2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.31 2004/05/21 15:41:23 drhodus Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/namei.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
71 #include <machine/limits.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
85 #include <sys/thread2.h>
87 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
89 static void insmntque (struct vnode *vp, struct mount *mp);
90 static void vclean (struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td);
91 static unsigned long numvnodes;
92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
94 enum vtype iftovt_tab[16] = {
95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
100 S_IFSOCK, S_IFIFO, S_IFMT,
103 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
105 static u_long wantfreevnodes = 25;
106 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
107 static u_long freevnodes = 0;
108 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
110 static int reassignbufcalls;
111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
112 static int reassignbufloops;
113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
114 static int reassignbufsortgood;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
116 static int reassignbufsortbad;
117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
118 static int reassignbufmethod = 1;
119 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
121 #ifdef ENABLE_VFS_IOOPT
123 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
126 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
127 struct lwkt_token mountlist_token;
128 struct lwkt_token mntvnode_token;
129 int nfs_mount_type = -1;
130 static struct lwkt_token mntid_token;
131 static struct lwkt_token vnode_free_list_token;
132 static struct lwkt_token spechash_token;
133 struct nfs_public nfs_pub; /* publicly exported FS */
134 static vm_zone_t vnode_zone;
137 * The workitem queue.
139 #define SYNCER_MAXDELAY 32
140 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
141 time_t syncdelay = 30; /* max time to delay syncing data */
142 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
143 "VFS data synchronization delay");
144 time_t filedelay = 30; /* time to delay syncing files */
145 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
146 "File synchronization delay");
147 time_t dirdelay = 29; /* time to delay syncing directories */
148 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
149 "Directory synchronization delay");
150 time_t metadelay = 28; /* time to delay syncing metadata */
151 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
152 "VFS metadata synchronization delay");
153 static int rushjob; /* number of slots to run ASAP */
154 static int stat_rush_requests; /* number of times I/O speeded up */
155 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
157 static int syncer_delayno = 0;
158 static long syncer_mask;
159 LIST_HEAD(synclist, vnode);
160 static struct synclist *syncer_workitem_pending;
163 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
164 &desiredvnodes, 0, "Maximum number of vnodes");
165 static int minvnodes;
166 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
167 &minvnodes, 0, "Minimum number of vnodes");
168 static int vnlru_nowhere = 0;
169 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
170 "Number of times the vnlru process ran without success");
172 static void vfs_free_addrlist (struct netexport *nep);
173 static int vfs_free_netcred (struct radix_node *rn, void *w);
174 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
175 struct export_args *argp);
177 #define VSHOULDFREE(vp) \
178 (!((vp)->v_flag & (VFREE|VDOOMED)) && \
179 !(vp)->v_holdcnt && !(vp)->v_usecount && \
180 (!(vp)->v_object || \
181 !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
183 #define VMIGHTFREE(vp) \
184 (((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) == 0 && \
185 cache_leaf_test(vp) == 0 && (vp)->v_usecount == 0)
187 #define VSHOULDBUSY(vp) \
188 (((vp)->v_flag & VFREE) && \
189 ((vp)->v_holdcnt || (vp)->v_usecount))
191 static void vbusy(struct vnode *vp);
192 static void vfree(struct vnode *vp);
193 static void vmaybefree(struct vnode *vp);
195 extern int dev_ref_debug;
198 * NOTE: the vnode interlock must be held on call.
201 vmaybefree(struct vnode *vp)
208 * Initialize the vnode management data structures.
215 * Desired vnodes is a result of the physical page count
216 * and the size of kernel's heap. It scales in proportion
217 * to the amount of available physical memory. This can
218 * cause trouble on 64-bit and large memory platforms.
220 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
222 min(maxproc + vmstats.v_page_count /4,
223 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
224 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
226 minvnodes = desiredvnodes / 4;
227 lwkt_token_init(&mountlist_token);
228 lwkt_token_init(&mntvnode_token);
229 lwkt_token_init(&mntid_token);
230 lwkt_token_init(&spechash_token);
231 TAILQ_INIT(&vnode_free_list);
232 lwkt_token_init(&vnode_free_list_token);
233 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
235 * Initialize the filesystem syncer.
237 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
239 syncer_maxdelay = syncer_mask + 1;
243 * Mark a mount point as busy. Used to synchronize access and to delay
244 * unmounting. Interlock is not released on failure.
247 vfs_busy(struct mount *mp, int flags, lwkt_tokref_t interlkp, struct thread *td)
251 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
252 if (flags & LK_NOWAIT)
254 mp->mnt_kern_flag |= MNTK_MWAIT;
256 * Since all busy locks are shared except the exclusive
257 * lock granted when unmounting, the only place that a
258 * wakeup needs to be done is at the release of the
259 * exclusive lock at the end of dounmount.
261 * note: interlkp is a serializer and thus can be safely
262 * held through any sleep
264 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
267 lkflags = LK_SHARED | LK_NOPAUSE;
269 lkflags |= LK_INTERLOCK;
270 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
271 panic("vfs_busy: unexpected lock failure");
276 * Free a busy filesystem.
279 vfs_unbusy(struct mount *mp, struct thread *td)
281 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
285 * Lookup a filesystem type, and if found allocate and initialize
286 * a mount structure for it.
288 * Devname is usually updated by mount(8) after booting.
291 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
293 struct thread *td = curthread; /* XXX */
294 struct vfsconf *vfsp;
297 if (fstypename == NULL)
299 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
300 if (!strcmp(vfsp->vfc_name, fstypename))
304 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
305 bzero((char *)mp, (u_long)sizeof(struct mount));
306 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
307 vfs_busy(mp, LK_NOWAIT, NULL, td);
308 TAILQ_INIT(&mp->mnt_nvnodelist);
309 TAILQ_INIT(&mp->mnt_reservedvnlist);
310 mp->mnt_nvnodelistsize = 0;
312 mp->mnt_op = vfsp->vfc_vfsops;
313 mp->mnt_flag = MNT_RDONLY;
314 mp->mnt_vnodecovered = NULLVP;
315 vfsp->vfc_refcount++;
316 mp->mnt_iosize_max = DFLTPHYS;
317 mp->mnt_stat.f_type = vfsp->vfc_typenum;
318 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
319 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
320 mp->mnt_stat.f_mntonname[0] = '/';
321 mp->mnt_stat.f_mntonname[1] = 0;
322 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
328 * Lookup a mount point by filesystem identifier.
337 lwkt_gettoken(&ilock, &mountlist_token);
338 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
339 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
340 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
344 lwkt_reltoken(&ilock);
349 * Get a new unique fsid. Try to make its val[0] unique, since this value
350 * will be used to create fake device numbers for stat(). Also try (but
351 * not so hard) make its val[0] unique mod 2^16, since some emulators only
352 * support 16-bit device numbers. We end up with unique val[0]'s for the
353 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
355 * Keep in mind that several mounts may be running in parallel. Starting
356 * the search one past where the previous search terminated is both a
357 * micro-optimization and a defense against returning the same fsid to
364 static u_int16_t mntid_base;
369 lwkt_gettoken(&ilock, &mntid_token);
370 mtype = mp->mnt_vfc->vfc_typenum;
371 tfsid.val[1] = mtype;
372 mtype = (mtype & 0xFF) << 24;
374 tfsid.val[0] = makeudev(255,
375 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
377 if (vfs_getvfs(&tfsid) == NULL)
380 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
381 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
382 lwkt_reltoken(&ilock);
386 * Knob to control the precision of file timestamps:
388 * 0 = seconds only; nanoseconds zeroed.
389 * 1 = seconds and nanoseconds, accurate within 1/HZ.
390 * 2 = seconds and nanoseconds, truncated to microseconds.
391 * >=3 = seconds and nanoseconds, maximum precision.
393 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
395 static int timestamp_precision = TSP_SEC;
396 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
397 ×tamp_precision, 0, "");
400 * Get a current timestamp.
404 struct timespec *tsp;
408 switch (timestamp_precision) {
410 tsp->tv_sec = time_second;
418 TIMEVAL_TO_TIMESPEC(&tv, tsp);
428 * Set vnode attributes to VNOVAL
436 vap->va_size = VNOVAL;
437 vap->va_bytes = VNOVAL;
438 vap->va_mode = VNOVAL;
439 vap->va_nlink = VNOVAL;
440 vap->va_uid = VNOVAL;
441 vap->va_gid = VNOVAL;
442 vap->va_fsid = VNOVAL;
443 vap->va_fileid = VNOVAL;
444 vap->va_blocksize = VNOVAL;
445 vap->va_rdev = VNOVAL;
446 vap->va_atime.tv_sec = VNOVAL;
447 vap->va_atime.tv_nsec = VNOVAL;
448 vap->va_mtime.tv_sec = VNOVAL;
449 vap->va_mtime.tv_nsec = VNOVAL;
450 vap->va_ctime.tv_sec = VNOVAL;
451 vap->va_ctime.tv_nsec = VNOVAL;
452 vap->va_flags = VNOVAL;
453 vap->va_gen = VNOVAL;
458 * This routine is called when we have too many vnodes. It attempts
459 * to free <count> vnodes and will potentially free vnodes that still
460 * have VM backing store (VM backing store is typically the cause
461 * of a vnode blowout so we want to do this). Therefore, this operation
462 * is not considered cheap.
464 * A number of conditions may prevent a vnode from being reclaimed.
465 * the buffer cache may have references on the vnode, a directory
466 * vnode may still have references due to the namei cache representing
467 * underlying files, or the vnode may be in active use. It is not
468 * desireable to reuse such vnodes. These conditions may cause the
469 * number of vnodes to reach some minimum value regardless of what
470 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
473 vlrureclaim(struct mount *mp)
484 * Calculate the trigger point, don't allow user
485 * screwups to blow us up. This prevents us from
486 * recycling vnodes with lots of resident pages. We
487 * aren't trying to free memory, we are trying to
490 usevnodes = desiredvnodes;
493 trigger = vmstats.v_page_count * 2 / usevnodes;
496 lwkt_gettoken(&ilock, &mntvnode_token);
497 count = mp->mnt_nvnodelistsize / 10 + 1;
498 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
502 * The VP will stick around while we hold mntvnode_token,
503 * at least until we block, so we can safely do an initial
504 * check. But we have to check again after obtaining
505 * the vnode interlock. vp->v_interlock points to stable
506 * storage so it's ok if the vp gets ripped out from
507 * under us while we are blocked.
509 if (vp->v_type == VNON ||
510 vp->v_type == VBAD ||
511 !VMIGHTFREE(vp) || /* critical path opt */
513 vp->v_object->resident_page_count >= trigger)
515 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
516 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
522 * Get the interlock, delay moving the node to the tail so
523 * we don't race against new additions to the mountlist.
525 lwkt_gettoken(&vlock, vp->v_interlock);
526 if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
527 lwkt_reltoken(&vlock);
530 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
531 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
536 if (vp->v_type == VNON ||
537 vp->v_type == VBAD ||
538 !VMIGHTFREE(vp) || /* critical path opt */
540 vp->v_object->resident_page_count >= trigger)
542 lwkt_reltoken(&vlock);
546 vgonel(vp, &vlock, curthread);
550 lwkt_reltoken(&ilock);
555 * Attempt to recycle vnodes in a context that is always safe to block.
556 * Calling vlrurecycle() from the bowels of file system code has some
557 * interesting deadlock problems.
559 static struct thread *vnlruthread;
560 static int vnlruproc_sig;
565 struct mount *mp, *nmp;
569 struct thread *td = curthread;
571 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
576 kproc_suspend_loop();
577 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
579 wakeup(&vnlruproc_sig);
580 tsleep(td, 0, "vlruwt", hz);
584 lwkt_gettoken(&ilock, &mountlist_token);
585 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
586 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
587 nmp = TAILQ_NEXT(mp, mnt_list);
590 done += vlrureclaim(mp);
591 lwkt_gettokref(&ilock);
592 nmp = TAILQ_NEXT(mp, mnt_list);
595 lwkt_reltoken(&ilock);
598 tsleep(td, 0, "vlrup", hz * 3);
604 static struct kproc_desc vnlru_kp = {
609 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
612 * Routines having to do with the management of the vnode table.
614 extern vop_t **dead_vnodeop_p;
617 * Return the next vnode from the free list.
620 getnewvnode(tag, mp, vops, vpp)
627 struct thread *td = curthread; /* XXX */
628 struct vnode *vp = NULL;
637 * Try to reuse vnodes if we hit the max. This situation only
638 * occurs in certain large-memory (2G+) situations. We cannot
639 * attempt to directly reclaim vnodes due to nasty recursion
642 while (numvnodes - freevnodes > desiredvnodes) {
643 if (vnlruproc_sig == 0) {
644 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
647 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
652 * Attempt to reuse a vnode already on the free list, allocating
653 * a new vnode if we can't find one or if we have not reached a
654 * good minimum for good LRU performance.
656 lwkt_gettoken(&ilock, &vnode_free_list_token);
657 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
660 for (count = 0; count < freevnodes; count++) {
664 * Pull the next vnode off the free list and do some
665 * sanity checks. Note that regardless of how we
666 * block, if freevnodes is non-zero there had better
667 * be something on the list.
669 vp = TAILQ_FIRST(&vnode_free_list);
671 panic("getnewvnode: free vnode isn't");
674 * Move the vnode to the end of the list so other
675 * processes do not double-block trying to recycle
676 * the same vnode (as an optimization), then get
679 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
680 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
683 * Skip vnodes that are in the process of being
684 * held or referenced. Since the act of adding or
685 * removing a vnode on the freelist requires a token
686 * and may block, the ref count may be adjusted
687 * prior to its addition or removal.
689 if (VSHOULDBUSY(vp)) {
696 * Obtain the vnode interlock and check that the
697 * vnode is still on the free list.
699 * This normally devolves into a degenerate case so
700 * it is optimal. Loop up if it isn't. Note that
701 * the vnode could be in the middle of being moved
702 * off the free list (the VSHOULDBUSY() check) and
703 * must be skipped if so.
705 lwkt_gettoken(&vlock, vp->v_interlock);
706 TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
707 freelst, v_freelist) {
711 if (vp != xvp || VSHOULDBUSY(vp)) {
717 * We now safely own the vnode. If the vnode has
718 * an object do not recycle it if its VM object
719 * has resident pages or references.
721 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
722 (object->resident_page_count || object->ref_count))
724 lwkt_reltoken(&vlock);
730 * We can almost reuse this vnode. But we don't want
731 * to recycle it if the vnode has children in the
732 * namecache because that breaks the namecache's
733 * path element chain. (YYY use nc_refs for the
736 KKASSERT(vp->v_flag & VFREE);
737 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
739 if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
740 cache_leaf_test(vp) >= 0) {
741 /* ok, we can reuse this vnode */
744 lwkt_reltoken(&vlock);
745 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
751 * If vp is non-NULL we hold it's interlock.
754 vp->v_flag |= VDOOMED;
755 vp->v_flag &= ~VFREE;
757 lwkt_reltoken(&ilock);
758 cache_purge(vp); /* YYY may block */
760 if (vp->v_type != VBAD) {
761 vgonel(vp, &vlock, td);
763 lwkt_reltoken(&vlock);
771 panic("cleaned vnode isn't");
774 panic("Clean vnode has pending I/O's");
784 vp->v_writecount = 0; /* XXX */
786 lwkt_reltoken(&ilock);
787 vp = zalloc(vnode_zone);
788 bzero(vp, sizeof(*vp));
789 vp->v_interlock = lwkt_token_pool_get(vp);
790 lwkt_token_init(&vp->v_pollinfo.vpi_token);
792 TAILQ_INIT(&vp->v_namecache);
796 TAILQ_INIT(&vp->v_cleanblkhd);
797 TAILQ_INIT(&vp->v_dirtyblkhd);
807 vfs_object_create(vp, td);
812 * Move a vnode from one mount queue to another.
821 lwkt_gettoken(&ilock, &mntvnode_token);
823 * Delete from old mount point vnode list, if on one.
825 if (vp->v_mount != NULL) {
826 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
827 ("bad mount point vnode list size"));
828 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
829 vp->v_mount->mnt_nvnodelistsize--;
832 * Insert into list of vnodes for the new mount point, if available.
834 if ((vp->v_mount = mp) == NULL) {
835 lwkt_reltoken(&ilock);
838 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
839 mp->mnt_nvnodelistsize++;
840 lwkt_reltoken(&ilock);
844 * Update outstanding I/O count and do wakeup if requested.
852 bp->b_flags &= ~B_WRITEINPROG;
853 if ((vp = bp->b_vp)) {
855 if (vp->v_numoutput < 0)
856 panic("vwakeup: neg numoutput");
857 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
858 vp->v_flag &= ~VBWAIT;
859 wakeup((caddr_t) &vp->v_numoutput);
865 * Flush out and invalidate all buffers associated with a vnode.
866 * Called with the underlying object locked.
869 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
870 int slpflag, int slptimeo)
873 struct buf *nbp, *blist;
878 if (flags & V_SAVE) {
880 while (vp->v_numoutput) {
881 vp->v_flag |= VBWAIT;
882 error = tsleep((caddr_t)&vp->v_numoutput,
883 slpflag, "vinvlbuf", slptimeo);
889 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
891 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
894 if (vp->v_numoutput > 0 ||
895 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
896 panic("vinvalbuf: dirty bufs");
902 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
904 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
908 for (bp = blist; bp; bp = nbp) {
909 nbp = TAILQ_NEXT(bp, b_vnbufs);
910 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
911 error = BUF_TIMELOCK(bp,
912 LK_EXCLUSIVE | LK_SLEEPFAIL,
913 "vinvalbuf", slpflag, slptimeo);
920 * XXX Since there are no node locks for NFS, I
921 * believe there is a slight chance that a delayed
922 * write will occur while sleeping just above, so
923 * check for it. Note that vfs_bio_awrite expects
924 * buffers to reside on a queue, while VOP_BWRITE and
927 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
930 if (bp->b_vp == vp) {
931 if (bp->b_flags & B_CLUSTEROK) {
936 bp->b_flags |= B_ASYNC;
937 VOP_BWRITE(bp->b_vp, bp);
941 (void) VOP_BWRITE(bp->b_vp, bp);
946 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
947 bp->b_flags &= ~B_ASYNC;
953 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
954 * have write I/O in-progress but if there is a VM object then the
955 * VM object can also have read-I/O in-progress.
958 while (vp->v_numoutput > 0) {
959 vp->v_flag |= VBWAIT;
960 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
962 if (VOP_GETVOBJECT(vp, &object) == 0) {
963 while (object->paging_in_progress)
964 vm_object_pip_sleep(object, "vnvlbx");
966 } while (vp->v_numoutput > 0);
971 * Destroy the copy in the VM cache, too.
973 lwkt_gettoken(&vlock, vp->v_interlock);
974 if (VOP_GETVOBJECT(vp, &object) == 0) {
975 vm_object_page_remove(object, 0, 0,
976 (flags & V_SAVE) ? TRUE : FALSE);
978 lwkt_reltoken(&vlock);
980 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
981 panic("vinvalbuf: flush failed");
986 * Truncate a file's buffer and pages to a specified length. This
987 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
991 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
999 * Round up to the *next* lbn.
1001 trunclbn = (length + blksize - 1) / blksize;
1008 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1009 nbp = TAILQ_NEXT(bp, b_vnbufs);
1010 if (bp->b_lblkno >= trunclbn) {
1011 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1012 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1016 bp->b_flags |= (B_INVAL | B_RELBUF);
1017 bp->b_flags &= ~B_ASYNC;
1022 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1023 (nbp->b_vp != vp) ||
1024 (nbp->b_flags & B_DELWRI))) {
1030 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1031 nbp = TAILQ_NEXT(bp, b_vnbufs);
1032 if (bp->b_lblkno >= trunclbn) {
1033 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1034 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1038 bp->b_flags |= (B_INVAL | B_RELBUF);
1039 bp->b_flags &= ~B_ASYNC;
1044 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1045 (nbp->b_vp != vp) ||
1046 (nbp->b_flags & B_DELWRI) == 0)) {
1055 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1056 nbp = TAILQ_NEXT(bp, b_vnbufs);
1057 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1058 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1059 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1063 if (bp->b_vp == vp) {
1064 bp->b_flags |= B_ASYNC;
1066 bp->b_flags &= ~B_ASYNC;
1068 VOP_BWRITE(bp->b_vp, bp);
1076 while (vp->v_numoutput > 0) {
1077 vp->v_flag |= VBWAIT;
1078 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1083 vnode_pager_setsize(vp, length);
1089 * Associate a buffer with a vnode.
1098 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1102 bp->b_dev = vn_todev(vp);
1104 * Insert onto list for new vnode.
1107 bp->b_xflags |= BX_VNCLEAN;
1108 bp->b_xflags &= ~BX_VNDIRTY;
1109 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1114 * Disassociate a buffer from a vnode.
1121 struct buflists *listheadp;
1124 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1127 * Delete from old vnode list, if on one.
1131 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1132 if (bp->b_xflags & BX_VNDIRTY)
1133 listheadp = &vp->v_dirtyblkhd;
1135 listheadp = &vp->v_cleanblkhd;
1136 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1137 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1139 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1140 vp->v_flag &= ~VONWORKLST;
1141 LIST_REMOVE(vp, v_synclist);
1144 bp->b_vp = (struct vnode *) 0;
1149 * The workitem queue.
1151 * It is useful to delay writes of file data and filesystem metadata
1152 * for tens of seconds so that quickly created and deleted files need
1153 * not waste disk bandwidth being created and removed. To realize this,
1154 * we append vnodes to a "workitem" queue. When running with a soft
1155 * updates implementation, most pending metadata dependencies should
1156 * not wait for more than a few seconds. Thus, mounted on block devices
1157 * are delayed only about a half the time that file data is delayed.
1158 * Similarly, directory updates are more critical, so are only delayed
1159 * about a third the time that file data is delayed. Thus, there are
1160 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1161 * one each second (driven off the filesystem syncer process). The
1162 * syncer_delayno variable indicates the next queue that is to be processed.
1163 * Items that need to be processed soon are placed in this queue:
1165 * syncer_workitem_pending[syncer_delayno]
1167 * A delay of fifteen seconds is done by placing the request fifteen
1168 * entries later in the queue:
1170 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1175 * Add an item to the syncer work queue.
1178 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1184 if (vp->v_flag & VONWORKLST) {
1185 LIST_REMOVE(vp, v_synclist);
1188 if (delay > syncer_maxdelay - 2)
1189 delay = syncer_maxdelay - 2;
1190 slot = (syncer_delayno + delay) & syncer_mask;
1192 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1193 vp->v_flag |= VONWORKLST;
1197 struct thread *updatethread;
1198 static void sched_sync (void);
1199 static struct kproc_desc up_kp = {
1204 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1207 * System filesystem synchronizer daemon.
1212 struct synclist *slp;
1216 struct thread *td = curthread;
1218 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1222 kproc_suspend_loop();
1224 starttime = time_second;
1227 * Push files whose dirty time has expired. Be careful
1228 * of interrupt race on slp queue.
1231 slp = &syncer_workitem_pending[syncer_delayno];
1232 syncer_delayno += 1;
1233 if (syncer_delayno == syncer_maxdelay)
1237 while ((vp = LIST_FIRST(slp)) != NULL) {
1238 if (VOP_ISLOCKED(vp, NULL) == 0) {
1239 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1240 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1241 VOP_UNLOCK(vp, NULL, 0, td);
1244 if (LIST_FIRST(slp) == vp) {
1246 * Note: v_tag VT_VFS vps can remain on the
1247 * worklist too with no dirty blocks, but
1248 * since sync_fsync() moves it to a different
1251 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1252 !vn_isdisk(vp, NULL))
1253 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1255 * Put us back on the worklist. The worklist
1256 * routine will remove us from our current
1257 * position and then add us back in at a later
1260 vn_syncer_add_to_worklist(vp, syncdelay);
1266 * Do soft update processing.
1269 (*bioops.io_sync)(NULL);
1272 * The variable rushjob allows the kernel to speed up the
1273 * processing of the filesystem syncer process. A rushjob
1274 * value of N tells the filesystem syncer to process the next
1275 * N seconds worth of work on its queue ASAP. Currently rushjob
1276 * is used by the soft update code to speed up the filesystem
1277 * syncer process when the incore state is getting so far
1278 * ahead of the disk that the kernel memory pool is being
1279 * threatened with exhaustion.
1286 * If it has taken us less than a second to process the
1287 * current work, then wait. Otherwise start right over
1288 * again. We can still lose time if any single round
1289 * takes more than two seconds, but it does not really
1290 * matter as we are just trying to generally pace the
1291 * filesystem activity.
1293 if (time_second == starttime)
1294 tsleep(&lbolt, 0, "syncer", 0);
1299 * Request the syncer daemon to speed up its work.
1300 * We never push it to speed up more than half of its
1301 * normal turn time, otherwise it could take over the cpu.
1303 * YYY wchan field protected by the BGL.
1309 if (updatethread->td_wchan == &lbolt) { /* YYY */
1310 unsleep(updatethread);
1311 lwkt_schedule(updatethread);
1314 if (rushjob < syncdelay / 2) {
1316 stat_rush_requests += 1;
1323 * Associate a p-buffer with a vnode.
1325 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1326 * with the buffer. i.e. the bp has not been linked into the vnode or
1335 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1338 bp->b_flags |= B_PAGING;
1339 bp->b_dev = vn_todev(vp);
1343 * Disassociate a p-buffer from a vnode.
1350 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1353 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1355 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1360 bp->b_vp = (struct vnode *) 0;
1361 bp->b_flags &= ~B_PAGING;
1365 pbreassignbuf(bp, newvp)
1367 struct vnode *newvp;
1369 if ((bp->b_flags & B_PAGING) == 0) {
1371 "pbreassignbuf() on non phys bp %p",
1379 * Reassign a buffer from one vnode to another.
1380 * Used to assign file specific control information
1381 * (indirect blocks) to the vnode to which they belong.
1384 reassignbuf(bp, newvp)
1386 struct vnode *newvp;
1388 struct buflists *listheadp;
1392 if (newvp == NULL) {
1393 printf("reassignbuf: NULL");
1399 * B_PAGING flagged buffers cannot be reassigned because their vp
1400 * is not fully linked in.
1402 if (bp->b_flags & B_PAGING)
1403 panic("cannot reassign paging buffer");
1407 * Delete from old vnode list, if on one.
1409 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1410 if (bp->b_xflags & BX_VNDIRTY)
1411 listheadp = &bp->b_vp->v_dirtyblkhd;
1413 listheadp = &bp->b_vp->v_cleanblkhd;
1414 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1415 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1416 if (bp->b_vp != newvp) {
1418 bp->b_vp = NULL; /* for clarification */
1422 * If dirty, put on list of dirty buffers; otherwise insert onto list
1425 if (bp->b_flags & B_DELWRI) {
1428 listheadp = &newvp->v_dirtyblkhd;
1429 if ((newvp->v_flag & VONWORKLST) == 0) {
1430 switch (newvp->v_type) {
1436 if (newvp->v_rdev &&
1437 newvp->v_rdev->si_mountpoint != NULL) {
1445 vn_syncer_add_to_worklist(newvp, delay);
1447 bp->b_xflags |= BX_VNDIRTY;
1448 tbp = TAILQ_FIRST(listheadp);
1450 bp->b_lblkno == 0 ||
1451 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1452 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1453 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1454 ++reassignbufsortgood;
1455 } else if (bp->b_lblkno < 0) {
1456 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1457 ++reassignbufsortgood;
1458 } else if (reassignbufmethod == 1) {
1460 * New sorting algorithm, only handle sequential case,
1461 * otherwise append to end (but before metadata)
1463 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1464 (tbp->b_xflags & BX_VNDIRTY)) {
1466 * Found the best place to insert the buffer
1468 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1469 ++reassignbufsortgood;
1472 * Missed, append to end, but before meta-data.
1473 * We know that the head buffer in the list is
1474 * not meta-data due to prior conditionals.
1476 * Indirect effects: NFS second stage write
1477 * tends to wind up here, giving maximum
1478 * distance between the unstable write and the
1481 tbp = TAILQ_LAST(listheadp, buflists);
1482 while (tbp && tbp->b_lblkno < 0)
1483 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1484 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1485 ++reassignbufsortbad;
1489 * Old sorting algorithm, scan queue and insert
1492 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1493 (ttbp->b_lblkno < bp->b_lblkno)) {
1497 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1500 bp->b_xflags |= BX_VNCLEAN;
1501 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1502 if ((newvp->v_flag & VONWORKLST) &&
1503 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1504 newvp->v_flag &= ~VONWORKLST;
1505 LIST_REMOVE(newvp, v_synclist);
1508 if (bp->b_vp != newvp) {
1516 * Create a vnode for a block device.
1517 * Used for mounting the root file system.
1520 bdevvp(dev_t dev, struct vnode **vpp)
1530 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1537 vp->v_udev = dev->si_udev;
1543 v_associate_rdev(struct vnode *vp, dev_t dev)
1547 if (dev == NULL || dev == NODEV)
1549 if (dev_is_good(dev) == 0)
1551 KKASSERT(vp->v_rdev == NULL);
1554 vp->v_rdev = reference_dev(dev);
1555 lwkt_gettoken(&ilock, &spechash_token);
1556 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
1557 lwkt_reltoken(&ilock);
1562 v_release_rdev(struct vnode *vp)
1567 if ((dev = vp->v_rdev) != NULL) {
1568 lwkt_gettoken(&ilock, &spechash_token);
1569 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1574 lwkt_reltoken(&ilock);
1579 * Add a vnode to the alias list hung off the dev_t. We only associate
1580 * the device number with the vnode. The actual device is not associated
1581 * until the vnode is opened (usually in spec_open()), and will be
1582 * disassociated on last close.
1585 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1587 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1588 panic("addaliasu on non-special vnode");
1589 nvp->v_udev = nvp_udev;
1593 * Grab a particular vnode from the free list, increment its
1594 * reference count and lock it. The vnode lock bit is set if the
1595 * vnode is being eliminated in vgone. The process is awakened
1596 * when the transition is completed, and an error returned to
1597 * indicate that the vnode is no longer usable (possibly having
1598 * been changed to a new file system type).
1600 * This code is very sensitive. We are depending on the vnode interlock
1601 * to be maintained through to the vn_lock() call, which means that we
1602 * cannot block which means that we cannot call vbusy() until after vn_lock().
1603 * If the interlock is not maintained, the VXLOCK check will not properly
1604 * interlock against a vclean()'s LK_DRAIN operation on the lock.
1607 vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
1613 * We need the interlock to safely modify the v_ fields. ZZZ it is
1614 * only legal to pass (1) the vnode's interlock and (2) only pass
1615 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1618 if ((flags & LK_INTERLOCK) == 0) {
1619 lwkt_gettoken(&vvlock, vp->v_interlock);
1624 * If the vnode is in the process of being cleaned out for
1625 * another use, we wait for the cleaning to finish and then
1626 * return failure. Cleaning is determined by checking that
1627 * the VXLOCK flag is set. It is possible for the vnode to be
1628 * self-referenced during the cleaning operation.
1630 if (vp->v_flag & VXLOCK) {
1631 if (vp->v_vxthread == curthread) {
1633 /* this can now occur in normal operation */
1634 log(LOG_INFO, "VXLOCK interlock avoided\n");
1637 vp->v_flag |= VXWANT;
1638 lwkt_reltoken(vlock);
1639 tsleep((caddr_t)vp, 0, "vget", 0);
1645 * Bump v_usecount to prevent the vnode from being recycled. The
1646 * usecount needs to be bumped before we successfully get our lock.
1649 if (flags & LK_TYPE_MASK) {
1650 if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
1652 * must expand vrele here because we do not want
1653 * to call VOP_INACTIVE if the reference count
1654 * drops back to zero since it was never really
1655 * active. We must remove it from the free list
1656 * before sleeping so that multiple processes do
1657 * not try to recycle it.
1659 lwkt_gettokref(vlock);
1662 lwkt_reltoken(vlock);
1666 if (VSHOULDBUSY(vp))
1667 vbusy(vp); /* interlock must be held on call */
1668 lwkt_reltoken(vlock);
1673 vref(struct vnode *vp)
1675 crit_enter(); /* YYY use crit section for moment / BGL protected */
1681 * Vnode put/release.
1682 * If count drops to zero, call inactive routine and return to freelist.
1685 vrele(struct vnode *vp)
1687 struct thread *td = curthread; /* XXX */
1690 KASSERT(vp != NULL && vp->v_usecount >= 0,
1691 ("vrele: null vp or <=0 v_usecount"));
1693 lwkt_gettoken(&vlock, vp->v_interlock);
1695 if (vp->v_usecount > 1) {
1697 lwkt_reltoken(&vlock);
1701 if (vp->v_usecount == 1) {
1704 * We must call VOP_INACTIVE with the node locked and the
1705 * usecount 0. If we are doing a vpu, the node is already
1706 * locked, but, in the case of vrele, we must explicitly lock
1707 * the vnode before calling VOP_INACTIVE.
1710 if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
1711 VOP_INACTIVE(vp, td);
1713 lwkt_reltoken(&vlock);
1716 vprint("vrele: negative ref count", vp);
1718 lwkt_reltoken(&vlock);
1719 panic("vrele: negative ref cnt");
1724 vput(struct vnode *vp)
1726 struct thread *td = curthread; /* XXX */
1729 KASSERT(vp != NULL, ("vput: null vp"));
1731 lwkt_gettoken(&vlock, vp->v_interlock);
1733 if (vp->v_usecount > 1) {
1735 VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
1739 if (vp->v_usecount == 1) {
1742 * We must call VOP_INACTIVE with the node locked.
1743 * If we are doing a vpu, the node is already locked,
1744 * so we just need to release the vnode mutex.
1746 VOP_INACTIVE(vp, td);
1748 lwkt_reltoken(&vlock);
1751 vprint("vput: negative ref count", vp);
1753 lwkt_reltoken(&vlock);
1754 panic("vput: negative ref cnt");
1759 * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1760 * be held but isn't.
1770 if (VSHOULDBUSY(vp))
1771 vbusy(vp); /* interlock must be held on call */
1776 * One less who cares about this vnode.
1784 lwkt_gettoken(&vlock, vp->v_interlock);
1785 if (vp->v_holdcnt <= 0)
1786 panic("vdrop: holdcnt");
1789 lwkt_reltoken(&vlock);
1795 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1796 int (*slowfunc)(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data),
1806 * Scan the vnodes on the mount's vnode list. Use a placemarker
1808 pvp = zalloc(vnode_zone);
1809 pvp->v_flag |= VPLACEMARKER;
1811 lwkt_gettoken(&ilock, &mntvnode_token);
1812 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1814 while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1816 * Move the placemarker and skip other placemarkers we
1817 * encounter. The nothing can get in our way so the
1818 * mount point on the vp must be valid.
1820 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1821 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1822 if (vp->v_flag & VPLACEMARKER)
1824 KKASSERT(vp->v_mount == mp);
1830 if ((r = fastfunc(mp, vp, data)) < 0)
1837 * Get the vnodes interlock and make sure it is still on the
1838 * mount list. Skip it if it has moved (we may encounter it
1839 * later). Then do the with-interlock test. The callback
1840 * is responsible for releasing the vnode interlock.
1842 * The interlock is type-stable.
1845 lwkt_gettoken(&vlock, vp->v_interlock);
1846 if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1847 printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1848 lwkt_reltoken(&vlock);
1851 if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1852 KKASSERT(lwkt_havetokref(&vlock) == 0);
1855 KKASSERT(lwkt_havetokref(&vlock) == 0);
1858 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1859 zfree(vnode_zone, pvp);
1860 lwkt_reltoken(&ilock);
1865 * Remove any vnodes in the vnode table belonging to mount point mp.
1867 * If FORCECLOSE is not specified, there should not be any active ones,
1868 * return error if any are found (nb: this is a user error, not a
1869 * system error). If FORCECLOSE is specified, detach any active vnodes
1872 * If WRITECLOSE is set, only flush out regular file vnodes open for
1875 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1877 * `rootrefs' specifies the base reference count for the root vnode
1878 * of this filesystem. The root vnode is considered busy if its
1879 * v_usecount exceeds this value. On a successful return, vflush()
1880 * will call vrele() on the root vnode exactly rootrefs times.
1881 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1885 static int busyprt = 0; /* print out busy vnodes */
1886 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1889 static int vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data);
1891 struct vflush_info {
1898 vflush(mp, rootrefs, flags)
1903 struct thread *td = curthread; /* XXX */
1904 struct vnode *rootvp = NULL;
1907 struct vflush_info vflush_info;
1910 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1911 ("vflush: bad args"));
1913 * Get the filesystem root vnode. We can vput() it
1914 * immediately, since with rootrefs > 0, it won't go away.
1916 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1921 vflush_info.busy = 0;
1922 vflush_info.flags = flags;
1923 vflush_info.td = td;
1924 vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
1926 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1928 * If just the root vnode is busy, and if its refcount
1929 * is equal to `rootrefs', then go ahead and kill it.
1931 lwkt_gettoken(&vlock, rootvp->v_interlock);
1932 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1933 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1934 if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1935 vgonel(rootvp, &vlock, td);
1936 vflush_info.busy = 0;
1938 lwkt_reltoken(&vlock);
1941 if (vflush_info.busy)
1943 for (; rootrefs > 0; rootrefs--)
1949 * The scan callback is made with an interlocked vnode.
1952 vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
1954 struct vflush_info *info = data;
1958 * Skip over a vnodes marked VSYSTEM.
1960 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1961 lwkt_reltoken(vlock);
1966 * If WRITECLOSE is set, flush out unlinked but still open
1967 * files (even if open only for reading) and regular file
1968 * vnodes open for writing.
1970 if ((info->flags & WRITECLOSE) &&
1971 (vp->v_type == VNON ||
1972 (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
1973 vattr.va_nlink > 0)) &&
1974 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1975 lwkt_reltoken(vlock);
1980 * With v_usecount == 0, all we need to do is clear out the
1981 * vnode data structures and we are done.
1983 if (vp->v_usecount == 0) {
1984 vgonel(vp, vlock, info->td);
1989 * If FORCECLOSE is set, forcibly close the vnode. For block
1990 * or character devices, revert to an anonymous device. For
1991 * all other files, just kill them.
1993 if (info->flags & FORCECLOSE) {
1994 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1995 vgonel(vp, vlock, info->td);
1997 vclean(vp, vlock, 0, info->td);
1998 vp->v_op = spec_vnodeop_p;
1999 insmntque(vp, (struct mount *) 0);
2005 vprint("vflush: busy vnode", vp);
2007 lwkt_reltoken(vlock);
2013 * Disassociate the underlying file system from a vnode.
2016 vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
2021 * Check to see if the vnode is in use. If so we have to reference it
2022 * before we clean it out so that its count cannot fall to zero and
2023 * generate a race against ourselves to recycle it.
2025 if ((active = vp->v_usecount))
2029 * Prevent the vnode from being recycled or brought into use while we
2032 if (vp->v_flag & VXLOCK)
2033 panic("vclean: deadlock");
2034 vp->v_flag |= VXLOCK;
2035 vp->v_vxthread = curthread;
2038 * Even if the count is zero, the VOP_INACTIVE routine may still
2039 * have the object locked while it cleans it out. The VOP_LOCK
2040 * ensures that the VOP_INACTIVE routine is done with its work.
2041 * For active vnodes, it ensures that no other activity can
2042 * occur while the underlying object is being cleaned out.
2044 * NOTE: we continue to hold the vnode interlock through to the
2047 VOP_LOCK(vp, NULL, LK_DRAIN, td);
2050 * Clean out any buffers associated with the vnode.
2052 vinvalbuf(vp, V_SAVE, td, 0, 0);
2053 VOP_DESTROYVOBJECT(vp);
2056 * If purging an active vnode, it must be closed and
2057 * deactivated before being reclaimed. Note that the
2058 * VOP_INACTIVE will unlock the vnode.
2061 if (flags & DOCLOSE)
2062 VOP_CLOSE(vp, FNONBLOCK, td);
2063 VOP_INACTIVE(vp, td);
2066 * Any other processes trying to obtain this lock must first
2067 * wait for VXLOCK to clear, then call the new lock operation.
2069 VOP_UNLOCK(vp, NULL, 0, td);
2072 * Reclaim the vnode.
2074 if (VOP_RECLAIM(vp, td))
2075 panic("vclean: cannot reclaim");
2079 * Inline copy of vrele() since VOP_INACTIVE
2080 * has already been called.
2082 if (--vp->v_usecount <= 0) {
2084 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2085 vprint("vclean: bad ref count", vp);
2086 panic("vclean: ref cnt");
2094 vp->v_vnlock = NULL;
2098 * Done with purge, notify sleepers of the grim news.
2100 vp->v_op = dead_vnodeop_p;
2103 vp->v_flag &= ~VXLOCK;
2104 vp->v_vxthread = NULL;
2105 if (vp->v_flag & VXWANT) {
2106 vp->v_flag &= ~VXWANT;
2107 wakeup((caddr_t) vp);
2109 lwkt_reltoken(vlock);
2113 * Eliminate all activity associated with the requested vnode
2114 * and with all vnodes aliased to the requested vnode.
2118 struct vop_revoke_args /* {
2123 struct vnode *vp, *vq;
2127 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2131 * If a vgone (or vclean) is already in progress,
2132 * wait until it is done and return.
2134 if (vp->v_flag & VXLOCK) {
2135 vp->v_flag |= VXWANT;
2136 /*lwkt_reltoken(vlock); ZZZ */
2137 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2142 * If the vnode has a device association, scrap all vnodes associated
2143 * with the device. Don't let the device disappear on us while we
2144 * are scrapping the vnodes.
2146 if (vp->v_type != VCHR && vp->v_type != VBLK)
2148 if ((dev = vp->v_rdev) == NULL) {
2149 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
2154 lwkt_gettoken(&ilock, &spechash_token);
2155 vq = SLIST_FIRST(&dev->si_hlist);
2156 lwkt_reltoken(&ilock);
2166 * Recycle an unused vnode to the front of the free list.
2167 * Release the passed interlock if the vnode will be recycled.
2170 vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
2174 lwkt_gettoken(&vlock, vp->v_interlock);
2175 if (vp->v_usecount == 0) {
2177 lwkt_reltoken(inter_lkp);
2178 vgonel(vp, &vlock, td);
2181 lwkt_reltoken(&vlock);
2186 * Eliminate all activity associated with a vnode
2187 * in preparation for reuse.
2190 vgone(struct vnode *vp)
2192 struct thread *td = curthread; /* XXX */
2195 lwkt_gettoken(&vlock, vp->v_interlock);
2196 vgonel(vp, &vlock, td);
2200 * vgone, with the vp interlock held.
2203 vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
2209 * If a vgone (or vclean) is already in progress,
2210 * wait until it is done and return.
2212 if (vp->v_flag & VXLOCK) {
2213 vp->v_flag |= VXWANT;
2214 lwkt_reltoken(vlock);
2215 tsleep((caddr_t)vp, 0, "vgone", 0);
2220 * Clean out the filesystem specific data.
2222 vclean(vp, vlock, DOCLOSE, td);
2223 lwkt_gettokref(vlock);
2226 * Delete from old mount point vnode list, if on one.
2228 if (vp->v_mount != NULL)
2229 insmntque(vp, (struct mount *)0);
2231 * If special device, remove it from special device alias list
2234 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2239 * If it is on the freelist and not already at the head,
2240 * move it to the head of the list. The test of the
2241 * VDOOMED flag and the reference count of zero is because
2242 * it will be removed from the free list by getnewvnode,
2243 * but will not have its reference count incremented until
2244 * after calling vgone. If the reference count were
2245 * incremented first, vgone would (incorrectly) try to
2246 * close the previous instance of the underlying object.
2248 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2250 lwkt_gettoken(&ilock, &vnode_free_list_token);
2251 if (vp->v_flag & VFREE)
2252 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2255 vp->v_flag |= VFREE;
2256 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2257 lwkt_reltoken(&ilock);
2261 lwkt_reltoken(vlock);
2265 * Lookup a vnode by device number.
2268 vfinddev(dev, type, vpp)
2276 lwkt_gettoken(&ilock, &spechash_token);
2277 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2278 if (type == vp->v_type) {
2280 lwkt_reltoken(&ilock);
2284 lwkt_reltoken(&ilock);
2289 * Calculate the total number of references to a special device. This
2290 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
2291 * an overloaded field. Since udev2dev can now return NODEV, we have
2292 * to check for a NULL v_rdev.
2295 count_dev(dev_t dev)
2301 if (SLIST_FIRST(&dev->si_hlist)) {
2302 lwkt_gettoken(&ilock, &spechash_token);
2303 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2304 count += vp->v_usecount;
2306 lwkt_reltoken(&ilock);
2312 count_udev(udev_t udev)
2316 if ((dev = udev2dev(udev, 0)) == NODEV)
2318 return(count_dev(dev));
2322 vcount(struct vnode *vp)
2324 if (vp->v_rdev == NULL)
2326 return(count_dev(vp->v_rdev));
2330 * Print out a description of a vnode.
2332 static char *typename[] =
2333 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2343 printf("%s: %p: ", label, (void *)vp);
2345 printf("%p: ", (void *)vp);
2346 printf("type %s, usecount %d, writecount %d, refcount %d,",
2347 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2350 if (vp->v_flag & VROOT)
2351 strcat(buf, "|VROOT");
2352 if (vp->v_flag & VTEXT)
2353 strcat(buf, "|VTEXT");
2354 if (vp->v_flag & VSYSTEM)
2355 strcat(buf, "|VSYSTEM");
2356 if (vp->v_flag & VXLOCK)
2357 strcat(buf, "|VXLOCK");
2358 if (vp->v_flag & VXWANT)
2359 strcat(buf, "|VXWANT");
2360 if (vp->v_flag & VBWAIT)
2361 strcat(buf, "|VBWAIT");
2362 if (vp->v_flag & VDOOMED)
2363 strcat(buf, "|VDOOMED");
2364 if (vp->v_flag & VFREE)
2365 strcat(buf, "|VFREE");
2366 if (vp->v_flag & VOBJBUF)
2367 strcat(buf, "|VOBJBUF");
2369 printf(" flags (%s)", &buf[1]);
2370 if (vp->v_data == NULL) {
2379 #include <ddb/ddb.h>
2381 * List all of the locked vnodes in the system.
2382 * Called when debugging the kernel.
2384 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2386 struct thread *td = curthread; /* XXX */
2388 struct mount *mp, *nmp;
2391 printf("Locked vnodes\n");
2392 lwkt_gettoken(&ilock, &mountlist_token);
2393 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2394 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
2395 nmp = TAILQ_NEXT(mp, mnt_list);
2398 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2399 if (VOP_ISLOCKED(vp, NULL))
2400 vprint((char *)0, vp);
2402 lwkt_gettokref(&ilock);
2403 nmp = TAILQ_NEXT(mp, mnt_list);
2406 lwkt_reltoken(&ilock);
2411 * Top level filesystem related information gathering.
2413 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2416 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2418 int *name = (int *)arg1 - 1; /* XXX */
2419 u_int namelen = arg2 + 1; /* XXX */
2420 struct vfsconf *vfsp;
2422 #if 1 || defined(COMPAT_PRELITE2)
2423 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2425 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2429 /* all sysctl names at this level are at least name and field */
2431 return (ENOTDIR); /* overloaded */
2432 if (name[0] != VFS_GENERIC) {
2433 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2434 if (vfsp->vfc_typenum == name[0])
2437 return (EOPNOTSUPP);
2438 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2439 oldp, oldlenp, newp, newlen, p));
2443 case VFS_MAXTYPENUM:
2446 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2449 return (ENOTDIR); /* overloaded */
2450 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2451 if (vfsp->vfc_typenum == name[2])
2454 return (EOPNOTSUPP);
2455 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2457 return (EOPNOTSUPP);
2460 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2461 "Generic filesystem");
2463 #if 1 || defined(COMPAT_PRELITE2)
2466 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2469 struct vfsconf *vfsp;
2470 struct ovfsconf ovfs;
2472 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2473 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2474 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2475 ovfs.vfc_index = vfsp->vfc_typenum;
2476 ovfs.vfc_refcount = vfsp->vfc_refcount;
2477 ovfs.vfc_flags = vfsp->vfc_flags;
2478 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2485 #endif /* 1 || COMPAT_PRELITE2 */
2488 #define KINFO_VNODESLOP 10
2490 * Dump vnode list (via sysctl).
2491 * Copyout address of vnode followed by vnode.
2495 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2497 struct proc *p = curproc; /* XXX */
2498 struct mount *mp, *nmp;
2499 struct vnode *nvp, *vp;
2504 #define VPTRSZ sizeof (struct vnode *)
2505 #define VNODESZ sizeof (struct vnode)
2508 if (!req->oldptr) /* Make an estimate */
2509 return (SYSCTL_OUT(req, 0,
2510 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2512 lwkt_gettoken(&ilock, &mountlist_token);
2513 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2514 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
2515 nmp = TAILQ_NEXT(mp, mnt_list);
2518 lwkt_gettoken(&jlock, &mntvnode_token);
2520 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2524 * Check that the vp is still associated with
2525 * this filesystem. RACE: could have been
2526 * recycled onto the same filesystem.
2528 if (vp->v_mount != mp)
2530 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2531 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2532 (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2533 lwkt_reltoken(&jlock);
2537 lwkt_reltoken(&jlock);
2538 lwkt_gettokref(&ilock);
2539 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */
2542 lwkt_reltoken(&ilock);
2550 * Exporting the vnode list on large systems causes them to crash.
2551 * Exporting the vnode list on medium systems causes sysctl to coredump.
2554 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2555 0, 0, sysctl_vnode, "S,vnode", "");
2559 * Check to see if a filesystem is mounted on a block device.
2562 vfs_mountedon(struct vnode *vp)
2566 if ((dev = vp->v_rdev) == NULL)
2567 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
2568 if (dev != NODEV && dev->si_mountpoint)
2574 * Unmount all filesystems. The list is traversed in reverse order
2575 * of mounting to avoid dependencies.
2581 struct thread *td = curthread;
2584 if (td->td_proc == NULL)
2585 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2588 * Since this only runs when rebooting, it is not interlocked.
2590 while(!TAILQ_EMPTY(&mountlist)) {
2591 mp = TAILQ_LAST(&mountlist, mntlist);
2592 error = dounmount(mp, MNT_FORCE, td);
2594 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2595 printf("unmount of %s failed (",
2596 mp->mnt_stat.f_mntonname);
2600 printf("%d)\n", error);
2602 /* The unmount has removed mp from the mountlist */
2608 * Build hash lists of net addresses and hang them off the mount point.
2609 * Called by ufs_mount() to set up the lists of export addresses.
2612 vfs_hang_addrlist(mp, nep, argp)
2614 struct netexport *nep;
2615 struct export_args *argp;
2618 struct radix_node_head *rnh;
2620 struct radix_node *rn;
2621 struct sockaddr *saddr, *smask = 0;
2625 if (argp->ex_addrlen == 0) {
2626 if (mp->mnt_flag & MNT_DEFEXPORTED)
2628 np = &nep->ne_defexported;
2629 np->netc_exflags = argp->ex_flags;
2630 np->netc_anon = argp->ex_anon;
2631 np->netc_anon.cr_ref = 1;
2632 mp->mnt_flag |= MNT_DEFEXPORTED;
2636 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2638 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2641 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2642 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2643 bzero((caddr_t) np, i);
2644 saddr = (struct sockaddr *) (np + 1);
2645 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2647 if (saddr->sa_len > argp->ex_addrlen)
2648 saddr->sa_len = argp->ex_addrlen;
2649 if (argp->ex_masklen) {
2650 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2651 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2654 if (smask->sa_len > argp->ex_masklen)
2655 smask->sa_len = argp->ex_masklen;
2657 i = saddr->sa_family;
2658 if ((rnh = nep->ne_rtable[i]) == 0) {
2660 * Seems silly to initialize every AF when most are not used,
2661 * do so on demand here
2663 for (dom = domains; dom; dom = dom->dom_next)
2664 if (dom->dom_family == i && dom->dom_rtattach) {
2665 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2669 if ((rnh = nep->ne_rtable[i]) == 0) {
2674 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2676 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2680 np->netc_exflags = argp->ex_flags;
2681 np->netc_anon = argp->ex_anon;
2682 np->netc_anon.cr_ref = 1;
2685 free(np, M_NETADDR);
2691 vfs_free_netcred(rn, w)
2692 struct radix_node *rn;
2695 struct radix_node_head *rnh = (struct radix_node_head *) w;
2697 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2698 free((caddr_t) rn, M_NETADDR);
2703 * Free the net address hash lists that are hanging off the mount points.
2706 vfs_free_addrlist(nep)
2707 struct netexport *nep;
2710 struct radix_node_head *rnh;
2712 for (i = 0; i <= AF_MAX; i++)
2713 if ((rnh = nep->ne_rtable[i])) {
2714 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2716 free((caddr_t) rnh, M_RTABLE);
2717 nep->ne_rtable[i] = 0;
2722 vfs_export(mp, nep, argp)
2724 struct netexport *nep;
2725 struct export_args *argp;
2729 if (argp->ex_flags & MNT_DELEXPORT) {
2730 if (mp->mnt_flag & MNT_EXPUBLIC) {
2731 vfs_setpublicfs(NULL, NULL, NULL);
2732 mp->mnt_flag &= ~MNT_EXPUBLIC;
2734 vfs_free_addrlist(nep);
2735 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2737 if (argp->ex_flags & MNT_EXPORTED) {
2738 if (argp->ex_flags & MNT_EXPUBLIC) {
2739 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2741 mp->mnt_flag |= MNT_EXPUBLIC;
2743 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2745 mp->mnt_flag |= MNT_EXPORTED;
2752 * Set the publicly exported filesystem (WebNFS). Currently, only
2753 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2756 vfs_setpublicfs(mp, nep, argp)
2758 struct netexport *nep;
2759 struct export_args *argp;
2766 * mp == NULL -> invalidate the current info, the FS is
2767 * no longer exported. May be called from either vfs_export
2768 * or unmount, so check if it hasn't already been done.
2771 if (nfs_pub.np_valid) {
2772 nfs_pub.np_valid = 0;
2773 if (nfs_pub.np_index != NULL) {
2774 FREE(nfs_pub.np_index, M_TEMP);
2775 nfs_pub.np_index = NULL;
2782 * Only one allowed at a time.
2784 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2788 * Get real filehandle for root of exported FS.
2790 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2791 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2793 if ((error = VFS_ROOT(mp, &rvp)))
2796 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2802 * If an indexfile was specified, pull it in.
2804 if (argp->ex_indexfile != NULL) {
2805 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2807 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2808 MAXNAMLEN, (size_t *)0);
2811 * Check for illegal filenames.
2813 for (cp = nfs_pub.np_index; *cp; cp++) {
2821 FREE(nfs_pub.np_index, M_TEMP);
2826 nfs_pub.np_mount = mp;
2827 nfs_pub.np_valid = 1;
2832 vfs_export_lookup(mp, nep, nam)
2834 struct netexport *nep;
2835 struct sockaddr *nam;
2838 struct radix_node_head *rnh;
2839 struct sockaddr *saddr;
2842 if (mp->mnt_flag & MNT_EXPORTED) {
2844 * Lookup in the export list first.
2848 rnh = nep->ne_rtable[saddr->sa_family];
2850 np = (struct netcred *)
2851 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2853 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2858 * If no address match, use the default if it exists.
2860 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2861 np = &nep->ne_defexported;
2867 * perform msync on all vnodes under a mount point. The mount point must
2868 * be locked. This code is also responsible for lazy-freeing unreferenced
2869 * vnodes whos VM objects no longer contain pages.
2871 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2873 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2874 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2875 lwkt_tokref_t vlock, void *data);
2878 vfs_msync(struct mount *mp, int flags)
2880 vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2884 * scan1 is a fast pre-check. There could be hundreds of thousands of
2885 * vnodes, we cannot afford to do anything heavy weight until we have a
2886 * fairly good indication that there is work to do.
2890 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2892 int flags = (int)data;
2894 if ((vp->v_flag & VXLOCK) == 0) {
2895 if (VSHOULDFREE(vp))
2897 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2898 (vp->v_flag & VOBJDIRTY) &&
2899 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2908 vfs_msync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
2912 int flags = (int)data;
2914 if (vp->v_flag & VXLOCK)
2917 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2918 (vp->v_flag & VOBJDIRTY) &&
2919 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2920 error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2922 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2923 vm_object_page_clean(obj, 0, 0,
2924 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2931 lwkt_reltoken(vlock);
2936 * Create the VM object needed for VMIO and mmap support. This
2937 * is done for all VREG files in the system. Some filesystems might
2938 * afford the additional metadata buffering capability of the
2939 * VMIO code by making the device node be VMIO mode also.
2941 * vp must be locked when vfs_object_create is called.
2944 vfs_object_create(struct vnode *vp, struct thread *td)
2946 return (VOP_CREATEVOBJECT(vp, td));
2950 * NOTE: the vnode interlock must be held during the call. We have to recheck
2951 * the VFREE flag since the vnode may have been removed from the free list
2952 * while we were blocked on vnode_free_list_token. The use or hold count
2953 * must have already been bumped by the caller.
2956 vbusy(struct vnode *vp)
2960 lwkt_gettoken(&ilock, &vnode_free_list_token);
2961 if ((vp->v_flag & VFREE) != 0) {
2962 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2964 vp->v_flag &= ~(VFREE|VAGE);
2966 lwkt_reltoken(&ilock);
2970 * NOTE: the vnode interlock must be held during the call. The use or hold
2971 * count must have already been bumped by the caller. We use a VINFREE to
2972 * interlock against other calls to vfree() which might occur while we
2973 * are blocked. The vnode cannot be reused until it has actually been
2974 * placed on the free list, so there are no other races even though the
2975 * use and hold counts are 0.
2978 vfree(struct vnode *vp)
2982 if ((vp->v_flag & VINFREE) == 0) {
2983 vp->v_flag |= VINFREE;
2984 lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
2985 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2986 if (vp->v_flag & VAGE) {
2987 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2989 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2992 vp->v_flag &= ~(VAGE|VINFREE);
2993 vp->v_flag |= VFREE;
2994 lwkt_reltoken(&ilock); /* can block */
3000 * Record a process's interest in events which might happen to
3001 * a vnode. Because poll uses the historic select-style interface
3002 * internally, this routine serves as both the ``check for any
3003 * pending events'' and the ``record my interest in future events''
3004 * functions. (These are done together, while the lock is held,
3005 * to avoid race conditions.)
3008 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3012 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3013 if (vp->v_pollinfo.vpi_revents & events) {
3015 * This leaves events we are not interested
3016 * in available for the other process which
3017 * which presumably had requested them
3018 * (otherwise they would never have been
3021 events &= vp->v_pollinfo.vpi_revents;
3022 vp->v_pollinfo.vpi_revents &= ~events;
3024 lwkt_reltoken(&ilock);
3027 vp->v_pollinfo.vpi_events |= events;
3028 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
3029 lwkt_reltoken(&ilock);
3034 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3035 * it is possible for us to miss an event due to race conditions, but
3036 * that condition is expected to be rare, so for the moment it is the
3037 * preferred interface.
3040 vn_pollevent(vp, events)
3046 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3047 if (vp->v_pollinfo.vpi_events & events) {
3049 * We clear vpi_events so that we don't
3050 * call selwakeup() twice if two events are
3051 * posted before the polling process(es) is
3052 * awakened. This also ensures that we take at
3053 * most one selwakeup() if the polling process
3054 * is no longer interested. However, it does
3055 * mean that only one event can be noticed at
3056 * a time. (Perhaps we should only clear those
3057 * event bits which we note?) XXX
3059 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
3060 vp->v_pollinfo.vpi_revents |= events;
3061 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3063 lwkt_reltoken(&ilock);
3067 * Wake up anyone polling on vp because it is being revoked.
3068 * This depends on dead_poll() returning POLLHUP for correct
3077 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3078 if (vp->v_pollinfo.vpi_events) {
3079 vp->v_pollinfo.vpi_events = 0;
3080 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3082 lwkt_reltoken(&ilock);
3088 * Routine to create and manage a filesystem syncer vnode.
3090 #define sync_close ((int (*) (struct vop_close_args *))nullop)
3091 static int sync_fsync (struct vop_fsync_args *);
3092 static int sync_inactive (struct vop_inactive_args *);
3093 static int sync_reclaim (struct vop_reclaim_args *);
3094 #define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
3095 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
3096 static int sync_print (struct vop_print_args *);
3097 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
3099 static vop_t **sync_vnodeop_p;
3100 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3101 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3102 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3103 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3104 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3105 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3106 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
3107 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
3108 { &vop_print_desc, (vop_t *) sync_print }, /* print */
3109 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
3112 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3113 { &sync_vnodeop_p, sync_vnodeop_entries };
3115 VNODEOP_SET(sync_vnodeop_opv_desc);
3118 * Create a new filesystem syncer vnode for the specified mount point.
3119 * This vnode is placed on the worklist and is responsible for sync'ing
3122 * NOTE: read-only mounts are also placed on the worklist. The filesystem
3123 * sync code is also responsible for cleaning up vnodes.
3126 vfs_allocate_syncvnode(struct mount *mp)
3129 static long start, incr, next;
3132 /* Allocate a new vnode */
3133 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3134 mp->mnt_syncer = NULL;
3139 * Place the vnode onto the syncer worklist. We attempt to
3140 * scatter them about on the list so that they will go off
3141 * at evenly distributed times even if all the filesystems
3142 * are mounted at once.
3145 if (next == 0 || next > syncer_maxdelay) {
3149 start = syncer_maxdelay / 2;
3150 incr = syncer_maxdelay;
3154 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3155 mp->mnt_syncer = vp;
3160 * Do a lazy sync of the filesystem.
3164 struct vop_fsync_args /* {
3166 struct ucred *a_cred;
3168 struct thread *a_td;
3171 struct vnode *syncvp = ap->a_vp;
3172 struct mount *mp = syncvp->v_mount;
3173 struct thread *td = ap->a_td;
3178 * We only need to do something if this is a lazy evaluation.
3180 if (ap->a_waitfor != MNT_LAZY)
3184 * Move ourselves to the back of the sync list.
3186 vn_syncer_add_to_worklist(syncvp, syncdelay);
3189 * Walk the list of vnodes pushing all that are dirty and
3190 * not already on the sync list, and freeing vnodes which have
3191 * no refs and whos VM objects are empty. vfs_msync() handles
3192 * the VM issues and must be called whether the mount is readonly
3195 lwkt_gettoken(&ilock, &mountlist_token);
3196 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3197 lwkt_reltoken(&ilock);
3200 if (mp->mnt_flag & MNT_RDONLY) {
3201 vfs_msync(mp, MNT_NOWAIT);
3203 asyncflag = mp->mnt_flag & MNT_ASYNC;
3204 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */
3205 vfs_msync(mp, MNT_NOWAIT);
3206 VFS_SYNC(mp, MNT_LAZY, td);
3208 mp->mnt_flag |= MNT_ASYNC;
3215 * The syncer vnode is no referenced.
3219 struct vop_inactive_args /* {
3230 * The syncer vnode is no longer needed and is being decommissioned.
3232 * Modifications to the worklist must be protected at splbio().
3236 struct vop_reclaim_args /* {
3240 struct vnode *vp = ap->a_vp;
3244 vp->v_mount->mnt_syncer = NULL;
3245 if (vp->v_flag & VONWORKLST) {
3246 LIST_REMOVE(vp, v_synclist);
3247 vp->v_flag &= ~VONWORKLST;
3255 * Print out a syncer vnode.
3259 struct vop_print_args /* {
3263 struct vnode *vp = ap->a_vp;
3265 printf("syncer vnode");
3266 if (vp->v_vnlock != NULL)
3267 lockmgr_printinfo(vp->v_vnlock);
3273 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
3274 * (or v_rdev might be NULL).
3277 vn_todev(struct vnode *vp)
3279 if (vp->v_type != VBLK && vp->v_type != VCHR)
3281 KKASSERT(vp->v_rdev != NULL);
3282 return (vp->v_rdev);
3286 * Check if vnode represents a disk device. The vnode does not need to be
3290 vn_isdisk(struct vnode *vp, int *errp)
3294 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3300 if ((dev = vp->v_rdev) == NULL)
3301 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
3302 if (dev == NULL || dev == NODEV) {
3307 if (dev_is_good(dev) == 0) {
3312 if ((dev_dflags(dev) & D_DISK) == 0) {
3324 struct nameidata *ndp;
3327 if (!(flags & NDF_NO_FREE_PNBUF) &&
3328 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3329 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3330 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3332 if (!(flags & NDF_NO_DNCP_RELE) &&
3333 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3335 cache_drop(ndp->ni_dncp);
3336 ndp->ni_dncp = NULL;
3338 if (!(flags & NDF_NO_NCP_RELE) &&
3339 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3341 cache_drop(ndp->ni_ncp);
3344 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3345 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3346 ndp->ni_dvp != ndp->ni_vp) {
3347 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
3349 if (!(flags & NDF_NO_DVP_RELE) &&
3350 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3354 if (!(flags & NDF_NO_VP_UNLOCK) &&
3355 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3356 VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
3358 if (!(flags & NDF_NO_VP_RELE) &&
3363 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3364 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3365 vrele(ndp->ni_startdir);
3366 ndp->ni_startdir = NULL;
3370 #ifdef DEBUG_VFS_LOCKS
3373 assert_vop_locked(struct vnode *vp, const char *str)
3376 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
3377 panic("%s: %p is not locked shared but should be", str, vp);
3382 assert_vop_unlocked(struct vnode *vp, const char *str)
3385 if (vp && IS_LOCKING_VFS(vp)) {
3386 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
3387 panic("%s: %p is locked but should not be", str, vp);