1/*
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
25 *
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
36 * SUCH DAMAGE.
37 *
38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
39 * $Id: vfs_subr.c,v 1.188 1999/02/25 05:22:29 dillon Exp $
40 */
41
42/*
43 * External virtual filesystem routines
44 */
45#include "opt_ddb.h"
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/conf.h>
50#include <sys/fcntl.h>
51#include <sys/kernel.h>
52#include <sys/proc.h>
53#include <sys/malloc.h>
54#include <sys/mount.h>
55#include <sys/socket.h>
56#include <sys/vnode.h>
57#include <sys/stat.h>
58#include <sys/buf.h>
59#include <sys/domain.h>
60#include <sys/dirent.h>
61#include <sys/vmmeter.h>
62
63#include <machine/limits.h>
64
65#include <vm/vm.h>
66#include <vm/vm_param.h>
67#include <vm/vm_prot.h>
68#include <vm/vm_object.h>
69#include <vm/vm_extern.h>
70#include <vm/pmap.h>
71#include <vm/vm_map.h>
72#include <vm/vm_page.h>
73#include <vm/vm_pager.h>
74#include <vm/vnode_pager.h>
75#include <vm/vm_zone.h>
76#include <sys/sysctl.h>
77
78#include <miscfs/specfs/specdev.h>
79
80static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
81
82static void insmntque __P((struct vnode *vp, struct mount *mp));
83static void vclean __P((struct vnode *vp, int flags, struct proc *p));
84static void vfree __P((struct vnode *));
85static void vgonel __P((struct vnode *vp, struct proc *p));
86static unsigned long numvnodes;
87SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
88
89enum vtype iftovt_tab[16] = {
90 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
91 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
92};
93int vttoif_tab[9] = {
94 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
95 S_IFSOCK, S_IFIFO, S_IFMT,
96};
97
98static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
99struct tobefreelist vnode_tobefree_list; /* vnode free list */
100
101static u_long wantfreevnodes = 25;
102SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
103static u_long freevnodes = 0;
104SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
105
106int vfs_ioopt = 0;
107#ifdef ENABLE_VFS_IOOPT
108SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
109#endif
110
111struct mntlist mountlist; /* mounted filesystem list */
112struct simplelock mountlist_slock;
113struct simplelock mntvnode_slock;
114int nfs_mount_type = -1;
115#ifndef NULL_SIMPLELOCKS
116static struct simplelock mntid_slock;
117static struct simplelock vnode_free_list_slock;
118static struct simplelock spechash_slock;
119#endif
120struct nfs_public nfs_pub; /* publicly exported FS */
121static vm_zone_t vnode_zone;
122
123/*
124 * The workitem queue.
125 */
126#define SYNCER_MAXDELAY 32
127static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
128time_t syncdelay = 30;
129int rushjob; /* number of slots to run ASAP */
130
131static int syncer_delayno = 0;
132static long syncer_mask;
133LIST_HEAD(synclist, vnode);
134static struct synclist *syncer_workitem_pending;
135
136int desiredvnodes;
137SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, "");
138
139static void vfs_free_addrlist __P((struct netexport *nep));
140static int vfs_free_netcred __P((struct radix_node *rn, void *w));
141static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
142 struct export_args *argp));
143
144/*
145 * Initialize the vnode management data structures.
146 */
147void
148vntblinit()
149{
150
151 desiredvnodes = maxproc + cnt.v_page_count / 4;
152 simple_lock_init(&mntvnode_slock);
153 simple_lock_init(&mntid_slock);
154 simple_lock_init(&spechash_slock);
155 TAILQ_INIT(&vnode_free_list);
156 TAILQ_INIT(&vnode_tobefree_list);
157 simple_lock_init(&vnode_free_list_slock);
158 CIRCLEQ_INIT(&mountlist);
159 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
160 /*
161 * Initialize the filesystem syncer.
162 */
163 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
164 &syncer_mask);
165 syncer_maxdelay = syncer_mask + 1;
166}
167
168/*
169 * Mark a mount point as busy. Used to synchronize access and to delay
170 * unmounting. Interlock is not released on failure.
171 */
172int
173vfs_busy(mp, flags, interlkp, p)
174 struct mount *mp;
175 int flags;
176 struct simplelock *interlkp;
177 struct proc *p;
178{
179 int lkflags;
180
181 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
182 if (flags & LK_NOWAIT)
183 return (ENOENT);
184 mp->mnt_kern_flag |= MNTK_MWAIT;
185 if (interlkp) {
186 simple_unlock(interlkp);
187 }
188 /*
189 * Since all busy locks are shared except the exclusive
190 * lock granted when unmounting, the only place that a
191 * wakeup needs to be done is at the release of the
192 * exclusive lock at the end of dounmount.
193 */
194 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
195 if (interlkp) {
196 simple_lock(interlkp);
197 }
198 return (ENOENT);
199 }
200 lkflags = LK_SHARED | LK_NOPAUSE;
201 if (interlkp)
202 lkflags |= LK_INTERLOCK;
203 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
204 panic("vfs_busy: unexpected lock failure");
205 return (0);
206}
207
208/*
209 * Free a busy filesystem.
210 */
211void
212vfs_unbusy(mp, p)
213 struct mount *mp;
214 struct proc *p;
215{
216
217 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
218}
219
220/*
221 * Lookup a filesystem type, and if found allocate and initialize
222 * a mount structure for it.
223 *
224 * Devname is usually updated by mount(8) after booting.
225 */
226int
227vfs_rootmountalloc(fstypename, devname, mpp)
228 char *fstypename;
229 char *devname;
230 struct mount **mpp;
231{
232 struct proc *p = curproc; /* XXX */
233 struct vfsconf *vfsp;
234 struct mount *mp;
235
236 if (fstypename == NULL)
237 return (ENODEV);
238 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
239 if (!strcmp(vfsp->vfc_name, fstypename))
240 break;
241 if (vfsp == NULL)
242 return (ENODEV);
243 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
244 bzero((char *)mp, (u_long)sizeof(struct mount));
245 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
246 (void)vfs_busy(mp, LK_NOWAIT, 0, p);
247 LIST_INIT(&mp->mnt_vnodelist);
248 mp->mnt_vfc = vfsp;
249 mp->mnt_op = vfsp->vfc_vfsops;
250 mp->mnt_flag = MNT_RDONLY;
251 mp->mnt_vnodecovered = NULLVP;
252 vfsp->vfc_refcount++;
253 mp->mnt_stat.f_type = vfsp->vfc_typenum;
254 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
255 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
256 mp->mnt_stat.f_mntonname[0] = '/';
257 mp->mnt_stat.f_mntonname[1] = 0;
258 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
259 *mpp = mp;
260 return (0);
261}
262
263/*
264 * Find an appropriate filesystem to use for the root. If a filesystem
265 * has not been preselected, walk through the list of known filesystems
266 * trying those that have mountroot routines, and try them until one
267 * works or we have tried them all.
268 */
269#ifdef notdef /* XXX JH */
270int
271lite2_vfs_mountroot()
272{
273 struct vfsconf *vfsp;
274 extern int (*lite2_mountroot) __P((void));
275 int error;
276
277 if (lite2_mountroot != NULL)
278 return ((*lite2_mountroot)());
279 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
280 if (vfsp->vfc_mountroot == NULL)
281 continue;
282 if ((error = (*vfsp->vfc_mountroot)()) == 0)
283 return (0);
284 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
285 }
286 return (ENODEV);
287}
288#endif
289
290/*
291 * Lookup a mount point by filesystem identifier.
292 */
293struct mount *
294vfs_getvfs(fsid)
295 fsid_t *fsid;
296{
297 register struct mount *mp;
298
299 simple_lock(&mountlist_slock);
300 for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
301 mp = mp->mnt_list.cqe_next) {
302 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
303 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
304 simple_unlock(&mountlist_slock);
305 return (mp);
306 }
307 }
308 simple_unlock(&mountlist_slock);
309 return ((struct mount *) 0);
310}
311
312/*
313 * Get a new unique fsid
314 */
315void
316vfs_getnewfsid(mp)
317 struct mount *mp;
318{
319 static u_short xxxfs_mntid;
320
321 fsid_t tfsid;
322 int mtype;
323
324 simple_lock(&mntid_slock);
325 mtype = mp->mnt_vfc->vfc_typenum;
326 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
327 mp->mnt_stat.f_fsid.val[1] = mtype;
328 if (xxxfs_mntid == 0)
329 ++xxxfs_mntid;
330 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
331 tfsid.val[1] = mtype;
332 if (mountlist.cqh_first != (void *)&mountlist) {
333 while (vfs_getvfs(&tfsid)) {
334 tfsid.val[0]++;
335 xxxfs_mntid++;
336 }
337 }
338 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
339 simple_unlock(&mntid_slock);
340}
341
342/*
343 * Set vnode attributes to VNOVAL
344 */
345void
346vattr_null(vap)
347 register struct vattr *vap;
348{
349
350 vap->va_type = VNON;
351 vap->va_size = VNOVAL;
352 vap->va_bytes = VNOVAL;
353 vap->va_mode = VNOVAL;
354 vap->va_nlink = VNOVAL;
355 vap->va_uid = VNOVAL;
356 vap->va_gid = VNOVAL;
357 vap->va_fsid = VNOVAL;
358 vap->va_fileid = VNOVAL;
359 vap->va_blocksize = VNOVAL;
360 vap->va_rdev = VNOVAL;
361 vap->va_atime.tv_sec = VNOVAL;
362 vap->va_atime.tv_nsec = VNOVAL;
363 vap->va_mtime.tv_sec = VNOVAL;
364 vap->va_mtime.tv_nsec = VNOVAL;
365 vap->va_ctime.tv_sec = VNOVAL;
366 vap->va_ctime.tv_nsec = VNOVAL;
367 vap->va_flags = VNOVAL;
368 vap->va_gen = VNOVAL;
369 vap->va_vaflags = 0;
370}
371
372/*
373 * Routines having to do with the management of the vnode table.
374 */
375extern vop_t **dead_vnodeop_p;
376
377/*
378 * Return the next vnode from the free list.
379 */
380int
381getnewvnode(tag, mp, vops, vpp)
382 enum vtagtype tag;
383 struct mount *mp;
384 vop_t **vops;
385 struct vnode **vpp;
386{
387 int s;
388 struct proc *p = curproc; /* XXX */
389 struct vnode *vp, *tvp, *nvp;
390 vm_object_t object;
391 TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
392
393 /*
394 * We take the least recently used vnode from the freelist
395 * if we can get it and it has no cached pages, and no
396 * namecache entries are relative to it.
397 * Otherwise we allocate a new vnode
398 */
399
400 s = splbio();
401 simple_lock(&vnode_free_list_slock);
402 TAILQ_INIT(&vnode_tmp_list);
403
404 for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
405 nvp = TAILQ_NEXT(vp, v_freelist);
406 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
407 if (vp->v_flag & VAGE) {
408 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
409 } else {
410 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
411 }
412 vp->v_flag &= ~(VTBFREE|VAGE);
413 vp->v_flag |= VFREE;
414 if (vp->v_usecount)
415 panic("tobe free vnode isn't");
416 freevnodes++;
417 }
418
419 if (wantfreevnodes && freevnodes < wantfreevnodes) {
420 vp = NULL;
421 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
422 /*
423 * XXX: this is only here to be backwards compatible
424 */
425 vp = NULL;
426 } else {
427 for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
428 nvp = TAILQ_NEXT(vp, v_freelist);
429 if (!simple_lock_try(&vp->v_interlock))
430 continue;
431 if (vp->v_usecount)
432 panic("free vnode isn't");
433
434 object = vp->v_object;
435 if (object && (object->resident_page_count || object->ref_count)) {
436 printf("object inconsistant state: RPC: %d, RC: %d\n",
437 object->resident_page_count, object->ref_count);
438 /* Don't recycle if it's caching some pages */
439 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
440 TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
441 continue;
442 } else if (LIST_FIRST(&vp->v_cache_src)) {
443 /* Don't recycle if active in the namecache */
444 simple_unlock(&vp->v_interlock);
445 continue;
446 } else {
447 break;
448 }
449 }
450 }
451
452 for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
453 nvp = TAILQ_NEXT(tvp, v_freelist);
454 TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
455 TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
456 simple_unlock(&tvp->v_interlock);
457 }
458
459 if (vp) {
460 vp->v_flag |= VDOOMED;
461 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
462 freevnodes--;
463 simple_unlock(&vnode_free_list_slock);
464 cache_purge(vp);
465 vp->v_lease = NULL;
466 if (vp->v_type != VBAD) {
467 vgonel(vp, p);
468 } else {
469 simple_unlock(&vp->v_interlock);
470 }
471
472#ifdef INVARIANTS
473 {
474 int s;
475
476 if (vp->v_data)
477 panic("cleaned vnode isn't");
478 s = splbio();
479 if (vp->v_numoutput)
480 panic("Clean vnode has pending I/O's");
481 splx(s);
482 }
483#endif
484 vp->v_flag = 0;
485 vp->v_lastr = 0;
486 vp->v_lastw = 0;
487 vp->v_lasta = 0;
488 vp->v_cstart = 0;
489 vp->v_clen = 0;
490 vp->v_socket = 0;
491 vp->v_writecount = 0; /* XXX */
492 vp->v_maxio = 0;
493 } else {
494 simple_unlock(&vnode_free_list_slock);
495 vp = (struct vnode *) zalloc(vnode_zone);
496 bzero((char *) vp, sizeof *vp);
497 simple_lock_init(&vp->v_interlock);
498 vp->v_dd = vp;
499 cache_purge(vp);
500 LIST_INIT(&vp->v_cache_src);
501 TAILQ_INIT(&vp->v_cache_dst);
502 numvnodes++;
503 }
504
505 TAILQ_INIT(&vp->v_cleanblkhd);
506 TAILQ_INIT(&vp->v_dirtyblkhd);
507 vp->v_type = VNON;
508 vp->v_tag = tag;
509 vp->v_op = vops;
510 insmntque(vp, mp);
511 *vpp = vp;
512 vp->v_usecount = 1;
513 vp->v_data = 0;
514 splx(s);
515
516 vfs_object_create(vp, p, p->p_ucred);
517 return (0);
518}
519
520/*
521 * Move a vnode from one mount queue to another.
522 */
523static void
524insmntque(vp, mp)
525 register struct vnode *vp;
526 register struct mount *mp;
527{
528
529 simple_lock(&mntvnode_slock);
530 /*
531 * Delete from old mount point vnode list, if on one.
532 */
533 if (vp->v_mount != NULL)
534 LIST_REMOVE(vp, v_mntvnodes);
535 /*
536 * Insert into list of vnodes for the new mount point, if available.
537 */
538 if ((vp->v_mount = mp) == NULL) {
539 simple_unlock(&mntvnode_slock);
540 return;
541 }
542 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
543 simple_unlock(&mntvnode_slock);
544}
545
546/*
547 * Update outstanding I/O count and do wakeup if requested.
548 */
549void
550vwakeup(bp)
551 register struct buf *bp;
552{
553 register struct vnode *vp;
554
555 bp->b_flags &= ~B_WRITEINPROG;
556 if ((vp = bp->b_vp)) {
557 vp->v_numoutput--;
558 if (vp->v_numoutput < 0)
559 panic("vwakeup: neg numoutput");
560 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
561 vp->v_flag &= ~VBWAIT;
562 wakeup((caddr_t) &vp->v_numoutput);
563 }
564 }
565}
566
567/*
568 * Flush out and invalidate all buffers associated with a vnode.
569 * Called with the underlying object locked.
570 */
571int
572vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
573 register struct vnode *vp;
574 int flags;
575 struct ucred *cred;
576 struct proc *p;
577 int slpflag, slptimeo;
578{
579 register struct buf *bp;
580 struct buf *nbp, *blist;
581 int s, error;
582 vm_object_t object;
583
584 if (flags & V_SAVE) {
585 s = splbio();
586 while (vp->v_numoutput) {
587 vp->v_flag |= VBWAIT;
588 error = tsleep((caddr_t)&vp->v_numoutput,
589 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
590 if (error) {
591 splx(s);
592 return (error);
593 }
594 }
595 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
596 splx(s);
597 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
598 return (error);
599 s = splbio();
600 if (vp->v_numoutput > 0 ||
601 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
602 panic("vinvalbuf: dirty bufs");
603 }
604 splx(s);
605 }
606 s = splbio();
607 for (;;) {
608 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
609 if (!blist)
610 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
611 if (!blist)
612 break;
613
614 for (bp = blist; bp; bp = nbp) {
615 nbp = TAILQ_NEXT(bp, b_vnbufs);
616 if (bp->b_flags & B_BUSY) {
617 bp->b_flags |= B_WANTED;
618 error = tsleep((caddr_t) bp,
619 slpflag | (PRIBIO + 4), "vinvalbuf",
620 slptimeo);
621 if (error) {
622 splx(s);
623 return (error);
624 }
625 break;
626 }
627 /*
628 * XXX Since there are no node locks for NFS, I
629 * believe there is a slight chance that a delayed
630 * write will occur while sleeping just above, so
631 * check for it. Note that vfs_bio_awrite expects
632 * buffers to reside on a queue, while VOP_BWRITE and
633 * brelse do not.
634 */
635 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
636 (flags & V_SAVE)) {
637
638 if (bp->b_vp == vp) {
639 if (bp->b_flags & B_CLUSTEROK) {
640 vfs_bio_awrite(bp);
641 } else {
642 bremfree(bp);
643 bp->b_flags |= (B_BUSY | B_ASYNC);
644 VOP_BWRITE(bp);
645 }
646 } else {
647 bremfree(bp);
648 bp->b_flags |= B_BUSY;
649 (void) VOP_BWRITE(bp);
650 }
651 break;
652 }
653 bremfree(bp);
654 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF | B_BUSY);
655 bp->b_flags &= ~B_ASYNC;
656 brelse(bp);
657 }
658 }
659
660 while (vp->v_numoutput > 0) {
661 vp->v_flag |= VBWAIT;
662 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
663 }
664
665 splx(s);
666
667 /*
668 * Destroy the copy in the VM cache, too.
669 */
670 simple_lock(&vp->v_interlock);
671 object = vp->v_object;
672 if (object != NULL) {
673 vm_object_page_remove(object, 0, 0,
674 (flags & V_SAVE) ? TRUE : FALSE);
675 }
676 simple_unlock(&vp->v_interlock);
677
678 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
679 panic("vinvalbuf: flush failed");
680 return (0);
681}
682
683/*
684 * Truncate a file's buffer and pages to a specified length. This
685 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
686 * sync activity.
687 */
688int
689vtruncbuf(vp, cred, p, length, blksize)
690 register struct vnode *vp;
691 struct ucred *cred;
692 struct proc *p;
693 off_t length;
694 int blksize;
695{
696 register struct buf *bp;
697 struct buf *nbp;
698 int s, anyfreed;
699 int trunclbn;
700
701 /*
702 * Round up to the *next* lbn.
703 */
704 trunclbn = (length + blksize - 1) / blksize;
705
706 s = splbio();
707restart:
708 anyfreed = 1;
709 for (;anyfreed;) {
710 anyfreed = 0;
711 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
712 nbp = TAILQ_NEXT(bp, b_vnbufs);
713 if (bp->b_lblkno >= trunclbn) {
714 if (bp->b_flags & B_BUSY) {
715 bp->b_flags |= B_WANTED;
716 tsleep(bp, PRIBIO + 4, "vtrb1", 0);
717 goto restart;
718 } else {
719 bremfree(bp);
720 bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
721 bp->b_flags &= ~B_ASYNC;
722 brelse(bp);
723 anyfreed = 1;
724 }
725 if (nbp && (((nbp->b_xflags & B_VNCLEAN) == 0)||
726 (nbp->b_vp != vp) ||
727 (nbp->b_flags & B_DELWRI))) {
728 goto restart;
729 }
730 }
731 }
732
733 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
734 nbp = TAILQ_NEXT(bp, b_vnbufs);
735 if (bp->b_lblkno >= trunclbn) {
736 if (bp->b_flags & B_BUSY) {
737 bp->b_flags |= B_WANTED;
738 tsleep(bp, PRIBIO + 4, "vtrb2", 0);
739 goto restart;
740 } else {
741 bremfree(bp);
742 bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
743 bp->b_flags &= ~B_ASYNC;
744 brelse(bp);
745 anyfreed = 1;
746 }
747 if (nbp && (((nbp->b_xflags & B_VNDIRTY) == 0)||
748 (nbp->b_vp != vp) ||
749 (nbp->b_flags & B_DELWRI) == 0)) {
750 goto restart;
751 }
752 }
753 }
754 }
755
756 if (length > 0) {
757restartsync:
758 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
759 nbp = TAILQ_NEXT(bp, b_vnbufs);
760 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
761 if (bp->b_flags & B_BUSY) {
762 bp->b_flags |= B_WANTED;
763 tsleep(bp, PRIBIO, "vtrb3", 0);
764 } else {
765 bremfree(bp);
766 bp->b_flags |= B_BUSY;
767 if (bp->b_vp == vp) {
768 bp->b_flags |= B_ASYNC;
769 } else {
770 bp->b_flags &= ~B_ASYNC;
771 }
772 VOP_BWRITE(bp);
773 }
774 goto restartsync;
775 }
776
777 }
778 }
779
780 while (vp->v_numoutput > 0) {
781 vp->v_flag |= VBWAIT;
782 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
783 }
784
785 splx(s);
786
787 vnode_pager_setsize(vp, length);
788
789 return (0);
790}
791
792/*
793 * Associate a buffer with a vnode.
794 */
795void
796bgetvp(vp, bp)
797 register struct vnode *vp;
798 register struct buf *bp;
799{
800 int s;
801
802 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
803
804 vhold(vp);
805 bp->b_vp = vp;
806 if (vp->v_type == VBLK || vp->v_type == VCHR)
807 bp->b_dev = vp->v_rdev;
808 else
809 bp->b_dev = NODEV;
810 /*
811 * Insert onto list for new vnode.
812 */
813 s = splbio();
814 bp->b_xflags |= B_VNCLEAN;
815 bp->b_xflags &= ~B_VNDIRTY;
816 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
817 splx(s);
818}
819
820/*
821 * Disassociate a buffer from a vnode.
822 */
823void
824brelvp(bp)
825 register struct buf *bp;
826{
827 struct vnode *vp;
828 struct buflists *listheadp;
829 int s;
830
831 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
832
833 /*
834 * Delete from old vnode list, if on one.
835 */
836 vp = bp->b_vp;
837 s = splbio();
838 if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
839 if (bp->b_xflags & B_VNDIRTY)
840 listheadp = &vp->v_dirtyblkhd;
841 else
842 listheadp = &vp->v_cleanblkhd;
843 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
844 bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
845 }
846 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
847 vp->v_flag &= ~VONWORKLST;
848 LIST_REMOVE(vp, v_synclist);
849 }
850 splx(s);
851 bp->b_vp = (struct vnode *) 0;
852 vdrop(vp);
853}
854
855/*
856 * The workitem queue.
857 *
858 * It is useful to delay writes of file data and filesystem metadata
859 * for tens of seconds so that quickly created and deleted files need
860 * not waste disk bandwidth being created and removed. To realize this,
861 * we append vnodes to a "workitem" queue. When running with a soft
862 * updates implementation, most pending metadata dependencies should
863 * not wait for more than a few seconds. Thus, mounted on block devices
864 * are delayed only about a half the time that file data is delayed.
865 * Similarly, directory updates are more critical, so are only delayed
866 * about a third the time that file data is delayed. Thus, there are
867 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
868 * one each second (driven off the filesystem syner process). The
869 * syncer_delayno variable indicates the next queue that is to be processed.
870 * Items that need to be processed soon are placed in this queue:
871 *
872 * syncer_workitem_pending[syncer_delayno]
873 *
874 * A delay of fifteen seconds is done by placing the request fifteen
875 * entries later in the queue:
876 *
877 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
878 *
879 */
880
881/*
882 * Add an item to the syncer work queue.
883 */
884static void
885vn_syncer_add_to_worklist(struct vnode *vp, int delay)
886{
887 int s, slot;
888
889 s = splbio();
890
891 if (vp->v_flag & VONWORKLST) {
892 LIST_REMOVE(vp, v_synclist);
893 }
894
895 if (delay > syncer_maxdelay - 2)
896 delay = syncer_maxdelay - 2;
897 slot = (syncer_delayno + delay) & syncer_mask;
898
899 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
900 vp->v_flag |= VONWORKLST;
901 splx(s);
902}
903
904struct proc *updateproc;
905static void sched_sync __P((void));
906static const struct kproc_desc up_kp = {
907 "syncer",
908 sched_sync,
909 &updateproc
910};
911SYSINIT_KT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
912
913/*
914 * System filesystem synchronizer daemon.
915 */
916void
917sched_sync(void)
918{
919 struct synclist *slp;
920 struct vnode *vp;
921 long starttime;
922 int s;
923 struct proc *p = updateproc;
924
925 for (;;) {
926 starttime = time_second;
927
928 /*
929 * Push files whose dirty time has expired. Be careful
930 * of interrupt race on slp queue.
931 */
932 s = splbio();
933 slp = &syncer_workitem_pending[syncer_delayno];
934 syncer_delayno += 1;
935 if (syncer_delayno == syncer_maxdelay)
936 syncer_delayno = 0;
937 splx(s);
938
939 while ((vp = LIST_FIRST(slp)) != NULL) {
940 if (VOP_ISLOCKED(vp) == 0) {
941 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
942 (void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
943 VOP_UNLOCK(vp, 0, p);
944 }
945 s = splbio();
946 if (LIST_FIRST(slp) == vp) {
947 /*
948 * Note: v_tag VT_VFS vps can remain on the
949 * worklist too with no dirty blocks, but
950 * since sync_fsync() moves it to a different
951 * slot we are safe.
952 */
953 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
954 vp->v_type != VBLK)
955 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
956 /*
957 * Put us back on the worklist. The worklist
958 * routine will remove us from our current
959 * position and then add us back in at a later
960 * position.
961 */
962 vn_syncer_add_to_worklist(vp, syncdelay);
963 }
964 splx(s);
965 }
966
967 /*
968 * Do soft update processing.
969 */
970 if (bioops.io_sync)
971 (*bioops.io_sync)(NULL);
972
973 /*
974 * The variable rushjob allows the kernel to speed up the
975 * processing of the filesystem syncer process. A rushjob
976 * value of N tells the filesystem syncer to process the next
977 * N seconds worth of work on its queue ASAP. Currently rushjob
978 * is used by the soft update code to speed up the filesystem
979 * syncer process when the incore state is getting so far
980 * ahead of the disk that the kernel memory pool is being
981 * threatened with exhaustion.
982 */
983 if (rushjob > 0) {
984 rushjob -= 1;
985 continue;
986 }
987 /*
988 * If it has taken us less than a second to process the
989 * current work, then wait. Otherwise start right over
990 * again. We can still lose time if any single round
991 * takes more than two seconds, but it does not really
992 * matter as we are just trying to generally pace the
993 * filesystem activity.
994 */
995 if (time_second == starttime)
996 tsleep(&lbolt, PPAUSE, "syncer", 0);
997 }
998}
999
1000/*
1001 * Associate a p-buffer with a vnode.
1002 *
1003 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1004 * with the buffer. i.e. the bp has not been linked into the vnode or
1005 * ref-counted.
1006 */
1007void
1008pbgetvp(vp, bp)
1009 register struct vnode *vp;
1010 register struct buf *bp;
1011{
1012
1013 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1014
1015 bp->b_vp = vp;
1016 bp->b_flags |= B_PAGING;
1017 if (vp->v_type == VBLK || vp->v_type == VCHR)
1018 bp->b_dev = vp->v_rdev;
1019 else
1020 bp->b_dev = NODEV;
1021}
1022
1023/*
1024 * Disassociate a p-buffer from a vnode.
1025 */
1026void
1027pbrelvp(bp)
1028 register struct buf *bp;
1029{
1030
1031 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1032
1033#if !defined(MAX_PERF)
1034 /* XXX REMOVE ME */
1035 if (bp->b_vnbufs.tqe_next != NULL) {
1036 panic(
1037 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1038 bp,
1039 (int)bp->b_flags
1040 );
1041 }
1042#endif
1043 bp->b_vp = (struct vnode *) 0;
1044 bp->b_flags &= ~B_PAGING;
1045}
1046
1047void
1048pbreassignbuf(bp, newvp)
1049 struct buf *bp;
1050 struct vnode *newvp;
1051{
1052#if !defined(MAX_PERF)
1053 if ((bp->b_flags & B_PAGING) == 0) {
1054 panic(
1055 "pbreassignbuf() on non phys bp %p",
1056 bp
1057 );
1058 }
1059#endif
1060 bp->b_vp = newvp;
1061}
1062
1063/*
1064 * Reassign a buffer from one vnode to another.
1065 * Used to assign file specific control information
1066 * (indirect blocks) to the vnode to which they belong.
1067 */
1068void
1069reassignbuf(bp, newvp)
1070 register struct buf *bp;
1071 register struct vnode *newvp;
1072{
1073 struct buflists *listheadp;
1074 int delay;
1075 int s;
1076
1077 if (newvp == NULL) {
1078 printf("reassignbuf: NULL");
1079 return;
1080 }
1081
1082#if !defined(MAX_PERF)
1083 /*
1084 * B_PAGING flagged buffers cannot be reassigned because their vp
1085 * is not fully linked in.
1086 */
1087 if (bp->b_flags & B_PAGING)
1088 panic("cannot reassign paging buffer");
1089#endif
1090
1091 s = splbio();
1092 /*
1093 * Delete from old vnode list, if on one.
1094 */
1095 if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
1096 if (bp->b_xflags & B_VNDIRTY)
1097 listheadp = &bp->b_vp->v_dirtyblkhd;
1098 else
1099 listheadp = &bp->b_vp->v_cleanblkhd;
1100 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1101 bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
1102 if (bp->b_vp != newvp) {
1103 vdrop(bp->b_vp);
1104 bp->b_vp = NULL; /* for clarification */
1105 }
1106 }
1107 /*
1108 * If dirty, put on list of dirty buffers; otherwise insert onto list
1109 * of clean buffers.
1110 */
1111 if (bp->b_flags & B_DELWRI) {
1112 struct buf *tbp;
1113
1114 listheadp = &newvp->v_dirtyblkhd;
1115 if ((newvp->v_flag & VONWORKLST) == 0) {
1116 switch (newvp->v_type) {
1117 case VDIR:
1118 delay = syncdelay / 3;
1119 break;
1120 case VBLK:
1121 if (newvp->v_specmountpoint != NULL) {
1122 delay = syncdelay / 2;
1123 break;
1124 }
1125 /* fall through */
1126 default:
1127 delay = syncdelay;
1128 }
1129 vn_syncer_add_to_worklist(newvp, delay);
1130 }
1131 bp->b_xflags |= B_VNDIRTY;
1132 tbp = TAILQ_FIRST(listheadp);
1133 if (tbp == NULL ||
1134 (bp->b_lblkno >= 0 && tbp->b_lblkno > bp->b_lblkno)) {
1135 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1136 } else {
1137 if (bp->b_lblkno >= 0) {
1138 struct buf *ttbp;
1139 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1140 (ttbp->b_lblkno < bp->b_lblkno)) {
1141 tbp = ttbp;
1142 }
1143 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1144 } else {
1145 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1146 }
1147 }
1148 } else {
1149 bp->b_xflags |= B_VNCLEAN;
1150 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1151 if ((newvp->v_flag & VONWORKLST) &&
1152 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1153 newvp->v_flag &= ~VONWORKLST;
1154 LIST_REMOVE(newvp, v_synclist);
1155 }
1156 }
1157 if (bp->b_vp != newvp) {
1158 bp->b_vp = newvp;
1159 vhold(bp->b_vp);
1160 }
1161 splx(s);
1162}
1163
1164/*
1165 * Create a vnode for a block device.
1166 * Used for mounting the root file system.
1167 */
1168int
1169bdevvp(dev, vpp)
1170 dev_t dev;
1171 struct vnode **vpp;
1172{
1173 register struct vnode *vp;
1174 struct vnode *nvp;
1175 int error;
1176
1177 /* XXX 255 is for mfs. */
1178 if (dev == NODEV || (major(dev) != 255 && (major(dev) >= nblkdev ||
1179 bdevsw[major(dev)] == NULL))) {
1180 *vpp = NULLVP;
1181 return (ENXIO);
1182 }
1183 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1184 if (error) {
1185 *vpp = NULLVP;
1186 return (error);
1187 }
1188 vp = nvp;
1189 vp->v_type = VBLK;
1190 if ((nvp = checkalias(vp, dev, (struct mount *)0)) != NULL) {
1191 vput(vp);
1192 vp = nvp;
1193 }
1194 *vpp = vp;
1195 return (0);
1196}
1197
1198/*
1199 * Check to see if the new vnode represents a special device
1200 * for which we already have a vnode (either because of
1201 * bdevvp() or because of a different vnode representing
1202 * the same block device). If such an alias exists, deallocate
1203 * the existing contents and return the aliased vnode. The
1204 * caller is responsible for filling it with its new contents.
1205 */
1206struct vnode *
1207checkalias(nvp, nvp_rdev, mp)
1208 register struct vnode *nvp;
1209 dev_t nvp_rdev;
1210 struct mount *mp;
1211{
1212 struct proc *p = curproc; /* XXX */
1213 struct vnode *vp;
1214 struct vnode **vpp;
1215 int rmaj = major(nvp_rdev);
1216
1217 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1218 return (NULLVP);
1219
1220 vpp = &speclisth[SPECHASH(nvp_rdev)];
1221loop:
1222 simple_lock(&spechash_slock);
1223 for (vp = *vpp; vp; vp = vp->v_specnext) {
1224 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1225 continue;
1226 /*
1227 * Alias, but not in use, so flush it out.
1228 * Only alias active device nodes.
1229 * Not sure why we don't re-use this like we do below.
1230 */
1231 simple_lock(&vp->v_interlock);
1232 if (vp->v_usecount == 0) {
1233 simple_unlock(&spechash_slock);
1234 vgonel(vp, p);
1235 goto loop;
1236 }
1237 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) {
1238 /*
1239 * It dissappeared, and we may have slept.
1240 * Restart from the beginning
1241 */
1242 simple_unlock(&spechash_slock);
1243 goto loop;
1244 }
1245 break;
1246 }
1247 /*
1248 * It would be a lot clearer what is going on here if
1249 * this had been expressed as:
1250 * if ( vp && (vp->v_tag == VT_NULL))
1251 * and the clauses had been swapped.
1252 */
1253 if (vp == NULL || vp->v_tag != VT_NON) {
1254 struct specinfo *sinfo;
1255
1256 /*
1257 * Put the new vnode into the hash chain.
1258 * and if there was an alias, connect them.
1259 */
1260 MALLOC(sinfo, struct specinfo *,
1261 sizeof(struct specinfo), M_VNODE, M_WAITOK);
1262 bzero(sinfo, sizeof(struct specinfo));
1263 nvp->v_specinfo = sinfo;
1264 sinfo->si_rdev = nvp_rdev;
1265 sinfo->si_hashchain = vpp;
1266 sinfo->si_specnext = *vpp;
1267 sinfo->si_bsize_phys = DEV_BSIZE;
1268 sinfo->si_bsize_best = BLKDEV_IOSIZE;
1269 sinfo->si_bsize_max = MAXBSIZE;
1270
1271 /*
1272 * Ask the device to fix up specinfo. Typically the
1273 * si_bsize_* parameters may need fixing up.
1274 */
1275
1276 if (nvp->v_type == VBLK && rmaj < nblkdev) {
1277 if (bdevsw[rmaj] && bdevsw[rmaj]->d_parms)
1278
1279 (*bdevsw[rmaj]->d_parms)(nvp_rdev, sinfo, DPARM_GET);
1280 } else if (nvp->v_type == VCHR && rmaj < nchrdev) {
1281 if (cdevsw[rmaj] && cdevsw[rmaj]->d_parms)
1282 (*cdevsw[rmaj]->d_parms)(nvp_rdev, sinfo, DPARM_GET);
1283 }
1284
1285 simple_unlock(&spechash_slock);
1286 *vpp = nvp;
1287 if (vp != NULLVP) {
1288 nvp->v_flag |= VALIASED;
1289 vp->v_flag |= VALIASED;
1290 vput(vp);
1291 }
1292 return (NULLVP);
1293 }
1294 /*
1295 * if ( vp && (vp->v_tag == VT_NULL))
1296 * We have a vnode alias, but it is a trashed.
1297 * Make it look like it's newley allocated. (by getnewvnode())
1298 * The caller should use this instead.
1299 */
1300 simple_unlock(&spechash_slock);
1301 VOP_UNLOCK(vp, 0, p);
1302 simple_lock(&vp->v_interlock);
1303 vclean(vp, 0, p);
1304 vp->v_op = nvp->v_op;
1305 vp->v_tag = nvp->v_tag;
1306 nvp->v_type = VNON;
1307 insmntque(vp, mp);
1308 return (vp);
1309}
1310
1311/*
1312 * Grab a particular vnode from the free list, increment its
1313 * reference count and lock it. The vnode lock bit is set the
1314 * vnode is being eliminated in vgone. The process is awakened
1315 * when the transition is completed, and an error returned to
1316 * indicate that the vnode is no longer usable (possibly having
1317 * been changed to a new file system type).
1318 */
1319int
1320vget(vp, flags, p)
1321 register struct vnode *vp;
1322 int flags;
1323 struct proc *p;
1324{
1325 int error;
1326
1327 /*
1328 * If the vnode is in the process of being cleaned out for
1329 * another use, we wait for the cleaning to finish and then
1330 * return failure. Cleaning is determined by checking that
1331 * the VXLOCK flag is set.
1332 */
1333 if ((flags & LK_INTERLOCK) == 0) {
1334 simple_lock(&vp->v_interlock);
1335 }
1336 if (vp->v_flag & VXLOCK) {
1337 vp->v_flag |= VXWANT;
1338 simple_unlock(&vp->v_interlock);
1339 tsleep((caddr_t)vp, PINOD, "vget", 0);
1340 return (ENOENT);
1341 }
1342
1343 vp->v_usecount++;
1344
1345 if (VSHOULDBUSY(vp))
1346 vbusy(vp);
1347 if (flags & LK_TYPE_MASK) {
1348 if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1349 /*
1350 * must expand vrele here because we do not want
1351 * to call VOP_INACTIVE if the reference count
1352 * drops back to zero since it was never really
1353 * active. We must remove it from the free list
1354 * before sleeping so that multiple processes do
1355 * not try to recycle it.
1356 */
1357 simple_lock(&vp->v_interlock);
1358 vp->v_usecount--;
1359 if (VSHOULDFREE(vp))
1360 vfree(vp);
1361 simple_unlock(&vp->v_interlock);
1362 }
1363 return (error);
1364 }
1365 simple_unlock(&vp->v_interlock);
1366 return (0);
1367}
1368
1369void
1370vref(struct vnode *vp)
1371{
1372 simple_lock(&vp->v_interlock);
1373 vp->v_usecount++;
1374 simple_unlock(&vp->v_interlock);
1375}
1376
1377/*
1378 * Vnode put/release.
1379 * If count drops to zero, call inactive routine and return to freelist.
1380 */
1381void
1382vrele(vp)
1383 struct vnode *vp;
1384{
1385 struct proc *p = curproc; /* XXX */
1386
1387 KASSERT(vp != NULL, ("vrele: null vp"));
1388
1389 simple_lock(&vp->v_interlock);
1390
1391 if (vp->v_usecount > 1) {
1392
1393 vp->v_usecount--;
1394 simple_unlock(&vp->v_interlock);
1395
1396 return;
1397 }
1398
1399 if (vp->v_usecount == 1) {
1400
1401 vp->v_usecount--;
1402 if (VSHOULDFREE(vp))
1403 vfree(vp);
1404 /*
1405 * If we are doing a vput, the node is already locked, and we must
1406 * call VOP_INACTIVE with the node locked. So, in the case of
1407 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1408 */
1409 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1410 VOP_INACTIVE(vp, p);
1411 }
1412
1413 } else {
1414#ifdef DIAGNOSTIC
1415 vprint("vrele: negative ref count", vp);
1416 simple_unlock(&vp->v_interlock);
1417#endif
1418 panic("vrele: negative ref cnt");
1419 }
1420}
1421
1422void
1423vput(vp)
1424 struct vnode *vp;
1425{
1426 struct proc *p = curproc; /* XXX */
1427
1428 KASSERT(vp != NULL, ("vput: null vp"));
1429
1430 simple_lock(&vp->v_interlock);
1431
1432 if (vp->v_usecount > 1) {
1433
1434 vp->v_usecount--;
1435 VOP_UNLOCK(vp, LK_INTERLOCK, p);
1436 return;
1437
1438 }
1439
1440 if (vp->v_usecount == 1) {
1441
1442 vp->v_usecount--;
1443 if (VSHOULDFREE(vp))
1444 vfree(vp);
1445 /*
1446 * If we are doing a vput, the node is already locked, and we must
1447 * call VOP_INACTIVE with the node locked. So, in the case of
1448 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1449 */
1450 simple_unlock(&vp->v_interlock);
1451 VOP_INACTIVE(vp, p);
1452
1453 } else {
1454#ifdef DIAGNOSTIC
1455 vprint("vput: negative ref count", vp);
1456#endif
1457 panic("vput: negative ref cnt");
1458 }
1459}
1460
1461/*
1462 * Somebody doesn't want the vnode recycled.
1463 */
1464void
1465vhold(vp)
1466 register struct vnode *vp;
1467{
1468 int s;
1469
1470 s = splbio();
1471 vp->v_holdcnt++;
1472 if (VSHOULDBUSY(vp))
1473 vbusy(vp);
1474 splx(s);
1475}
1476
1477/*
1478 * One less who cares about this vnode.
1479 */
1480void
1481vdrop(vp)
1482 register struct vnode *vp;
1483{
1484 int s;
1485
1486 s = splbio();
1487 if (vp->v_holdcnt <= 0)
1488 panic("vdrop: holdcnt");
1489 vp->v_holdcnt--;
1490 if (VSHOULDFREE(vp))
1491 vfree(vp);
1492 splx(s);
1493}
1494
1495/*
1496 * Remove any vnodes in the vnode table belonging to mount point mp.
1497 *
1498 * If MNT_NOFORCE is specified, there should not be any active ones,
1499 * return error if any are found (nb: this is a user error, not a
1500 * system error). If MNT_FORCE is specified, detach any active vnodes
1501 * that are found.
1502 */
1503#ifdef DIAGNOSTIC
1504static int busyprt = 0; /* print out busy vnodes */
1505SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1506#endif
1507
1508int
1509vflush(mp, skipvp, flags)
1510 struct mount *mp;
1511 struct vnode *skipvp;
1512 int flags;
1513{
1514 struct proc *p = curproc; /* XXX */
1515 struct vnode *vp, *nvp;
1516 int busy = 0;
1517
1518 simple_lock(&mntvnode_slock);
1519loop:
1520 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
1521 /*
1522 * Make sure this vnode wasn't reclaimed in getnewvnode().
1523 * Start over if it has (it won't be on the list anymore).
1524 */
1525 if (vp->v_mount != mp)
1526 goto loop;
1527 nvp = vp->v_mntvnodes.le_next;
1528 /*
1529 * Skip over a selected vnode.
1530 */
1531 if (vp == skipvp)
1532 continue;
1533
1534 simple_lock(&vp->v_interlock);
1535 /*
1536 * Skip over a vnodes marked VSYSTEM.
1537 */
1538 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1539 simple_unlock(&vp->v_interlock);
1540 continue;
1541 }
1542 /*
1543 * If WRITECLOSE is set, only flush out regular file vnodes
1544 * open for writing.
1545 */
1546 if ((flags & WRITECLOSE) &&
1547 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1548 simple_unlock(&vp->v_interlock);
1549 continue;
1550 }
1551
1552 /*
1553 * With v_usecount == 0, all we need to do is clear out the
1554 * vnode data structures and we are done.
1555 */
1556 if (vp->v_usecount == 0) {
1557 simple_unlock(&mntvnode_slock);
1558 vgonel(vp, p);
1559 simple_lock(&mntvnode_slock);
1560 continue;
1561 }
1562
1563 /*
1564 * If FORCECLOSE is set, forcibly close the vnode. For block
1565 * or character devices, revert to an anonymous device. For
1566 * all other files, just kill them.
1567 */
1568 if (flags & FORCECLOSE) {
1569 simple_unlock(&mntvnode_slock);
1570 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1571 vgonel(vp, p);
1572 } else {
1573 vclean(vp, 0, p);
1574 vp->v_op = spec_vnodeop_p;
1575 insmntque(vp, (struct mount *) 0);
1576 }
1577 simple_lock(&mntvnode_slock);
1578 continue;
1579 }
1580#ifdef DIAGNOSTIC
1581 if (busyprt)
1582 vprint("vflush: busy vnode", vp);
1583#endif
1584 simple_unlock(&vp->v_interlock);
1585 busy++;
1586 }
1587 simple_unlock(&mntvnode_slock);
1588 if (busy)
1589 return (EBUSY);
1590 return (0);
1591}
1592
1593/*
1594 * Disassociate the underlying file system from a vnode.
1595 */
1596static void
1597vclean(vp, flags, p)
1598 struct vnode *vp;
1599 int flags;
1600 struct proc *p;
1601{
1602 int active;
1603 vm_object_t obj;
1604
1605 /*
1606 * Check to see if the vnode is in use. If so we have to reference it
1607 * before we clean it out so that its count cannot fall to zero and
1608 * generate a race against ourselves to recycle it.
1609 */
1610 if ((active = vp->v_usecount))
1611 vp->v_usecount++;
1612
1613 /*
1614 * Prevent the vnode from being recycled or brought into use while we
1615 * clean it out.
1616 */
1617 if (vp->v_flag & VXLOCK)
1618 panic("vclean: deadlock");
1619 vp->v_flag |= VXLOCK;
1620 /*
1621 * Even if the count is zero, the VOP_INACTIVE routine may still
1622 * have the object locked while it cleans it out. The VOP_LOCK
1623 * ensures that the VOP_INACTIVE routine is done with its work.
1624 * For active vnodes, it ensures that no other activity can
1625 * occur while the underlying object is being cleaned out.
1626 */
1627 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1628
1629 /*
1630 * Clean out any buffers associated with the vnode.
1631 */
1632 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1633 if ((obj = vp->v_object) != NULL) {
1634 if (obj->ref_count == 0) {
1635 /*
1636 * This is a normal way of shutting down the object/vnode
1637 * association.
1638 */
1639 vm_object_terminate(obj);
1640 } else {
1641 /*
1642 * Woe to the process that tries to page now :-).
1643 */
1644 vm_pager_deallocate(obj);
1645 }
1646 }
1647
1648 /*
1649 * If purging an active vnode, it must be closed and
1650 * deactivated before being reclaimed. Note that the
1651 * VOP_INACTIVE will unlock the vnode.
1652 */
1653 if (active) {
1654 if (flags & DOCLOSE)
1655 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1656 VOP_INACTIVE(vp, p);
1657 } else {
1658 /*
1659 * Any other processes trying to obtain this lock must first
1660 * wait for VXLOCK to clear, then call the new lock operation.
1661 */
1662 VOP_UNLOCK(vp, 0, p);
1663 }
1664 /*
1665 * Reclaim the vnode.
1666 */
1667 if (VOP_RECLAIM(vp, p))
1668 panic("vclean: cannot reclaim");
1669
1670 if (active)
1671 vrele(vp);
1672
1673 cache_purge(vp);
1674 if (vp->v_vnlock) {
1675#if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */
1676#ifdef DIAGNOSTIC
1677 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0)
1678 vprint("vclean: lock not drained", vp);
1679#endif
1680#endif
1681 FREE(vp->v_vnlock, M_VNODE);
1682 vp->v_vnlock = NULL;
1683 }
1684
1685 if (VSHOULDFREE(vp))
1686 vfree(vp);
1687
1688 /*
1689 * Done with purge, notify sleepers of the grim news.
1690 */
1691 vp->v_op = dead_vnodeop_p;
1692 vn_pollgone(vp);
1693 vp->v_tag = VT_NON;
1694 vp->v_flag &= ~VXLOCK;
1695 if (vp->v_flag & VXWANT) {
1696 vp->v_flag &= ~VXWANT;
1697 wakeup((caddr_t) vp);
1698 }
1699}
1700
1701/*
1702 * Eliminate all activity associated with the requested vnode
1703 * and with all vnodes aliased to the requested vnode.
1704 */
1705int
1706vop_revoke(ap)
1707 struct vop_revoke_args /* {
1708 struct vnode *a_vp;
1709 int a_flags;
1710 } */ *ap;
1711{
1712 struct vnode *vp, *vq;
1713 struct proc *p = curproc; /* XXX */
1714
1715 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1716
1717 vp = ap->a_vp;
1718 simple_lock(&vp->v_interlock);
1719
1720 if (vp->v_flag & VALIASED) {
1721 /*
1722 * If a vgone (or vclean) is already in progress,
1723 * wait until it is done and return.
1724 */
1725 if (vp->v_flag & VXLOCK) {
1726 vp->v_flag |= VXWANT;
1727 simple_unlock(&vp->v_interlock);
1728 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1729 return (0);
1730 }
1731 /*
1732 * Ensure that vp will not be vgone'd while we
1733 * are eliminating its aliases.
1734 */
1735 vp->v_flag |= VXLOCK;
1736 simple_unlock(&vp->v_interlock);
1737 while (vp->v_flag & VALIASED) {
1738 simple_lock(&spechash_slock);
1739 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1740 if (vq->v_rdev != vp->v_rdev ||
1741 vq->v_type != vp->v_type || vp == vq)
1742 continue;
1743 simple_unlock(&spechash_slock);
1744 vgone(vq);
1745 break;
1746 }
1747 if (vq == NULLVP) {
1748 simple_unlock(&spechash_slock);
1749 }
1750 }
1751 /*
1752 * Remove the lock so that vgone below will
1753 * really eliminate the vnode after which time
1754 * vgone will awaken any sleepers.
1755 */
1756 simple_lock(&vp->v_interlock);
1757 vp->v_flag &= ~VXLOCK;
1758 if (vp->v_flag & VXWANT) {
1759 vp->v_flag &= ~VXWANT;
1760 wakeup(vp);
1761 }
1762 }
1763 vgonel(vp, p);
1764 return (0);
1765}
1766
1767/*
1768 * Recycle an unused vnode to the front of the free list.
1769 * Release the passed interlock if the vnode will be recycled.
1770 */
1771int
1772vrecycle(vp, inter_lkp, p)
1773 struct vnode *vp;
1774 struct simplelock *inter_lkp;
1775 struct proc *p;
1776{
1777
1778 simple_lock(&vp->v_interlock);
1779 if (vp->v_usecount == 0) {
1780 if (inter_lkp) {
1781 simple_unlock(inter_lkp);
1782 }
1783 vgonel(vp, p);
1784 return (1);
1785 }
1786 simple_unlock(&vp->v_interlock);
1787 return (0);
1788}
1789
1790/*
1791 * Eliminate all activity associated with a vnode
1792 * in preparation for reuse.
1793 */
1794void
1795vgone(vp)
1796 register struct vnode *vp;
1797{
1798 struct proc *p = curproc; /* XXX */
1799
1800 simple_lock(&vp->v_interlock);
1801 vgonel(vp, p);
1802}
1803
1804/*
1805 * vgone, with the vp interlock held.
1806 */
1807static void
1808vgonel(vp, p)
1809 struct vnode *vp;
1810 struct proc *p;
1811{
1812 int s;
1813 struct vnode *vq;
1814 struct vnode *vx;
1815
1816 /*
1817 * If a vgone (or vclean) is already in progress,
1818 * wait until it is done and return.
1819 */
1820 if (vp->v_flag & VXLOCK) {
1821 vp->v_flag |= VXWANT;
1822 simple_unlock(&vp->v_interlock);
1823 tsleep((caddr_t)vp, PINOD, "vgone", 0);
1824 return;
1825 }
1826
1827 /*
1828 * Clean out the filesystem specific data.
1829 */
1830 vclean(vp, DOCLOSE, p);
1831 simple_lock(&vp->v_interlock);
1832
1833 /*
1834 * Delete from old mount point vnode list, if on one.
1835 */
1836 if (vp->v_mount != NULL)
1837 insmntque(vp, (struct mount *)0);
1838 /*
1839 * If special device, remove it from special device alias list
1840 * if it is on one.
1841 */
1842 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1843 simple_lock(&spechash_slock);
1844 if (*vp->v_hashchain == vp) {
1845 *vp->v_hashchain = vp->v_specnext;
1846 } else {
1847 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1848 if (vq->v_specnext != vp)
1849 continue;
1850 vq->v_specnext = vp->v_specnext;
1851 break;
1852 }
1853 if (vq == NULL)
1854 panic("missing bdev");
1855 }
1856 if (vp->v_flag & VALIASED) {
1857 vx = NULL;
1858 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1859 if (vq->v_rdev != vp->v_rdev ||
1860 vq->v_type != vp->v_type)
1861 continue;
1862 if (vx)
1863 break;
1864 vx = vq;
1865 }
1866 if (vx == NULL)
1867 panic("missing alias");
1868 if (vq == NULL)
1869 vx->v_flag &= ~VALIASED;
1870 vp->v_flag &= ~VALIASED;
1871 }
1872 simple_unlock(&spechash_slock);
1873 FREE(vp->v_specinfo, M_VNODE);
1874 vp->v_specinfo = NULL;
1875 }
1876
1877 /*
1878 * If it is on the freelist and not already at the head,
1879 * move it to the head of the list. The test of the back
1880 * pointer and the reference count of zero is because
1881 * it will be removed from the free list by getnewvnode,
1882 * but will not have its reference count incremented until
1883 * after calling vgone. If the reference count were
1884 * incremented first, vgone would (incorrectly) try to
1885 * close the previous instance of the underlying object.
1886 */
1887 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1888 s = splbio();
1889 simple_lock(&vnode_free_list_slock);
1890 if (vp->v_flag & VFREE) {
1891 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1892 } else if (vp->v_flag & VTBFREE) {
1893 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1894 vp->v_flag &= ~VTBFREE;
1895 freevnodes++;
1896 } else
1897 freevnodes++;
1898 vp->v_flag |= VFREE;
1899 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1900 simple_unlock(&vnode_free_list_slock);
1901 splx(s);
1902 }
1903
1904 vp->v_type = VBAD;
1905 simple_unlock(&vp->v_interlock);
1906}
1907
1908/*
1909 * Lookup a vnode by device number.
1910 */
1911int
1912vfinddev(dev, type, vpp)
1913 dev_t dev;
1914 enum vtype type;
1915 struct vnode **vpp;
1916{
1917 register struct vnode *vp;
1918 int rc = 0;
1919
1920 simple_lock(&spechash_slock);
1921 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1922 if (dev != vp->v_rdev || type != vp->v_type)
1923 continue;
1924 *vpp = vp;
1925 rc = 1;
1926 break;
1927 }
1928 simple_unlock(&spechash_slock);
1929 return (rc);
1930}
1931
1932/*
1933 * Calculate the total number of references to a special device.
1934 */
1935int
1936vcount(vp)
1937 register struct vnode *vp;
1938{
1939 struct vnode *vq, *vnext;
1940 int count;
1941
1942loop:
1943 if ((vp->v_flag & VALIASED) == 0)
1944 return (vp->v_usecount);
1945 simple_lock(&spechash_slock);
1946 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1947 vnext = vq->v_specnext;
1948 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1949 continue;
1950 /*
1951 * Alias, but not in use, so flush it out.
1952 */
1953 if (vq->v_usecount == 0 && vq != vp) {
1954 simple_unlock(&spechash_slock);
1955 vgone(vq);
1956 goto loop;
1957 }
1958 count += vq->v_usecount;
1959 }
1960 simple_unlock(&spechash_slock);
1961 return (count);
1962}
1963/*
1964 * Print out a description of a vnode.
1965 */
1966static char *typename[] =
1967{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1968
1969void
1970vprint(label, vp)
1971 char *label;
1972 register struct vnode *vp;
1973{
1974 char buf[96];
1975
1976 if (label != NULL)
1977 printf("%s: %p: ", label, (void *)vp);
1978 else
1979 printf("%p: ", (void *)vp);
1980 printf("type %s, usecount %d, writecount %d, refcount %d,",
1981 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1982 vp->v_holdcnt);
1983 buf[0] = '\0';
1984 if (vp->v_flag & VROOT)
1985 strcat(buf, "|VROOT");
1986 if (vp->v_flag & VTEXT)
1987 strcat(buf, "|VTEXT");
1988 if (vp->v_flag & VSYSTEM)
1989 strcat(buf, "|VSYSTEM");
1990 if (vp->v_flag & VXLOCK)
1991 strcat(buf, "|VXLOCK");
1992 if (vp->v_flag & VXWANT)
1993 strcat(buf, "|VXWANT");
1994 if (vp->v_flag & VBWAIT)
1995 strcat(buf, "|VBWAIT");
1996 if (vp->v_flag & VALIASED)
1997 strcat(buf, "|VALIASED");
1998 if (vp->v_flag & VDOOMED)
1999 strcat(buf, "|VDOOMED");
2000 if (vp->v_flag & VFREE)
2001 strcat(buf, "|VFREE");
2002 if (vp->v_flag & VOBJBUF)
2003 strcat(buf, "|VOBJBUF");
2004 if (buf[0] != '\0')
2005 printf(" flags (%s)", &buf[1]);
2006 if (vp->v_data == NULL) {
2007 printf("\n");
2008 } else {
2009 printf("\n\t");
2010 VOP_PRINT(vp);
2011 }
2012}
2013
2014#ifdef DDB
2015#include <ddb/ddb.h>
2016/*
2017 * List all of the locked vnodes in the system.
2018 * Called when debugging the kernel.
2019 */
2020DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2021{
2022 struct proc *p = curproc; /* XXX */
2023 struct mount *mp, *nmp;
2024 struct vnode *vp;
2025
2026 printf("Locked vnodes\n");
2027 simple_lock(&mountlist_slock);
2028 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2029 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2030 nmp = mp->mnt_list.cqe_next;
2031 continue;
2032 }
2033 for (vp = mp->mnt_vnodelist.lh_first;
2034 vp != NULL;
2035 vp = vp->v_mntvnodes.le_next) {
2036 if (VOP_ISLOCKED(vp))
2037 vprint((char *)0, vp);
2038 }
2039 simple_lock(&mountlist_slock);
2040 nmp = mp->mnt_list.cqe_next;
2041 vfs_unbusy(mp, p);
2042 }
2043 simple_unlock(&mountlist_slock);
2044}
2045#endif
2046
2047/*
2048 * Top level filesystem related information gathering.
2049 */
2050static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
2051
2052static int
2053vfs_sysctl SYSCTL_HANDLER_ARGS
2054{
2055 int *name = (int *)arg1 - 1; /* XXX */
2056 u_int namelen = arg2 + 1; /* XXX */
2057 struct vfsconf *vfsp;
2058
2059#if 1 || defined(COMPAT_PRELITE2)
2060 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2061 if (namelen == 1)
2062 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2063#endif
2064
2065#ifdef notyet
2066 /* all sysctl names at this level are at least name and field */
2067 if (namelen < 2)
2068 return (ENOTDIR); /* overloaded */
2069 if (name[0] != VFS_GENERIC) {
2070 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2071 if (vfsp->vfc_typenum == name[0])
2072 break;
2073 if (vfsp == NULL)
2074 return (EOPNOTSUPP);
2075 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2076 oldp, oldlenp, newp, newlen, p));
2077 }
2078#endif
2079 switch (name[1]) {
2080 case VFS_MAXTYPENUM:
2081 if (namelen != 2)
2082 return (ENOTDIR);
2083 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2084 case VFS_CONF:
2085 if (namelen != 3)
2086 return (ENOTDIR); /* overloaded */
2087 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2088 if (vfsp->vfc_typenum == name[2])
2089 break;
2090 if (vfsp == NULL)
2091 return (EOPNOTSUPP);
2092 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2093 }
2094 return (EOPNOTSUPP);
2095}
2096
2097SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2098 "Generic filesystem");
2099
2100#if 1 || defined(COMPAT_PRELITE2)
2101
2102static int
2103sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2104{
2105 int error;
2106 struct vfsconf *vfsp;
2107 struct ovfsconf ovfs;
2108
2109 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2110 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2111 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2112 ovfs.vfc_index = vfsp->vfc_typenum;
2113 ovfs.vfc_refcount = vfsp->vfc_refcount;
2114 ovfs.vfc_flags = vfsp->vfc_flags;
2115 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2116 if (error)
2117 return error;
2118 }
2119 return 0;
2120}
2121
2122#endif /* 1 || COMPAT_PRELITE2 */
2123
2124#if 0
2125#define KINFO_VNODESLOP 10
2126/*
2127 * Dump vnode list (via sysctl).
2128 * Copyout address of vnode followed by vnode.
2129 */
2130/* ARGSUSED */
2131static int
2132sysctl_vnode SYSCTL_HANDLER_ARGS
2133{
2134 struct proc *p = curproc; /* XXX */
2135 struct mount *mp, *nmp;
2136 struct vnode *nvp, *vp;
2137 int error;
2138
2139#define VPTRSZ sizeof (struct vnode *)
2140#define VNODESZ sizeof (struct vnode)
2141
2142 req->lock = 0;
2143 if (!req->oldptr) /* Make an estimate */
2144 return (SYSCTL_OUT(req, 0,
2145 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2146
2147 simple_lock(&mountlist_slock);
2148 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2149 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2150 nmp = mp->mnt_list.cqe_next;
2151 continue;
2152 }
2153again:
2154 simple_lock(&mntvnode_slock);
2155 for (vp = mp->mnt_vnodelist.lh_first;
2156 vp != NULL;
2157 vp = nvp) {
2158 /*
2159 * Check that the vp is still associated with
2160 * this filesystem. RACE: could have been
2161 * recycled onto the same filesystem.
2162 */
2163 if (vp->v_mount != mp) {
2164 simple_unlock(&mntvnode_slock);
2165 goto again;
2166 }
2167 nvp = vp->v_mntvnodes.le_next;
2168 simple_unlock(&mntvnode_slock);
2169 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2170 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2171 return (error);
2172 simple_lock(&mntvnode_slock);
2173 }
2174 simple_unlock(&mntvnode_slock);
2175 simple_lock(&mountlist_slock);
2176 nmp = mp->mnt_list.cqe_next;
2177 vfs_unbusy(mp, p);
2178 }
2179 simple_unlock(&mountlist_slock);
2180
2181 return (0);
2182}
2183#endif
2184
2185/*
2186 * XXX
2187 * Exporting the vnode list on large systems causes them to crash.
2188 * Exporting the vnode list on medium systems causes sysctl to coredump.
2189 */
2190#if 0
2191SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2192 0, 0, sysctl_vnode, "S,vnode", "");
2193#endif
2194
2195/*
2196 * Check to see if a filesystem is mounted on a block device.
2197 */
2198int
2199vfs_mountedon(vp)
2200 struct vnode *vp;
2201{
2202 struct vnode *vq;
2203 int error = 0;
2204
2205 if (vp->v_specmountpoint != NULL)
2206 return (EBUSY);
2207 if (vp->v_flag & VALIASED) {
2208 simple_lock(&spechash_slock);
2209 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2210 if (vq->v_rdev != vp->v_rdev ||
2211 vq->v_type != vp->v_type)
2212 continue;
2213 if (vq->v_specmountpoint != NULL) {
2214 error = EBUSY;
2215 break;
2216 }
2217 }
2218 simple_unlock(&spechash_slock);
2219 }
2220 return (error);
2221}
2222
2223/*
2224 * Unmount all filesystems. The list is traversed in reverse order
2225 * of mounting to avoid dependencies.
2226 */
2227void
2228vfs_unmountall()
2229{
2230 struct mount *mp, *nmp;
2231 struct proc *p;
2232 int error;
2233
2234 if (curproc != NULL)
2235 p = curproc;
2236 else
2237 p = initproc; /* XXX XXX should this be proc0? */
2238 /*
2239 * Since this only runs when rebooting, it is not interlocked.
2240 */
2241 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2242 nmp = mp->mnt_list.cqe_prev;
2243 error = dounmount(mp, MNT_FORCE, p);
2244 if (error) {
2245 printf("unmount of %s failed (",
2246 mp->mnt_stat.f_mntonname);
2247 if (error == EBUSY)
2248 printf("BUSY)\n");
2249 else
2250 printf("%d)\n", error);
2251 }
2252 }
2253}
2254
2255/*
2256 * Build hash lists of net addresses and hang them off the mount point.
2257 * Called by ufs_mount() to set up the lists of export addresses.
2258 */
2259static int
2260vfs_hang_addrlist(mp, nep, argp)
2261 struct mount *mp;
2262 struct netexport *nep;
2263 struct export_args *argp;
2264{
2265 register struct netcred *np;
2266 register struct radix_node_head *rnh;
2267 register int i;
2268 struct radix_node *rn;
2269 struct sockaddr *saddr, *smask = 0;
2270 struct domain *dom;
2271 int error;
2272
2273 if (argp->ex_addrlen == 0) {
2274 if (mp->mnt_flag & MNT_DEFEXPORTED)
2275 return (EPERM);
2276 np = &nep->ne_defexported;
2277 np->netc_exflags = argp->ex_flags;
2278 np->netc_anon = argp->ex_anon;
2279 np->netc_anon.cr_ref = 1;
2280 mp->mnt_flag |= MNT_DEFEXPORTED;
2281 return (0);
2282 }
2283 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2284 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2285 bzero((caddr_t) np, i);
2286 saddr = (struct sockaddr *) (np + 1);
2287 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2288 goto out;
2289 if (saddr->sa_len > argp->ex_addrlen)
2290 saddr->sa_len = argp->ex_addrlen;
2291 if (argp->ex_masklen) {
2292 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2293 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2294 if (error)
2295 goto out;
2296 if (smask->sa_len > argp->ex_masklen)
2297 smask->sa_len = argp->ex_masklen;
2298 }
2299 i = saddr->sa_family;
2300 if ((rnh = nep->ne_rtable[i]) == 0) {
2301 /*
2302 * Seems silly to initialize every AF when most are not used,
2303 * do so on demand here
2304 */
2305 for (dom = domains; dom; dom = dom->dom_next)
2306 if (dom->dom_family == i && dom->dom_rtattach) {
2307 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2308 dom->dom_rtoffset);
2309 break;
2310 }
2311 if ((rnh = nep->ne_rtable[i]) == 0) {
2312 error = ENOBUFS;
2313 goto out;
2314 }
2315 }
2316 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2317 np->netc_rnodes);
2318 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2319 error = EPERM;
2320 goto out;
2321 }
2322 np->netc_exflags = argp->ex_flags;
2323 np->netc_anon = argp->ex_anon;
2324 np->netc_anon.cr_ref = 1;
2325 return (0);
2326out:
2327 free(np, M_NETADDR);
2328 return (error);
2329}
2330
2331/* ARGSUSED */
2332static int
2333vfs_free_netcred(rn, w)
2334 struct radix_node *rn;
2335 void *w;
2336{
2337 register struct radix_node_head *rnh = (struct radix_node_head *) w;
2338
2339 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2340 free((caddr_t) rn, M_NETADDR);
2341 return (0);
2342}
2343
2344/*
2345 * Free the net address hash lists that are hanging off the mount points.
2346 */
2347static void
2348vfs_free_addrlist(nep)
2349 struct netexport *nep;
2350{
2351 register int i;
2352 register struct radix_node_head *rnh;
2353
2354 for (i = 0; i <= AF_MAX; i++)
2355 if ((rnh = nep->ne_rtable[i])) {
2356 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2357 (caddr_t) rnh);
2358 free((caddr_t) rnh, M_RTABLE);
2359 nep->ne_rtable[i] = 0;
2360 }
2361}
2362
2363int
2364vfs_export(mp, nep, argp)
2365 struct mount *mp;
2366 struct netexport *nep;
2367 struct export_args *argp;
2368{
2369 int error;
2370
2371 if (argp->ex_flags & MNT_DELEXPORT) {
2372 if (mp->mnt_flag & MNT_EXPUBLIC) {
2373 vfs_setpublicfs(NULL, NULL, NULL);
2374 mp->mnt_flag &= ~MNT_EXPUBLIC;
2375 }
2376 vfs_free_addrlist(nep);
2377 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2378 }
2379 if (argp->ex_flags & MNT_EXPORTED) {
2380 if (argp->ex_flags & MNT_EXPUBLIC) {
2381 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2382 return (error);
2383 mp->mnt_flag |= MNT_EXPUBLIC;
2384 }
2385 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2386 return (error);
2387 mp->mnt_flag |= MNT_EXPORTED;
2388 }
2389 return (0);
2390}
2391
2392
2393/*
2394 * Set the publicly exported filesystem (WebNFS). Currently, only
2395 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2396 */
2397int
2398vfs_setpublicfs(mp, nep, argp)
2399 struct mount *mp;
2400 struct netexport *nep;
2401 struct export_args *argp;
2402{
2403 int error;
2404 struct vnode *rvp;
2405 char *cp;
2406
2407 /*
2408 * mp == NULL -> invalidate the current info, the FS is
2409 * no longer exported. May be called from either vfs_export
2410 * or unmount, so check if it hasn't already been done.
2411 */
2412 if (mp == NULL) {
2413 if (nfs_pub.np_valid) {
2414 nfs_pub.np_valid = 0;
2415 if (nfs_pub.np_index != NULL) {
2416 FREE(nfs_pub.np_index, M_TEMP);
2417 nfs_pub.np_index = NULL;
2418 }
2419 }
2420 return (0);
2421 }
2422
2423 /*
2424 * Only one allowed at a time.
2425 */
2426 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2427 return (EBUSY);
2428
2429 /*
2430 * Get real filehandle for root of exported FS.
2431 */
2432 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2433 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2434
2435 if ((error = VFS_ROOT(mp, &rvp)))
2436 return (error);
2437
2438 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2439 return (error);
2440
2441 vput(rvp);
2442
2443 /*
2444 * If an indexfile was specified, pull it in.
2445 */
2446 if (argp->ex_indexfile != NULL) {
2447 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2448 M_WAITOK);
2449 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2450 MAXNAMLEN, (size_t *)0);
2451 if (!error) {
2452 /*
2453 * Check for illegal filenames.
2454 */
2455 for (cp = nfs_pub.np_index; *cp; cp++) {
2456 if (*cp == '/') {
2457 error = EINVAL;
2458 break;
2459 }
2460 }
2461 }
2462 if (error) {
2463 FREE(nfs_pub.np_index, M_TEMP);
2464 return (error);
2465 }
2466 }
2467
2468 nfs_pub.np_mount = mp;
2469 nfs_pub.np_valid = 1;
2470 return (0);
2471}
2472
2473struct netcred *
2474vfs_export_lookup(mp, nep, nam)
2475 register struct mount *mp;
2476 struct netexport *nep;
2477 struct sockaddr *nam;
2478{
2479 register struct netcred *np;
2480 register struct radix_node_head *rnh;
2481 struct sockaddr *saddr;
2482
2483 np = NULL;
2484 if (mp->mnt_flag & MNT_EXPORTED) {
2485 /*
2486 * Lookup in the export list first.
2487 */
2488 if (nam != NULL) {
2489 saddr = nam;
2490 rnh = nep->ne_rtable[saddr->sa_family];
2491 if (rnh != NULL) {
2492 np = (struct netcred *)
2493 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2494 rnh);
2495 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2496 np = NULL;
2497 }
2498 }
2499 /*
2500 * If no address match, use the default if it exists.
2501 */
2502 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2503 np = &nep->ne_defexported;
2504 }
2505 return (np);
2506}
2507
2508/*
2509 * perform msync on all vnodes under a mount point
2510 * the mount point must be locked.
2511 */
2512void
2513vfs_msync(struct mount *mp, int flags) {
2514 struct vnode *vp, *nvp;
2515 struct vm_object *obj;
2516 int anyio, tries;
2517
2518 tries = 5;
2519loop:
2520 anyio = 0;
2521 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
2522
2523 nvp = vp->v_mntvnodes.le_next;
2524
2525 if (vp->v_mount != mp) {
2526 goto loop;
2527 }
2528
2529 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2530 continue;
2531
2532 if (flags != MNT_WAIT) {
2533 obj = vp->v_object;
2534 if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2535 continue;
2536 if (VOP_ISLOCKED(vp))
2537 continue;
2538 }
2539
2540 simple_lock(&vp->v_interlock);
2541 if (vp->v_object &&
2542 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2543 if (!vget(vp,
2544 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2545 if (vp->v_object) {
2546 vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : 0);
2547 anyio = 1;
2548 }
2549 vput(vp);
2550 }
2551 } else {
2552 simple_unlock(&vp->v_interlock);
2553 }
2554 }
2555 if (anyio && (--tries > 0))
2556 goto loop;
2557}
2558
2559/*
2560 * Create the VM object needed for VMIO and mmap support. This
2561 * is done for all VREG files in the system. Some filesystems might
2562 * afford the additional metadata buffering capability of the
2563 * VMIO code by making the device node be VMIO mode also.
2564 *
2565 * vp must be locked when vfs_object_create is called.
2566 */
2567int
2568vfs_object_create(vp, p, cred)
2569 struct vnode *vp;
2570 struct proc *p;
2571 struct ucred *cred;
2572{
2573 struct vattr vat;
2574 vm_object_t object;
2575 int error = 0;
2576
2577 if ((vp->v_type != VREG) && (vp->v_type != VBLK))
2578 return 0;
2579
2580retry:
2581 if ((object = vp->v_object) == NULL) {
2582 if (vp->v_type == VREG) {
2583 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2584 goto retn;
2585 object = vnode_pager_alloc(vp, vat.va_size, 0, 0);
2586 } else if (major(vp->v_rdev) < nblkdev &&
2587 bdevsw[major(vp->v_rdev)] != NULL) {
2588 /*
2589 * This simply allocates the biggest object possible
2590 * for a VBLK vnode. This should be fixed, but doesn't
2591 * cause any problems (yet).
2592 */
2593 object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0);
2594 } else {
2595 goto retn;
2596 }
2597 /*
2598 * Dereference the reference we just created. This assumes
2599 * that the object is associated with the vp.
2600 */
2601 object->ref_count--;
2602 vp->v_usecount--;
2603 } else {
2604 if (object->flags & OBJ_DEAD) {
2605 VOP_UNLOCK(vp, 0, p);
2606 tsleep(object, PVM, "vodead", 0);
2607 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2608 goto retry;
2609 }
2610 }
2611
2612 KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object"));
2613 vp->v_flag |= VOBJBUF;
2614
2615retn:
2616 return error;
2617}
2618
2619static void
2620vfree(vp)
2621 struct vnode *vp;
2622{
2623 int s;
2624
2625 s = splbio();
2626 simple_lock(&vnode_free_list_slock);
2627 if (vp->v_flag & VTBFREE) {
2628 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2629 vp->v_flag &= ~VTBFREE;
2630 }
2631 if (vp->v_flag & VAGE) {
2632 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2633 } else {
2634 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2635 }
2636 freevnodes++;
2637 simple_unlock(&vnode_free_list_slock);
2638 vp->v_flag &= ~VAGE;
2639 vp->v_flag |= VFREE;
2640 splx(s);
2641}
2642
2643void
2644vbusy(vp)
2645 struct vnode *vp;
2646{
2647 int s;
2648
2649 s = splbio();
2650 simple_lock(&vnode_free_list_slock);
2651 if (vp->v_flag & VTBFREE) {
2652 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2653 vp->v_flag &= ~VTBFREE;
2654 } else {
2655 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2656 freevnodes--;
2657 }
2658 simple_unlock(&vnode_free_list_slock);
2659 vp->v_flag &= ~(VFREE|VAGE);
2660 splx(s);
2661}
2662
2663/*
2664 * Record a process's interest in events which might happen to
2665 * a vnode. Because poll uses the historic select-style interface
2666 * internally, this routine serves as both the ``check for any
2667 * pending events'' and the ``record my interest in future events''
2668 * functions. (These are done together, while the lock is held,
2669 * to avoid race conditions.)
2670 */
2671int
2672vn_pollrecord(vp, p, events)
2673 struct vnode *vp;
2674 struct proc *p;
2675 short events;
2676{
2677 simple_lock(&vp->v_pollinfo.vpi_lock);
2678 if (vp->v_pollinfo.vpi_revents & events) {
2679 /*
2680 * This leaves events we are not interested
2681 * in available for the other process which
2682 * which presumably had requested them
2683 * (otherwise they would never have been
2684 * recorded).
2685 */
2686 events &= vp->v_pollinfo.vpi_revents;
2687 vp->v_pollinfo.vpi_revents &= ~events;
2688
2689 simple_unlock(&vp->v_pollinfo.vpi_lock);
2690 return events;
2691 }
2692 vp->v_pollinfo.vpi_events |= events;
2693 selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2694 simple_unlock(&vp->v_pollinfo.vpi_lock);
2695 return 0;
2696}
2697
2698/*
2699 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2700 * it is possible for us to miss an event due to race conditions, but
2701 * that condition is expected to be rare, so for the moment it is the
2702 * preferred interface.
2703 */
2704void
2705vn_pollevent(vp, events)
2706 struct vnode *vp;
2707 short events;
2708{
2709 simple_lock(&vp->v_pollinfo.vpi_lock);
2710 if (vp->v_pollinfo.vpi_events & events) {
2711 /*
2712 * We clear vpi_events so that we don't
2713 * call selwakeup() twice if two events are
2714 * posted before the polling process(es) is
2715 * awakened. This also ensures that we take at
2716 * most one selwakeup() if the polling process
2717 * is no longer interested. However, it does
2718 * mean that only one event can be noticed at
2719 * a time. (Perhaps we should only clear those
2720 * event bits which we note?) XXX
2721 */
2722 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2723 vp->v_pollinfo.vpi_revents |= events;
2724 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2725 }
2726 simple_unlock(&vp->v_pollinfo.vpi_lock);
2727}
2728
2729/*
2730 * Wake up anyone polling on vp because it is being revoked.
2731 * This depends on dead_poll() returning POLLHUP for correct
2732 * behavior.
2733 */
2734void
2735vn_pollgone(vp)
2736 struct vnode *vp;
2737{
2738 simple_lock(&vp->v_pollinfo.vpi_lock);
2739 if (vp->v_pollinfo.vpi_events) {
2740 vp->v_pollinfo.vpi_events = 0;
2741 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2742 }
2743 simple_unlock(&vp->v_pollinfo.vpi_lock);
2744}
2745
2746
2747
2748/*
2749 * Routine to create and manage a filesystem syncer vnode.
2750 */
2751#define sync_close ((int (*) __P((struct vop_close_args *)))nullop)
2752static int sync_fsync __P((struct vop_fsync_args *));
2753static int sync_inactive __P((struct vop_inactive_args *));
2754static int sync_reclaim __P((struct vop_reclaim_args *));
2755#define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock)
2756#define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock)
2757static int sync_print __P((struct vop_print_args *));
2758#define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2759
2760static vop_t **sync_vnodeop_p;
2761static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2762 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2763 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2764 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2765 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2766 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2767 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2768 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2769 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2770 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2771 { NULL, NULL }
2772};
2773static struct vnodeopv_desc sync_vnodeop_opv_desc =
2774 { &sync_vnodeop_p, sync_vnodeop_entries };
2775
2776VNODEOP_SET(sync_vnodeop_opv_desc);
2777
2778/*
2779 * Create a new filesystem syncer vnode for the specified mount point.
2780 */
2781int
2782vfs_allocate_syncvnode(mp)
2783 struct mount *mp;
2784{
2785 struct vnode *vp;
2786 static long start, incr, next;
2787 int error;
2788
2789 /* Allocate a new vnode */
2790 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2791 mp->mnt_syncer = NULL;
2792 return (error);
2793 }
2794 vp->v_type = VNON;
2795 /*
2796 * Place the vnode onto the syncer worklist. We attempt to
2797 * scatter them about on the list so that they will go off
2798 * at evenly distributed times even if all the filesystems
2799 * are mounted at once.
2800 */
2801 next += incr;
2802 if (next == 0 || next > syncer_maxdelay) {
2803 start /= 2;
2804 incr /= 2;
2805 if (start == 0) {
2806 start = syncer_maxdelay / 2;
2807 incr = syncer_maxdelay;
2808 }
2809 next = start;
2810 }
2811 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2812 mp->mnt_syncer = vp;
2813 return (0);
2814}
2815
2816/*
2817 * Do a lazy sync of the filesystem.
2818 */
2819static int
2820sync_fsync(ap)
2821 struct vop_fsync_args /* {
2822 struct vnode *a_vp;
2823 struct ucred *a_cred;
2824 int a_waitfor;
2825 struct proc *a_p;
2826 } */ *ap;
2827{
2828 struct vnode *syncvp = ap->a_vp;
2829 struct mount *mp = syncvp->v_mount;
2830 struct proc *p = ap->a_p;
2831 int asyncflag;
2832
2833 /*
2834 * We only need to do something if this is a lazy evaluation.
2835 */
2836 if (ap->a_waitfor != MNT_LAZY)
2837 return (0);
2838
2839 /*
2840 * Move ourselves to the back of the sync list.
2841 */
2842 vn_syncer_add_to_worklist(syncvp, syncdelay);
2843
2844 /*
2845 * Walk the list of vnodes pushing all that are dirty and
2846 * not already on the sync list.
2847 */
2848 simple_lock(&mountlist_slock);
2849 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2850 simple_unlock(&mountlist_slock);
2851 return (0);
2852 }
2853 asyncflag = mp->mnt_flag & MNT_ASYNC;
2854 mp->mnt_flag &= ~MNT_ASYNC;
2855 vfs_msync(mp, MNT_NOWAIT);
2856 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2857 if (asyncflag)
2858 mp->mnt_flag |= MNT_ASYNC;
2859 vfs_unbusy(mp, p);
2860 return (0);
2861}
2862
2863/*
2864 * The syncer vnode is no referenced.
2865 */
2866static int
2867sync_inactive(ap)
2868 struct vop_inactive_args /* {
2869 struct vnode *a_vp;
2870 struct proc *a_p;
2871 } */ *ap;
2872{
2873
2874 vgone(ap->a_vp);
2875 return (0);
2876}
2877
2878/*
2879 * The syncer vnode is no longer needed and is being decommissioned.
2880 *
2881 * Modifications to the worklist must be protected at splbio().
2882 */
2883static int
2884sync_reclaim(ap)
2885 struct vop_reclaim_args /* {
2886 struct vnode *a_vp;
2887 } */ *ap;
2888{
2889 struct vnode *vp = ap->a_vp;
2890 int s;
2891
2892 s = splbio();
2893 vp->v_mount->mnt_syncer = NULL;
2894 if (vp->v_flag & VONWORKLST) {
2895 LIST_REMOVE(vp, v_synclist);
2896 vp->v_flag &= ~VONWORKLST;
2897 }
2898 splx(s);
2899
2900 return (0);
2901}
2902
2903/*
2904 * Print out a syncer vnode.
2905 */
2906static int
2907sync_print(ap)
2908 struct vop_print_args /* {
2909 struct vnode *a_vp;
2910 } */ *ap;
2911{
2912 struct vnode *vp = ap->a_vp;
2913
2914 printf("syncer vnode");
2915 if (vp->v_vnlock != NULL)
2916 lockmgr_printinfo(vp->v_vnlock);
2917 printf("\n");
2918 return (0);
2919}
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
25 *
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
36 * SUCH DAMAGE.
37 *
38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
39 * $Id: vfs_subr.c,v 1.188 1999/02/25 05:22:29 dillon Exp $
40 */
41
42/*
43 * External virtual filesystem routines
44 */
45#include "opt_ddb.h"
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/conf.h>
50#include <sys/fcntl.h>
51#include <sys/kernel.h>
52#include <sys/proc.h>
53#include <sys/malloc.h>
54#include <sys/mount.h>
55#include <sys/socket.h>
56#include <sys/vnode.h>
57#include <sys/stat.h>
58#include <sys/buf.h>
59#include <sys/domain.h>
60#include <sys/dirent.h>
61#include <sys/vmmeter.h>
62
63#include <machine/limits.h>
64
65#include <vm/vm.h>
66#include <vm/vm_param.h>
67#include <vm/vm_prot.h>
68#include <vm/vm_object.h>
69#include <vm/vm_extern.h>
70#include <vm/pmap.h>
71#include <vm/vm_map.h>
72#include <vm/vm_page.h>
73#include <vm/vm_pager.h>
74#include <vm/vnode_pager.h>
75#include <vm/vm_zone.h>
76#include <sys/sysctl.h>
77
78#include <miscfs/specfs/specdev.h>
79
80static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
81
82static void insmntque __P((struct vnode *vp, struct mount *mp));
83static void vclean __P((struct vnode *vp, int flags, struct proc *p));
84static void vfree __P((struct vnode *));
85static void vgonel __P((struct vnode *vp, struct proc *p));
86static unsigned long numvnodes;
87SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
88
89enum vtype iftovt_tab[16] = {
90 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
91 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
92};
93int vttoif_tab[9] = {
94 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
95 S_IFSOCK, S_IFIFO, S_IFMT,
96};
97
98static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
99struct tobefreelist vnode_tobefree_list; /* vnode free list */
100
101static u_long wantfreevnodes = 25;
102SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
103static u_long freevnodes = 0;
104SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
105
106int vfs_ioopt = 0;
107#ifdef ENABLE_VFS_IOOPT
108SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
109#endif
110
111struct mntlist mountlist; /* mounted filesystem list */
112struct simplelock mountlist_slock;
113struct simplelock mntvnode_slock;
114int nfs_mount_type = -1;
115#ifndef NULL_SIMPLELOCKS
116static struct simplelock mntid_slock;
117static struct simplelock vnode_free_list_slock;
118static struct simplelock spechash_slock;
119#endif
120struct nfs_public nfs_pub; /* publicly exported FS */
121static vm_zone_t vnode_zone;
122
123/*
124 * The workitem queue.
125 */
126#define SYNCER_MAXDELAY 32
127static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
128time_t syncdelay = 30;
129int rushjob; /* number of slots to run ASAP */
130
131static int syncer_delayno = 0;
132static long syncer_mask;
133LIST_HEAD(synclist, vnode);
134static struct synclist *syncer_workitem_pending;
135
136int desiredvnodes;
137SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, "");
138
139static void vfs_free_addrlist __P((struct netexport *nep));
140static int vfs_free_netcred __P((struct radix_node *rn, void *w));
141static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
142 struct export_args *argp));
143
144/*
145 * Initialize the vnode management data structures.
146 */
147void
148vntblinit()
149{
150
151 desiredvnodes = maxproc + cnt.v_page_count / 4;
152 simple_lock_init(&mntvnode_slock);
153 simple_lock_init(&mntid_slock);
154 simple_lock_init(&spechash_slock);
155 TAILQ_INIT(&vnode_free_list);
156 TAILQ_INIT(&vnode_tobefree_list);
157 simple_lock_init(&vnode_free_list_slock);
158 CIRCLEQ_INIT(&mountlist);
159 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
160 /*
161 * Initialize the filesystem syncer.
162 */
163 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
164 &syncer_mask);
165 syncer_maxdelay = syncer_mask + 1;
166}
167
168/*
169 * Mark a mount point as busy. Used to synchronize access and to delay
170 * unmounting. Interlock is not released on failure.
171 */
172int
173vfs_busy(mp, flags, interlkp, p)
174 struct mount *mp;
175 int flags;
176 struct simplelock *interlkp;
177 struct proc *p;
178{
179 int lkflags;
180
181 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
182 if (flags & LK_NOWAIT)
183 return (ENOENT);
184 mp->mnt_kern_flag |= MNTK_MWAIT;
185 if (interlkp) {
186 simple_unlock(interlkp);
187 }
188 /*
189 * Since all busy locks are shared except the exclusive
190 * lock granted when unmounting, the only place that a
191 * wakeup needs to be done is at the release of the
192 * exclusive lock at the end of dounmount.
193 */
194 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
195 if (interlkp) {
196 simple_lock(interlkp);
197 }
198 return (ENOENT);
199 }
200 lkflags = LK_SHARED | LK_NOPAUSE;
201 if (interlkp)
202 lkflags |= LK_INTERLOCK;
203 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
204 panic("vfs_busy: unexpected lock failure");
205 return (0);
206}
207
208/*
209 * Free a busy filesystem.
210 */
211void
212vfs_unbusy(mp, p)
213 struct mount *mp;
214 struct proc *p;
215{
216
217 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
218}
219
220/*
221 * Lookup a filesystem type, and if found allocate and initialize
222 * a mount structure for it.
223 *
224 * Devname is usually updated by mount(8) after booting.
225 */
226int
227vfs_rootmountalloc(fstypename, devname, mpp)
228 char *fstypename;
229 char *devname;
230 struct mount **mpp;
231{
232 struct proc *p = curproc; /* XXX */
233 struct vfsconf *vfsp;
234 struct mount *mp;
235
236 if (fstypename == NULL)
237 return (ENODEV);
238 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
239 if (!strcmp(vfsp->vfc_name, fstypename))
240 break;
241 if (vfsp == NULL)
242 return (ENODEV);
243 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
244 bzero((char *)mp, (u_long)sizeof(struct mount));
245 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
246 (void)vfs_busy(mp, LK_NOWAIT, 0, p);
247 LIST_INIT(&mp->mnt_vnodelist);
248 mp->mnt_vfc = vfsp;
249 mp->mnt_op = vfsp->vfc_vfsops;
250 mp->mnt_flag = MNT_RDONLY;
251 mp->mnt_vnodecovered = NULLVP;
252 vfsp->vfc_refcount++;
253 mp->mnt_stat.f_type = vfsp->vfc_typenum;
254 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
255 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
256 mp->mnt_stat.f_mntonname[0] = '/';
257 mp->mnt_stat.f_mntonname[1] = 0;
258 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
259 *mpp = mp;
260 return (0);
261}
262
263/*
264 * Find an appropriate filesystem to use for the root. If a filesystem
265 * has not been preselected, walk through the list of known filesystems
266 * trying those that have mountroot routines, and try them until one
267 * works or we have tried them all.
268 */
269#ifdef notdef /* XXX JH */
270int
271lite2_vfs_mountroot()
272{
273 struct vfsconf *vfsp;
274 extern int (*lite2_mountroot) __P((void));
275 int error;
276
277 if (lite2_mountroot != NULL)
278 return ((*lite2_mountroot)());
279 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
280 if (vfsp->vfc_mountroot == NULL)
281 continue;
282 if ((error = (*vfsp->vfc_mountroot)()) == 0)
283 return (0);
284 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
285 }
286 return (ENODEV);
287}
288#endif
289
290/*
291 * Lookup a mount point by filesystem identifier.
292 */
293struct mount *
294vfs_getvfs(fsid)
295 fsid_t *fsid;
296{
297 register struct mount *mp;
298
299 simple_lock(&mountlist_slock);
300 for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
301 mp = mp->mnt_list.cqe_next) {
302 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
303 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
304 simple_unlock(&mountlist_slock);
305 return (mp);
306 }
307 }
308 simple_unlock(&mountlist_slock);
309 return ((struct mount *) 0);
310}
311
312/*
313 * Get a new unique fsid
314 */
315void
316vfs_getnewfsid(mp)
317 struct mount *mp;
318{
319 static u_short xxxfs_mntid;
320
321 fsid_t tfsid;
322 int mtype;
323
324 simple_lock(&mntid_slock);
325 mtype = mp->mnt_vfc->vfc_typenum;
326 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
327 mp->mnt_stat.f_fsid.val[1] = mtype;
328 if (xxxfs_mntid == 0)
329 ++xxxfs_mntid;
330 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
331 tfsid.val[1] = mtype;
332 if (mountlist.cqh_first != (void *)&mountlist) {
333 while (vfs_getvfs(&tfsid)) {
334 tfsid.val[0]++;
335 xxxfs_mntid++;
336 }
337 }
338 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
339 simple_unlock(&mntid_slock);
340}
341
342/*
343 * Set vnode attributes to VNOVAL
344 */
345void
346vattr_null(vap)
347 register struct vattr *vap;
348{
349
350 vap->va_type = VNON;
351 vap->va_size = VNOVAL;
352 vap->va_bytes = VNOVAL;
353 vap->va_mode = VNOVAL;
354 vap->va_nlink = VNOVAL;
355 vap->va_uid = VNOVAL;
356 vap->va_gid = VNOVAL;
357 vap->va_fsid = VNOVAL;
358 vap->va_fileid = VNOVAL;
359 vap->va_blocksize = VNOVAL;
360 vap->va_rdev = VNOVAL;
361 vap->va_atime.tv_sec = VNOVAL;
362 vap->va_atime.tv_nsec = VNOVAL;
363 vap->va_mtime.tv_sec = VNOVAL;
364 vap->va_mtime.tv_nsec = VNOVAL;
365 vap->va_ctime.tv_sec = VNOVAL;
366 vap->va_ctime.tv_nsec = VNOVAL;
367 vap->va_flags = VNOVAL;
368 vap->va_gen = VNOVAL;
369 vap->va_vaflags = 0;
370}
371
372/*
373 * Routines having to do with the management of the vnode table.
374 */
375extern vop_t **dead_vnodeop_p;
376
377/*
378 * Return the next vnode from the free list.
379 */
380int
381getnewvnode(tag, mp, vops, vpp)
382 enum vtagtype tag;
383 struct mount *mp;
384 vop_t **vops;
385 struct vnode **vpp;
386{
387 int s;
388 struct proc *p = curproc; /* XXX */
389 struct vnode *vp, *tvp, *nvp;
390 vm_object_t object;
391 TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
392
393 /*
394 * We take the least recently used vnode from the freelist
395 * if we can get it and it has no cached pages, and no
396 * namecache entries are relative to it.
397 * Otherwise we allocate a new vnode
398 */
399
400 s = splbio();
401 simple_lock(&vnode_free_list_slock);
402 TAILQ_INIT(&vnode_tmp_list);
403
404 for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
405 nvp = TAILQ_NEXT(vp, v_freelist);
406 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
407 if (vp->v_flag & VAGE) {
408 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
409 } else {
410 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
411 }
412 vp->v_flag &= ~(VTBFREE|VAGE);
413 vp->v_flag |= VFREE;
414 if (vp->v_usecount)
415 panic("tobe free vnode isn't");
416 freevnodes++;
417 }
418
419 if (wantfreevnodes && freevnodes < wantfreevnodes) {
420 vp = NULL;
421 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
422 /*
423 * XXX: this is only here to be backwards compatible
424 */
425 vp = NULL;
426 } else {
427 for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
428 nvp = TAILQ_NEXT(vp, v_freelist);
429 if (!simple_lock_try(&vp->v_interlock))
430 continue;
431 if (vp->v_usecount)
432 panic("free vnode isn't");
433
434 object = vp->v_object;
435 if (object && (object->resident_page_count || object->ref_count)) {
436 printf("object inconsistant state: RPC: %d, RC: %d\n",
437 object->resident_page_count, object->ref_count);
438 /* Don't recycle if it's caching some pages */
439 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
440 TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
441 continue;
442 } else if (LIST_FIRST(&vp->v_cache_src)) {
443 /* Don't recycle if active in the namecache */
444 simple_unlock(&vp->v_interlock);
445 continue;
446 } else {
447 break;
448 }
449 }
450 }
451
452 for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
453 nvp = TAILQ_NEXT(tvp, v_freelist);
454 TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
455 TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
456 simple_unlock(&tvp->v_interlock);
457 }
458
459 if (vp) {
460 vp->v_flag |= VDOOMED;
461 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
462 freevnodes--;
463 simple_unlock(&vnode_free_list_slock);
464 cache_purge(vp);
465 vp->v_lease = NULL;
466 if (vp->v_type != VBAD) {
467 vgonel(vp, p);
468 } else {
469 simple_unlock(&vp->v_interlock);
470 }
471
472#ifdef INVARIANTS
473 {
474 int s;
475
476 if (vp->v_data)
477 panic("cleaned vnode isn't");
478 s = splbio();
479 if (vp->v_numoutput)
480 panic("Clean vnode has pending I/O's");
481 splx(s);
482 }
483#endif
484 vp->v_flag = 0;
485 vp->v_lastr = 0;
486 vp->v_lastw = 0;
487 vp->v_lasta = 0;
488 vp->v_cstart = 0;
489 vp->v_clen = 0;
490 vp->v_socket = 0;
491 vp->v_writecount = 0; /* XXX */
492 vp->v_maxio = 0;
493 } else {
494 simple_unlock(&vnode_free_list_slock);
495 vp = (struct vnode *) zalloc(vnode_zone);
496 bzero((char *) vp, sizeof *vp);
497 simple_lock_init(&vp->v_interlock);
498 vp->v_dd = vp;
499 cache_purge(vp);
500 LIST_INIT(&vp->v_cache_src);
501 TAILQ_INIT(&vp->v_cache_dst);
502 numvnodes++;
503 }
504
505 TAILQ_INIT(&vp->v_cleanblkhd);
506 TAILQ_INIT(&vp->v_dirtyblkhd);
507 vp->v_type = VNON;
508 vp->v_tag = tag;
509 vp->v_op = vops;
510 insmntque(vp, mp);
511 *vpp = vp;
512 vp->v_usecount = 1;
513 vp->v_data = 0;
514 splx(s);
515
516 vfs_object_create(vp, p, p->p_ucred);
517 return (0);
518}
519
520/*
521 * Move a vnode from one mount queue to another.
522 */
523static void
524insmntque(vp, mp)
525 register struct vnode *vp;
526 register struct mount *mp;
527{
528
529 simple_lock(&mntvnode_slock);
530 /*
531 * Delete from old mount point vnode list, if on one.
532 */
533 if (vp->v_mount != NULL)
534 LIST_REMOVE(vp, v_mntvnodes);
535 /*
536 * Insert into list of vnodes for the new mount point, if available.
537 */
538 if ((vp->v_mount = mp) == NULL) {
539 simple_unlock(&mntvnode_slock);
540 return;
541 }
542 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
543 simple_unlock(&mntvnode_slock);
544}
545
546/*
547 * Update outstanding I/O count and do wakeup if requested.
548 */
549void
550vwakeup(bp)
551 register struct buf *bp;
552{
553 register struct vnode *vp;
554
555 bp->b_flags &= ~B_WRITEINPROG;
556 if ((vp = bp->b_vp)) {
557 vp->v_numoutput--;
558 if (vp->v_numoutput < 0)
559 panic("vwakeup: neg numoutput");
560 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
561 vp->v_flag &= ~VBWAIT;
562 wakeup((caddr_t) &vp->v_numoutput);
563 }
564 }
565}
566
567/*
568 * Flush out and invalidate all buffers associated with a vnode.
569 * Called with the underlying object locked.
570 */
571int
572vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
573 register struct vnode *vp;
574 int flags;
575 struct ucred *cred;
576 struct proc *p;
577 int slpflag, slptimeo;
578{
579 register struct buf *bp;
580 struct buf *nbp, *blist;
581 int s, error;
582 vm_object_t object;
583
584 if (flags & V_SAVE) {
585 s = splbio();
586 while (vp->v_numoutput) {
587 vp->v_flag |= VBWAIT;
588 error = tsleep((caddr_t)&vp->v_numoutput,
589 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
590 if (error) {
591 splx(s);
592 return (error);
593 }
594 }
595 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
596 splx(s);
597 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
598 return (error);
599 s = splbio();
600 if (vp->v_numoutput > 0 ||
601 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
602 panic("vinvalbuf: dirty bufs");
603 }
604 splx(s);
605 }
606 s = splbio();
607 for (;;) {
608 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
609 if (!blist)
610 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
611 if (!blist)
612 break;
613
614 for (bp = blist; bp; bp = nbp) {
615 nbp = TAILQ_NEXT(bp, b_vnbufs);
616 if (bp->b_flags & B_BUSY) {
617 bp->b_flags |= B_WANTED;
618 error = tsleep((caddr_t) bp,
619 slpflag | (PRIBIO + 4), "vinvalbuf",
620 slptimeo);
621 if (error) {
622 splx(s);
623 return (error);
624 }
625 break;
626 }
627 /*
628 * XXX Since there are no node locks for NFS, I
629 * believe there is a slight chance that a delayed
630 * write will occur while sleeping just above, so
631 * check for it. Note that vfs_bio_awrite expects
632 * buffers to reside on a queue, while VOP_BWRITE and
633 * brelse do not.
634 */
635 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
636 (flags & V_SAVE)) {
637
638 if (bp->b_vp == vp) {
639 if (bp->b_flags & B_CLUSTEROK) {
640 vfs_bio_awrite(bp);
641 } else {
642 bremfree(bp);
643 bp->b_flags |= (B_BUSY | B_ASYNC);
644 VOP_BWRITE(bp);
645 }
646 } else {
647 bremfree(bp);
648 bp->b_flags |= B_BUSY;
649 (void) VOP_BWRITE(bp);
650 }
651 break;
652 }
653 bremfree(bp);
654 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF | B_BUSY);
655 bp->b_flags &= ~B_ASYNC;
656 brelse(bp);
657 }
658 }
659
660 while (vp->v_numoutput > 0) {
661 vp->v_flag |= VBWAIT;
662 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
663 }
664
665 splx(s);
666
667 /*
668 * Destroy the copy in the VM cache, too.
669 */
670 simple_lock(&vp->v_interlock);
671 object = vp->v_object;
672 if (object != NULL) {
673 vm_object_page_remove(object, 0, 0,
674 (flags & V_SAVE) ? TRUE : FALSE);
675 }
676 simple_unlock(&vp->v_interlock);
677
678 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
679 panic("vinvalbuf: flush failed");
680 return (0);
681}
682
683/*
684 * Truncate a file's buffer and pages to a specified length. This
685 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
686 * sync activity.
687 */
688int
689vtruncbuf(vp, cred, p, length, blksize)
690 register struct vnode *vp;
691 struct ucred *cred;
692 struct proc *p;
693 off_t length;
694 int blksize;
695{
696 register struct buf *bp;
697 struct buf *nbp;
698 int s, anyfreed;
699 int trunclbn;
700
701 /*
702 * Round up to the *next* lbn.
703 */
704 trunclbn = (length + blksize - 1) / blksize;
705
706 s = splbio();
707restart:
708 anyfreed = 1;
709 for (;anyfreed;) {
710 anyfreed = 0;
711 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
712 nbp = TAILQ_NEXT(bp, b_vnbufs);
713 if (bp->b_lblkno >= trunclbn) {
714 if (bp->b_flags & B_BUSY) {
715 bp->b_flags |= B_WANTED;
716 tsleep(bp, PRIBIO + 4, "vtrb1", 0);
717 goto restart;
718 } else {
719 bremfree(bp);
720 bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
721 bp->b_flags &= ~B_ASYNC;
722 brelse(bp);
723 anyfreed = 1;
724 }
725 if (nbp && (((nbp->b_xflags & B_VNCLEAN) == 0)||
726 (nbp->b_vp != vp) ||
727 (nbp->b_flags & B_DELWRI))) {
728 goto restart;
729 }
730 }
731 }
732
733 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
734 nbp = TAILQ_NEXT(bp, b_vnbufs);
735 if (bp->b_lblkno >= trunclbn) {
736 if (bp->b_flags & B_BUSY) {
737 bp->b_flags |= B_WANTED;
738 tsleep(bp, PRIBIO + 4, "vtrb2", 0);
739 goto restart;
740 } else {
741 bremfree(bp);
742 bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
743 bp->b_flags &= ~B_ASYNC;
744 brelse(bp);
745 anyfreed = 1;
746 }
747 if (nbp && (((nbp->b_xflags & B_VNDIRTY) == 0)||
748 (nbp->b_vp != vp) ||
749 (nbp->b_flags & B_DELWRI) == 0)) {
750 goto restart;
751 }
752 }
753 }
754 }
755
756 if (length > 0) {
757restartsync:
758 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
759 nbp = TAILQ_NEXT(bp, b_vnbufs);
760 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
761 if (bp->b_flags & B_BUSY) {
762 bp->b_flags |= B_WANTED;
763 tsleep(bp, PRIBIO, "vtrb3", 0);
764 } else {
765 bremfree(bp);
766 bp->b_flags |= B_BUSY;
767 if (bp->b_vp == vp) {
768 bp->b_flags |= B_ASYNC;
769 } else {
770 bp->b_flags &= ~B_ASYNC;
771 }
772 VOP_BWRITE(bp);
773 }
774 goto restartsync;
775 }
776
777 }
778 }
779
780 while (vp->v_numoutput > 0) {
781 vp->v_flag |= VBWAIT;
782 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
783 }
784
785 splx(s);
786
787 vnode_pager_setsize(vp, length);
788
789 return (0);
790}
791
792/*
793 * Associate a buffer with a vnode.
794 */
795void
796bgetvp(vp, bp)
797 register struct vnode *vp;
798 register struct buf *bp;
799{
800 int s;
801
802 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
803
804 vhold(vp);
805 bp->b_vp = vp;
806 if (vp->v_type == VBLK || vp->v_type == VCHR)
807 bp->b_dev = vp->v_rdev;
808 else
809 bp->b_dev = NODEV;
810 /*
811 * Insert onto list for new vnode.
812 */
813 s = splbio();
814 bp->b_xflags |= B_VNCLEAN;
815 bp->b_xflags &= ~B_VNDIRTY;
816 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
817 splx(s);
818}
819
820/*
821 * Disassociate a buffer from a vnode.
822 */
823void
824brelvp(bp)
825 register struct buf *bp;
826{
827 struct vnode *vp;
828 struct buflists *listheadp;
829 int s;
830
831 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
832
833 /*
834 * Delete from old vnode list, if on one.
835 */
836 vp = bp->b_vp;
837 s = splbio();
838 if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
839 if (bp->b_xflags & B_VNDIRTY)
840 listheadp = &vp->v_dirtyblkhd;
841 else
842 listheadp = &vp->v_cleanblkhd;
843 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
844 bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
845 }
846 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
847 vp->v_flag &= ~VONWORKLST;
848 LIST_REMOVE(vp, v_synclist);
849 }
850 splx(s);
851 bp->b_vp = (struct vnode *) 0;
852 vdrop(vp);
853}
854
855/*
856 * The workitem queue.
857 *
858 * It is useful to delay writes of file data and filesystem metadata
859 * for tens of seconds so that quickly created and deleted files need
860 * not waste disk bandwidth being created and removed. To realize this,
861 * we append vnodes to a "workitem" queue. When running with a soft
862 * updates implementation, most pending metadata dependencies should
863 * not wait for more than a few seconds. Thus, mounted on block devices
864 * are delayed only about a half the time that file data is delayed.
865 * Similarly, directory updates are more critical, so are only delayed
866 * about a third the time that file data is delayed. Thus, there are
867 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
868 * one each second (driven off the filesystem syner process). The
869 * syncer_delayno variable indicates the next queue that is to be processed.
870 * Items that need to be processed soon are placed in this queue:
871 *
872 * syncer_workitem_pending[syncer_delayno]
873 *
874 * A delay of fifteen seconds is done by placing the request fifteen
875 * entries later in the queue:
876 *
877 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
878 *
879 */
880
881/*
882 * Add an item to the syncer work queue.
883 */
884static void
885vn_syncer_add_to_worklist(struct vnode *vp, int delay)
886{
887 int s, slot;
888
889 s = splbio();
890
891 if (vp->v_flag & VONWORKLST) {
892 LIST_REMOVE(vp, v_synclist);
893 }
894
895 if (delay > syncer_maxdelay - 2)
896 delay = syncer_maxdelay - 2;
897 slot = (syncer_delayno + delay) & syncer_mask;
898
899 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
900 vp->v_flag |= VONWORKLST;
901 splx(s);
902}
903
904struct proc *updateproc;
905static void sched_sync __P((void));
906static const struct kproc_desc up_kp = {
907 "syncer",
908 sched_sync,
909 &updateproc
910};
911SYSINIT_KT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
912
913/*
914 * System filesystem synchronizer daemon.
915 */
916void
917sched_sync(void)
918{
919 struct synclist *slp;
920 struct vnode *vp;
921 long starttime;
922 int s;
923 struct proc *p = updateproc;
924
925 for (;;) {
926 starttime = time_second;
927
928 /*
929 * Push files whose dirty time has expired. Be careful
930 * of interrupt race on slp queue.
931 */
932 s = splbio();
933 slp = &syncer_workitem_pending[syncer_delayno];
934 syncer_delayno += 1;
935 if (syncer_delayno == syncer_maxdelay)
936 syncer_delayno = 0;
937 splx(s);
938
939 while ((vp = LIST_FIRST(slp)) != NULL) {
940 if (VOP_ISLOCKED(vp) == 0) {
941 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
942 (void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
943 VOP_UNLOCK(vp, 0, p);
944 }
945 s = splbio();
946 if (LIST_FIRST(slp) == vp) {
947 /*
948 * Note: v_tag VT_VFS vps can remain on the
949 * worklist too with no dirty blocks, but
950 * since sync_fsync() moves it to a different
951 * slot we are safe.
952 */
953 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
954 vp->v_type != VBLK)
955 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
956 /*
957 * Put us back on the worklist. The worklist
958 * routine will remove us from our current
959 * position and then add us back in at a later
960 * position.
961 */
962 vn_syncer_add_to_worklist(vp, syncdelay);
963 }
964 splx(s);
965 }
966
967 /*
968 * Do soft update processing.
969 */
970 if (bioops.io_sync)
971 (*bioops.io_sync)(NULL);
972
973 /*
974 * The variable rushjob allows the kernel to speed up the
975 * processing of the filesystem syncer process. A rushjob
976 * value of N tells the filesystem syncer to process the next
977 * N seconds worth of work on its queue ASAP. Currently rushjob
978 * is used by the soft update code to speed up the filesystem
979 * syncer process when the incore state is getting so far
980 * ahead of the disk that the kernel memory pool is being
981 * threatened with exhaustion.
982 */
983 if (rushjob > 0) {
984 rushjob -= 1;
985 continue;
986 }
987 /*
988 * If it has taken us less than a second to process the
989 * current work, then wait. Otherwise start right over
990 * again. We can still lose time if any single round
991 * takes more than two seconds, but it does not really
992 * matter as we are just trying to generally pace the
993 * filesystem activity.
994 */
995 if (time_second == starttime)
996 tsleep(&lbolt, PPAUSE, "syncer", 0);
997 }
998}
999
1000/*
1001 * Associate a p-buffer with a vnode.
1002 *
1003 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1004 * with the buffer. i.e. the bp has not been linked into the vnode or
1005 * ref-counted.
1006 */
1007void
1008pbgetvp(vp, bp)
1009 register struct vnode *vp;
1010 register struct buf *bp;
1011{
1012
1013 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1014
1015 bp->b_vp = vp;
1016 bp->b_flags |= B_PAGING;
1017 if (vp->v_type == VBLK || vp->v_type == VCHR)
1018 bp->b_dev = vp->v_rdev;
1019 else
1020 bp->b_dev = NODEV;
1021}
1022
1023/*
1024 * Disassociate a p-buffer from a vnode.
1025 */
1026void
1027pbrelvp(bp)
1028 register struct buf *bp;
1029{
1030
1031 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1032
1033#if !defined(MAX_PERF)
1034 /* XXX REMOVE ME */
1035 if (bp->b_vnbufs.tqe_next != NULL) {
1036 panic(
1037 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1038 bp,
1039 (int)bp->b_flags
1040 );
1041 }
1042#endif
1043 bp->b_vp = (struct vnode *) 0;
1044 bp->b_flags &= ~B_PAGING;
1045}
1046
1047void
1048pbreassignbuf(bp, newvp)
1049 struct buf *bp;
1050 struct vnode *newvp;
1051{
1052#if !defined(MAX_PERF)
1053 if ((bp->b_flags & B_PAGING) == 0) {
1054 panic(
1055 "pbreassignbuf() on non phys bp %p",
1056 bp
1057 );
1058 }
1059#endif
1060 bp->b_vp = newvp;
1061}
1062
1063/*
1064 * Reassign a buffer from one vnode to another.
1065 * Used to assign file specific control information
1066 * (indirect blocks) to the vnode to which they belong.
1067 */
1068void
1069reassignbuf(bp, newvp)
1070 register struct buf *bp;
1071 register struct vnode *newvp;
1072{
1073 struct buflists *listheadp;
1074 int delay;
1075 int s;
1076
1077 if (newvp == NULL) {
1078 printf("reassignbuf: NULL");
1079 return;
1080 }
1081
1082#if !defined(MAX_PERF)
1083 /*
1084 * B_PAGING flagged buffers cannot be reassigned because their vp
1085 * is not fully linked in.
1086 */
1087 if (bp->b_flags & B_PAGING)
1088 panic("cannot reassign paging buffer");
1089#endif
1090
1091 s = splbio();
1092 /*
1093 * Delete from old vnode list, if on one.
1094 */
1095 if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
1096 if (bp->b_xflags & B_VNDIRTY)
1097 listheadp = &bp->b_vp->v_dirtyblkhd;
1098 else
1099 listheadp = &bp->b_vp->v_cleanblkhd;
1100 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1101 bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
1102 if (bp->b_vp != newvp) {
1103 vdrop(bp->b_vp);
1104 bp->b_vp = NULL; /* for clarification */
1105 }
1106 }
1107 /*
1108 * If dirty, put on list of dirty buffers; otherwise insert onto list
1109 * of clean buffers.
1110 */
1111 if (bp->b_flags & B_DELWRI) {
1112 struct buf *tbp;
1113
1114 listheadp = &newvp->v_dirtyblkhd;
1115 if ((newvp->v_flag & VONWORKLST) == 0) {
1116 switch (newvp->v_type) {
1117 case VDIR:
1118 delay = syncdelay / 3;
1119 break;
1120 case VBLK:
1121 if (newvp->v_specmountpoint != NULL) {
1122 delay = syncdelay / 2;
1123 break;
1124 }
1125 /* fall through */
1126 default:
1127 delay = syncdelay;
1128 }
1129 vn_syncer_add_to_worklist(newvp, delay);
1130 }
1131 bp->b_xflags |= B_VNDIRTY;
1132 tbp = TAILQ_FIRST(listheadp);
1133 if (tbp == NULL ||
1134 (bp->b_lblkno >= 0 && tbp->b_lblkno > bp->b_lblkno)) {
1135 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1136 } else {
1137 if (bp->b_lblkno >= 0) {
1138 struct buf *ttbp;
1139 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1140 (ttbp->b_lblkno < bp->b_lblkno)) {
1141 tbp = ttbp;
1142 }
1143 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1144 } else {
1145 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1146 }
1147 }
1148 } else {
1149 bp->b_xflags |= B_VNCLEAN;
1150 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1151 if ((newvp->v_flag & VONWORKLST) &&
1152 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1153 newvp->v_flag &= ~VONWORKLST;
1154 LIST_REMOVE(newvp, v_synclist);
1155 }
1156 }
1157 if (bp->b_vp != newvp) {
1158 bp->b_vp = newvp;
1159 vhold(bp->b_vp);
1160 }
1161 splx(s);
1162}
1163
1164/*
1165 * Create a vnode for a block device.
1166 * Used for mounting the root file system.
1167 */
1168int
1169bdevvp(dev, vpp)
1170 dev_t dev;
1171 struct vnode **vpp;
1172{
1173 register struct vnode *vp;
1174 struct vnode *nvp;
1175 int error;
1176
1177 /* XXX 255 is for mfs. */
1178 if (dev == NODEV || (major(dev) != 255 && (major(dev) >= nblkdev ||
1179 bdevsw[major(dev)] == NULL))) {
1180 *vpp = NULLVP;
1181 return (ENXIO);
1182 }
1183 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1184 if (error) {
1185 *vpp = NULLVP;
1186 return (error);
1187 }
1188 vp = nvp;
1189 vp->v_type = VBLK;
1190 if ((nvp = checkalias(vp, dev, (struct mount *)0)) != NULL) {
1191 vput(vp);
1192 vp = nvp;
1193 }
1194 *vpp = vp;
1195 return (0);
1196}
1197
1198/*
1199 * Check to see if the new vnode represents a special device
1200 * for which we already have a vnode (either because of
1201 * bdevvp() or because of a different vnode representing
1202 * the same block device). If such an alias exists, deallocate
1203 * the existing contents and return the aliased vnode. The
1204 * caller is responsible for filling it with its new contents.
1205 */
1206struct vnode *
1207checkalias(nvp, nvp_rdev, mp)
1208 register struct vnode *nvp;
1209 dev_t nvp_rdev;
1210 struct mount *mp;
1211{
1212 struct proc *p = curproc; /* XXX */
1213 struct vnode *vp;
1214 struct vnode **vpp;
1215 int rmaj = major(nvp_rdev);
1216
1217 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1218 return (NULLVP);
1219
1220 vpp = &speclisth[SPECHASH(nvp_rdev)];
1221loop:
1222 simple_lock(&spechash_slock);
1223 for (vp = *vpp; vp; vp = vp->v_specnext) {
1224 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1225 continue;
1226 /*
1227 * Alias, but not in use, so flush it out.
1228 * Only alias active device nodes.
1229 * Not sure why we don't re-use this like we do below.
1230 */
1231 simple_lock(&vp->v_interlock);
1232 if (vp->v_usecount == 0) {
1233 simple_unlock(&spechash_slock);
1234 vgonel(vp, p);
1235 goto loop;
1236 }
1237 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) {
1238 /*
1239 * It dissappeared, and we may have slept.
1240 * Restart from the beginning
1241 */
1242 simple_unlock(&spechash_slock);
1243 goto loop;
1244 }
1245 break;
1246 }
1247 /*
1248 * It would be a lot clearer what is going on here if
1249 * this had been expressed as:
1250 * if ( vp && (vp->v_tag == VT_NULL))
1251 * and the clauses had been swapped.
1252 */
1253 if (vp == NULL || vp->v_tag != VT_NON) {
1254 struct specinfo *sinfo;
1255
1256 /*
1257 * Put the new vnode into the hash chain.
1258 * and if there was an alias, connect them.
1259 */
1260 MALLOC(sinfo, struct specinfo *,
1261 sizeof(struct specinfo), M_VNODE, M_WAITOK);
1262 bzero(sinfo, sizeof(struct specinfo));
1263 nvp->v_specinfo = sinfo;
1264 sinfo->si_rdev = nvp_rdev;
1265 sinfo->si_hashchain = vpp;
1266 sinfo->si_specnext = *vpp;
1267 sinfo->si_bsize_phys = DEV_BSIZE;
1268 sinfo->si_bsize_best = BLKDEV_IOSIZE;
1269 sinfo->si_bsize_max = MAXBSIZE;
1270
1271 /*
1272 * Ask the device to fix up specinfo. Typically the
1273 * si_bsize_* parameters may need fixing up.
1274 */
1275
1276 if (nvp->v_type == VBLK && rmaj < nblkdev) {
1277 if (bdevsw[rmaj] && bdevsw[rmaj]->d_parms)
1278
1279 (*bdevsw[rmaj]->d_parms)(nvp_rdev, sinfo, DPARM_GET);
1280 } else if (nvp->v_type == VCHR && rmaj < nchrdev) {
1281 if (cdevsw[rmaj] && cdevsw[rmaj]->d_parms)
1282 (*cdevsw[rmaj]->d_parms)(nvp_rdev, sinfo, DPARM_GET);
1283 }
1284
1285 simple_unlock(&spechash_slock);
1286 *vpp = nvp;
1287 if (vp != NULLVP) {
1288 nvp->v_flag |= VALIASED;
1289 vp->v_flag |= VALIASED;
1290 vput(vp);
1291 }
1292 return (NULLVP);
1293 }
1294 /*
1295 * if ( vp && (vp->v_tag == VT_NULL))
1296 * We have a vnode alias, but it is a trashed.
1297 * Make it look like it's newley allocated. (by getnewvnode())
1298 * The caller should use this instead.
1299 */
1300 simple_unlock(&spechash_slock);
1301 VOP_UNLOCK(vp, 0, p);
1302 simple_lock(&vp->v_interlock);
1303 vclean(vp, 0, p);
1304 vp->v_op = nvp->v_op;
1305 vp->v_tag = nvp->v_tag;
1306 nvp->v_type = VNON;
1307 insmntque(vp, mp);
1308 return (vp);
1309}
1310
1311/*
1312 * Grab a particular vnode from the free list, increment its
1313 * reference count and lock it. The vnode lock bit is set the
1314 * vnode is being eliminated in vgone. The process is awakened
1315 * when the transition is completed, and an error returned to
1316 * indicate that the vnode is no longer usable (possibly having
1317 * been changed to a new file system type).
1318 */
1319int
1320vget(vp, flags, p)
1321 register struct vnode *vp;
1322 int flags;
1323 struct proc *p;
1324{
1325 int error;
1326
1327 /*
1328 * If the vnode is in the process of being cleaned out for
1329 * another use, we wait for the cleaning to finish and then
1330 * return failure. Cleaning is determined by checking that
1331 * the VXLOCK flag is set.
1332 */
1333 if ((flags & LK_INTERLOCK) == 0) {
1334 simple_lock(&vp->v_interlock);
1335 }
1336 if (vp->v_flag & VXLOCK) {
1337 vp->v_flag |= VXWANT;
1338 simple_unlock(&vp->v_interlock);
1339 tsleep((caddr_t)vp, PINOD, "vget", 0);
1340 return (ENOENT);
1341 }
1342
1343 vp->v_usecount++;
1344
1345 if (VSHOULDBUSY(vp))
1346 vbusy(vp);
1347 if (flags & LK_TYPE_MASK) {
1348 if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1349 /*
1350 * must expand vrele here because we do not want
1351 * to call VOP_INACTIVE if the reference count
1352 * drops back to zero since it was never really
1353 * active. We must remove it from the free list
1354 * before sleeping so that multiple processes do
1355 * not try to recycle it.
1356 */
1357 simple_lock(&vp->v_interlock);
1358 vp->v_usecount--;
1359 if (VSHOULDFREE(vp))
1360 vfree(vp);
1361 simple_unlock(&vp->v_interlock);
1362 }
1363 return (error);
1364 }
1365 simple_unlock(&vp->v_interlock);
1366 return (0);
1367}
1368
1369void
1370vref(struct vnode *vp)
1371{
1372 simple_lock(&vp->v_interlock);
1373 vp->v_usecount++;
1374 simple_unlock(&vp->v_interlock);
1375}
1376
1377/*
1378 * Vnode put/release.
1379 * If count drops to zero, call inactive routine and return to freelist.
1380 */
1381void
1382vrele(vp)
1383 struct vnode *vp;
1384{
1385 struct proc *p = curproc; /* XXX */
1386
1387 KASSERT(vp != NULL, ("vrele: null vp"));
1388
1389 simple_lock(&vp->v_interlock);
1390
1391 if (vp->v_usecount > 1) {
1392
1393 vp->v_usecount--;
1394 simple_unlock(&vp->v_interlock);
1395
1396 return;
1397 }
1398
1399 if (vp->v_usecount == 1) {
1400
1401 vp->v_usecount--;
1402 if (VSHOULDFREE(vp))
1403 vfree(vp);
1404 /*
1405 * If we are doing a vput, the node is already locked, and we must
1406 * call VOP_INACTIVE with the node locked. So, in the case of
1407 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1408 */
1409 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1410 VOP_INACTIVE(vp, p);
1411 }
1412
1413 } else {
1414#ifdef DIAGNOSTIC
1415 vprint("vrele: negative ref count", vp);
1416 simple_unlock(&vp->v_interlock);
1417#endif
1418 panic("vrele: negative ref cnt");
1419 }
1420}
1421
1422void
1423vput(vp)
1424 struct vnode *vp;
1425{
1426 struct proc *p = curproc; /* XXX */
1427
1428 KASSERT(vp != NULL, ("vput: null vp"));
1429
1430 simple_lock(&vp->v_interlock);
1431
1432 if (vp->v_usecount > 1) {
1433
1434 vp->v_usecount--;
1435 VOP_UNLOCK(vp, LK_INTERLOCK, p);
1436 return;
1437
1438 }
1439
1440 if (vp->v_usecount == 1) {
1441
1442 vp->v_usecount--;
1443 if (VSHOULDFREE(vp))
1444 vfree(vp);
1445 /*
1446 * If we are doing a vput, the node is already locked, and we must
1447 * call VOP_INACTIVE with the node locked. So, in the case of
1448 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1449 */
1450 simple_unlock(&vp->v_interlock);
1451 VOP_INACTIVE(vp, p);
1452
1453 } else {
1454#ifdef DIAGNOSTIC
1455 vprint("vput: negative ref count", vp);
1456#endif
1457 panic("vput: negative ref cnt");
1458 }
1459}
1460
1461/*
1462 * Somebody doesn't want the vnode recycled.
1463 */
1464void
1465vhold(vp)
1466 register struct vnode *vp;
1467{
1468 int s;
1469
1470 s = splbio();
1471 vp->v_holdcnt++;
1472 if (VSHOULDBUSY(vp))
1473 vbusy(vp);
1474 splx(s);
1475}
1476
1477/*
1478 * One less who cares about this vnode.
1479 */
1480void
1481vdrop(vp)
1482 register struct vnode *vp;
1483{
1484 int s;
1485
1486 s = splbio();
1487 if (vp->v_holdcnt <= 0)
1488 panic("vdrop: holdcnt");
1489 vp->v_holdcnt--;
1490 if (VSHOULDFREE(vp))
1491 vfree(vp);
1492 splx(s);
1493}
1494
1495/*
1496 * Remove any vnodes in the vnode table belonging to mount point mp.
1497 *
1498 * If MNT_NOFORCE is specified, there should not be any active ones,
1499 * return error if any are found (nb: this is a user error, not a
1500 * system error). If MNT_FORCE is specified, detach any active vnodes
1501 * that are found.
1502 */
1503#ifdef DIAGNOSTIC
1504static int busyprt = 0; /* print out busy vnodes */
1505SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1506#endif
1507
1508int
1509vflush(mp, skipvp, flags)
1510 struct mount *mp;
1511 struct vnode *skipvp;
1512 int flags;
1513{
1514 struct proc *p = curproc; /* XXX */
1515 struct vnode *vp, *nvp;
1516 int busy = 0;
1517
1518 simple_lock(&mntvnode_slock);
1519loop:
1520 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
1521 /*
1522 * Make sure this vnode wasn't reclaimed in getnewvnode().
1523 * Start over if it has (it won't be on the list anymore).
1524 */
1525 if (vp->v_mount != mp)
1526 goto loop;
1527 nvp = vp->v_mntvnodes.le_next;
1528 /*
1529 * Skip over a selected vnode.
1530 */
1531 if (vp == skipvp)
1532 continue;
1533
1534 simple_lock(&vp->v_interlock);
1535 /*
1536 * Skip over a vnodes marked VSYSTEM.
1537 */
1538 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1539 simple_unlock(&vp->v_interlock);
1540 continue;
1541 }
1542 /*
1543 * If WRITECLOSE is set, only flush out regular file vnodes
1544 * open for writing.
1545 */
1546 if ((flags & WRITECLOSE) &&
1547 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1548 simple_unlock(&vp->v_interlock);
1549 continue;
1550 }
1551
1552 /*
1553 * With v_usecount == 0, all we need to do is clear out the
1554 * vnode data structures and we are done.
1555 */
1556 if (vp->v_usecount == 0) {
1557 simple_unlock(&mntvnode_slock);
1558 vgonel(vp, p);
1559 simple_lock(&mntvnode_slock);
1560 continue;
1561 }
1562
1563 /*
1564 * If FORCECLOSE is set, forcibly close the vnode. For block
1565 * or character devices, revert to an anonymous device. For
1566 * all other files, just kill them.
1567 */
1568 if (flags & FORCECLOSE) {
1569 simple_unlock(&mntvnode_slock);
1570 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1571 vgonel(vp, p);
1572 } else {
1573 vclean(vp, 0, p);
1574 vp->v_op = spec_vnodeop_p;
1575 insmntque(vp, (struct mount *) 0);
1576 }
1577 simple_lock(&mntvnode_slock);
1578 continue;
1579 }
1580#ifdef DIAGNOSTIC
1581 if (busyprt)
1582 vprint("vflush: busy vnode", vp);
1583#endif
1584 simple_unlock(&vp->v_interlock);
1585 busy++;
1586 }
1587 simple_unlock(&mntvnode_slock);
1588 if (busy)
1589 return (EBUSY);
1590 return (0);
1591}
1592
1593/*
1594 * Disassociate the underlying file system from a vnode.
1595 */
1596static void
1597vclean(vp, flags, p)
1598 struct vnode *vp;
1599 int flags;
1600 struct proc *p;
1601{
1602 int active;
1603 vm_object_t obj;
1604
1605 /*
1606 * Check to see if the vnode is in use. If so we have to reference it
1607 * before we clean it out so that its count cannot fall to zero and
1608 * generate a race against ourselves to recycle it.
1609 */
1610 if ((active = vp->v_usecount))
1611 vp->v_usecount++;
1612
1613 /*
1614 * Prevent the vnode from being recycled or brought into use while we
1615 * clean it out.
1616 */
1617 if (vp->v_flag & VXLOCK)
1618 panic("vclean: deadlock");
1619 vp->v_flag |= VXLOCK;
1620 /*
1621 * Even if the count is zero, the VOP_INACTIVE routine may still
1622 * have the object locked while it cleans it out. The VOP_LOCK
1623 * ensures that the VOP_INACTIVE routine is done with its work.
1624 * For active vnodes, it ensures that no other activity can
1625 * occur while the underlying object is being cleaned out.
1626 */
1627 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1628
1629 /*
1630 * Clean out any buffers associated with the vnode.
1631 */
1632 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1633 if ((obj = vp->v_object) != NULL) {
1634 if (obj->ref_count == 0) {
1635 /*
1636 * This is a normal way of shutting down the object/vnode
1637 * association.
1638 */
1639 vm_object_terminate(obj);
1640 } else {
1641 /*
1642 * Woe to the process that tries to page now :-).
1643 */
1644 vm_pager_deallocate(obj);
1645 }
1646 }
1647
1648 /*
1649 * If purging an active vnode, it must be closed and
1650 * deactivated before being reclaimed. Note that the
1651 * VOP_INACTIVE will unlock the vnode.
1652 */
1653 if (active) {
1654 if (flags & DOCLOSE)
1655 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1656 VOP_INACTIVE(vp, p);
1657 } else {
1658 /*
1659 * Any other processes trying to obtain this lock must first
1660 * wait for VXLOCK to clear, then call the new lock operation.
1661 */
1662 VOP_UNLOCK(vp, 0, p);
1663 }
1664 /*
1665 * Reclaim the vnode.
1666 */
1667 if (VOP_RECLAIM(vp, p))
1668 panic("vclean: cannot reclaim");
1669
1670 if (active)
1671 vrele(vp);
1672
1673 cache_purge(vp);
1674 if (vp->v_vnlock) {
1675#if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */
1676#ifdef DIAGNOSTIC
1677 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0)
1678 vprint("vclean: lock not drained", vp);
1679#endif
1680#endif
1681 FREE(vp->v_vnlock, M_VNODE);
1682 vp->v_vnlock = NULL;
1683 }
1684
1685 if (VSHOULDFREE(vp))
1686 vfree(vp);
1687
1688 /*
1689 * Done with purge, notify sleepers of the grim news.
1690 */
1691 vp->v_op = dead_vnodeop_p;
1692 vn_pollgone(vp);
1693 vp->v_tag = VT_NON;
1694 vp->v_flag &= ~VXLOCK;
1695 if (vp->v_flag & VXWANT) {
1696 vp->v_flag &= ~VXWANT;
1697 wakeup((caddr_t) vp);
1698 }
1699}
1700
1701/*
1702 * Eliminate all activity associated with the requested vnode
1703 * and with all vnodes aliased to the requested vnode.
1704 */
1705int
1706vop_revoke(ap)
1707 struct vop_revoke_args /* {
1708 struct vnode *a_vp;
1709 int a_flags;
1710 } */ *ap;
1711{
1712 struct vnode *vp, *vq;
1713 struct proc *p = curproc; /* XXX */
1714
1715 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1716
1717 vp = ap->a_vp;
1718 simple_lock(&vp->v_interlock);
1719
1720 if (vp->v_flag & VALIASED) {
1721 /*
1722 * If a vgone (or vclean) is already in progress,
1723 * wait until it is done and return.
1724 */
1725 if (vp->v_flag & VXLOCK) {
1726 vp->v_flag |= VXWANT;
1727 simple_unlock(&vp->v_interlock);
1728 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1729 return (0);
1730 }
1731 /*
1732 * Ensure that vp will not be vgone'd while we
1733 * are eliminating its aliases.
1734 */
1735 vp->v_flag |= VXLOCK;
1736 simple_unlock(&vp->v_interlock);
1737 while (vp->v_flag & VALIASED) {
1738 simple_lock(&spechash_slock);
1739 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1740 if (vq->v_rdev != vp->v_rdev ||
1741 vq->v_type != vp->v_type || vp == vq)
1742 continue;
1743 simple_unlock(&spechash_slock);
1744 vgone(vq);
1745 break;
1746 }
1747 if (vq == NULLVP) {
1748 simple_unlock(&spechash_slock);
1749 }
1750 }
1751 /*
1752 * Remove the lock so that vgone below will
1753 * really eliminate the vnode after which time
1754 * vgone will awaken any sleepers.
1755 */
1756 simple_lock(&vp->v_interlock);
1757 vp->v_flag &= ~VXLOCK;
1758 if (vp->v_flag & VXWANT) {
1759 vp->v_flag &= ~VXWANT;
1760 wakeup(vp);
1761 }
1762 }
1763 vgonel(vp, p);
1764 return (0);
1765}
1766
1767/*
1768 * Recycle an unused vnode to the front of the free list.
1769 * Release the passed interlock if the vnode will be recycled.
1770 */
1771int
1772vrecycle(vp, inter_lkp, p)
1773 struct vnode *vp;
1774 struct simplelock *inter_lkp;
1775 struct proc *p;
1776{
1777
1778 simple_lock(&vp->v_interlock);
1779 if (vp->v_usecount == 0) {
1780 if (inter_lkp) {
1781 simple_unlock(inter_lkp);
1782 }
1783 vgonel(vp, p);
1784 return (1);
1785 }
1786 simple_unlock(&vp->v_interlock);
1787 return (0);
1788}
1789
1790/*
1791 * Eliminate all activity associated with a vnode
1792 * in preparation for reuse.
1793 */
1794void
1795vgone(vp)
1796 register struct vnode *vp;
1797{
1798 struct proc *p = curproc; /* XXX */
1799
1800 simple_lock(&vp->v_interlock);
1801 vgonel(vp, p);
1802}
1803
1804/*
1805 * vgone, with the vp interlock held.
1806 */
1807static void
1808vgonel(vp, p)
1809 struct vnode *vp;
1810 struct proc *p;
1811{
1812 int s;
1813 struct vnode *vq;
1814 struct vnode *vx;
1815
1816 /*
1817 * If a vgone (or vclean) is already in progress,
1818 * wait until it is done and return.
1819 */
1820 if (vp->v_flag & VXLOCK) {
1821 vp->v_flag |= VXWANT;
1822 simple_unlock(&vp->v_interlock);
1823 tsleep((caddr_t)vp, PINOD, "vgone", 0);
1824 return;
1825 }
1826
1827 /*
1828 * Clean out the filesystem specific data.
1829 */
1830 vclean(vp, DOCLOSE, p);
1831 simple_lock(&vp->v_interlock);
1832
1833 /*
1834 * Delete from old mount point vnode list, if on one.
1835 */
1836 if (vp->v_mount != NULL)
1837 insmntque(vp, (struct mount *)0);
1838 /*
1839 * If special device, remove it from special device alias list
1840 * if it is on one.
1841 */
1842 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1843 simple_lock(&spechash_slock);
1844 if (*vp->v_hashchain == vp) {
1845 *vp->v_hashchain = vp->v_specnext;
1846 } else {
1847 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1848 if (vq->v_specnext != vp)
1849 continue;
1850 vq->v_specnext = vp->v_specnext;
1851 break;
1852 }
1853 if (vq == NULL)
1854 panic("missing bdev");
1855 }
1856 if (vp->v_flag & VALIASED) {
1857 vx = NULL;
1858 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1859 if (vq->v_rdev != vp->v_rdev ||
1860 vq->v_type != vp->v_type)
1861 continue;
1862 if (vx)
1863 break;
1864 vx = vq;
1865 }
1866 if (vx == NULL)
1867 panic("missing alias");
1868 if (vq == NULL)
1869 vx->v_flag &= ~VALIASED;
1870 vp->v_flag &= ~VALIASED;
1871 }
1872 simple_unlock(&spechash_slock);
1873 FREE(vp->v_specinfo, M_VNODE);
1874 vp->v_specinfo = NULL;
1875 }
1876
1877 /*
1878 * If it is on the freelist and not already at the head,
1879 * move it to the head of the list. The test of the back
1880 * pointer and the reference count of zero is because
1881 * it will be removed from the free list by getnewvnode,
1882 * but will not have its reference count incremented until
1883 * after calling vgone. If the reference count were
1884 * incremented first, vgone would (incorrectly) try to
1885 * close the previous instance of the underlying object.
1886 */
1887 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1888 s = splbio();
1889 simple_lock(&vnode_free_list_slock);
1890 if (vp->v_flag & VFREE) {
1891 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1892 } else if (vp->v_flag & VTBFREE) {
1893 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1894 vp->v_flag &= ~VTBFREE;
1895 freevnodes++;
1896 } else
1897 freevnodes++;
1898 vp->v_flag |= VFREE;
1899 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1900 simple_unlock(&vnode_free_list_slock);
1901 splx(s);
1902 }
1903
1904 vp->v_type = VBAD;
1905 simple_unlock(&vp->v_interlock);
1906}
1907
1908/*
1909 * Lookup a vnode by device number.
1910 */
1911int
1912vfinddev(dev, type, vpp)
1913 dev_t dev;
1914 enum vtype type;
1915 struct vnode **vpp;
1916{
1917 register struct vnode *vp;
1918 int rc = 0;
1919
1920 simple_lock(&spechash_slock);
1921 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1922 if (dev != vp->v_rdev || type != vp->v_type)
1923 continue;
1924 *vpp = vp;
1925 rc = 1;
1926 break;
1927 }
1928 simple_unlock(&spechash_slock);
1929 return (rc);
1930}
1931
1932/*
1933 * Calculate the total number of references to a special device.
1934 */
1935int
1936vcount(vp)
1937 register struct vnode *vp;
1938{
1939 struct vnode *vq, *vnext;
1940 int count;
1941
1942loop:
1943 if ((vp->v_flag & VALIASED) == 0)
1944 return (vp->v_usecount);
1945 simple_lock(&spechash_slock);
1946 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1947 vnext = vq->v_specnext;
1948 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1949 continue;
1950 /*
1951 * Alias, but not in use, so flush it out.
1952 */
1953 if (vq->v_usecount == 0 && vq != vp) {
1954 simple_unlock(&spechash_slock);
1955 vgone(vq);
1956 goto loop;
1957 }
1958 count += vq->v_usecount;
1959 }
1960 simple_unlock(&spechash_slock);
1961 return (count);
1962}
1963/*
1964 * Print out a description of a vnode.
1965 */
1966static char *typename[] =
1967{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1968
1969void
1970vprint(label, vp)
1971 char *label;
1972 register struct vnode *vp;
1973{
1974 char buf[96];
1975
1976 if (label != NULL)
1977 printf("%s: %p: ", label, (void *)vp);
1978 else
1979 printf("%p: ", (void *)vp);
1980 printf("type %s, usecount %d, writecount %d, refcount %d,",
1981 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1982 vp->v_holdcnt);
1983 buf[0] = '\0';
1984 if (vp->v_flag & VROOT)
1985 strcat(buf, "|VROOT");
1986 if (vp->v_flag & VTEXT)
1987 strcat(buf, "|VTEXT");
1988 if (vp->v_flag & VSYSTEM)
1989 strcat(buf, "|VSYSTEM");
1990 if (vp->v_flag & VXLOCK)
1991 strcat(buf, "|VXLOCK");
1992 if (vp->v_flag & VXWANT)
1993 strcat(buf, "|VXWANT");
1994 if (vp->v_flag & VBWAIT)
1995 strcat(buf, "|VBWAIT");
1996 if (vp->v_flag & VALIASED)
1997 strcat(buf, "|VALIASED");
1998 if (vp->v_flag & VDOOMED)
1999 strcat(buf, "|VDOOMED");
2000 if (vp->v_flag & VFREE)
2001 strcat(buf, "|VFREE");
2002 if (vp->v_flag & VOBJBUF)
2003 strcat(buf, "|VOBJBUF");
2004 if (buf[0] != '\0')
2005 printf(" flags (%s)", &buf[1]);
2006 if (vp->v_data == NULL) {
2007 printf("\n");
2008 } else {
2009 printf("\n\t");
2010 VOP_PRINT(vp);
2011 }
2012}
2013
2014#ifdef DDB
2015#include <ddb/ddb.h>
2016/*
2017 * List all of the locked vnodes in the system.
2018 * Called when debugging the kernel.
2019 */
2020DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2021{
2022 struct proc *p = curproc; /* XXX */
2023 struct mount *mp, *nmp;
2024 struct vnode *vp;
2025
2026 printf("Locked vnodes\n");
2027 simple_lock(&mountlist_slock);
2028 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2029 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2030 nmp = mp->mnt_list.cqe_next;
2031 continue;
2032 }
2033 for (vp = mp->mnt_vnodelist.lh_first;
2034 vp != NULL;
2035 vp = vp->v_mntvnodes.le_next) {
2036 if (VOP_ISLOCKED(vp))
2037 vprint((char *)0, vp);
2038 }
2039 simple_lock(&mountlist_slock);
2040 nmp = mp->mnt_list.cqe_next;
2041 vfs_unbusy(mp, p);
2042 }
2043 simple_unlock(&mountlist_slock);
2044}
2045#endif
2046
2047/*
2048 * Top level filesystem related information gathering.
2049 */
2050static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
2051
2052static int
2053vfs_sysctl SYSCTL_HANDLER_ARGS
2054{
2055 int *name = (int *)arg1 - 1; /* XXX */
2056 u_int namelen = arg2 + 1; /* XXX */
2057 struct vfsconf *vfsp;
2058
2059#if 1 || defined(COMPAT_PRELITE2)
2060 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2061 if (namelen == 1)
2062 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2063#endif
2064
2065#ifdef notyet
2066 /* all sysctl names at this level are at least name and field */
2067 if (namelen < 2)
2068 return (ENOTDIR); /* overloaded */
2069 if (name[0] != VFS_GENERIC) {
2070 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2071 if (vfsp->vfc_typenum == name[0])
2072 break;
2073 if (vfsp == NULL)
2074 return (EOPNOTSUPP);
2075 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2076 oldp, oldlenp, newp, newlen, p));
2077 }
2078#endif
2079 switch (name[1]) {
2080 case VFS_MAXTYPENUM:
2081 if (namelen != 2)
2082 return (ENOTDIR);
2083 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2084 case VFS_CONF:
2085 if (namelen != 3)
2086 return (ENOTDIR); /* overloaded */
2087 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2088 if (vfsp->vfc_typenum == name[2])
2089 break;
2090 if (vfsp == NULL)
2091 return (EOPNOTSUPP);
2092 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2093 }
2094 return (EOPNOTSUPP);
2095}
2096
2097SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2098 "Generic filesystem");
2099
2100#if 1 || defined(COMPAT_PRELITE2)
2101
2102static int
2103sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2104{
2105 int error;
2106 struct vfsconf *vfsp;
2107 struct ovfsconf ovfs;
2108
2109 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2110 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2111 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2112 ovfs.vfc_index = vfsp->vfc_typenum;
2113 ovfs.vfc_refcount = vfsp->vfc_refcount;
2114 ovfs.vfc_flags = vfsp->vfc_flags;
2115 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2116 if (error)
2117 return error;
2118 }
2119 return 0;
2120}
2121
2122#endif /* 1 || COMPAT_PRELITE2 */
2123
2124#if 0
2125#define KINFO_VNODESLOP 10
2126/*
2127 * Dump vnode list (via sysctl).
2128 * Copyout address of vnode followed by vnode.
2129 */
2130/* ARGSUSED */
2131static int
2132sysctl_vnode SYSCTL_HANDLER_ARGS
2133{
2134 struct proc *p = curproc; /* XXX */
2135 struct mount *mp, *nmp;
2136 struct vnode *nvp, *vp;
2137 int error;
2138
2139#define VPTRSZ sizeof (struct vnode *)
2140#define VNODESZ sizeof (struct vnode)
2141
2142 req->lock = 0;
2143 if (!req->oldptr) /* Make an estimate */
2144 return (SYSCTL_OUT(req, 0,
2145 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2146
2147 simple_lock(&mountlist_slock);
2148 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2149 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2150 nmp = mp->mnt_list.cqe_next;
2151 continue;
2152 }
2153again:
2154 simple_lock(&mntvnode_slock);
2155 for (vp = mp->mnt_vnodelist.lh_first;
2156 vp != NULL;
2157 vp = nvp) {
2158 /*
2159 * Check that the vp is still associated with
2160 * this filesystem. RACE: could have been
2161 * recycled onto the same filesystem.
2162 */
2163 if (vp->v_mount != mp) {
2164 simple_unlock(&mntvnode_slock);
2165 goto again;
2166 }
2167 nvp = vp->v_mntvnodes.le_next;
2168 simple_unlock(&mntvnode_slock);
2169 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2170 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2171 return (error);
2172 simple_lock(&mntvnode_slock);
2173 }
2174 simple_unlock(&mntvnode_slock);
2175 simple_lock(&mountlist_slock);
2176 nmp = mp->mnt_list.cqe_next;
2177 vfs_unbusy(mp, p);
2178 }
2179 simple_unlock(&mountlist_slock);
2180
2181 return (0);
2182}
2183#endif
2184
2185/*
2186 * XXX
2187 * Exporting the vnode list on large systems causes them to crash.
2188 * Exporting the vnode list on medium systems causes sysctl to coredump.
2189 */
2190#if 0
2191SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2192 0, 0, sysctl_vnode, "S,vnode", "");
2193#endif
2194
2195/*
2196 * Check to see if a filesystem is mounted on a block device.
2197 */
2198int
2199vfs_mountedon(vp)
2200 struct vnode *vp;
2201{
2202 struct vnode *vq;
2203 int error = 0;
2204
2205 if (vp->v_specmountpoint != NULL)
2206 return (EBUSY);
2207 if (vp->v_flag & VALIASED) {
2208 simple_lock(&spechash_slock);
2209 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2210 if (vq->v_rdev != vp->v_rdev ||
2211 vq->v_type != vp->v_type)
2212 continue;
2213 if (vq->v_specmountpoint != NULL) {
2214 error = EBUSY;
2215 break;
2216 }
2217 }
2218 simple_unlock(&spechash_slock);
2219 }
2220 return (error);
2221}
2222
2223/*
2224 * Unmount all filesystems. The list is traversed in reverse order
2225 * of mounting to avoid dependencies.
2226 */
2227void
2228vfs_unmountall()
2229{
2230 struct mount *mp, *nmp;
2231 struct proc *p;
2232 int error;
2233
2234 if (curproc != NULL)
2235 p = curproc;
2236 else
2237 p = initproc; /* XXX XXX should this be proc0? */
2238 /*
2239 * Since this only runs when rebooting, it is not interlocked.
2240 */
2241 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2242 nmp = mp->mnt_list.cqe_prev;
2243 error = dounmount(mp, MNT_FORCE, p);
2244 if (error) {
2245 printf("unmount of %s failed (",
2246 mp->mnt_stat.f_mntonname);
2247 if (error == EBUSY)
2248 printf("BUSY)\n");
2249 else
2250 printf("%d)\n", error);
2251 }
2252 }
2253}
2254
2255/*
2256 * Build hash lists of net addresses and hang them off the mount point.
2257 * Called by ufs_mount() to set up the lists of export addresses.
2258 */
2259static int
2260vfs_hang_addrlist(mp, nep, argp)
2261 struct mount *mp;
2262 struct netexport *nep;
2263 struct export_args *argp;
2264{
2265 register struct netcred *np;
2266 register struct radix_node_head *rnh;
2267 register int i;
2268 struct radix_node *rn;
2269 struct sockaddr *saddr, *smask = 0;
2270 struct domain *dom;
2271 int error;
2272
2273 if (argp->ex_addrlen == 0) {
2274 if (mp->mnt_flag & MNT_DEFEXPORTED)
2275 return (EPERM);
2276 np = &nep->ne_defexported;
2277 np->netc_exflags = argp->ex_flags;
2278 np->netc_anon = argp->ex_anon;
2279 np->netc_anon.cr_ref = 1;
2280 mp->mnt_flag |= MNT_DEFEXPORTED;
2281 return (0);
2282 }
2283 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2284 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2285 bzero((caddr_t) np, i);
2286 saddr = (struct sockaddr *) (np + 1);
2287 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2288 goto out;
2289 if (saddr->sa_len > argp->ex_addrlen)
2290 saddr->sa_len = argp->ex_addrlen;
2291 if (argp->ex_masklen) {
2292 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2293 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2294 if (error)
2295 goto out;
2296 if (smask->sa_len > argp->ex_masklen)
2297 smask->sa_len = argp->ex_masklen;
2298 }
2299 i = saddr->sa_family;
2300 if ((rnh = nep->ne_rtable[i]) == 0) {
2301 /*
2302 * Seems silly to initialize every AF when most are not used,
2303 * do so on demand here
2304 */
2305 for (dom = domains; dom; dom = dom->dom_next)
2306 if (dom->dom_family == i && dom->dom_rtattach) {
2307 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2308 dom->dom_rtoffset);
2309 break;
2310 }
2311 if ((rnh = nep->ne_rtable[i]) == 0) {
2312 error = ENOBUFS;
2313 goto out;
2314 }
2315 }
2316 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2317 np->netc_rnodes);
2318 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2319 error = EPERM;
2320 goto out;
2321 }
2322 np->netc_exflags = argp->ex_flags;
2323 np->netc_anon = argp->ex_anon;
2324 np->netc_anon.cr_ref = 1;
2325 return (0);
2326out:
2327 free(np, M_NETADDR);
2328 return (error);
2329}
2330
2331/* ARGSUSED */
2332static int
2333vfs_free_netcred(rn, w)
2334 struct radix_node *rn;
2335 void *w;
2336{
2337 register struct radix_node_head *rnh = (struct radix_node_head *) w;
2338
2339 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2340 free((caddr_t) rn, M_NETADDR);
2341 return (0);
2342}
2343
2344/*
2345 * Free the net address hash lists that are hanging off the mount points.
2346 */
2347static void
2348vfs_free_addrlist(nep)
2349 struct netexport *nep;
2350{
2351 register int i;
2352 register struct radix_node_head *rnh;
2353
2354 for (i = 0; i <= AF_MAX; i++)
2355 if ((rnh = nep->ne_rtable[i])) {
2356 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2357 (caddr_t) rnh);
2358 free((caddr_t) rnh, M_RTABLE);
2359 nep->ne_rtable[i] = 0;
2360 }
2361}
2362
2363int
2364vfs_export(mp, nep, argp)
2365 struct mount *mp;
2366 struct netexport *nep;
2367 struct export_args *argp;
2368{
2369 int error;
2370
2371 if (argp->ex_flags & MNT_DELEXPORT) {
2372 if (mp->mnt_flag & MNT_EXPUBLIC) {
2373 vfs_setpublicfs(NULL, NULL, NULL);
2374 mp->mnt_flag &= ~MNT_EXPUBLIC;
2375 }
2376 vfs_free_addrlist(nep);
2377 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2378 }
2379 if (argp->ex_flags & MNT_EXPORTED) {
2380 if (argp->ex_flags & MNT_EXPUBLIC) {
2381 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2382 return (error);
2383 mp->mnt_flag |= MNT_EXPUBLIC;
2384 }
2385 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2386 return (error);
2387 mp->mnt_flag |= MNT_EXPORTED;
2388 }
2389 return (0);
2390}
2391
2392
2393/*
2394 * Set the publicly exported filesystem (WebNFS). Currently, only
2395 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2396 */
2397int
2398vfs_setpublicfs(mp, nep, argp)
2399 struct mount *mp;
2400 struct netexport *nep;
2401 struct export_args *argp;
2402{
2403 int error;
2404 struct vnode *rvp;
2405 char *cp;
2406
2407 /*
2408 * mp == NULL -> invalidate the current info, the FS is
2409 * no longer exported. May be called from either vfs_export
2410 * or unmount, so check if it hasn't already been done.
2411 */
2412 if (mp == NULL) {
2413 if (nfs_pub.np_valid) {
2414 nfs_pub.np_valid = 0;
2415 if (nfs_pub.np_index != NULL) {
2416 FREE(nfs_pub.np_index, M_TEMP);
2417 nfs_pub.np_index = NULL;
2418 }
2419 }
2420 return (0);
2421 }
2422
2423 /*
2424 * Only one allowed at a time.
2425 */
2426 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2427 return (EBUSY);
2428
2429 /*
2430 * Get real filehandle for root of exported FS.
2431 */
2432 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2433 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2434
2435 if ((error = VFS_ROOT(mp, &rvp)))
2436 return (error);
2437
2438 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2439 return (error);
2440
2441 vput(rvp);
2442
2443 /*
2444 * If an indexfile was specified, pull it in.
2445 */
2446 if (argp->ex_indexfile != NULL) {
2447 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2448 M_WAITOK);
2449 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2450 MAXNAMLEN, (size_t *)0);
2451 if (!error) {
2452 /*
2453 * Check for illegal filenames.
2454 */
2455 for (cp = nfs_pub.np_index; *cp; cp++) {
2456 if (*cp == '/') {
2457 error = EINVAL;
2458 break;
2459 }
2460 }
2461 }
2462 if (error) {
2463 FREE(nfs_pub.np_index, M_TEMP);
2464 return (error);
2465 }
2466 }
2467
2468 nfs_pub.np_mount = mp;
2469 nfs_pub.np_valid = 1;
2470 return (0);
2471}
2472
2473struct netcred *
2474vfs_export_lookup(mp, nep, nam)
2475 register struct mount *mp;
2476 struct netexport *nep;
2477 struct sockaddr *nam;
2478{
2479 register struct netcred *np;
2480 register struct radix_node_head *rnh;
2481 struct sockaddr *saddr;
2482
2483 np = NULL;
2484 if (mp->mnt_flag & MNT_EXPORTED) {
2485 /*
2486 * Lookup in the export list first.
2487 */
2488 if (nam != NULL) {
2489 saddr = nam;
2490 rnh = nep->ne_rtable[saddr->sa_family];
2491 if (rnh != NULL) {
2492 np = (struct netcred *)
2493 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2494 rnh);
2495 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2496 np = NULL;
2497 }
2498 }
2499 /*
2500 * If no address match, use the default if it exists.
2501 */
2502 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2503 np = &nep->ne_defexported;
2504 }
2505 return (np);
2506}
2507
2508/*
2509 * perform msync on all vnodes under a mount point
2510 * the mount point must be locked.
2511 */
2512void
2513vfs_msync(struct mount *mp, int flags) {
2514 struct vnode *vp, *nvp;
2515 struct vm_object *obj;
2516 int anyio, tries;
2517
2518 tries = 5;
2519loop:
2520 anyio = 0;
2521 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
2522
2523 nvp = vp->v_mntvnodes.le_next;
2524
2525 if (vp->v_mount != mp) {
2526 goto loop;
2527 }
2528
2529 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2530 continue;
2531
2532 if (flags != MNT_WAIT) {
2533 obj = vp->v_object;
2534 if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2535 continue;
2536 if (VOP_ISLOCKED(vp))
2537 continue;
2538 }
2539
2540 simple_lock(&vp->v_interlock);
2541 if (vp->v_object &&
2542 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2543 if (!vget(vp,
2544 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2545 if (vp->v_object) {
2546 vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : 0);
2547 anyio = 1;
2548 }
2549 vput(vp);
2550 }
2551 } else {
2552 simple_unlock(&vp->v_interlock);
2553 }
2554 }
2555 if (anyio && (--tries > 0))
2556 goto loop;
2557}
2558
2559/*
2560 * Create the VM object needed for VMIO and mmap support. This
2561 * is done for all VREG files in the system. Some filesystems might
2562 * afford the additional metadata buffering capability of the
2563 * VMIO code by making the device node be VMIO mode also.
2564 *
2565 * vp must be locked when vfs_object_create is called.
2566 */
2567int
2568vfs_object_create(vp, p, cred)
2569 struct vnode *vp;
2570 struct proc *p;
2571 struct ucred *cred;
2572{
2573 struct vattr vat;
2574 vm_object_t object;
2575 int error = 0;
2576
2577 if ((vp->v_type != VREG) && (vp->v_type != VBLK))
2578 return 0;
2579
2580retry:
2581 if ((object = vp->v_object) == NULL) {
2582 if (vp->v_type == VREG) {
2583 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2584 goto retn;
2585 object = vnode_pager_alloc(vp, vat.va_size, 0, 0);
2586 } else if (major(vp->v_rdev) < nblkdev &&
2587 bdevsw[major(vp->v_rdev)] != NULL) {
2588 /*
2589 * This simply allocates the biggest object possible
2590 * for a VBLK vnode. This should be fixed, but doesn't
2591 * cause any problems (yet).
2592 */
2593 object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0);
2594 } else {
2595 goto retn;
2596 }
2597 /*
2598 * Dereference the reference we just created. This assumes
2599 * that the object is associated with the vp.
2600 */
2601 object->ref_count--;
2602 vp->v_usecount--;
2603 } else {
2604 if (object->flags & OBJ_DEAD) {
2605 VOP_UNLOCK(vp, 0, p);
2606 tsleep(object, PVM, "vodead", 0);
2607 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2608 goto retry;
2609 }
2610 }
2611
2612 KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object"));
2613 vp->v_flag |= VOBJBUF;
2614
2615retn:
2616 return error;
2617}
2618
2619static void
2620vfree(vp)
2621 struct vnode *vp;
2622{
2623 int s;
2624
2625 s = splbio();
2626 simple_lock(&vnode_free_list_slock);
2627 if (vp->v_flag & VTBFREE) {
2628 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2629 vp->v_flag &= ~VTBFREE;
2630 }
2631 if (vp->v_flag & VAGE) {
2632 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2633 } else {
2634 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2635 }
2636 freevnodes++;
2637 simple_unlock(&vnode_free_list_slock);
2638 vp->v_flag &= ~VAGE;
2639 vp->v_flag |= VFREE;
2640 splx(s);
2641}
2642
2643void
2644vbusy(vp)
2645 struct vnode *vp;
2646{
2647 int s;
2648
2649 s = splbio();
2650 simple_lock(&vnode_free_list_slock);
2651 if (vp->v_flag & VTBFREE) {
2652 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2653 vp->v_flag &= ~VTBFREE;
2654 } else {
2655 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2656 freevnodes--;
2657 }
2658 simple_unlock(&vnode_free_list_slock);
2659 vp->v_flag &= ~(VFREE|VAGE);
2660 splx(s);
2661}
2662
2663/*
2664 * Record a process's interest in events which might happen to
2665 * a vnode. Because poll uses the historic select-style interface
2666 * internally, this routine serves as both the ``check for any
2667 * pending events'' and the ``record my interest in future events''
2668 * functions. (These are done together, while the lock is held,
2669 * to avoid race conditions.)
2670 */
2671int
2672vn_pollrecord(vp, p, events)
2673 struct vnode *vp;
2674 struct proc *p;
2675 short events;
2676{
2677 simple_lock(&vp->v_pollinfo.vpi_lock);
2678 if (vp->v_pollinfo.vpi_revents & events) {
2679 /*
2680 * This leaves events we are not interested
2681 * in available for the other process which
2682 * which presumably had requested them
2683 * (otherwise they would never have been
2684 * recorded).
2685 */
2686 events &= vp->v_pollinfo.vpi_revents;
2687 vp->v_pollinfo.vpi_revents &= ~events;
2688
2689 simple_unlock(&vp->v_pollinfo.vpi_lock);
2690 return events;
2691 }
2692 vp->v_pollinfo.vpi_events |= events;
2693 selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2694 simple_unlock(&vp->v_pollinfo.vpi_lock);
2695 return 0;
2696}
2697
2698/*
2699 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2700 * it is possible for us to miss an event due to race conditions, but
2701 * that condition is expected to be rare, so for the moment it is the
2702 * preferred interface.
2703 */
2704void
2705vn_pollevent(vp, events)
2706 struct vnode *vp;
2707 short events;
2708{
2709 simple_lock(&vp->v_pollinfo.vpi_lock);
2710 if (vp->v_pollinfo.vpi_events & events) {
2711 /*
2712 * We clear vpi_events so that we don't
2713 * call selwakeup() twice if two events are
2714 * posted before the polling process(es) is
2715 * awakened. This also ensures that we take at
2716 * most one selwakeup() if the polling process
2717 * is no longer interested. However, it does
2718 * mean that only one event can be noticed at
2719 * a time. (Perhaps we should only clear those
2720 * event bits which we note?) XXX
2721 */
2722 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2723 vp->v_pollinfo.vpi_revents |= events;
2724 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2725 }
2726 simple_unlock(&vp->v_pollinfo.vpi_lock);
2727}
2728
2729/*
2730 * Wake up anyone polling on vp because it is being revoked.
2731 * This depends on dead_poll() returning POLLHUP for correct
2732 * behavior.
2733 */
2734void
2735vn_pollgone(vp)
2736 struct vnode *vp;
2737{
2738 simple_lock(&vp->v_pollinfo.vpi_lock);
2739 if (vp->v_pollinfo.vpi_events) {
2740 vp->v_pollinfo.vpi_events = 0;
2741 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2742 }
2743 simple_unlock(&vp->v_pollinfo.vpi_lock);
2744}
2745
2746
2747
2748/*
2749 * Routine to create and manage a filesystem syncer vnode.
2750 */
2751#define sync_close ((int (*) __P((struct vop_close_args *)))nullop)
2752static int sync_fsync __P((struct vop_fsync_args *));
2753static int sync_inactive __P((struct vop_inactive_args *));
2754static int sync_reclaim __P((struct vop_reclaim_args *));
2755#define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock)
2756#define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock)
2757static int sync_print __P((struct vop_print_args *));
2758#define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2759
2760static vop_t **sync_vnodeop_p;
2761static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2762 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2763 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2764 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2765 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2766 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2767 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2768 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2769 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2770 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2771 { NULL, NULL }
2772};
2773static struct vnodeopv_desc sync_vnodeop_opv_desc =
2774 { &sync_vnodeop_p, sync_vnodeop_entries };
2775
2776VNODEOP_SET(sync_vnodeop_opv_desc);
2777
2778/*
2779 * Create a new filesystem syncer vnode for the specified mount point.
2780 */
2781int
2782vfs_allocate_syncvnode(mp)
2783 struct mount *mp;
2784{
2785 struct vnode *vp;
2786 static long start, incr, next;
2787 int error;
2788
2789 /* Allocate a new vnode */
2790 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2791 mp->mnt_syncer = NULL;
2792 return (error);
2793 }
2794 vp->v_type = VNON;
2795 /*
2796 * Place the vnode onto the syncer worklist. We attempt to
2797 * scatter them about on the list so that they will go off
2798 * at evenly distributed times even if all the filesystems
2799 * are mounted at once.
2800 */
2801 next += incr;
2802 if (next == 0 || next > syncer_maxdelay) {
2803 start /= 2;
2804 incr /= 2;
2805 if (start == 0) {
2806 start = syncer_maxdelay / 2;
2807 incr = syncer_maxdelay;
2808 }
2809 next = start;
2810 }
2811 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2812 mp->mnt_syncer = vp;
2813 return (0);
2814}
2815
2816/*
2817 * Do a lazy sync of the filesystem.
2818 */
2819static int
2820sync_fsync(ap)
2821 struct vop_fsync_args /* {
2822 struct vnode *a_vp;
2823 struct ucred *a_cred;
2824 int a_waitfor;
2825 struct proc *a_p;
2826 } */ *ap;
2827{
2828 struct vnode *syncvp = ap->a_vp;
2829 struct mount *mp = syncvp->v_mount;
2830 struct proc *p = ap->a_p;
2831 int asyncflag;
2832
2833 /*
2834 * We only need to do something if this is a lazy evaluation.
2835 */
2836 if (ap->a_waitfor != MNT_LAZY)
2837 return (0);
2838
2839 /*
2840 * Move ourselves to the back of the sync list.
2841 */
2842 vn_syncer_add_to_worklist(syncvp, syncdelay);
2843
2844 /*
2845 * Walk the list of vnodes pushing all that are dirty and
2846 * not already on the sync list.
2847 */
2848 simple_lock(&mountlist_slock);
2849 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2850 simple_unlock(&mountlist_slock);
2851 return (0);
2852 }
2853 asyncflag = mp->mnt_flag & MNT_ASYNC;
2854 mp->mnt_flag &= ~MNT_ASYNC;
2855 vfs_msync(mp, MNT_NOWAIT);
2856 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2857 if (asyncflag)
2858 mp->mnt_flag |= MNT_ASYNC;
2859 vfs_unbusy(mp, p);
2860 return (0);
2861}
2862
2863/*
2864 * The syncer vnode is no referenced.
2865 */
2866static int
2867sync_inactive(ap)
2868 struct vop_inactive_args /* {
2869 struct vnode *a_vp;
2870 struct proc *a_p;
2871 } */ *ap;
2872{
2873
2874 vgone(ap->a_vp);
2875 return (0);
2876}
2877
2878/*
2879 * The syncer vnode is no longer needed and is being decommissioned.
2880 *
2881 * Modifications to the worklist must be protected at splbio().
2882 */
2883static int
2884sync_reclaim(ap)
2885 struct vop_reclaim_args /* {
2886 struct vnode *a_vp;
2887 } */ *ap;
2888{
2889 struct vnode *vp = ap->a_vp;
2890 int s;
2891
2892 s = splbio();
2893 vp->v_mount->mnt_syncer = NULL;
2894 if (vp->v_flag & VONWORKLST) {
2895 LIST_REMOVE(vp, v_synclist);
2896 vp->v_flag &= ~VONWORKLST;
2897 }
2898 splx(s);
2899
2900 return (0);
2901}
2902
2903/*
2904 * Print out a syncer vnode.
2905 */
2906static int
2907sync_print(ap)
2908 struct vop_print_args /* {
2909 struct vnode *a_vp;
2910 } */ *ap;
2911{
2912 struct vnode *vp = ap->a_vp;
2913
2914 printf("syncer vnode");
2915 if (vp->v_vnlock != NULL)
2916 lockmgr_printinfo(vp->v_vnlock);
2917 printf("\n");
2918 return (0);
2919}