))
207 KKASSERT(vp->v_usecount > 0);
214 * Hold a vnode, preventing it from being recycled (unless it is already
215 * undergoing a recyclement or already has been recycled).
217 * NOTE: vhold() will defer removal of the vnode from the freelist.
220 vhold(struct vnode *vp)
222 atomic_add_int(&vp->v_holdcnt, 1);
226 vdrop(struct vnode *vp)
229 if (vp->v_holdcnt == 1) {
231 if (vshouldfree(vp, 0))
235 KKASSERT(vp->v_holdcnt > 0);
240 /****************************************************************
241 * VX LOCKING FUNCTIONS *
242 ****************************************************************
244 * These functions lock vnodes for reclamation and deactivation related
245 * activities. Only vp->v_lock, the top layer of the VFS, is locked.
246 * You must be holding a normal reference in order to be able to safely
247 * call vx_lock() and vx_unlock().
249 * vx_get() also differs from vget() in that it does not clear the
250 * VINACTIVE bit on a vnode.
253 #define VXLOCKFLAGS (LK_EXCLUSIVE|LK_RETRY)
254 #define VXLOCKFLAGS_NB (LK_EXCLUSIVE|LK_NOWAIT)
257 __vxlock(struct vnode *vp, int flags)
259 return(lockmgr(&vp->v_lock, flags));
263 __vxunlock(struct vnode *vp)
265 lockmgr(&vp->v_lock, LK_RELEASE);
269 vx_lock(struct vnode *vp)
271 return(__vxlock(vp, VXLOCKFLAGS));
275 vx_unlock(struct vnode *vp)
281 vx_get(struct vnode *vp)
286 if ((error = __vxlock(vp, VXLOCKFLAGS)) != 0)
292 vx_get_nonblock(struct vnode *vp)
297 if ((error = __vxlock(vp, VXLOCKFLAGS_NB)) != 0)
303 vx_put(struct vnode *vp)
309 /****************************************************************
310 * VNODE ACQUISITION FUNCTIONS *
311 ****************************************************************
313 * vget() and vput() access a vnode for the intent of executing an
314 * operation other then a reclamation or deactivation. vget() will ref
315 * and lock the vnode, vput() will unlock and deref the vnode.
316 * The VOP_*() locking functions are used.
318 * CALLING VGET IS MANDATORY PRIOR TO ANY MODIFYING OPERATION ON A VNODE.
319 * This is because vget handles the VINACTIVE interlock and is responsible
320 * for clearing the bit. If the bit is not cleared inode updates may not
323 * Special cases: If vget()'s locking operation fails the vrele() call may
324 * cause the vnode to be deactivated (VOP_INACTIVE called). However, this
325 * never occurs if the vnode is in a reclaimed state. Vnodes in reclaimed
326 * states always return an error code of ENOENT.
328 * Special cases: vput() will unlock and, if it is the last reference,
329 * deactivate the vnode. The deactivation uses a separate non-layered
330 * VX lock after the normal unlock. XXX make it more efficient.
333 vget(struct vnode *vp, int flags)
339 if (flags & LK_TYPE_MASK) {
340 if ((error = vn_lock(vp, flags)) != 0) {
342 } else if (vp->v_flag & VRECLAIMED) {
347 vp->v_flag &= ~VINACTIVE;
351 panic("vget() called with no lock specified!");
352 error = ENOENT; /* not reached, compiler opt */
359 vput(struct vnode *vp)
366 vsetflags(struct vnode *vp, int flags)
374 vclrflags(struct vnode *vp, int flags)
377 vp->v_flag &= ~flags;
382 * Obtain a new vnode from the freelist, allocating more if necessary.
383 * The returned vnode is VX locked & refd.
386 allocvnode(int lktimeout, int lkflags)
392 * Try to reuse vnodes if we hit the max. This situation only
393 * occurs in certain large-memory (2G+) situations. We cannot
394 * attempt to directly reclaim vnodes due to nasty recursion
397 while (numvnodes - freevnodes > desiredvnodes)
404 * Attempt to reuse a vnode already on the free list, allocating
405 * a new vnode if we can't find one or if we have not reached a
406 * good minimum for good LRU performance.
408 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
411 for (count = 0; count < freevnodes; count++) {
415 * Pull the next vnode off the free list and do some
416 * sanity checks. Note that regardless of how we
417 * block, if freevnodes is non-zero there had better
418 * be something on the list.
420 vp = TAILQ_FIRST(&vnode_free_list);
422 panic("getnewvnode: free vnode isn't");
425 * Note the lack of a critical section. We vx_get()
426 * the vnode before we check it for validity, reducing
427 * the number of checks we have to make. The vx_get()
428 * will also pull the vnode off of the freelist.
436 * Can this vnode be recycled? It must be in a
437 * VINACTIVE state with only our reference to it.
438 * (vx_get(), unlike vget(), does not reactivate
439 * the vnode). vx_put() will recycle it onto the
440 * end of the freelist.
442 if ((vp->v_flag & VINACTIVE) == 0 ||
443 vp->v_holdcnt || vp->v_usecount != 1) {
450 * Ok, we can reclaim the vnode if it isn't already
451 * in a reclaimed state. If the reclamation fails,
452 * or if someone else is referencing the vnode after
453 * we have vgone()'d it, we stop here.
455 if ((vp->v_flag & VRECLAIMED) == 0) {
457 if ((vp->v_flag & VRECLAIMED) == 0 ||
458 vp->v_holdcnt || vp->v_usecount != 1) {
464 KKASSERT(vp->v_flag & VINACTIVE);
474 * If we have a vp it will be refd and VX locked.
479 panic("cleaned vnode isn't");
480 if (vp->v_track_read.bk_active + vp->v_track_write.bk_active)
481 panic("Clean vnode has pending I/O's");
482 KKASSERT(vp->v_mount == NULL);
491 vp->v_writecount = 0; /* XXX */
492 lockreinit(&vp->v_lock, "vnode", lktimeout, lkflags);
493 KKASSERT(TAILQ_FIRST(&vp->v_namecache) == NULL);
496 * A brand-new vnode (we could use malloc() here I think) XXX
498 vp = malloc(sizeof(struct vnode), M_VNODE, M_WAITOK|M_ZERO);
499 lwkt_token_init(&vp->v_pollinfo.vpi_token);
500 lockinit(&vp->v_lock, "vnode", lktimeout, lkflags);
501 TAILQ_INIT(&vp->v_namecache);
504 * short cut around vfreeing it and looping, just set it up
505 * as if we had pulled a reclaimed vnode off the freelist
506 * and reinitialized it.
509 if (__vxlock(vp, VXLOCKFLAGS))
510 panic("getnewvnode: __vxlock failed");
514 RB_INIT(&vp->v_rbclean_tree);
515 RB_INIT(&vp->v_rbdirty_tree);
516 RB_INIT(&vp->v_rbhash_tree);
517 vp->v_filesize = NOOFFSET;
522 KKASSERT(vp->v_mount == NULL);