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 1.23 2003/11/03 17:11:21 dillon 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, int flags, struct thread *td);
91 static unsigned long numvnodes;
92 static void vlruvp(struct vnode *vp);
93 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
95 enum vtype iftovt_tab[16] = {
96 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
97 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
100 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
101 S_IFSOCK, S_IFIFO, S_IFMT,
104 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
106 static u_long wantfreevnodes = 25;
107 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
108 static u_long freevnodes = 0;
109 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
111 static int reassignbufcalls;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
113 static int reassignbufloops;
114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
115 static int reassignbufsortgood;
116 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
117 static int reassignbufsortbad;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
119 static int reassignbufmethod = 1;
120 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
122 #ifdef ENABLE_VFS_IOOPT
124 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
127 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
128 struct lwkt_token mountlist_token;
129 struct lwkt_token mntvnode_token;
130 int nfs_mount_type = -1;
131 static struct lwkt_token mntid_token;
132 static struct lwkt_token vnode_free_list_token;
133 static struct lwkt_token spechash_token;
134 struct nfs_public nfs_pub; /* publicly exported FS */
135 static vm_zone_t vnode_zone;
138 * The workitem queue.
140 #define SYNCER_MAXDELAY 32
141 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
142 time_t syncdelay = 30; /* max time to delay syncing data */
143 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
144 "VFS data synchronization delay");
145 time_t filedelay = 30; /* time to delay syncing files */
146 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
147 "File synchronization delay");
148 time_t dirdelay = 29; /* time to delay syncing directories */
149 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
150 "Directory synchronization delay");
151 time_t metadelay = 28; /* time to delay syncing metadata */
152 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
153 "VFS metadata synchronization delay");
154 static int rushjob; /* number of slots to run ASAP */
155 static int stat_rush_requests; /* number of times I/O speeded up */
156 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
158 static int syncer_delayno = 0;
159 static long syncer_mask;
160 LIST_HEAD(synclist, vnode);
161 static struct synclist *syncer_workitem_pending;
164 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
165 &desiredvnodes, 0, "Maximum number of vnodes");
166 static int minvnodes;
167 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
168 &minvnodes, 0, "Minimum number of vnodes");
169 static int vnlru_nowhere = 0;
170 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
171 "Number of times the vnlru process ran without success");
173 static void vfs_free_addrlist (struct netexport *nep);
174 static int vfs_free_netcred (struct radix_node *rn, void *w);
175 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
176 struct export_args *argp);
179 * Initialize the vnode management data structures.
186 * Desired vnodes is a result of the physical page count
187 * and the size of kernel's heap. It scales in proportion
188 * to the amount of available physical memory. This can
189 * cause trouble on 64-bit and large memory platforms.
191 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
193 min(maxproc + vmstats.v_page_count /4,
194 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
195 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
197 minvnodes = desiredvnodes / 4;
198 lwkt_inittoken(&mntvnode_token);
199 lwkt_inittoken(&mntid_token);
200 lwkt_inittoken(&spechash_token);
201 TAILQ_INIT(&vnode_free_list);
202 lwkt_inittoken(&vnode_free_list_token);
203 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
205 * Initialize the filesystem syncer.
207 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
209 syncer_maxdelay = syncer_mask + 1;
213 * Mark a mount point as busy. Used to synchronize access and to delay
214 * unmounting. Interlock is not released on failure.
217 vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp,
222 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
223 if (flags & LK_NOWAIT)
225 mp->mnt_kern_flag |= MNTK_MWAIT;
227 lwkt_reltoken(interlkp);
230 * Since all busy locks are shared except the exclusive
231 * lock granted when unmounting, the only place that a
232 * wakeup needs to be done is at the release of the
233 * exclusive lock at the end of dounmount.
235 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
237 lwkt_gettoken(interlkp);
241 lkflags = LK_SHARED | LK_NOPAUSE;
243 lkflags |= LK_INTERLOCK;
244 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
245 panic("vfs_busy: unexpected lock failure");
250 * Free a busy filesystem.
253 vfs_unbusy(struct mount *mp, struct thread *td)
255 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
259 * Lookup a filesystem type, and if found allocate and initialize
260 * a mount structure for it.
262 * Devname is usually updated by mount(8) after booting.
265 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
267 struct thread *td = curthread; /* XXX */
268 struct vfsconf *vfsp;
271 if (fstypename == NULL)
273 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
274 if (!strcmp(vfsp->vfc_name, fstypename))
278 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
279 bzero((char *)mp, (u_long)sizeof(struct mount));
280 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
281 (void)vfs_busy(mp, LK_NOWAIT, 0, td);
282 TAILQ_INIT(&mp->mnt_nvnodelist);
283 TAILQ_INIT(&mp->mnt_reservedvnlist);
284 mp->mnt_nvnodelistsize = 0;
286 mp->mnt_op = vfsp->vfc_vfsops;
287 mp->mnt_flag = MNT_RDONLY;
288 mp->mnt_vnodecovered = NULLVP;
289 vfsp->vfc_refcount++;
290 mp->mnt_iosize_max = DFLTPHYS;
291 mp->mnt_stat.f_type = vfsp->vfc_typenum;
292 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
293 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
294 mp->mnt_stat.f_mntonname[0] = '/';
295 mp->mnt_stat.f_mntonname[1] = 0;
296 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
302 * Find an appropriate filesystem to use for the root. If a filesystem
303 * has not been preselected, walk through the list of known filesystems
304 * trying those that have mountroot routines, and try them until one
305 * works or we have tried them all.
307 #ifdef notdef /* XXX JH */
309 lite2_vfs_mountroot()
311 struct vfsconf *vfsp;
312 extern int (*lite2_mountroot) (void);
315 if (lite2_mountroot != NULL)
316 return ((*lite2_mountroot)());
317 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
318 if (vfsp->vfc_mountroot == NULL)
320 if ((error = (*vfsp->vfc_mountroot)()) == 0)
322 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
329 * Lookup a mount point by filesystem identifier.
337 lwkt_gettoken(&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]) {
341 lwkt_reltoken(&mountlist_token);
345 lwkt_reltoken(&mountlist_token);
346 return ((struct mount *) 0);
350 * Get a new unique fsid. Try to make its val[0] unique, since this value
351 * will be used to create fake device numbers for stat(). Also try (but
352 * not so hard) make its val[0] unique mod 2^16, since some emulators only
353 * support 16-bit device numbers. We end up with unique val[0]'s for the
354 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
356 * Keep in mind that several mounts may be running in parallel. Starting
357 * the search one past where the previous search terminated is both a
358 * micro-optimization and a defense against returning the same fsid to
365 static u_int16_t mntid_base;
369 lwkt_gettoken(&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(&mntid_token);
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)
483 * Calculate the trigger point, don't allow user
484 * screwups to blow us up. This prevents us from
485 * recycling vnodes with lots of resident pages. We
486 * aren't trying to free memory, we are trying to
489 usevnodes = desiredvnodes;
492 trigger = vmstats.v_page_count * 2 / usevnodes;
495 gen = lwkt_gettoken(&mntvnode_token);
496 count = mp->mnt_nvnodelistsize / 10 + 1;
497 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
498 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
499 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
501 if (vp->v_type != VNON &&
502 vp->v_type != VBAD &&
503 VMIGHTFREE(vp) && /* critical path opt */
504 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger)
506 lwkt_gettoken(&vp->v_interlock);
507 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) {
508 if (VMIGHTFREE(vp)) {
509 vgonel(vp, curthread);
512 lwkt_reltoken(&vp->v_interlock);
515 lwkt_reltoken(&vp->v_interlock);
520 lwkt_reltoken(&mntvnode_token);
525 * Attempt to recycle vnodes in a context that is always safe to block.
526 * Calling vlrurecycle() from the bowels of file system code has some
527 * interesting deadlock problems.
529 static struct thread *vnlruthread;
530 static int vnlruproc_sig;
535 struct mount *mp, *nmp;
538 struct thread *td = curthread;
540 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
545 kproc_suspend_loop();
546 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
548 wakeup(&vnlruproc_sig);
549 tsleep(td, 0, "vlruwt", hz);
553 lwkt_gettoken(&mountlist_token);
554 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
555 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
556 nmp = TAILQ_NEXT(mp, mnt_list);
559 done += vlrureclaim(mp);
560 lwkt_gettoken(&mountlist_token);
561 nmp = TAILQ_NEXT(mp, mnt_list);
564 lwkt_reltoken(&mountlist_token);
567 tsleep(td, 0, "vlrup", hz * 3);
573 static struct kproc_desc vnlru_kp = {
578 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
581 * Routines having to do with the management of the vnode table.
583 extern vop_t **dead_vnodeop_p;
586 * Return the next vnode from the free list.
589 getnewvnode(tag, mp, vops, vpp)
598 struct thread *td = curthread; /* XXX */
599 struct vnode *vp = NULL;
605 * Try to reuse vnodes if we hit the max. This situation only
606 * occurs in certain large-memory (2G+) situations. We cannot
607 * attempt to directly reclaim vnodes due to nasty recursion
610 while (numvnodes - freevnodes > desiredvnodes) {
611 if (vnlruproc_sig == 0) {
612 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
615 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
620 * Attempt to reuse a vnode already on the free list, allocating
621 * a new vnode if we can't find one or if we have not reached a
622 * good minimum for good LRU performance.
624 gen = lwkt_gettoken(&vnode_free_list_token);
625 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
628 for (count = 0; count < freevnodes; count++) {
629 vp = TAILQ_FIRST(&vnode_free_list);
630 if (vp == NULL || vp->v_usecount)
631 panic("getnewvnode: free vnode isn't");
634 * Get the vnode's interlock, then re-obtain
635 * vnode_free_list_token in case we lost it. If we
636 * did lose it while getting the vnode interlock,
637 * even if we got it back again, then retry.
639 vgen = lwkt_gettoken(&vp->v_interlock);
640 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
642 lwkt_reltoken(&vp->v_interlock);
648 * Whew! We have both tokens. Since we didn't lose
649 * the free list VFREE had better still be set. But
650 * we aren't out of the woods yet. We have to get
651 * the object (may block). If the vnode is not
652 * suitable then move it to the end of the list
653 * if we can. If we can't move it to the end of the
656 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
657 (object->resident_page_count || object->ref_count))
659 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 &&
660 lwkt_gentoken(&vnode_free_list_token, &gen) == 0
662 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
663 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
667 lwkt_reltoken(&vp->v_interlock);
673 * Still not out of the woods. VOBJECT might have
674 * blocked, if we did not retain our tokens we have
677 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 ||
678 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
683 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
684 KKASSERT(vp->v_flag & VFREE);
687 * If we have children in the namecache we cannot
688 * reuse the vnode yet because it will break the
689 * namecache chain (YYY use nc_refs for the check?)
691 if (TAILQ_FIRST(&vp->v_namecache)) {
692 if (cache_leaf_test(vp) < 0) {
693 lwkt_reltoken(&vp->v_interlock);
694 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
704 vp->v_flag |= VDOOMED;
705 vp->v_flag &= ~VFREE;
707 lwkt_reltoken(&vnode_free_list_token);
708 cache_purge(vp); /* YYY may block */
710 if (vp->v_type != VBAD) {
713 lwkt_reltoken(&vp->v_interlock);
721 panic("cleaned vnode isn't");
724 panic("Clean vnode has pending I/O's");
734 vp->v_writecount = 0; /* XXX */
736 lwkt_reltoken(&vnode_free_list_token);
737 vp = (struct vnode *) zalloc(vnode_zone);
738 bzero((char *) vp, sizeof *vp);
739 lwkt_inittoken(&vp->v_interlock);
742 TAILQ_INIT(&vp->v_namecache);
746 TAILQ_INIT(&vp->v_cleanblkhd);
747 TAILQ_INIT(&vp->v_dirtyblkhd);
757 vfs_object_create(vp, td);
762 * Move a vnode from one mount queue to another.
770 lwkt_gettoken(&mntvnode_token);
772 * Delete from old mount point vnode list, if on one.
774 if (vp->v_mount != NULL) {
775 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
776 ("bad mount point vnode list size"));
777 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
778 vp->v_mount->mnt_nvnodelistsize--;
781 * Insert into list of vnodes for the new mount point, if available.
783 if ((vp->v_mount = mp) == NULL) {
784 lwkt_reltoken(&mntvnode_token);
787 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
788 mp->mnt_nvnodelistsize++;
789 lwkt_reltoken(&mntvnode_token);
793 * Update outstanding I/O count and do wakeup if requested.
801 bp->b_flags &= ~B_WRITEINPROG;
802 if ((vp = bp->b_vp)) {
804 if (vp->v_numoutput < 0)
805 panic("vwakeup: neg numoutput");
806 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
807 vp->v_flag &= ~VBWAIT;
808 wakeup((caddr_t) &vp->v_numoutput);
814 * Flush out and invalidate all buffers associated with a vnode.
815 * Called with the underlying object locked.
818 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
819 int slpflag, int slptimeo)
822 struct buf *nbp, *blist;
826 if (flags & V_SAVE) {
828 while (vp->v_numoutput) {
829 vp->v_flag |= VBWAIT;
830 error = tsleep((caddr_t)&vp->v_numoutput,
831 slpflag, "vinvlbuf", slptimeo);
837 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
839 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
842 if (vp->v_numoutput > 0 ||
843 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
844 panic("vinvalbuf: dirty bufs");
850 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
852 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
856 for (bp = blist; bp; bp = nbp) {
857 nbp = TAILQ_NEXT(bp, b_vnbufs);
858 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
859 error = BUF_TIMELOCK(bp,
860 LK_EXCLUSIVE | LK_SLEEPFAIL,
861 "vinvalbuf", slpflag, slptimeo);
868 * XXX Since there are no node locks for NFS, I
869 * believe there is a slight chance that a delayed
870 * write will occur while sleeping just above, so
871 * check for it. Note that vfs_bio_awrite expects
872 * buffers to reside on a queue, while VOP_BWRITE and
875 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
878 if (bp->b_vp == vp) {
879 if (bp->b_flags & B_CLUSTEROK) {
884 bp->b_flags |= B_ASYNC;
885 VOP_BWRITE(bp->b_vp, bp);
889 (void) VOP_BWRITE(bp->b_vp, bp);
894 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
895 bp->b_flags &= ~B_ASYNC;
901 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
902 * have write I/O in-progress but if there is a VM object then the
903 * VM object can also have read-I/O in-progress.
906 while (vp->v_numoutput > 0) {
907 vp->v_flag |= VBWAIT;
908 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
910 if (VOP_GETVOBJECT(vp, &object) == 0) {
911 while (object->paging_in_progress)
912 vm_object_pip_sleep(object, "vnvlbx");
914 } while (vp->v_numoutput > 0);
919 * Destroy the copy in the VM cache, too.
921 lwkt_gettoken(&vp->v_interlock);
922 if (VOP_GETVOBJECT(vp, &object) == 0) {
923 vm_object_page_remove(object, 0, 0,
924 (flags & V_SAVE) ? TRUE : FALSE);
926 lwkt_reltoken(&vp->v_interlock);
928 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
929 panic("vinvalbuf: flush failed");
934 * Truncate a file's buffer and pages to a specified length. This
935 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
939 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
947 * Round up to the *next* lbn.
949 trunclbn = (length + blksize - 1) / blksize;
956 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
957 nbp = TAILQ_NEXT(bp, b_vnbufs);
958 if (bp->b_lblkno >= trunclbn) {
959 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
960 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
964 bp->b_flags |= (B_INVAL | B_RELBUF);
965 bp->b_flags &= ~B_ASYNC;
970 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
972 (nbp->b_flags & B_DELWRI))) {
978 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
979 nbp = TAILQ_NEXT(bp, b_vnbufs);
980 if (bp->b_lblkno >= trunclbn) {
981 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
982 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
986 bp->b_flags |= (B_INVAL | B_RELBUF);
987 bp->b_flags &= ~B_ASYNC;
992 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
994 (nbp->b_flags & B_DELWRI) == 0)) {
1003 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1004 nbp = TAILQ_NEXT(bp, b_vnbufs);
1005 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1006 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1007 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1011 if (bp->b_vp == vp) {
1012 bp->b_flags |= B_ASYNC;
1014 bp->b_flags &= ~B_ASYNC;
1016 VOP_BWRITE(bp->b_vp, bp);
1024 while (vp->v_numoutput > 0) {
1025 vp->v_flag |= VBWAIT;
1026 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1031 vnode_pager_setsize(vp, length);
1037 * Associate a buffer with a vnode.
1046 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1050 bp->b_dev = vn_todev(vp);
1052 * Insert onto list for new vnode.
1055 bp->b_xflags |= BX_VNCLEAN;
1056 bp->b_xflags &= ~BX_VNDIRTY;
1057 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1062 * Disassociate a buffer from a vnode.
1069 struct buflists *listheadp;
1072 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1075 * Delete from old vnode list, if on one.
1079 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1080 if (bp->b_xflags & BX_VNDIRTY)
1081 listheadp = &vp->v_dirtyblkhd;
1083 listheadp = &vp->v_cleanblkhd;
1084 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1085 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1087 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1088 vp->v_flag &= ~VONWORKLST;
1089 LIST_REMOVE(vp, v_synclist);
1092 bp->b_vp = (struct vnode *) 0;
1097 * The workitem queue.
1099 * It is useful to delay writes of file data and filesystem metadata
1100 * for tens of seconds so that quickly created and deleted files need
1101 * not waste disk bandwidth being created and removed. To realize this,
1102 * we append vnodes to a "workitem" queue. When running with a soft
1103 * updates implementation, most pending metadata dependencies should
1104 * not wait for more than a few seconds. Thus, mounted on block devices
1105 * are delayed only about a half the time that file data is delayed.
1106 * Similarly, directory updates are more critical, so are only delayed
1107 * about a third the time that file data is delayed. Thus, there are
1108 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1109 * one each second (driven off the filesystem syncer process). The
1110 * syncer_delayno variable indicates the next queue that is to be processed.
1111 * Items that need to be processed soon are placed in this queue:
1113 * syncer_workitem_pending[syncer_delayno]
1115 * A delay of fifteen seconds is done by placing the request fifteen
1116 * entries later in the queue:
1118 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1123 * Add an item to the syncer work queue.
1126 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1132 if (vp->v_flag & VONWORKLST) {
1133 LIST_REMOVE(vp, v_synclist);
1136 if (delay > syncer_maxdelay - 2)
1137 delay = syncer_maxdelay - 2;
1138 slot = (syncer_delayno + delay) & syncer_mask;
1140 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1141 vp->v_flag |= VONWORKLST;
1145 struct thread *updatethread;
1146 static void sched_sync (void);
1147 static struct kproc_desc up_kp = {
1152 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1155 * System filesystem synchronizer daemon.
1160 struct synclist *slp;
1164 struct thread *td = curthread;
1166 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1170 kproc_suspend_loop();
1172 starttime = time_second;
1175 * Push files whose dirty time has expired. Be careful
1176 * of interrupt race on slp queue.
1179 slp = &syncer_workitem_pending[syncer_delayno];
1180 syncer_delayno += 1;
1181 if (syncer_delayno == syncer_maxdelay)
1185 while ((vp = LIST_FIRST(slp)) != NULL) {
1186 if (VOP_ISLOCKED(vp, NULL) == 0) {
1187 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1188 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1189 VOP_UNLOCK(vp, 0, td);
1192 if (LIST_FIRST(slp) == vp) {
1194 * Note: v_tag VT_VFS vps can remain on the
1195 * worklist too with no dirty blocks, but
1196 * since sync_fsync() moves it to a different
1199 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1200 !vn_isdisk(vp, NULL))
1201 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1203 * Put us back on the worklist. The worklist
1204 * routine will remove us from our current
1205 * position and then add us back in at a later
1208 vn_syncer_add_to_worklist(vp, syncdelay);
1214 * Do soft update processing.
1217 (*bioops.io_sync)(NULL);
1220 * The variable rushjob allows the kernel to speed up the
1221 * processing of the filesystem syncer process. A rushjob
1222 * value of N tells the filesystem syncer to process the next
1223 * N seconds worth of work on its queue ASAP. Currently rushjob
1224 * is used by the soft update code to speed up the filesystem
1225 * syncer process when the incore state is getting so far
1226 * ahead of the disk that the kernel memory pool is being
1227 * threatened with exhaustion.
1234 * If it has taken us less than a second to process the
1235 * current work, then wait. Otherwise start right over
1236 * again. We can still lose time if any single round
1237 * takes more than two seconds, but it does not really
1238 * matter as we are just trying to generally pace the
1239 * filesystem activity.
1241 if (time_second == starttime)
1242 tsleep(&lbolt, 0, "syncer", 0);
1247 * Request the syncer daemon to speed up its work.
1248 * We never push it to speed up more than half of its
1249 * normal turn time, otherwise it could take over the cpu.
1251 * YYY wchan field protected by the BGL.
1257 if (updatethread->td_wchan == &lbolt) { /* YYY */
1258 unsleep(updatethread);
1259 lwkt_schedule(updatethread);
1262 if (rushjob < syncdelay / 2) {
1264 stat_rush_requests += 1;
1271 * Associate a p-buffer with a vnode.
1273 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1274 * with the buffer. i.e. the bp has not been linked into the vnode or
1283 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1286 bp->b_flags |= B_PAGING;
1287 bp->b_dev = vn_todev(vp);
1291 * Disassociate a p-buffer from a vnode.
1298 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1301 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1303 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1308 bp->b_vp = (struct vnode *) 0;
1309 bp->b_flags &= ~B_PAGING;
1313 pbreassignbuf(bp, newvp)
1315 struct vnode *newvp;
1317 if ((bp->b_flags & B_PAGING) == 0) {
1319 "pbreassignbuf() on non phys bp %p",
1327 * Reassign a buffer from one vnode to another.
1328 * Used to assign file specific control information
1329 * (indirect blocks) to the vnode to which they belong.
1332 reassignbuf(bp, newvp)
1334 struct vnode *newvp;
1336 struct buflists *listheadp;
1340 if (newvp == NULL) {
1341 printf("reassignbuf: NULL");
1347 * B_PAGING flagged buffers cannot be reassigned because their vp
1348 * is not fully linked in.
1350 if (bp->b_flags & B_PAGING)
1351 panic("cannot reassign paging buffer");
1355 * Delete from old vnode list, if on one.
1357 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1358 if (bp->b_xflags & BX_VNDIRTY)
1359 listheadp = &bp->b_vp->v_dirtyblkhd;
1361 listheadp = &bp->b_vp->v_cleanblkhd;
1362 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1363 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1364 if (bp->b_vp != newvp) {
1366 bp->b_vp = NULL; /* for clarification */
1370 * If dirty, put on list of dirty buffers; otherwise insert onto list
1373 if (bp->b_flags & B_DELWRI) {
1376 listheadp = &newvp->v_dirtyblkhd;
1377 if ((newvp->v_flag & VONWORKLST) == 0) {
1378 switch (newvp->v_type) {
1384 if (newvp->v_specmountpoint != NULL) {
1392 vn_syncer_add_to_worklist(newvp, delay);
1394 bp->b_xflags |= BX_VNDIRTY;
1395 tbp = TAILQ_FIRST(listheadp);
1397 bp->b_lblkno == 0 ||
1398 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1399 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1400 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1401 ++reassignbufsortgood;
1402 } else if (bp->b_lblkno < 0) {
1403 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1404 ++reassignbufsortgood;
1405 } else if (reassignbufmethod == 1) {
1407 * New sorting algorithm, only handle sequential case,
1408 * otherwise append to end (but before metadata)
1410 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1411 (tbp->b_xflags & BX_VNDIRTY)) {
1413 * Found the best place to insert the buffer
1415 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1416 ++reassignbufsortgood;
1419 * Missed, append to end, but before meta-data.
1420 * We know that the head buffer in the list is
1421 * not meta-data due to prior conditionals.
1423 * Indirect effects: NFS second stage write
1424 * tends to wind up here, giving maximum
1425 * distance between the unstable write and the
1428 tbp = TAILQ_LAST(listheadp, buflists);
1429 while (tbp && tbp->b_lblkno < 0)
1430 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1431 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1432 ++reassignbufsortbad;
1436 * Old sorting algorithm, scan queue and insert
1439 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1440 (ttbp->b_lblkno < bp->b_lblkno)) {
1444 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1447 bp->b_xflags |= BX_VNCLEAN;
1448 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1449 if ((newvp->v_flag & VONWORKLST) &&
1450 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1451 newvp->v_flag &= ~VONWORKLST;
1452 LIST_REMOVE(newvp, v_synclist);
1455 if (bp->b_vp != newvp) {
1463 * Create a vnode for a block device.
1464 * Used for mounting the root file system.
1479 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1492 * Add a vnode to the alias list hung off the dev_t.
1494 * The reason for this gunk is that multiple vnodes can reference
1495 * the same physical device, so checking vp->v_usecount to see
1496 * how many users there are is inadequate; the v_usecount for
1497 * the vnodes need to be accumulated. vcount() does that.
1500 addaliasu(struct vnode *nvp, udev_t nvp_rdev)
1504 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1505 panic("addaliasu on non-special vnode");
1506 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
1515 addalias(struct vnode *nvp, dev_t dev)
1518 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1519 panic("addalias on non-special vnode");
1522 lwkt_gettoken(&spechash_token);
1523 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1524 lwkt_reltoken(&spechash_token);
1528 * Grab a particular vnode from the free list, increment its
1529 * reference count and lock it. The vnode lock bit is set if the
1530 * vnode is being eliminated in vgone. The process is awakened
1531 * when the transition is completed, and an error returned to
1532 * indicate that the vnode is no longer usable (possibly having
1533 * been changed to a new file system type).
1544 * If the vnode is in the process of being cleaned out for
1545 * another use, we wait for the cleaning to finish and then
1546 * return failure. Cleaning is determined by checking that
1547 * the VXLOCK flag is set.
1549 if ((flags & LK_INTERLOCK) == 0) {
1550 lwkt_gettoken(&vp->v_interlock);
1552 if (vp->v_flag & VXLOCK) {
1553 if (vp->v_vxproc == curproc) {
1555 /* this can now occur in normal operation */
1556 log(LOG_INFO, "VXLOCK interlock avoided\n");
1559 vp->v_flag |= VXWANT;
1560 lwkt_reltoken(&vp->v_interlock);
1561 tsleep((caddr_t)vp, 0, "vget", 0);
1568 if (VSHOULDBUSY(vp))
1570 if (flags & LK_TYPE_MASK) {
1571 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1573 * must expand vrele here because we do not want
1574 * to call VOP_INACTIVE if the reference count
1575 * drops back to zero since it was never really
1576 * active. We must remove it from the free list
1577 * before sleeping so that multiple processes do
1578 * not try to recycle it.
1580 lwkt_gettoken(&vp->v_interlock);
1582 if (VSHOULDFREE(vp))
1586 lwkt_reltoken(&vp->v_interlock);
1590 lwkt_reltoken(&vp->v_interlock);
1595 vref(struct vnode *vp)
1597 lwkt_gettoken(&vp->v_interlock);
1599 lwkt_reltoken(&vp->v_interlock);
1603 * Vnode put/release.
1604 * If count drops to zero, call inactive routine and return to freelist.
1607 vrele(struct vnode *vp)
1609 struct thread *td = curthread; /* XXX */
1611 KASSERT(vp != NULL, ("vrele: null vp"));
1613 lwkt_gettoken(&vp->v_interlock);
1615 if (vp->v_usecount > 1) {
1618 lwkt_reltoken(&vp->v_interlock);
1623 if (vp->v_usecount == 1) {
1626 * We must call VOP_INACTIVE with the node locked.
1627 * If we are doing a vpu, the node is already locked,
1628 * but, in the case of vrele, we must explicitly lock
1629 * the vnode before calling VOP_INACTIVE
1632 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1633 VOP_INACTIVE(vp, td);
1634 if (VSHOULDFREE(vp))
1640 vprint("vrele: negative ref count", vp);
1641 lwkt_reltoken(&vp->v_interlock);
1643 panic("vrele: negative ref cnt");
1648 vput(struct vnode *vp)
1650 struct thread *td = curthread; /* XXX */
1652 KASSERT(vp != NULL, ("vput: null vp"));
1654 lwkt_gettoken(&vp->v_interlock);
1656 if (vp->v_usecount > 1) {
1658 VOP_UNLOCK(vp, LK_INTERLOCK, td);
1662 if (vp->v_usecount == 1) {
1665 * We must call VOP_INACTIVE with the node locked.
1666 * If we are doing a vpu, the node is already locked,
1667 * so we just need to release the vnode mutex.
1669 lwkt_reltoken(&vp->v_interlock);
1670 VOP_INACTIVE(vp, td);
1671 if (VSHOULDFREE(vp))
1677 vprint("vput: negative ref count", vp);
1679 panic("vput: negative ref cnt");
1684 * Somebody doesn't want the vnode recycled.
1694 if (VSHOULDBUSY(vp))
1700 * One less who cares about this vnode.
1709 if (vp->v_holdcnt <= 0)
1710 panic("vdrop: holdcnt");
1712 if (VSHOULDFREE(vp))
1718 * Remove any vnodes in the vnode table belonging to mount point mp.
1720 * If FORCECLOSE is not specified, there should not be any active ones,
1721 * return error if any are found (nb: this is a user error, not a
1722 * system error). If FORCECLOSE is specified, detach any active vnodes
1725 * If WRITECLOSE is set, only flush out regular file vnodes open for
1728 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1730 * `rootrefs' specifies the base reference count for the root vnode
1731 * of this filesystem. The root vnode is considered busy if its
1732 * v_usecount exceeds this value. On a successful return, vflush()
1733 * will call vrele() on the root vnode exactly rootrefs times.
1734 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1738 static int busyprt = 0; /* print out busy vnodes */
1739 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1743 vflush(mp, rootrefs, flags)
1748 struct thread *td = curthread; /* XXX */
1749 struct vnode *vp, *nvp, *rootvp = NULL;
1751 int busy = 0, error;
1754 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1755 ("vflush: bad args"));
1757 * Get the filesystem root vnode. We can vput() it
1758 * immediately, since with rootrefs > 0, it won't go away.
1760 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1764 lwkt_gettoken(&mntvnode_token);
1766 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1768 * Make sure this vnode wasn't reclaimed in getnewvnode().
1769 * Start over if it has (it won't be on the list anymore).
1771 if (vp->v_mount != mp)
1773 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1775 lwkt_gettoken(&vp->v_interlock);
1777 * Skip over a vnodes marked VSYSTEM.
1779 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1780 lwkt_reltoken(&vp->v_interlock);
1784 * If WRITECLOSE is set, flush out unlinked but still open
1785 * files (even if open only for reading) and regular file
1786 * vnodes open for writing.
1788 if ((flags & WRITECLOSE) &&
1789 (vp->v_type == VNON ||
1790 (VOP_GETATTR(vp, &vattr, td) == 0 &&
1791 vattr.va_nlink > 0)) &&
1792 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1793 lwkt_reltoken(&vp->v_interlock);
1798 * With v_usecount == 0, all we need to do is clear out the
1799 * vnode data structures and we are done.
1801 if (vp->v_usecount == 0) {
1802 lwkt_reltoken(&mntvnode_token);
1804 lwkt_gettoken(&mntvnode_token);
1809 * If FORCECLOSE is set, forcibly close the vnode. For block
1810 * or character devices, revert to an anonymous device. For
1811 * all other files, just kill them.
1813 if (flags & FORCECLOSE) {
1814 lwkt_reltoken(&mntvnode_token);
1815 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1819 vp->v_op = spec_vnodeop_p;
1820 insmntque(vp, (struct mount *) 0);
1822 lwkt_gettoken(&mntvnode_token);
1827 vprint("vflush: busy vnode", vp);
1829 lwkt_reltoken(&vp->v_interlock);
1832 lwkt_reltoken(&mntvnode_token);
1833 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1835 * If just the root vnode is busy, and if its refcount
1836 * is equal to `rootrefs', then go ahead and kill it.
1838 lwkt_gettoken(&rootvp->v_interlock);
1839 KASSERT(busy > 0, ("vflush: not busy"));
1840 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1841 if (busy == 1 && rootvp->v_usecount == rootrefs) {
1845 lwkt_reltoken(&rootvp->v_interlock);
1849 for (; rootrefs > 0; rootrefs--)
1855 * We do not want to recycle the vnode too quickly.
1857 * XXX we can't move vp's around the nvnodelist without really screwing
1858 * up the efficiency of filesystem SYNC and friends. This code is
1859 * disabled until we fix the syncing code's scanning algorithm.
1862 vlruvp(struct vnode *vp)
1867 if ((mp = vp->v_mount) != NULL) {
1868 lwkt_gettoken(&mntvnode_token);
1869 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1870 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1871 lwkt_reltoken(&mntvnode_token);
1877 * Disassociate the underlying file system from a vnode.
1880 vclean(struct vnode *vp, int flags, struct thread *td)
1885 * Check to see if the vnode is in use. If so we have to reference it
1886 * before we clean it out so that its count cannot fall to zero and
1887 * generate a race against ourselves to recycle it.
1889 if ((active = vp->v_usecount))
1893 * Prevent the vnode from being recycled or brought into use while we
1896 if (vp->v_flag & VXLOCK)
1897 panic("vclean: deadlock");
1898 vp->v_flag |= VXLOCK;
1899 vp->v_vxproc = curproc;
1901 * Even if the count is zero, the VOP_INACTIVE routine may still
1902 * have the object locked while it cleans it out. The VOP_LOCK
1903 * ensures that the VOP_INACTIVE routine is done with its work.
1904 * For active vnodes, it ensures that no other activity can
1905 * occur while the underlying object is being cleaned out.
1907 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
1910 * Clean out any buffers associated with the vnode.
1912 vinvalbuf(vp, V_SAVE, td, 0, 0);
1914 VOP_DESTROYVOBJECT(vp);
1917 * If purging an active vnode, it must be closed and
1918 * deactivated before being reclaimed. Note that the
1919 * VOP_INACTIVE will unlock the vnode.
1922 if (flags & DOCLOSE)
1923 VOP_CLOSE(vp, FNONBLOCK, td);
1924 VOP_INACTIVE(vp, td);
1927 * Any other processes trying to obtain this lock must first
1928 * wait for VXLOCK to clear, then call the new lock operation.
1930 VOP_UNLOCK(vp, 0, td);
1933 * Reclaim the vnode.
1935 if (VOP_RECLAIM(vp, td))
1936 panic("vclean: cannot reclaim");
1940 * Inline copy of vrele() since VOP_INACTIVE
1941 * has already been called.
1943 lwkt_gettoken(&vp->v_interlock);
1944 if (--vp->v_usecount <= 0) {
1946 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1947 vprint("vclean: bad ref count", vp);
1948 panic("vclean: ref cnt");
1953 lwkt_reltoken(&vp->v_interlock);
1957 vp->v_vnlock = NULL;
1959 if (VSHOULDFREE(vp))
1963 * Done with purge, notify sleepers of the grim news.
1965 vp->v_op = dead_vnodeop_p;
1968 vp->v_flag &= ~VXLOCK;
1969 vp->v_vxproc = NULL;
1970 if (vp->v_flag & VXWANT) {
1971 vp->v_flag &= ~VXWANT;
1972 wakeup((caddr_t) vp);
1977 * Eliminate all activity associated with the requested vnode
1978 * and with all vnodes aliased to the requested vnode.
1982 struct vop_revoke_args /* {
1987 struct vnode *vp, *vq;
1990 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1994 * If a vgone (or vclean) is already in progress,
1995 * wait until it is done and return.
1997 if (vp->v_flag & VXLOCK) {
1998 vp->v_flag |= VXWANT;
1999 lwkt_reltoken(&vp->v_interlock);
2000 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2005 lwkt_gettoken(&spechash_token);
2006 vq = SLIST_FIRST(&dev->si_hlist);
2007 lwkt_reltoken(&spechash_token);
2016 * Recycle an unused vnode to the front of the free list.
2017 * Release the passed interlock if the vnode will be recycled.
2020 vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td)
2022 lwkt_gettoken(&vp->v_interlock);
2023 if (vp->v_usecount == 0) {
2025 lwkt_reltoken(inter_lkp);
2030 lwkt_reltoken(&vp->v_interlock);
2035 * Eliminate all activity associated with a vnode
2036 * in preparation for reuse.
2039 vgone(struct vnode *vp)
2041 struct thread *td = curthread; /* XXX */
2043 lwkt_gettoken(&vp->v_interlock);
2048 * vgone, with the vp interlock held.
2051 vgonel(struct vnode *vp, struct thread *td)
2056 * If a vgone (or vclean) is already in progress,
2057 * wait until it is done and return.
2059 if (vp->v_flag & VXLOCK) {
2060 vp->v_flag |= VXWANT;
2061 lwkt_reltoken(&vp->v_interlock);
2062 tsleep((caddr_t)vp, 0, "vgone", 0);
2067 * Clean out the filesystem specific data.
2069 vclean(vp, DOCLOSE, td);
2070 lwkt_gettoken(&vp->v_interlock);
2073 * Delete from old mount point vnode list, if on one.
2075 if (vp->v_mount != NULL)
2076 insmntque(vp, (struct mount *)0);
2078 * If special device, remove it from special device alias list
2081 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2082 lwkt_gettoken(&spechash_token);
2083 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2084 freedev(vp->v_rdev);
2085 lwkt_reltoken(&spechash_token);
2090 * If it is on the freelist and not already at the head,
2091 * move it to the head of the list. The test of the
2092 * VDOOMED flag and the reference count of zero is because
2093 * it will be removed from the free list by getnewvnode,
2094 * but will not have its reference count incremented until
2095 * after calling vgone. If the reference count were
2096 * incremented first, vgone would (incorrectly) try to
2097 * close the previous instance of the underlying object.
2099 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2101 lwkt_gettoken(&vnode_free_list_token);
2102 if (vp->v_flag & VFREE)
2103 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2106 vp->v_flag |= VFREE;
2107 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2108 lwkt_reltoken(&vnode_free_list_token);
2113 lwkt_reltoken(&vp->v_interlock);
2117 * Lookup a vnode by device number.
2120 vfinddev(dev, type, vpp)
2127 lwkt_gettoken(&spechash_token);
2128 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2129 if (type == vp->v_type) {
2131 lwkt_reltoken(&spechash_token);
2135 lwkt_reltoken(&spechash_token);
2140 * Calculate the total number of references to a special device.
2150 lwkt_gettoken(&spechash_token);
2151 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2152 count += vq->v_usecount;
2153 lwkt_reltoken(&spechash_token);
2158 * Same as above, but using the dev_t as argument
2167 vp = SLIST_FIRST(&dev->si_hlist);
2174 * Print out a description of a vnode.
2176 static char *typename[] =
2177 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2187 printf("%s: %p: ", label, (void *)vp);
2189 printf("%p: ", (void *)vp);
2190 printf("type %s, usecount %d, writecount %d, refcount %d,",
2191 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2194 if (vp->v_flag & VROOT)
2195 strcat(buf, "|VROOT");
2196 if (vp->v_flag & VTEXT)
2197 strcat(buf, "|VTEXT");
2198 if (vp->v_flag & VSYSTEM)
2199 strcat(buf, "|VSYSTEM");
2200 if (vp->v_flag & VXLOCK)
2201 strcat(buf, "|VXLOCK");
2202 if (vp->v_flag & VXWANT)
2203 strcat(buf, "|VXWANT");
2204 if (vp->v_flag & VBWAIT)
2205 strcat(buf, "|VBWAIT");
2206 if (vp->v_flag & VDOOMED)
2207 strcat(buf, "|VDOOMED");
2208 if (vp->v_flag & VFREE)
2209 strcat(buf, "|VFREE");
2210 if (vp->v_flag & VOBJBUF)
2211 strcat(buf, "|VOBJBUF");
2213 printf(" flags (%s)", &buf[1]);
2214 if (vp->v_data == NULL) {
2223 #include <ddb/ddb.h>
2225 * List all of the locked vnodes in the system.
2226 * Called when debugging the kernel.
2228 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2230 struct thread *td = curthread; /* XXX */
2231 struct mount *mp, *nmp;
2234 printf("Locked vnodes\n");
2235 lwkt_gettoken(&mountlist_token);
2236 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2237 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
2238 nmp = TAILQ_NEXT(mp, mnt_list);
2241 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2242 if (VOP_ISLOCKED(vp, NULL))
2243 vprint((char *)0, vp);
2245 lwkt_gettoken(&mountlist_token);
2246 nmp = TAILQ_NEXT(mp, mnt_list);
2249 lwkt_reltoken(&mountlist_token);
2254 * Top level filesystem related information gathering.
2256 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2259 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2261 int *name = (int *)arg1 - 1; /* XXX */
2262 u_int namelen = arg2 + 1; /* XXX */
2263 struct vfsconf *vfsp;
2265 #if 1 || defined(COMPAT_PRELITE2)
2266 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2268 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2272 /* all sysctl names at this level are at least name and field */
2274 return (ENOTDIR); /* overloaded */
2275 if (name[0] != VFS_GENERIC) {
2276 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2277 if (vfsp->vfc_typenum == name[0])
2280 return (EOPNOTSUPP);
2281 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2282 oldp, oldlenp, newp, newlen, p));
2286 case VFS_MAXTYPENUM:
2289 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2292 return (ENOTDIR); /* overloaded */
2293 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2294 if (vfsp->vfc_typenum == name[2])
2297 return (EOPNOTSUPP);
2298 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2300 return (EOPNOTSUPP);
2303 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2304 "Generic filesystem");
2306 #if 1 || defined(COMPAT_PRELITE2)
2309 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2312 struct vfsconf *vfsp;
2313 struct ovfsconf ovfs;
2315 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2316 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2317 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2318 ovfs.vfc_index = vfsp->vfc_typenum;
2319 ovfs.vfc_refcount = vfsp->vfc_refcount;
2320 ovfs.vfc_flags = vfsp->vfc_flags;
2321 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2328 #endif /* 1 || COMPAT_PRELITE2 */
2331 #define KINFO_VNODESLOP 10
2333 * Dump vnode list (via sysctl).
2334 * Copyout address of vnode followed by vnode.
2338 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2340 struct proc *p = curproc; /* XXX */
2341 struct mount *mp, *nmp;
2342 struct vnode *nvp, *vp;
2345 #define VPTRSZ sizeof (struct vnode *)
2346 #define VNODESZ sizeof (struct vnode)
2349 if (!req->oldptr) /* Make an estimate */
2350 return (SYSCTL_OUT(req, 0,
2351 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2353 lwkt_gettoken(&mountlist_token);
2354 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2355 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) {
2356 nmp = TAILQ_NEXT(mp, mnt_list);
2360 lwkt_gettoken(&mntvnode_token);
2361 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2365 * Check that the vp is still associated with
2366 * this filesystem. RACE: could have been
2367 * recycled onto the same filesystem.
2369 if (vp->v_mount != mp) {
2370 lwkt_reltoken(&mntvnode_token);
2373 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2374 lwkt_reltoken(&mntvnode_token);
2375 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2376 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2378 lwkt_gettoken(&mntvnode_token);
2380 lwkt_reltoken(&mntvnode_token);
2381 lwkt_gettoken(&mountlist_token);
2382 nmp = TAILQ_NEXT(mp, mnt_list);
2385 lwkt_reltoken(&mountlist_token);
2393 * Exporting the vnode list on large systems causes them to crash.
2394 * Exporting the vnode list on medium systems causes sysctl to coredump.
2397 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2398 0, 0, sysctl_vnode, "S,vnode", "");
2402 * Check to see if a filesystem is mounted on a block device.
2409 if (vp->v_specmountpoint != NULL)
2415 * Unmount all filesystems. The list is traversed in reverse order
2416 * of mounting to avoid dependencies.
2422 struct thread *td = curthread;
2425 if (td->td_proc == NULL)
2426 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2429 * Since this only runs when rebooting, it is not interlocked.
2431 while(!TAILQ_EMPTY(&mountlist)) {
2432 mp = TAILQ_LAST(&mountlist, mntlist);
2433 error = dounmount(mp, MNT_FORCE, td);
2435 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2436 printf("unmount of %s failed (",
2437 mp->mnt_stat.f_mntonname);
2441 printf("%d)\n", error);
2443 /* The unmount has removed mp from the mountlist */
2449 * Build hash lists of net addresses and hang them off the mount point.
2450 * Called by ufs_mount() to set up the lists of export addresses.
2453 vfs_hang_addrlist(mp, nep, argp)
2455 struct netexport *nep;
2456 struct export_args *argp;
2459 struct radix_node_head *rnh;
2461 struct radix_node *rn;
2462 struct sockaddr *saddr, *smask = 0;
2466 if (argp->ex_addrlen == 0) {
2467 if (mp->mnt_flag & MNT_DEFEXPORTED)
2469 np = &nep->ne_defexported;
2470 np->netc_exflags = argp->ex_flags;
2471 np->netc_anon = argp->ex_anon;
2472 np->netc_anon.cr_ref = 1;
2473 mp->mnt_flag |= MNT_DEFEXPORTED;
2477 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2479 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2482 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2483 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2484 bzero((caddr_t) np, i);
2485 saddr = (struct sockaddr *) (np + 1);
2486 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2488 if (saddr->sa_len > argp->ex_addrlen)
2489 saddr->sa_len = argp->ex_addrlen;
2490 if (argp->ex_masklen) {
2491 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2492 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2495 if (smask->sa_len > argp->ex_masklen)
2496 smask->sa_len = argp->ex_masklen;
2498 i = saddr->sa_family;
2499 if ((rnh = nep->ne_rtable[i]) == 0) {
2501 * Seems silly to initialize every AF when most are not used,
2502 * do so on demand here
2504 for (dom = domains; dom; dom = dom->dom_next)
2505 if (dom->dom_family == i && dom->dom_rtattach) {
2506 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2510 if ((rnh = nep->ne_rtable[i]) == 0) {
2515 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2517 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2521 np->netc_exflags = argp->ex_flags;
2522 np->netc_anon = argp->ex_anon;
2523 np->netc_anon.cr_ref = 1;
2526 free(np, M_NETADDR);
2532 vfs_free_netcred(rn, w)
2533 struct radix_node *rn;
2536 struct radix_node_head *rnh = (struct radix_node_head *) w;
2538 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2539 free((caddr_t) rn, M_NETADDR);
2544 * Free the net address hash lists that are hanging off the mount points.
2547 vfs_free_addrlist(nep)
2548 struct netexport *nep;
2551 struct radix_node_head *rnh;
2553 for (i = 0; i <= AF_MAX; i++)
2554 if ((rnh = nep->ne_rtable[i])) {
2555 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2557 free((caddr_t) rnh, M_RTABLE);
2558 nep->ne_rtable[i] = 0;
2563 vfs_export(mp, nep, argp)
2565 struct netexport *nep;
2566 struct export_args *argp;
2570 if (argp->ex_flags & MNT_DELEXPORT) {
2571 if (mp->mnt_flag & MNT_EXPUBLIC) {
2572 vfs_setpublicfs(NULL, NULL, NULL);
2573 mp->mnt_flag &= ~MNT_EXPUBLIC;
2575 vfs_free_addrlist(nep);
2576 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2578 if (argp->ex_flags & MNT_EXPORTED) {
2579 if (argp->ex_flags & MNT_EXPUBLIC) {
2580 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2582 mp->mnt_flag |= MNT_EXPUBLIC;
2584 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2586 mp->mnt_flag |= MNT_EXPORTED;
2593 * Set the publicly exported filesystem (WebNFS). Currently, only
2594 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2597 vfs_setpublicfs(mp, nep, argp)
2599 struct netexport *nep;
2600 struct export_args *argp;
2607 * mp == NULL -> invalidate the current info, the FS is
2608 * no longer exported. May be called from either vfs_export
2609 * or unmount, so check if it hasn't already been done.
2612 if (nfs_pub.np_valid) {
2613 nfs_pub.np_valid = 0;
2614 if (nfs_pub.np_index != NULL) {
2615 FREE(nfs_pub.np_index, M_TEMP);
2616 nfs_pub.np_index = NULL;
2623 * Only one allowed at a time.
2625 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2629 * Get real filehandle for root of exported FS.
2631 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2632 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2634 if ((error = VFS_ROOT(mp, &rvp)))
2637 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2643 * If an indexfile was specified, pull it in.
2645 if (argp->ex_indexfile != NULL) {
2646 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2648 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2649 MAXNAMLEN, (size_t *)0);
2652 * Check for illegal filenames.
2654 for (cp = nfs_pub.np_index; *cp; cp++) {
2662 FREE(nfs_pub.np_index, M_TEMP);
2667 nfs_pub.np_mount = mp;
2668 nfs_pub.np_valid = 1;
2673 vfs_export_lookup(mp, nep, nam)
2675 struct netexport *nep;
2676 struct sockaddr *nam;
2679 struct radix_node_head *rnh;
2680 struct sockaddr *saddr;
2683 if (mp->mnt_flag & MNT_EXPORTED) {
2685 * Lookup in the export list first.
2689 rnh = nep->ne_rtable[saddr->sa_family];
2691 np = (struct netcred *)
2692 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2694 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2699 * If no address match, use the default if it exists.
2701 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2702 np = &nep->ne_defexported;
2708 * perform msync on all vnodes under a mount point
2709 * the mount point must be locked.
2712 vfs_msync(struct mount *mp, int flags)
2714 struct thread *td = curthread; /* XXX */
2715 struct vnode *vp, *nvp;
2716 struct vm_object *obj;
2720 lwkt_gettoken(&mntvnode_token);
2722 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2723 if (vp->v_mount != mp) {
2728 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2730 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2734 * There could be hundreds of thousands of vnodes, we cannot
2735 * afford to do anything heavy-weight until we have a fairly
2736 * good indication that there is something to do.
2738 if ((vp->v_flag & VOBJDIRTY) &&
2739 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2740 lwkt_reltoken(&mntvnode_token);
2742 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
2743 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2744 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2748 lwkt_gettoken(&mntvnode_token);
2749 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2756 lwkt_reltoken(&mntvnode_token);
2760 * Create the VM object needed for VMIO and mmap support. This
2761 * is done for all VREG files in the system. Some filesystems might
2762 * afford the additional metadata buffering capability of the
2763 * VMIO code by making the device node be VMIO mode also.
2765 * vp must be locked when vfs_object_create is called.
2768 vfs_object_create(struct vnode *vp, struct thread *td)
2770 return (VOP_CREATEVOBJECT(vp, td));
2780 lwkt_gettoken(&vnode_free_list_token);
2781 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2782 if (vp->v_flag & VAGE) {
2783 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2785 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2788 lwkt_reltoken(&vnode_free_list_token);
2789 vp->v_flag &= ~VAGE;
2790 vp->v_flag |= VFREE;
2801 lwkt_gettoken(&vnode_free_list_token);
2802 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2803 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2805 lwkt_reltoken(&vnode_free_list_token);
2806 vp->v_flag &= ~(VFREE|VAGE);
2811 * Record a process's interest in events which might happen to
2812 * a vnode. Because poll uses the historic select-style interface
2813 * internally, this routine serves as both the ``check for any
2814 * pending events'' and the ``record my interest in future events''
2815 * functions. (These are done together, while the lock is held,
2816 * to avoid race conditions.)
2819 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
2821 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2822 if (vp->v_pollinfo.vpi_revents & events) {
2824 * This leaves events we are not interested
2825 * in available for the other process which
2826 * which presumably had requested them
2827 * (otherwise they would never have been
2830 events &= vp->v_pollinfo.vpi_revents;
2831 vp->v_pollinfo.vpi_revents &= ~events;
2833 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2836 vp->v_pollinfo.vpi_events |= events;
2837 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
2838 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2843 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2844 * it is possible for us to miss an event due to race conditions, but
2845 * that condition is expected to be rare, so for the moment it is the
2846 * preferred interface.
2849 vn_pollevent(vp, events)
2853 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2854 if (vp->v_pollinfo.vpi_events & events) {
2856 * We clear vpi_events so that we don't
2857 * call selwakeup() twice if two events are
2858 * posted before the polling process(es) is
2859 * awakened. This also ensures that we take at
2860 * most one selwakeup() if the polling process
2861 * is no longer interested. However, it does
2862 * mean that only one event can be noticed at
2863 * a time. (Perhaps we should only clear those
2864 * event bits which we note?) XXX
2866 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2867 vp->v_pollinfo.vpi_revents |= events;
2868 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2870 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2874 * Wake up anyone polling on vp because it is being revoked.
2875 * This depends on dead_poll() returning POLLHUP for correct
2882 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2883 if (vp->v_pollinfo.vpi_events) {
2884 vp->v_pollinfo.vpi_events = 0;
2885 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2887 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2893 * Routine to create and manage a filesystem syncer vnode.
2895 #define sync_close ((int (*) (struct vop_close_args *))nullop)
2896 static int sync_fsync (struct vop_fsync_args *);
2897 static int sync_inactive (struct vop_inactive_args *);
2898 static int sync_reclaim (struct vop_reclaim_args *);
2899 #define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
2900 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
2901 static int sync_print (struct vop_print_args *);
2902 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
2904 static vop_t **sync_vnodeop_p;
2905 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2906 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2907 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2908 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2909 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2910 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2911 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2912 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2913 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2914 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2917 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2918 { &sync_vnodeop_p, sync_vnodeop_entries };
2920 VNODEOP_SET(sync_vnodeop_opv_desc);
2923 * Create a new filesystem syncer vnode for the specified mount point.
2926 vfs_allocate_syncvnode(mp)
2930 static long start, incr, next;
2933 /* Allocate a new vnode */
2934 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2935 mp->mnt_syncer = NULL;
2940 * Place the vnode onto the syncer worklist. We attempt to
2941 * scatter them about on the list so that they will go off
2942 * at evenly distributed times even if all the filesystems
2943 * are mounted at once.
2946 if (next == 0 || next > syncer_maxdelay) {
2950 start = syncer_maxdelay / 2;
2951 incr = syncer_maxdelay;
2955 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2956 mp->mnt_syncer = vp;
2961 * Do a lazy sync of the filesystem.
2965 struct vop_fsync_args /* {
2967 struct ucred *a_cred;
2969 struct thread *a_td;
2972 struct vnode *syncvp = ap->a_vp;
2973 struct mount *mp = syncvp->v_mount;
2974 struct thread *td = ap->a_td;
2978 * We only need to do something if this is a lazy evaluation.
2980 if (ap->a_waitfor != MNT_LAZY)
2984 * Move ourselves to the back of the sync list.
2986 vn_syncer_add_to_worklist(syncvp, syncdelay);
2989 * Walk the list of vnodes pushing all that are dirty and
2990 * not already on the sync list.
2992 lwkt_gettoken(&mountlist_token);
2993 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) {
2994 lwkt_reltoken(&mountlist_token);
2997 asyncflag = mp->mnt_flag & MNT_ASYNC;
2998 mp->mnt_flag &= ~MNT_ASYNC;
2999 vfs_msync(mp, MNT_NOWAIT);
3000 VFS_SYNC(mp, MNT_LAZY, td);
3002 mp->mnt_flag |= MNT_ASYNC;
3008 * The syncer vnode is no referenced.
3012 struct vop_inactive_args /* {
3023 * The syncer vnode is no longer needed and is being decommissioned.
3025 * Modifications to the worklist must be protected at splbio().
3029 struct vop_reclaim_args /* {
3033 struct vnode *vp = ap->a_vp;
3037 vp->v_mount->mnt_syncer = NULL;
3038 if (vp->v_flag & VONWORKLST) {
3039 LIST_REMOVE(vp, v_synclist);
3040 vp->v_flag &= ~VONWORKLST;
3048 * Print out a syncer vnode.
3052 struct vop_print_args /* {
3056 struct vnode *vp = ap->a_vp;
3058 printf("syncer vnode");
3059 if (vp->v_vnlock != NULL)
3060 lockmgr_printinfo(vp->v_vnlock);
3066 * extract the dev_t from a VBLK or VCHR
3072 if (vp->v_type != VBLK && vp->v_type != VCHR)
3074 return (vp->v_rdev);
3078 * Check if vnode represents a disk device
3085 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3090 if (vp->v_rdev == NULL) {
3095 if (!dev_dport(vp->v_rdev)) {
3100 if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
3112 struct nameidata *ndp;
3115 if (!(flags & NDF_NO_FREE_PNBUF) &&
3116 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3117 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3118 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3120 if (!(flags & NDF_NO_DNCP_RELE) &&
3121 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3123 cache_drop(ndp->ni_dncp);
3124 ndp->ni_dncp = NULL;
3126 if (!(flags & NDF_NO_NCP_RELE) &&
3127 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3129 cache_drop(ndp->ni_ncp);
3132 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3133 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3134 ndp->ni_dvp != ndp->ni_vp) {
3135 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
3137 if (!(flags & NDF_NO_DVP_RELE) &&
3138 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3142 if (!(flags & NDF_NO_VP_UNLOCK) &&
3143 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3144 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
3146 if (!(flags & NDF_NO_VP_RELE) &&
3151 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3152 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3153 vrele(ndp->ni_startdir);
3154 ndp->ni_startdir = NULL;