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 1.12 2003/07/19 21:14:39 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>
77 #include <vm/vm_map.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vnode_pager.h>
81 #include <vm/vm_zone.h>
84 #include <sys/thread2.h>
86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
88 static void insmntque __P((struct vnode *vp, struct mount *mp));
89 static void vclean __P((struct vnode *vp, int flags, struct thread *td));
90 static unsigned long numvnodes;
91 static void vlruvp(struct vnode *vp);
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, "");
120 static int nameileafonly = 0;
121 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
123 #ifdef ENABLE_VFS_IOOPT
125 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
128 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
129 struct lwkt_token mountlist_token;
130 struct lwkt_token mntvnode_token;
131 int nfs_mount_type = -1;
132 static struct lwkt_token mntid_token;
133 static struct lwkt_token vnode_free_list_token;
134 static struct lwkt_token spechash_token;
135 struct nfs_public nfs_pub; /* publicly exported FS */
136 static vm_zone_t vnode_zone;
139 * The workitem queue.
141 #define SYNCER_MAXDELAY 32
142 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
143 time_t syncdelay = 30; /* max time to delay syncing data */
144 time_t filedelay = 30; /* time to delay syncing files */
145 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
146 time_t dirdelay = 29; /* time to delay syncing directories */
147 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
148 time_t metadelay = 28; /* time to delay syncing metadata */
149 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
150 static int rushjob; /* number of slots to run ASAP */
151 static int stat_rush_requests; /* number of times I/O speeded up */
152 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
154 static int syncer_delayno = 0;
155 static long syncer_mask;
156 LIST_HEAD(synclist, vnode);
157 static struct synclist *syncer_workitem_pending;
160 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
161 &desiredvnodes, 0, "Maximum number of vnodes");
162 static int minvnodes;
163 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
164 &minvnodes, 0, "Minimum number of vnodes");
165 static int vnlru_nowhere = 0;
166 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
167 "Number of times the vnlru process ran without success");
169 static void vfs_free_addrlist __P((struct netexport *nep));
170 static int vfs_free_netcred __P((struct radix_node *rn, void *w));
171 static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
172 struct export_args *argp));
175 * Initialize the vnode management data structures.
181 desiredvnodes = maxproc + vmstats.v_page_count / 4;
182 minvnodes = desiredvnodes / 4;
183 lwkt_inittoken(&mntvnode_token);
184 lwkt_inittoken(&mntid_token);
185 lwkt_inittoken(&spechash_token);
186 TAILQ_INIT(&vnode_free_list);
187 lwkt_inittoken(&vnode_free_list_token);
188 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
190 * Initialize the filesystem syncer.
192 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
194 syncer_maxdelay = syncer_mask + 1;
198 * Mark a mount point as busy. Used to synchronize access and to delay
199 * unmounting. Interlock is not released on failure.
202 vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp,
207 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
208 if (flags & LK_NOWAIT)
210 mp->mnt_kern_flag |= MNTK_MWAIT;
212 lwkt_reltoken(interlkp);
215 * Since all busy locks are shared except the exclusive
216 * lock granted when unmounting, the only place that a
217 * wakeup needs to be done is at the release of the
218 * exclusive lock at the end of dounmount.
220 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
222 lwkt_gettoken(interlkp);
226 lkflags = LK_SHARED | LK_NOPAUSE;
228 lkflags |= LK_INTERLOCK;
229 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
230 panic("vfs_busy: unexpected lock failure");
235 * Free a busy filesystem.
238 vfs_unbusy(struct mount *mp, struct thread *td)
240 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
244 * Lookup a filesystem type, and if found allocate and initialize
245 * a mount structure for it.
247 * Devname is usually updated by mount(8) after booting.
250 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
252 struct thread *td = curthread; /* XXX */
253 struct vfsconf *vfsp;
256 if (fstypename == NULL)
258 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
259 if (!strcmp(vfsp->vfc_name, fstypename))
263 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
264 bzero((char *)mp, (u_long)sizeof(struct mount));
265 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
266 (void)vfs_busy(mp, LK_NOWAIT, 0, td);
267 TAILQ_INIT(&mp->mnt_nvnodelist);
268 TAILQ_INIT(&mp->mnt_reservedvnlist);
269 mp->mnt_nvnodelistsize = 0;
271 mp->mnt_op = vfsp->vfc_vfsops;
272 mp->mnt_flag = MNT_RDONLY;
273 mp->mnt_vnodecovered = NULLVP;
274 vfsp->vfc_refcount++;
275 mp->mnt_iosize_max = DFLTPHYS;
276 mp->mnt_stat.f_type = vfsp->vfc_typenum;
277 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
278 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
279 mp->mnt_stat.f_mntonname[0] = '/';
280 mp->mnt_stat.f_mntonname[1] = 0;
281 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
287 * Find an appropriate filesystem to use for the root. If a filesystem
288 * has not been preselected, walk through the list of known filesystems
289 * trying those that have mountroot routines, and try them until one
290 * works or we have tried them all.
292 #ifdef notdef /* XXX JH */
294 lite2_vfs_mountroot()
296 struct vfsconf *vfsp;
297 extern int (*lite2_mountroot) __P((void));
300 if (lite2_mountroot != NULL)
301 return ((*lite2_mountroot)());
302 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
303 if (vfsp->vfc_mountroot == NULL)
305 if ((error = (*vfsp->vfc_mountroot)()) == 0)
307 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
314 * Lookup a mount point by filesystem identifier.
320 register struct mount *mp;
322 lwkt_gettoken(&mountlist_token);
323 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
324 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
325 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
326 lwkt_reltoken(&mountlist_token);
330 lwkt_reltoken(&mountlist_token);
331 return ((struct mount *) 0);
335 * Get a new unique fsid. Try to make its val[0] unique, since this value
336 * will be used to create fake device numbers for stat(). Also try (but
337 * not so hard) make its val[0] unique mod 2^16, since some emulators only
338 * support 16-bit device numbers. We end up with unique val[0]'s for the
339 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
341 * Keep in mind that several mounts may be running in parallel. Starting
342 * the search one past where the previous search terminated is both a
343 * micro-optimization and a defense against returning the same fsid to
350 static u_int16_t mntid_base;
354 lwkt_gettoken(&mntid_token);
355 mtype = mp->mnt_vfc->vfc_typenum;
356 tfsid.val[1] = mtype;
357 mtype = (mtype & 0xFF) << 24;
359 tfsid.val[0] = makeudev(255,
360 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
362 if (vfs_getvfs(&tfsid) == NULL)
365 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
366 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
367 lwkt_reltoken(&mntid_token);
371 * Knob to control the precision of file timestamps:
373 * 0 = seconds only; nanoseconds zeroed.
374 * 1 = seconds and nanoseconds, accurate within 1/HZ.
375 * 2 = seconds and nanoseconds, truncated to microseconds.
376 * >=3 = seconds and nanoseconds, maximum precision.
378 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
380 static int timestamp_precision = TSP_SEC;
381 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
382 ×tamp_precision, 0, "");
385 * Get a current timestamp.
389 struct timespec *tsp;
393 switch (timestamp_precision) {
395 tsp->tv_sec = time_second;
403 TIMEVAL_TO_TIMESPEC(&tv, tsp);
413 * Set vnode attributes to VNOVAL
417 register struct vattr *vap;
421 vap->va_size = VNOVAL;
422 vap->va_bytes = VNOVAL;
423 vap->va_mode = VNOVAL;
424 vap->va_nlink = VNOVAL;
425 vap->va_uid = VNOVAL;
426 vap->va_gid = VNOVAL;
427 vap->va_fsid = VNOVAL;
428 vap->va_fileid = VNOVAL;
429 vap->va_blocksize = VNOVAL;
430 vap->va_rdev = VNOVAL;
431 vap->va_atime.tv_sec = VNOVAL;
432 vap->va_atime.tv_nsec = VNOVAL;
433 vap->va_mtime.tv_sec = VNOVAL;
434 vap->va_mtime.tv_nsec = VNOVAL;
435 vap->va_ctime.tv_sec = VNOVAL;
436 vap->va_ctime.tv_nsec = VNOVAL;
437 vap->va_flags = VNOVAL;
438 vap->va_gen = VNOVAL;
443 * This routine is called when we have too many vnodes. It attempts
444 * to free <count> vnodes and will potentially free vnodes that still
445 * have VM backing store (VM backing store is typically the cause
446 * of a vnode blowout so we want to do this). Therefore, this operation
447 * is not considered cheap.
449 * A number of conditions may prevent a vnode from being reclaimed.
450 * the buffer cache may have references on the vnode, a directory
451 * vnode may still have references due to the namei cache representing
452 * underlying files, or the vnode may be in active use. It is not
453 * desireable to reuse such vnodes. These conditions may cause the
454 * number of vnodes to reach some minimum value regardless of what
455 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
458 vlrureclaim(struct mount *mp)
468 * Calculate the trigger point, don't allow user
469 * screwups to blow us up. This prevents us from
470 * recycling vnodes with lots of resident pages. We
471 * aren't trying to free memory, we are trying to
474 usevnodes = desiredvnodes;
477 trigger = vmstats.v_page_count * 2 / usevnodes;
480 gen = lwkt_gettoken(&mntvnode_token);
481 count = mp->mnt_nvnodelistsize / 10 + 1;
482 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
483 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
484 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
486 if (vp->v_type != VNON &&
487 vp->v_type != VBAD &&
488 VMIGHTFREE(vp) && /* critical path opt */
489 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger)
491 lwkt_gettoken(&vp->v_interlock);
492 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) {
493 if (VMIGHTFREE(vp)) {
494 vgonel(vp, curthread);
497 lwkt_reltoken(&vp->v_interlock);
500 lwkt_reltoken(&vp->v_interlock);
505 lwkt_reltoken(&mntvnode_token);
510 * Attempt to recycle vnodes in a context that is always safe to block.
511 * Calling vlrurecycle() from the bowels of file system code has some
512 * interesting deadlock problems.
514 static struct thread *vnlruthread;
515 static int vnlruproc_sig;
520 struct mount *mp, *nmp;
523 struct thread *td = curthread;
525 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
530 kproc_suspend_loop();
531 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
533 wakeup(&vnlruproc_sig);
534 tsleep(td, 0, "vlruwt", hz);
538 lwkt_gettoken(&mountlist_token);
539 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
540 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
541 nmp = TAILQ_NEXT(mp, mnt_list);
544 done += vlrureclaim(mp);
545 lwkt_gettoken(&mountlist_token);
546 nmp = TAILQ_NEXT(mp, mnt_list);
549 lwkt_reltoken(&mountlist_token);
552 tsleep(td, 0, "vlrup", hz * 3);
558 static struct kproc_desc vnlru_kp = {
563 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
566 * Routines having to do with the management of the vnode table.
568 extern vop_t **dead_vnodeop_p;
571 * Return the next vnode from the free list.
574 getnewvnode(tag, mp, vops, vpp)
583 struct thread *td = curthread; /* XXX */
584 struct vnode *vp = NULL;
590 * Try to reuse vnodes if we hit the max. This situation only
591 * occurs in certain large-memory (2G+) situations. We cannot
592 * attempt to directly reclaim vnodes due to nasty recursion
595 while (numvnodes - freevnodes > desiredvnodes) {
596 if (vnlruproc_sig == 0) {
597 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
600 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
605 * Attempt to reuse a vnode already on the free list, allocating
606 * a new vnode if we can't find one or if we have not reached a
607 * good minimum for good LRU performance.
609 gen = lwkt_gettoken(&vnode_free_list_token);
610 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
613 for (count = 0; count < freevnodes; count++) {
614 vp = TAILQ_FIRST(&vnode_free_list);
615 if (vp == NULL || vp->v_usecount)
616 panic("getnewvnode: free vnode isn't");
619 * Get the vnode's interlock, then re-obtain
620 * vnode_free_list_token in case we lost it. If we
621 * did lose it while getting the vnode interlock,
622 * even if we got it back again, then retry.
624 vgen = lwkt_gettoken(&vp->v_interlock);
625 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
627 lwkt_reltoken(&vp->v_interlock);
633 * Whew! We have both tokens. Since we didn't lose
634 * the free list VFREE had better still be set. But
635 * we aren't out of the woods yet. We have to get
636 * the object (may block). If the vnode is not
637 * suitable then move it to the end of the list
638 * if we can. If we can't move it to the end of the
641 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
642 (object->resident_page_count || object->ref_count))
644 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 &&
645 lwkt_gentoken(&vnode_free_list_token, &gen) == 0
647 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
648 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
652 lwkt_reltoken(&vp->v_interlock);
658 * Still not out of the woods. VOBJECT might have
659 * blocked, if we did not retain our tokens we have
662 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 ||
663 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
668 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
669 KKASSERT(vp->v_flag & VFREE);
671 if (LIST_FIRST(&vp->v_cache_src)) {
673 * note: nameileafonly sysctl is temporary,
674 * for debugging only, and will eventually be
677 if (nameileafonly > 0) {
679 * Do not reuse namei-cached directory
680 * vnodes that have cached
683 if (cache_leaf_test(vp) < 0) {
684 lwkt_reltoken(&vp->v_interlock);
685 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
689 } else if (nameileafonly < 0 ||
690 vmiodirenable == 0) {
692 * Do not reuse namei-cached directory
693 * vnodes if nameileafonly is -1 or
694 * if VMIO backing for directories is
695 * turned off (otherwise we reuse them
698 lwkt_reltoken(&vp->v_interlock);
699 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
709 vp->v_flag |= VDOOMED;
710 vp->v_flag &= ~VFREE;
712 lwkt_reltoken(&vnode_free_list_token);
713 cache_purge(vp); /* YYY may block */
715 if (vp->v_type != VBAD) {
718 lwkt_reltoken(&vp->v_interlock);
726 panic("cleaned vnode isn't");
729 panic("Clean vnode has pending I/O's");
739 vp->v_writecount = 0; /* XXX */
741 lwkt_reltoken(&vnode_free_list_token);
742 vp = (struct vnode *) zalloc(vnode_zone);
743 bzero((char *) vp, sizeof *vp);
744 lwkt_inittoken(&vp->v_interlock);
747 LIST_INIT(&vp->v_cache_src);
748 TAILQ_INIT(&vp->v_cache_dst);
752 TAILQ_INIT(&vp->v_cleanblkhd);
753 TAILQ_INIT(&vp->v_dirtyblkhd);
763 vfs_object_create(vp, td);
768 * Move a vnode from one mount queue to another.
772 register struct vnode *vp;
773 register struct mount *mp;
776 lwkt_gettoken(&mntvnode_token);
778 * Delete from old mount point vnode list, if on one.
780 if (vp->v_mount != NULL) {
781 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
782 ("bad mount point vnode list size"));
783 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
784 vp->v_mount->mnt_nvnodelistsize--;
787 * Insert into list of vnodes for the new mount point, if available.
789 if ((vp->v_mount = mp) == NULL) {
790 lwkt_reltoken(&mntvnode_token);
793 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
794 mp->mnt_nvnodelistsize++;
795 lwkt_reltoken(&mntvnode_token);
799 * Update outstanding I/O count and do wakeup if requested.
803 register struct buf *bp;
805 register struct vnode *vp;
807 bp->b_flags &= ~B_WRITEINPROG;
808 if ((vp = bp->b_vp)) {
810 if (vp->v_numoutput < 0)
811 panic("vwakeup: neg numoutput");
812 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
813 vp->v_flag &= ~VBWAIT;
814 wakeup((caddr_t) &vp->v_numoutput);
820 * Flush out and invalidate all buffers associated with a vnode.
821 * Called with the underlying object locked.
824 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
825 int slpflag, int slptimeo)
827 register struct buf *bp;
828 struct buf *nbp, *blist;
832 if (flags & V_SAVE) {
834 while (vp->v_numoutput) {
835 vp->v_flag |= VBWAIT;
836 error = tsleep((caddr_t)&vp->v_numoutput,
837 slpflag, "vinvlbuf", slptimeo);
843 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
845 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
848 if (vp->v_numoutput > 0 ||
849 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
850 panic("vinvalbuf: dirty bufs");
856 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
858 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
862 for (bp = blist; bp; bp = nbp) {
863 nbp = TAILQ_NEXT(bp, b_vnbufs);
864 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
865 error = BUF_TIMELOCK(bp,
866 LK_EXCLUSIVE | LK_SLEEPFAIL,
867 "vinvalbuf", slpflag, slptimeo);
874 * XXX Since there are no node locks for NFS, I
875 * believe there is a slight chance that a delayed
876 * write will occur while sleeping just above, so
877 * check for it. Note that vfs_bio_awrite expects
878 * buffers to reside on a queue, while VOP_BWRITE and
881 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
884 if (bp->b_vp == vp) {
885 if (bp->b_flags & B_CLUSTEROK) {
890 bp->b_flags |= B_ASYNC;
891 VOP_BWRITE(bp->b_vp, bp);
895 (void) VOP_BWRITE(bp->b_vp, bp);
900 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
901 bp->b_flags &= ~B_ASYNC;
907 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
908 * have write I/O in-progress but if there is a VM object then the
909 * VM object can also have read-I/O in-progress.
912 while (vp->v_numoutput > 0) {
913 vp->v_flag |= VBWAIT;
914 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
916 if (VOP_GETVOBJECT(vp, &object) == 0) {
917 while (object->paging_in_progress)
918 vm_object_pip_sleep(object, "vnvlbx");
920 } while (vp->v_numoutput > 0);
925 * Destroy the copy in the VM cache, too.
927 lwkt_gettoken(&vp->v_interlock);
928 if (VOP_GETVOBJECT(vp, &object) == 0) {
929 vm_object_page_remove(object, 0, 0,
930 (flags & V_SAVE) ? TRUE : FALSE);
932 lwkt_reltoken(&vp->v_interlock);
934 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
935 panic("vinvalbuf: flush failed");
940 * Truncate a file's buffer and pages to a specified length. This
941 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
945 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
953 * Round up to the *next* lbn.
955 trunclbn = (length + blksize - 1) / blksize;
962 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
963 nbp = TAILQ_NEXT(bp, b_vnbufs);
964 if (bp->b_lblkno >= trunclbn) {
965 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
966 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
970 bp->b_flags |= (B_INVAL | B_RELBUF);
971 bp->b_flags &= ~B_ASYNC;
976 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
978 (nbp->b_flags & B_DELWRI))) {
984 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
985 nbp = TAILQ_NEXT(bp, b_vnbufs);
986 if (bp->b_lblkno >= trunclbn) {
987 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
988 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
992 bp->b_flags |= (B_INVAL | B_RELBUF);
993 bp->b_flags &= ~B_ASYNC;
998 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1000 (nbp->b_flags & B_DELWRI) == 0)) {
1009 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1010 nbp = TAILQ_NEXT(bp, b_vnbufs);
1011 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1012 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1013 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1017 if (bp->b_vp == vp) {
1018 bp->b_flags |= B_ASYNC;
1020 bp->b_flags &= ~B_ASYNC;
1022 VOP_BWRITE(bp->b_vp, bp);
1030 while (vp->v_numoutput > 0) {
1031 vp->v_flag |= VBWAIT;
1032 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1037 vnode_pager_setsize(vp, length);
1043 * Associate a buffer with a vnode.
1047 register struct vnode *vp;
1048 register struct buf *bp;
1052 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1056 bp->b_dev = vn_todev(vp);
1058 * Insert onto list for new vnode.
1061 bp->b_xflags |= BX_VNCLEAN;
1062 bp->b_xflags &= ~BX_VNDIRTY;
1063 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1068 * Disassociate a buffer from a vnode.
1072 register struct buf *bp;
1075 struct buflists *listheadp;
1078 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1081 * Delete from old vnode list, if on one.
1085 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1086 if (bp->b_xflags & BX_VNDIRTY)
1087 listheadp = &vp->v_dirtyblkhd;
1089 listheadp = &vp->v_cleanblkhd;
1090 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1091 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1093 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1094 vp->v_flag &= ~VONWORKLST;
1095 LIST_REMOVE(vp, v_synclist);
1098 bp->b_vp = (struct vnode *) 0;
1103 * The workitem queue.
1105 * It is useful to delay writes of file data and filesystem metadata
1106 * for tens of seconds so that quickly created and deleted files need
1107 * not waste disk bandwidth being created and removed. To realize this,
1108 * we append vnodes to a "workitem" queue. When running with a soft
1109 * updates implementation, most pending metadata dependencies should
1110 * not wait for more than a few seconds. Thus, mounted on block devices
1111 * are delayed only about a half the time that file data is delayed.
1112 * Similarly, directory updates are more critical, so are only delayed
1113 * about a third the time that file data is delayed. Thus, there are
1114 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1115 * one each second (driven off the filesystem syncer process). The
1116 * syncer_delayno variable indicates the next queue that is to be processed.
1117 * Items that need to be processed soon are placed in this queue:
1119 * syncer_workitem_pending[syncer_delayno]
1121 * A delay of fifteen seconds is done by placing the request fifteen
1122 * entries later in the queue:
1124 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1129 * Add an item to the syncer work queue.
1132 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1138 if (vp->v_flag & VONWORKLST) {
1139 LIST_REMOVE(vp, v_synclist);
1142 if (delay > syncer_maxdelay - 2)
1143 delay = syncer_maxdelay - 2;
1144 slot = (syncer_delayno + delay) & syncer_mask;
1146 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1147 vp->v_flag |= VONWORKLST;
1151 struct thread *updatethread;
1152 static void sched_sync __P((void));
1153 static struct kproc_desc up_kp = {
1158 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1161 * System filesystem synchronizer daemon.
1166 struct synclist *slp;
1170 struct thread *td = curthread;
1172 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1176 kproc_suspend_loop();
1178 starttime = time_second;
1181 * Push files whose dirty time has expired. Be careful
1182 * of interrupt race on slp queue.
1185 slp = &syncer_workitem_pending[syncer_delayno];
1186 syncer_delayno += 1;
1187 if (syncer_delayno == syncer_maxdelay)
1191 while ((vp = LIST_FIRST(slp)) != NULL) {
1192 if (VOP_ISLOCKED(vp, NULL) == 0) {
1193 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1194 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1195 VOP_UNLOCK(vp, 0, td);
1198 if (LIST_FIRST(slp) == vp) {
1200 * Note: v_tag VT_VFS vps can remain on the
1201 * worklist too with no dirty blocks, but
1202 * since sync_fsync() moves it to a different
1205 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1206 !vn_isdisk(vp, NULL))
1207 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1209 * Put us back on the worklist. The worklist
1210 * routine will remove us from our current
1211 * position and then add us back in at a later
1214 vn_syncer_add_to_worklist(vp, syncdelay);
1220 * Do soft update processing.
1223 (*bioops.io_sync)(NULL);
1226 * The variable rushjob allows the kernel to speed up the
1227 * processing of the filesystem syncer process. A rushjob
1228 * value of N tells the filesystem syncer to process the next
1229 * N seconds worth of work on its queue ASAP. Currently rushjob
1230 * is used by the soft update code to speed up the filesystem
1231 * syncer process when the incore state is getting so far
1232 * ahead of the disk that the kernel memory pool is being
1233 * threatened with exhaustion.
1240 * If it has taken us less than a second to process the
1241 * current work, then wait. Otherwise start right over
1242 * again. We can still lose time if any single round
1243 * takes more than two seconds, but it does not really
1244 * matter as we are just trying to generally pace the
1245 * filesystem activity.
1247 if (time_second == starttime)
1248 tsleep(&lbolt, 0, "syncer", 0);
1253 * Request the syncer daemon to speed up its work.
1254 * We never push it to speed up more than half of its
1255 * normal turn time, otherwise it could take over the cpu.
1257 * YYY wchan field protected by the BGL.
1263 if (updatethread->td_wchan == &lbolt) { /* YYY */
1264 unsleep(updatethread);
1265 lwkt_schedule(updatethread);
1268 if (rushjob < syncdelay / 2) {
1270 stat_rush_requests += 1;
1277 * Associate a p-buffer with a vnode.
1279 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1280 * with the buffer. i.e. the bp has not been linked into the vnode or
1285 register struct vnode *vp;
1286 register struct buf *bp;
1289 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1292 bp->b_flags |= B_PAGING;
1293 bp->b_dev = vn_todev(vp);
1297 * Disassociate a p-buffer from a vnode.
1301 register struct buf *bp;
1304 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1307 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1309 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1314 bp->b_vp = (struct vnode *) 0;
1315 bp->b_flags &= ~B_PAGING;
1319 pbreassignbuf(bp, newvp)
1321 struct vnode *newvp;
1323 if ((bp->b_flags & B_PAGING) == 0) {
1325 "pbreassignbuf() on non phys bp %p",
1333 * Reassign a buffer from one vnode to another.
1334 * Used to assign file specific control information
1335 * (indirect blocks) to the vnode to which they belong.
1338 reassignbuf(bp, newvp)
1339 register struct buf *bp;
1340 register struct vnode *newvp;
1342 struct buflists *listheadp;
1346 if (newvp == NULL) {
1347 printf("reassignbuf: NULL");
1353 * B_PAGING flagged buffers cannot be reassigned because their vp
1354 * is not fully linked in.
1356 if (bp->b_flags & B_PAGING)
1357 panic("cannot reassign paging buffer");
1361 * Delete from old vnode list, if on one.
1363 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1364 if (bp->b_xflags & BX_VNDIRTY)
1365 listheadp = &bp->b_vp->v_dirtyblkhd;
1367 listheadp = &bp->b_vp->v_cleanblkhd;
1368 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1369 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1370 if (bp->b_vp != newvp) {
1372 bp->b_vp = NULL; /* for clarification */
1376 * If dirty, put on list of dirty buffers; otherwise insert onto list
1379 if (bp->b_flags & B_DELWRI) {
1382 listheadp = &newvp->v_dirtyblkhd;
1383 if ((newvp->v_flag & VONWORKLST) == 0) {
1384 switch (newvp->v_type) {
1390 if (newvp->v_specmountpoint != NULL) {
1398 vn_syncer_add_to_worklist(newvp, delay);
1400 bp->b_xflags |= BX_VNDIRTY;
1401 tbp = TAILQ_FIRST(listheadp);
1403 bp->b_lblkno == 0 ||
1404 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1405 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1406 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1407 ++reassignbufsortgood;
1408 } else if (bp->b_lblkno < 0) {
1409 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1410 ++reassignbufsortgood;
1411 } else if (reassignbufmethod == 1) {
1413 * New sorting algorithm, only handle sequential case,
1414 * otherwise append to end (but before metadata)
1416 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1417 (tbp->b_xflags & BX_VNDIRTY)) {
1419 * Found the best place to insert the buffer
1421 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1422 ++reassignbufsortgood;
1425 * Missed, append to end, but before meta-data.
1426 * We know that the head buffer in the list is
1427 * not meta-data due to prior conditionals.
1429 * Indirect effects: NFS second stage write
1430 * tends to wind up here, giving maximum
1431 * distance between the unstable write and the
1434 tbp = TAILQ_LAST(listheadp, buflists);
1435 while (tbp && tbp->b_lblkno < 0)
1436 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1437 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1438 ++reassignbufsortbad;
1442 * Old sorting algorithm, scan queue and insert
1445 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1446 (ttbp->b_lblkno < bp->b_lblkno)) {
1450 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1453 bp->b_xflags |= BX_VNCLEAN;
1454 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1455 if ((newvp->v_flag & VONWORKLST) &&
1456 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1457 newvp->v_flag &= ~VONWORKLST;
1458 LIST_REMOVE(newvp, v_synclist);
1461 if (bp->b_vp != newvp) {
1469 * Create a vnode for a block device.
1470 * Used for mounting the root file system.
1477 register struct vnode *vp;
1485 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1498 * Add vnode to the alias list hung off the dev_t.
1500 * The reason for this gunk is that multiple vnodes can reference
1501 * the same physical device, so checking vp->v_usecount to see
1502 * how many users there are is inadequate; the v_usecount for
1503 * the vnodes need to be accumulated. vcount() does that.
1506 addaliasu(nvp, nvp_rdev)
1511 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1512 panic("addaliasu on non-special vnode");
1513 addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
1522 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1523 panic("addalias on non-special vnode");
1526 lwkt_gettoken(&spechash_token);
1527 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1528 lwkt_reltoken(&spechash_token);
1532 * Grab a particular vnode from the free list, increment its
1533 * reference count and lock it. The vnode lock bit is set if the
1534 * vnode is being eliminated in vgone. The process is awakened
1535 * when the transition is completed, and an error returned to
1536 * indicate that the vnode is no longer usable (possibly having
1537 * been changed to a new file system type).
1548 * If the vnode is in the process of being cleaned out for
1549 * another use, we wait for the cleaning to finish and then
1550 * return failure. Cleaning is determined by checking that
1551 * the VXLOCK flag is set.
1553 if ((flags & LK_INTERLOCK) == 0) {
1554 lwkt_gettoken(&vp->v_interlock);
1556 if (vp->v_flag & VXLOCK) {
1557 if (vp->v_vxproc == curproc) {
1559 /* this can now occur in normal operation */
1560 log(LOG_INFO, "VXLOCK interlock avoided\n");
1563 vp->v_flag |= VXWANT;
1564 lwkt_reltoken(&vp->v_interlock);
1565 tsleep((caddr_t)vp, 0, "vget", 0);
1572 if (VSHOULDBUSY(vp))
1574 if (flags & LK_TYPE_MASK) {
1575 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1577 * must expand vrele here because we do not want
1578 * to call VOP_INACTIVE if the reference count
1579 * drops back to zero since it was never really
1580 * active. We must remove it from the free list
1581 * before sleeping so that multiple processes do
1582 * not try to recycle it.
1584 lwkt_gettoken(&vp->v_interlock);
1586 if (VSHOULDFREE(vp))
1590 lwkt_reltoken(&vp->v_interlock);
1594 lwkt_reltoken(&vp->v_interlock);
1599 vref(struct vnode *vp)
1601 lwkt_gettoken(&vp->v_interlock);
1603 lwkt_reltoken(&vp->v_interlock);
1607 * Vnode put/release.
1608 * If count drops to zero, call inactive routine and return to freelist.
1611 vrele(struct vnode *vp)
1613 struct thread *td = curthread; /* XXX */
1615 KASSERT(vp != NULL, ("vrele: null vp"));
1617 lwkt_gettoken(&vp->v_interlock);
1619 if (vp->v_usecount > 1) {
1622 lwkt_reltoken(&vp->v_interlock);
1627 if (vp->v_usecount == 1) {
1630 * We must call VOP_INACTIVE with the node locked.
1631 * If we are doing a vpu, the node is already locked,
1632 * but, in the case of vrele, we must explicitly lock
1633 * the vnode before calling VOP_INACTIVE
1636 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1637 VOP_INACTIVE(vp, td);
1638 if (VSHOULDFREE(vp))
1644 vprint("vrele: negative ref count", vp);
1645 lwkt_reltoken(&vp->v_interlock);
1647 panic("vrele: negative ref cnt");
1652 vput(struct vnode *vp)
1654 struct thread *td = curthread; /* XXX */
1656 KASSERT(vp != NULL, ("vput: null vp"));
1658 lwkt_gettoken(&vp->v_interlock);
1660 if (vp->v_usecount > 1) {
1662 VOP_UNLOCK(vp, LK_INTERLOCK, td);
1666 if (vp->v_usecount == 1) {
1669 * We must call VOP_INACTIVE with the node locked.
1670 * If we are doing a vpu, the node is already locked,
1671 * so we just need to release the vnode mutex.
1673 lwkt_reltoken(&vp->v_interlock);
1674 VOP_INACTIVE(vp, td);
1675 if (VSHOULDFREE(vp))
1681 vprint("vput: negative ref count", vp);
1683 panic("vput: negative ref cnt");
1688 * Somebody doesn't want the vnode recycled.
1692 register struct vnode *vp;
1698 if (VSHOULDBUSY(vp))
1704 * One less who cares about this vnode.
1708 register struct vnode *vp;
1713 if (vp->v_holdcnt <= 0)
1714 panic("vdrop: holdcnt");
1716 if (VSHOULDFREE(vp))
1722 * Remove any vnodes in the vnode table belonging to mount point mp.
1724 * If FORCECLOSE is not specified, there should not be any active ones,
1725 * return error if any are found (nb: this is a user error, not a
1726 * system error). If FORCECLOSE is specified, detach any active vnodes
1729 * If WRITECLOSE is set, only flush out regular file vnodes open for
1732 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1734 * `rootrefs' specifies the base reference count for the root vnode
1735 * of this filesystem. The root vnode is considered busy if its
1736 * v_usecount exceeds this value. On a successful return, vflush()
1737 * will call vrele() on the root vnode exactly rootrefs times.
1738 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1742 static int busyprt = 0; /* print out busy vnodes */
1743 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1747 vflush(mp, rootrefs, flags)
1752 struct thread *td = curthread; /* XXX */
1753 struct vnode *vp, *nvp, *rootvp = NULL;
1755 int busy = 0, error;
1758 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1759 ("vflush: bad args"));
1761 * Get the filesystem root vnode. We can vput() it
1762 * immediately, since with rootrefs > 0, it won't go away.
1764 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1768 lwkt_gettoken(&mntvnode_token);
1770 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1772 * Make sure this vnode wasn't reclaimed in getnewvnode().
1773 * Start over if it has (it won't be on the list anymore).
1775 if (vp->v_mount != mp)
1777 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1779 lwkt_gettoken(&vp->v_interlock);
1781 * Skip over a vnodes marked VSYSTEM.
1783 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1784 lwkt_reltoken(&vp->v_interlock);
1788 * If WRITECLOSE is set, flush out unlinked but still open
1789 * files (even if open only for reading) and regular file
1790 * vnodes open for writing.
1792 if ((flags & WRITECLOSE) &&
1793 (vp->v_type == VNON ||
1794 (VOP_GETATTR(vp, &vattr, td) == 0 &&
1795 vattr.va_nlink > 0)) &&
1796 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1797 lwkt_reltoken(&vp->v_interlock);
1802 * With v_usecount == 0, all we need to do is clear out the
1803 * vnode data structures and we are done.
1805 if (vp->v_usecount == 0) {
1806 lwkt_reltoken(&mntvnode_token);
1808 lwkt_gettoken(&mntvnode_token);
1813 * If FORCECLOSE is set, forcibly close the vnode. For block
1814 * or character devices, revert to an anonymous device. For
1815 * all other files, just kill them.
1817 if (flags & FORCECLOSE) {
1818 lwkt_reltoken(&mntvnode_token);
1819 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1823 vp->v_op = spec_vnodeop_p;
1824 insmntque(vp, (struct mount *) 0);
1826 lwkt_gettoken(&mntvnode_token);
1831 vprint("vflush: busy vnode", vp);
1833 lwkt_reltoken(&vp->v_interlock);
1836 lwkt_reltoken(&mntvnode_token);
1837 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1839 * If just the root vnode is busy, and if its refcount
1840 * is equal to `rootrefs', then go ahead and kill it.
1842 lwkt_gettoken(&rootvp->v_interlock);
1843 KASSERT(busy > 0, ("vflush: not busy"));
1844 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1845 if (busy == 1 && rootvp->v_usecount == rootrefs) {
1849 lwkt_reltoken(&rootvp->v_interlock);
1853 for (; rootrefs > 0; rootrefs--)
1859 * We do not want to recycle the vnode too quickly.
1861 * XXX we can't move vp's around the nvnodelist without really screwing
1862 * up the efficiency of filesystem SYNC and friends. This code is
1863 * disabled until we fix the syncing code's scanning algorithm.
1866 vlruvp(struct vnode *vp)
1871 if ((mp = vp->v_mount) != NULL) {
1872 lwkt_gettoken(&mntvnode_token);
1873 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1874 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1875 lwkt_reltoken(&mntvnode_token);
1881 * Disassociate the underlying file system from a vnode.
1884 vclean(struct vnode *vp, int flags, struct thread *td)
1889 * Check to see if the vnode is in use. If so we have to reference it
1890 * before we clean it out so that its count cannot fall to zero and
1891 * generate a race against ourselves to recycle it.
1893 if ((active = vp->v_usecount))
1897 * Prevent the vnode from being recycled or brought into use while we
1900 if (vp->v_flag & VXLOCK)
1901 panic("vclean: deadlock");
1902 vp->v_flag |= VXLOCK;
1903 vp->v_vxproc = curproc;
1905 * Even if the count is zero, the VOP_INACTIVE routine may still
1906 * have the object locked while it cleans it out. The VOP_LOCK
1907 * ensures that the VOP_INACTIVE routine is done with its work.
1908 * For active vnodes, it ensures that no other activity can
1909 * occur while the underlying object is being cleaned out.
1911 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
1914 * Clean out any buffers associated with the vnode.
1916 vinvalbuf(vp, V_SAVE, td, 0, 0);
1918 VOP_DESTROYVOBJECT(vp);
1921 * If purging an active vnode, it must be closed and
1922 * deactivated before being reclaimed. Note that the
1923 * VOP_INACTIVE will unlock the vnode.
1926 if (flags & DOCLOSE)
1927 VOP_CLOSE(vp, FNONBLOCK, td);
1928 VOP_INACTIVE(vp, td);
1931 * Any other processes trying to obtain this lock must first
1932 * wait for VXLOCK to clear, then call the new lock operation.
1934 VOP_UNLOCK(vp, 0, td);
1937 * Reclaim the vnode.
1939 if (VOP_RECLAIM(vp, td))
1940 panic("vclean: cannot reclaim");
1944 * Inline copy of vrele() since VOP_INACTIVE
1945 * has already been called.
1947 lwkt_gettoken(&vp->v_interlock);
1948 if (--vp->v_usecount <= 0) {
1950 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1951 vprint("vclean: bad ref count", vp);
1952 panic("vclean: ref cnt");
1957 lwkt_reltoken(&vp->v_interlock);
1961 vp->v_vnlock = NULL;
1963 if (VSHOULDFREE(vp))
1967 * Done with purge, notify sleepers of the grim news.
1969 vp->v_op = dead_vnodeop_p;
1972 vp->v_flag &= ~VXLOCK;
1973 vp->v_vxproc = NULL;
1974 if (vp->v_flag & VXWANT) {
1975 vp->v_flag &= ~VXWANT;
1976 wakeup((caddr_t) vp);
1981 * Eliminate all activity associated with the requested vnode
1982 * and with all vnodes aliased to the requested vnode.
1986 struct vop_revoke_args /* {
1991 struct vnode *vp, *vq;
1994 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1998 * If a vgone (or vclean) is already in progress,
1999 * wait until it is done and return.
2001 if (vp->v_flag & VXLOCK) {
2002 vp->v_flag |= VXWANT;
2003 lwkt_reltoken(&vp->v_interlock);
2004 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2009 lwkt_gettoken(&spechash_token);
2010 vq = SLIST_FIRST(&dev->si_hlist);
2011 lwkt_reltoken(&spechash_token);
2020 * Recycle an unused vnode to the front of the free list.
2021 * Release the passed interlock if the vnode will be recycled.
2024 vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td)
2026 lwkt_gettoken(&vp->v_interlock);
2027 if (vp->v_usecount == 0) {
2029 lwkt_reltoken(inter_lkp);
2034 lwkt_reltoken(&vp->v_interlock);
2039 * Eliminate all activity associated with a vnode
2040 * in preparation for reuse.
2043 vgone(struct vnode *vp)
2045 struct thread *td = curthread; /* XXX */
2047 lwkt_gettoken(&vp->v_interlock);
2052 * vgone, with the vp interlock held.
2055 vgonel(struct vnode *vp, struct thread *td)
2060 * If a vgone (or vclean) is already in progress,
2061 * wait until it is done and return.
2063 if (vp->v_flag & VXLOCK) {
2064 vp->v_flag |= VXWANT;
2065 lwkt_reltoken(&vp->v_interlock);
2066 tsleep((caddr_t)vp, 0, "vgone", 0);
2071 * Clean out the filesystem specific data.
2073 vclean(vp, DOCLOSE, td);
2074 lwkt_gettoken(&vp->v_interlock);
2077 * Delete from old mount point vnode list, if on one.
2079 if (vp->v_mount != NULL)
2080 insmntque(vp, (struct mount *)0);
2082 * If special device, remove it from special device alias list
2085 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2086 lwkt_gettoken(&spechash_token);
2087 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2088 freedev(vp->v_rdev);
2089 lwkt_reltoken(&spechash_token);
2094 * If it is on the freelist and not already at the head,
2095 * move it to the head of the list. The test of the
2096 * VDOOMED flag and the reference count of zero is because
2097 * it will be removed from the free list by getnewvnode,
2098 * but will not have its reference count incremented until
2099 * after calling vgone. If the reference count were
2100 * incremented first, vgone would (incorrectly) try to
2101 * close the previous instance of the underlying object.
2103 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2105 lwkt_gettoken(&vnode_free_list_token);
2106 if (vp->v_flag & VFREE)
2107 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2110 vp->v_flag |= VFREE;
2111 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2112 lwkt_reltoken(&vnode_free_list_token);
2117 lwkt_reltoken(&vp->v_interlock);
2121 * Lookup a vnode by device number.
2124 vfinddev(dev, type, vpp)
2131 lwkt_gettoken(&spechash_token);
2132 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2133 if (type == vp->v_type) {
2135 lwkt_reltoken(&spechash_token);
2139 lwkt_reltoken(&spechash_token);
2144 * Calculate the total number of references to a special device.
2154 lwkt_gettoken(&spechash_token);
2155 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2156 count += vq->v_usecount;
2157 lwkt_reltoken(&spechash_token);
2162 * Same as above, but using the dev_t as argument
2171 vp = SLIST_FIRST(&dev->si_hlist);
2178 * Print out a description of a vnode.
2180 static char *typename[] =
2181 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2191 printf("%s: %p: ", label, (void *)vp);
2193 printf("%p: ", (void *)vp);
2194 printf("type %s, usecount %d, writecount %d, refcount %d,",
2195 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2198 if (vp->v_flag & VROOT)
2199 strcat(buf, "|VROOT");
2200 if (vp->v_flag & VTEXT)
2201 strcat(buf, "|VTEXT");
2202 if (vp->v_flag & VSYSTEM)
2203 strcat(buf, "|VSYSTEM");
2204 if (vp->v_flag & VXLOCK)
2205 strcat(buf, "|VXLOCK");
2206 if (vp->v_flag & VXWANT)
2207 strcat(buf, "|VXWANT");
2208 if (vp->v_flag & VBWAIT)
2209 strcat(buf, "|VBWAIT");
2210 if (vp->v_flag & VDOOMED)
2211 strcat(buf, "|VDOOMED");
2212 if (vp->v_flag & VFREE)
2213 strcat(buf, "|VFREE");
2214 if (vp->v_flag & VOBJBUF)
2215 strcat(buf, "|VOBJBUF");
2217 printf(" flags (%s)", &buf[1]);
2218 if (vp->v_data == NULL) {
2227 #include <ddb/ddb.h>
2229 * List all of the locked vnodes in the system.
2230 * Called when debugging the kernel.
2232 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2234 struct thread *td = curthread; /* XXX */
2235 struct mount *mp, *nmp;
2238 printf("Locked vnodes\n");
2239 lwkt_gettoken(&mountlist_token);
2240 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2241 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
2242 nmp = TAILQ_NEXT(mp, mnt_list);
2245 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2246 if (VOP_ISLOCKED(vp, NULL))
2247 vprint((char *)0, vp);
2249 lwkt_gettoken(&mountlist_token);
2250 nmp = TAILQ_NEXT(mp, mnt_list);
2253 lwkt_reltoken(&mountlist_token);
2258 * Top level filesystem related information gathering.
2260 static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2263 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2265 int *name = (int *)arg1 - 1; /* XXX */
2266 u_int namelen = arg2 + 1; /* XXX */
2267 struct vfsconf *vfsp;
2269 #if 1 || defined(COMPAT_PRELITE2)
2270 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2272 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2276 /* all sysctl names at this level are at least name and field */
2278 return (ENOTDIR); /* overloaded */
2279 if (name[0] != VFS_GENERIC) {
2280 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2281 if (vfsp->vfc_typenum == name[0])
2284 return (EOPNOTSUPP);
2285 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2286 oldp, oldlenp, newp, newlen, p));
2290 case VFS_MAXTYPENUM:
2293 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2296 return (ENOTDIR); /* overloaded */
2297 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2298 if (vfsp->vfc_typenum == name[2])
2301 return (EOPNOTSUPP);
2302 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2304 return (EOPNOTSUPP);
2307 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2308 "Generic filesystem");
2310 #if 1 || defined(COMPAT_PRELITE2)
2313 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2316 struct vfsconf *vfsp;
2317 struct ovfsconf ovfs;
2319 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2320 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2321 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2322 ovfs.vfc_index = vfsp->vfc_typenum;
2323 ovfs.vfc_refcount = vfsp->vfc_refcount;
2324 ovfs.vfc_flags = vfsp->vfc_flags;
2325 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2332 #endif /* 1 || COMPAT_PRELITE2 */
2335 #define KINFO_VNODESLOP 10
2337 * Dump vnode list (via sysctl).
2338 * Copyout address of vnode followed by vnode.
2342 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2344 struct proc *p = curproc; /* XXX */
2345 struct mount *mp, *nmp;
2346 struct vnode *nvp, *vp;
2349 #define VPTRSZ sizeof (struct vnode *)
2350 #define VNODESZ sizeof (struct vnode)
2353 if (!req->oldptr) /* Make an estimate */
2354 return (SYSCTL_OUT(req, 0,
2355 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2357 lwkt_gettoken(&mountlist_token);
2358 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2359 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) {
2360 nmp = TAILQ_NEXT(mp, mnt_list);
2364 lwkt_gettoken(&mntvnode_token);
2365 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2369 * Check that the vp is still associated with
2370 * this filesystem. RACE: could have been
2371 * recycled onto the same filesystem.
2373 if (vp->v_mount != mp) {
2374 lwkt_reltoken(&mntvnode_token);
2377 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2378 lwkt_reltoken(&mntvnode_token);
2379 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2380 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2382 lwkt_gettoken(&mntvnode_token);
2384 lwkt_reltoken(&mntvnode_token);
2385 lwkt_gettoken(&mountlist_token);
2386 nmp = TAILQ_NEXT(mp, mnt_list);
2389 lwkt_reltoken(&mountlist_token);
2397 * Exporting the vnode list on large systems causes them to crash.
2398 * Exporting the vnode list on medium systems causes sysctl to coredump.
2401 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2402 0, 0, sysctl_vnode, "S,vnode", "");
2406 * Check to see if a filesystem is mounted on a block device.
2413 if (vp->v_specmountpoint != NULL)
2419 * Unmount all filesystems. The list is traversed in reverse order
2420 * of mounting to avoid dependencies.
2426 struct thread *td = curthread;
2429 if (td->td_proc == NULL)
2430 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2433 * Since this only runs when rebooting, it is not interlocked.
2435 while(!TAILQ_EMPTY(&mountlist)) {
2436 mp = TAILQ_LAST(&mountlist, mntlist);
2437 error = dounmount(mp, MNT_FORCE, td);
2439 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2440 printf("unmount of %s failed (",
2441 mp->mnt_stat.f_mntonname);
2445 printf("%d)\n", error);
2447 /* The unmount has removed mp from the mountlist */
2453 * Build hash lists of net addresses and hang them off the mount point.
2454 * Called by ufs_mount() to set up the lists of export addresses.
2457 vfs_hang_addrlist(mp, nep, argp)
2459 struct netexport *nep;
2460 struct export_args *argp;
2462 register struct netcred *np;
2463 register struct radix_node_head *rnh;
2465 struct radix_node *rn;
2466 struct sockaddr *saddr, *smask = 0;
2470 if (argp->ex_addrlen == 0) {
2471 if (mp->mnt_flag & MNT_DEFEXPORTED)
2473 np = &nep->ne_defexported;
2474 np->netc_exflags = argp->ex_flags;
2475 np->netc_anon = argp->ex_anon;
2476 np->netc_anon.cr_ref = 1;
2477 mp->mnt_flag |= MNT_DEFEXPORTED;
2481 if (argp->ex_addrlen > MLEN)
2484 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2485 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2486 bzero((caddr_t) np, i);
2487 saddr = (struct sockaddr *) (np + 1);
2488 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2490 if (saddr->sa_len > argp->ex_addrlen)
2491 saddr->sa_len = argp->ex_addrlen;
2492 if (argp->ex_masklen) {
2493 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2494 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2497 if (smask->sa_len > argp->ex_masklen)
2498 smask->sa_len = argp->ex_masklen;
2500 i = saddr->sa_family;
2501 if ((rnh = nep->ne_rtable[i]) == 0) {
2503 * Seems silly to initialize every AF when most are not used,
2504 * do so on demand here
2506 for (dom = domains; dom; dom = dom->dom_next)
2507 if (dom->dom_family == i && dom->dom_rtattach) {
2508 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2512 if ((rnh = nep->ne_rtable[i]) == 0) {
2517 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2519 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2523 np->netc_exflags = argp->ex_flags;
2524 np->netc_anon = argp->ex_anon;
2525 np->netc_anon.cr_ref = 1;
2528 free(np, M_NETADDR);
2534 vfs_free_netcred(rn, w)
2535 struct radix_node *rn;
2538 register struct radix_node_head *rnh = (struct radix_node_head *) w;
2540 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2541 free((caddr_t) rn, M_NETADDR);
2546 * Free the net address hash lists that are hanging off the mount points.
2549 vfs_free_addrlist(nep)
2550 struct netexport *nep;
2553 register struct radix_node_head *rnh;
2555 for (i = 0; i <= AF_MAX; i++)
2556 if ((rnh = nep->ne_rtable[i])) {
2557 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2559 free((caddr_t) rnh, M_RTABLE);
2560 nep->ne_rtable[i] = 0;
2565 vfs_export(mp, nep, argp)
2567 struct netexport *nep;
2568 struct export_args *argp;
2572 if (argp->ex_flags & MNT_DELEXPORT) {
2573 if (mp->mnt_flag & MNT_EXPUBLIC) {
2574 vfs_setpublicfs(NULL, NULL, NULL);
2575 mp->mnt_flag &= ~MNT_EXPUBLIC;
2577 vfs_free_addrlist(nep);
2578 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2580 if (argp->ex_flags & MNT_EXPORTED) {
2581 if (argp->ex_flags & MNT_EXPUBLIC) {
2582 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2584 mp->mnt_flag |= MNT_EXPUBLIC;
2586 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2588 mp->mnt_flag |= MNT_EXPORTED;
2595 * Set the publicly exported filesystem (WebNFS). Currently, only
2596 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2599 vfs_setpublicfs(mp, nep, argp)
2601 struct netexport *nep;
2602 struct export_args *argp;
2609 * mp == NULL -> invalidate the current info, the FS is
2610 * no longer exported. May be called from either vfs_export
2611 * or unmount, so check if it hasn't already been done.
2614 if (nfs_pub.np_valid) {
2615 nfs_pub.np_valid = 0;
2616 if (nfs_pub.np_index != NULL) {
2617 FREE(nfs_pub.np_index, M_TEMP);
2618 nfs_pub.np_index = NULL;
2625 * Only one allowed at a time.
2627 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2631 * Get real filehandle for root of exported FS.
2633 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2634 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2636 if ((error = VFS_ROOT(mp, &rvp)))
2639 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2645 * If an indexfile was specified, pull it in.
2647 if (argp->ex_indexfile != NULL) {
2648 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2650 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2651 MAXNAMLEN, (size_t *)0);
2654 * Check for illegal filenames.
2656 for (cp = nfs_pub.np_index; *cp; cp++) {
2664 FREE(nfs_pub.np_index, M_TEMP);
2669 nfs_pub.np_mount = mp;
2670 nfs_pub.np_valid = 1;
2675 vfs_export_lookup(mp, nep, nam)
2676 register struct mount *mp;
2677 struct netexport *nep;
2678 struct sockaddr *nam;
2680 register struct netcred *np;
2681 register struct radix_node_head *rnh;
2682 struct sockaddr *saddr;
2685 if (mp->mnt_flag & MNT_EXPORTED) {
2687 * Lookup in the export list first.
2691 rnh = nep->ne_rtable[saddr->sa_family];
2693 np = (struct netcred *)
2694 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2696 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2701 * If no address match, use the default if it exists.
2703 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2704 np = &nep->ne_defexported;
2710 * perform msync on all vnodes under a mount point
2711 * the mount point must be locked.
2714 vfs_msync(struct mount *mp, int flags)
2716 struct thread *td = curthread; /* XXX */
2717 struct vnode *vp, *nvp;
2718 struct vm_object *obj;
2722 lwkt_gettoken(&mntvnode_token);
2724 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2725 if (vp->v_mount != mp) {
2730 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2732 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2736 * There could be hundreds of thousands of vnodes, we cannot
2737 * afford to do anything heavy-weight until we have a fairly
2738 * good indication that there is something to do.
2740 if ((vp->v_flag & VOBJDIRTY) &&
2741 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2742 lwkt_reltoken(&mntvnode_token);
2744 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
2745 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2746 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2750 lwkt_gettoken(&mntvnode_token);
2751 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2758 lwkt_reltoken(&mntvnode_token);
2762 * Create the VM object needed for VMIO and mmap support. This
2763 * is done for all VREG files in the system. Some filesystems might
2764 * afford the additional metadata buffering capability of the
2765 * VMIO code by making the device node be VMIO mode also.
2767 * vp must be locked when vfs_object_create is called.
2770 vfs_object_create(struct vnode *vp, struct thread *td)
2772 return (VOP_CREATEVOBJECT(vp, td));
2782 lwkt_gettoken(&vnode_free_list_token);
2783 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2784 if (vp->v_flag & VAGE) {
2785 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2787 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2790 lwkt_reltoken(&vnode_free_list_token);
2791 vp->v_flag &= ~VAGE;
2792 vp->v_flag |= VFREE;
2803 lwkt_gettoken(&vnode_free_list_token);
2804 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2805 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2807 lwkt_reltoken(&vnode_free_list_token);
2808 vp->v_flag &= ~(VFREE|VAGE);
2813 * Record a process's interest in events which might happen to
2814 * a vnode. Because poll uses the historic select-style interface
2815 * internally, this routine serves as both the ``check for any
2816 * pending events'' and the ``record my interest in future events''
2817 * functions. (These are done together, while the lock is held,
2818 * to avoid race conditions.)
2821 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
2823 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2824 if (vp->v_pollinfo.vpi_revents & events) {
2826 * This leaves events we are not interested
2827 * in available for the other process which
2828 * which presumably had requested them
2829 * (otherwise they would never have been
2832 events &= vp->v_pollinfo.vpi_revents;
2833 vp->v_pollinfo.vpi_revents &= ~events;
2835 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2838 vp->v_pollinfo.vpi_events |= events;
2839 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
2840 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2845 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2846 * it is possible for us to miss an event due to race conditions, but
2847 * that condition is expected to be rare, so for the moment it is the
2848 * preferred interface.
2851 vn_pollevent(vp, events)
2855 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2856 if (vp->v_pollinfo.vpi_events & events) {
2858 * We clear vpi_events so that we don't
2859 * call selwakeup() twice if two events are
2860 * posted before the polling process(es) is
2861 * awakened. This also ensures that we take at
2862 * most one selwakeup() if the polling process
2863 * is no longer interested. However, it does
2864 * mean that only one event can be noticed at
2865 * a time. (Perhaps we should only clear those
2866 * event bits which we note?) XXX
2868 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2869 vp->v_pollinfo.vpi_revents |= events;
2870 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2872 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2876 * Wake up anyone polling on vp because it is being revoked.
2877 * This depends on dead_poll() returning POLLHUP for correct
2884 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2885 if (vp->v_pollinfo.vpi_events) {
2886 vp->v_pollinfo.vpi_events = 0;
2887 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2889 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2895 * Routine to create and manage a filesystem syncer vnode.
2897 #define sync_close ((int (*) __P((struct vop_close_args *)))nullop)
2898 static int sync_fsync __P((struct vop_fsync_args *));
2899 static int sync_inactive __P((struct vop_inactive_args *));
2900 static int sync_reclaim __P((struct vop_reclaim_args *));
2901 #define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock)
2902 #define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock)
2903 static int sync_print __P((struct vop_print_args *));
2904 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2906 static vop_t **sync_vnodeop_p;
2907 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2908 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2909 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2910 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2911 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2912 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2913 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2914 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2915 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2916 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2919 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2920 { &sync_vnodeop_p, sync_vnodeop_entries };
2922 VNODEOP_SET(sync_vnodeop_opv_desc);
2925 * Create a new filesystem syncer vnode for the specified mount point.
2928 vfs_allocate_syncvnode(mp)
2932 static long start, incr, next;
2935 /* Allocate a new vnode */
2936 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2937 mp->mnt_syncer = NULL;
2942 * Place the vnode onto the syncer worklist. We attempt to
2943 * scatter them about on the list so that they will go off
2944 * at evenly distributed times even if all the filesystems
2945 * are mounted at once.
2948 if (next == 0 || next > syncer_maxdelay) {
2952 start = syncer_maxdelay / 2;
2953 incr = syncer_maxdelay;
2957 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2958 mp->mnt_syncer = vp;
2963 * Do a lazy sync of the filesystem.
2967 struct vop_fsync_args /* {
2969 struct ucred *a_cred;
2971 struct thread *a_td;
2974 struct vnode *syncvp = ap->a_vp;
2975 struct mount *mp = syncvp->v_mount;
2976 struct thread *td = ap->a_td;
2980 * We only need to do something if this is a lazy evaluation.
2982 if (ap->a_waitfor != MNT_LAZY)
2986 * Move ourselves to the back of the sync list.
2988 vn_syncer_add_to_worklist(syncvp, syncdelay);
2991 * Walk the list of vnodes pushing all that are dirty and
2992 * not already on the sync list.
2994 lwkt_gettoken(&mountlist_token);
2995 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) {
2996 lwkt_reltoken(&mountlist_token);
2999 asyncflag = mp->mnt_flag & MNT_ASYNC;
3000 mp->mnt_flag &= ~MNT_ASYNC;
3001 vfs_msync(mp, MNT_NOWAIT);
3002 VFS_SYNC(mp, MNT_LAZY, td);
3004 mp->mnt_flag |= MNT_ASYNC;
3010 * The syncer vnode is no referenced.
3014 struct vop_inactive_args /* {
3025 * The syncer vnode is no longer needed and is being decommissioned.
3027 * Modifications to the worklist must be protected at splbio().
3031 struct vop_reclaim_args /* {
3035 struct vnode *vp = ap->a_vp;
3039 vp->v_mount->mnt_syncer = NULL;
3040 if (vp->v_flag & VONWORKLST) {
3041 LIST_REMOVE(vp, v_synclist);
3042 vp->v_flag &= ~VONWORKLST;
3050 * Print out a syncer vnode.
3054 struct vop_print_args /* {
3058 struct vnode *vp = ap->a_vp;
3060 printf("syncer vnode");
3061 if (vp->v_vnlock != NULL)
3062 lockmgr_printinfo(vp->v_vnlock);
3068 * extract the dev_t from a VBLK or VCHR
3074 if (vp->v_type != VBLK && vp->v_type != VCHR)
3076 return (vp->v_rdev);
3080 * Check if vnode represents a disk device
3087 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3092 if (vp->v_rdev == NULL) {
3097 if (!devsw(vp->v_rdev)) {
3102 if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) {
3114 struct nameidata *ndp;
3117 if (!(flags & NDF_NO_FREE_PNBUF) &&
3118 (ndp->ni_cnd.cn_flags & HASBUF)) {
3119 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3120 ndp->ni_cnd.cn_flags &= ~HASBUF;
3122 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3123 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3124 ndp->ni_dvp != ndp->ni_vp)
3125 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
3126 if (!(flags & NDF_NO_DVP_RELE) &&
3127 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3131 if (!(flags & NDF_NO_VP_UNLOCK) &&
3132 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3133 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
3134 if (!(flags & NDF_NO_VP_RELE) &&
3139 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3140 (ndp->ni_cnd.cn_flags & SAVESTART)) {
3141 vrele(ndp->ni_startdir);
3142 ndp->ni_startdir = NULL;