#include reorganization
[dragonfly.git] / sys / kern / vfs_subr.c
CommitLineData
984263bc
MD
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 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
402ed7e1 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.18 2003/08/26 21:09:02 rob Exp $
984263bc
MD
41 */
42
43/*
44 * External virtual filesystem routines
45 */
46#include "opt_ddb.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/buf.h>
51#include <sys/conf.h>
52#include <sys/dirent.h>
53#include <sys/domain.h>
54#include <sys/eventhandler.h>
55#include <sys/fcntl.h>
56#include <sys/kernel.h>
57#include <sys/kthread.h>
58#include <sys/malloc.h>
59#include <sys/mbuf.h>
60#include <sys/mount.h>
984263bc 61#include <sys/proc.h>
dadab5e9 62#include <sys/namei.h>
984263bc
MD
63#include <sys/reboot.h>
64#include <sys/socket.h>
65#include <sys/stat.h>
66#include <sys/sysctl.h>
67#include <sys/syslog.h>
68#include <sys/vmmeter.h>
69#include <sys/vnode.h>
70
71#include <machine/limits.h>
72
73#include <vm/vm.h>
74#include <vm/vm_object.h>
75#include <vm/vm_extern.h>
76#include <vm/pmap.h>
77#include <vm/vm_map.h>
78#include <vm/vm_page.h>
79#include <vm/vm_pager.h>
80#include <vm/vnode_pager.h>
81#include <vm/vm_zone.h>
82
3020e3be 83#include <sys/buf2.h>
f5865223 84#include <sys/thread2.h>
3020e3be 85
984263bc
MD
86static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
87
402ed7e1
RG
88static void insmntque (struct vnode *vp, struct mount *mp);
89static void vclean (struct vnode *vp, int flags, struct thread *td);
984263bc
MD
90static unsigned long numvnodes;
91static void vlruvp(struct vnode *vp);
92SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93
94enum vtype iftovt_tab[16] = {
95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
97};
98int vttoif_tab[9] = {
99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
100 S_IFSOCK, S_IFIFO, S_IFMT,
101};
102
103static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
104
105static u_long wantfreevnodes = 25;
106SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
107static u_long freevnodes = 0;
108SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
109
110static int reassignbufcalls;
111SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
112static int reassignbufloops;
113SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
114static int reassignbufsortgood;
115SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
116static int reassignbufsortbad;
117SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
118static int reassignbufmethod = 1;
119SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
120static int nameileafonly = 0;
121SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
122
123#ifdef ENABLE_VFS_IOOPT
124int vfs_ioopt = 0;
125SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
126#endif
127
128struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
8a8d5d85
MD
129struct lwkt_token mountlist_token;
130struct lwkt_token mntvnode_token;
984263bc 131int nfs_mount_type = -1;
8a8d5d85
MD
132static struct lwkt_token mntid_token;
133static struct lwkt_token vnode_free_list_token;
134static struct lwkt_token spechash_token;
984263bc
MD
135struct nfs_public nfs_pub; /* publicly exported FS */
136static vm_zone_t vnode_zone;
137
138/*
139 * The workitem queue.
140 */
141#define SYNCER_MAXDELAY 32
142static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
143time_t syncdelay = 30; /* max time to delay syncing data */
144time_t filedelay = 30; /* time to delay syncing files */
145SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
146time_t dirdelay = 29; /* time to delay syncing directories */
147SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
148time_t metadelay = 28; /* time to delay syncing metadata */
149SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
150static int rushjob; /* number of slots to run ASAP */
151static int stat_rush_requests; /* number of times I/O speeded up */
152SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
153
154static int syncer_delayno = 0;
155static long syncer_mask;
156LIST_HEAD(synclist, vnode);
157static struct synclist *syncer_workitem_pending;
158
159int desiredvnodes;
160SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
161 &desiredvnodes, 0, "Maximum number of vnodes");
162static int minvnodes;
163SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
164 &minvnodes, 0, "Minimum number of vnodes");
165static int vnlru_nowhere = 0;
166SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
167 "Number of times the vnlru process ran without success");
168
402ed7e1
RG
169static void vfs_free_addrlist (struct netexport *nep);
170static int vfs_free_netcred (struct radix_node *rn, void *w);
171static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
172 struct export_args *argp);
984263bc
MD
173
174/*
175 * Initialize the vnode management data structures.
176 */
177void
178vntblinit()
179{
180
12e4aaff 181 desiredvnodes = maxproc + vmstats.v_page_count / 4;
984263bc 182 minvnodes = desiredvnodes / 4;
8a8d5d85
MD
183 lwkt_inittoken(&mntvnode_token);
184 lwkt_inittoken(&mntid_token);
185 lwkt_inittoken(&spechash_token);
984263bc 186 TAILQ_INIT(&vnode_free_list);
8a8d5d85 187 lwkt_inittoken(&vnode_free_list_token);
984263bc
MD
188 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
189 /*
190 * Initialize the filesystem syncer.
191 */
192 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
193 &syncer_mask);
194 syncer_maxdelay = syncer_mask + 1;
195}
196
197/*
198 * Mark a mount point as busy. Used to synchronize access and to delay
199 * unmounting. Interlock is not released on failure.
200 */
201int
8a8d5d85 202vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp,
dadab5e9 203 struct thread *td)
984263bc
MD
204{
205 int lkflags;
206
207 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
208 if (flags & LK_NOWAIT)
209 return (ENOENT);
210 mp->mnt_kern_flag |= MNTK_MWAIT;
211 if (interlkp) {
8a8d5d85 212 lwkt_reltoken(interlkp);
984263bc
MD
213 }
214 /*
215 * Since all busy locks are shared except the exclusive
216 * lock granted when unmounting, the only place that a
217 * wakeup needs to be done is at the release of the
218 * exclusive lock at the end of dounmount.
219 */
377d4740 220 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
984263bc 221 if (interlkp) {
8a8d5d85 222 lwkt_gettoken(interlkp);
984263bc
MD
223 }
224 return (ENOENT);
225 }
226 lkflags = LK_SHARED | LK_NOPAUSE;
227 if (interlkp)
228 lkflags |= LK_INTERLOCK;
dadab5e9 229 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
984263bc
MD
230 panic("vfs_busy: unexpected lock failure");
231 return (0);
232}
233
234/*
235 * Free a busy filesystem.
236 */
237void
dadab5e9 238vfs_unbusy(struct mount *mp, struct thread *td)
984263bc 239{
dadab5e9 240 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
984263bc
MD
241}
242
243/*
244 * Lookup a filesystem type, and if found allocate and initialize
245 * a mount structure for it.
246 *
247 * Devname is usually updated by mount(8) after booting.
248 */
249int
dadab5e9 250vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
984263bc 251{
dadab5e9 252 struct thread *td = curthread; /* XXX */
984263bc
MD
253 struct vfsconf *vfsp;
254 struct mount *mp;
255
256 if (fstypename == NULL)
257 return (ENODEV);
258 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
259 if (!strcmp(vfsp->vfc_name, fstypename))
260 break;
261 if (vfsp == NULL)
262 return (ENODEV);
263 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
264 bzero((char *)mp, (u_long)sizeof(struct mount));
377d4740 265 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
dadab5e9 266 (void)vfs_busy(mp, LK_NOWAIT, 0, td);
984263bc
MD
267 TAILQ_INIT(&mp->mnt_nvnodelist);
268 TAILQ_INIT(&mp->mnt_reservedvnlist);
269 mp->mnt_nvnodelistsize = 0;
270 mp->mnt_vfc = vfsp;
271 mp->mnt_op = vfsp->vfc_vfsops;
272 mp->mnt_flag = MNT_RDONLY;
273 mp->mnt_vnodecovered = NULLVP;
274 vfsp->vfc_refcount++;
275 mp->mnt_iosize_max = DFLTPHYS;
276 mp->mnt_stat.f_type = vfsp->vfc_typenum;
277 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
278 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
279 mp->mnt_stat.f_mntonname[0] = '/';
280 mp->mnt_stat.f_mntonname[1] = 0;
281 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
282 *mpp = mp;
283 return (0);
284}
285
286/*
287 * Find an appropriate filesystem to use for the root. If a filesystem
288 * has not been preselected, walk through the list of known filesystems
289 * trying those that have mountroot routines, and try them until one
290 * works or we have tried them all.
291 */
292#ifdef notdef /* XXX JH */
293int
294lite2_vfs_mountroot()
295{
296 struct vfsconf *vfsp;
402ed7e1 297 extern int (*lite2_mountroot) (void);
984263bc
MD
298 int error;
299
300 if (lite2_mountroot != NULL)
301 return ((*lite2_mountroot)());
302 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
303 if (vfsp->vfc_mountroot == NULL)
304 continue;
305 if ((error = (*vfsp->vfc_mountroot)()) == 0)
306 return (0);
307 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
308 }
309 return (ENODEV);
310}
311#endif
312
313/*
314 * Lookup a mount point by filesystem identifier.
315 */
316struct mount *
317vfs_getvfs(fsid)
318 fsid_t *fsid;
319{
1fd87d54 320 struct mount *mp;
984263bc 321
8a8d5d85 322 lwkt_gettoken(&mountlist_token);
984263bc
MD
323 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
324 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
325 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
8a8d5d85 326 lwkt_reltoken(&mountlist_token);
984263bc
MD
327 return (mp);
328 }
329 }
8a8d5d85 330 lwkt_reltoken(&mountlist_token);
984263bc
MD
331 return ((struct mount *) 0);
332}
333
334/*
335 * Get a new unique fsid. Try to make its val[0] unique, since this value
336 * will be used to create fake device numbers for stat(). Also try (but
337 * not so hard) make its val[0] unique mod 2^16, since some emulators only
338 * support 16-bit device numbers. We end up with unique val[0]'s for the
339 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
340 *
341 * Keep in mind that several mounts may be running in parallel. Starting
342 * the search one past where the previous search terminated is both a
343 * micro-optimization and a defense against returning the same fsid to
344 * different mounts.
345 */
346void
347vfs_getnewfsid(mp)
348 struct mount *mp;
349{
350 static u_int16_t mntid_base;
351 fsid_t tfsid;
352 int mtype;
353
8a8d5d85 354 lwkt_gettoken(&mntid_token);
984263bc
MD
355 mtype = mp->mnt_vfc->vfc_typenum;
356 tfsid.val[1] = mtype;
357 mtype = (mtype & 0xFF) << 24;
358 for (;;) {
359 tfsid.val[0] = makeudev(255,
360 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
361 mntid_base++;
362 if (vfs_getvfs(&tfsid) == NULL)
363 break;
364 }
365 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
366 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
8a8d5d85 367 lwkt_reltoken(&mntid_token);
984263bc
MD
368}
369
370/*
371 * Knob to control the precision of file timestamps:
372 *
373 * 0 = seconds only; nanoseconds zeroed.
374 * 1 = seconds and nanoseconds, accurate within 1/HZ.
375 * 2 = seconds and nanoseconds, truncated to microseconds.
376 * >=3 = seconds and nanoseconds, maximum precision.
377 */
378enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
379
380static int timestamp_precision = TSP_SEC;
381SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
382 &timestamp_precision, 0, "");
383
384/*
385 * Get a current timestamp.
386 */
387void
388vfs_timestamp(tsp)
389 struct timespec *tsp;
390{
391 struct timeval tv;
392
393 switch (timestamp_precision) {
394 case TSP_SEC:
395 tsp->tv_sec = time_second;
396 tsp->tv_nsec = 0;
397 break;
398 case TSP_HZ:
399 getnanotime(tsp);
400 break;
401 case TSP_USEC:
402 microtime(&tv);
403 TIMEVAL_TO_TIMESPEC(&tv, tsp);
404 break;
405 case TSP_NSEC:
406 default:
407 nanotime(tsp);
408 break;
409 }
410}
411
412/*
413 * Set vnode attributes to VNOVAL
414 */
415void
416vattr_null(vap)
1fd87d54 417 struct vattr *vap;
984263bc
MD
418{
419
420 vap->va_type = VNON;
421 vap->va_size = VNOVAL;
422 vap->va_bytes = VNOVAL;
423 vap->va_mode = VNOVAL;
424 vap->va_nlink = VNOVAL;
425 vap->va_uid = VNOVAL;
426 vap->va_gid = VNOVAL;
427 vap->va_fsid = VNOVAL;
428 vap->va_fileid = VNOVAL;
429 vap->va_blocksize = VNOVAL;
430 vap->va_rdev = VNOVAL;
431 vap->va_atime.tv_sec = VNOVAL;
432 vap->va_atime.tv_nsec = VNOVAL;
433 vap->va_mtime.tv_sec = VNOVAL;
434 vap->va_mtime.tv_nsec = VNOVAL;
435 vap->va_ctime.tv_sec = VNOVAL;
436 vap->va_ctime.tv_nsec = VNOVAL;
437 vap->va_flags = VNOVAL;
438 vap->va_gen = VNOVAL;
439 vap->va_vaflags = 0;
440}
441
442/*
443 * This routine is called when we have too many vnodes. It attempts
444 * to free <count> vnodes and will potentially free vnodes that still
445 * have VM backing store (VM backing store is typically the cause
446 * of a vnode blowout so we want to do this). Therefore, this operation
447 * is not considered cheap.
448 *
449 * A number of conditions may prevent a vnode from being reclaimed.
450 * the buffer cache may have references on the vnode, a directory
451 * vnode may still have references due to the namei cache representing
452 * underlying files, or the vnode may be in active use. It is not
453 * desireable to reuse such vnodes. These conditions may cause the
454 * number of vnodes to reach some minimum value regardless of what
455 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
456 */
457static int
458vlrureclaim(struct mount *mp)
459{
460 struct vnode *vp;
461 int done;
462 int trigger;
463 int usevnodes;
464 int count;
8a8d5d85 465 int gen;
984263bc
MD
466
467 /*
468 * Calculate the trigger point, don't allow user
469 * screwups to blow us up. This prevents us from
470 * recycling vnodes with lots of resident pages. We
471 * aren't trying to free memory, we are trying to
472 * free vnodes.
473 */
474 usevnodes = desiredvnodes;
475 if (usevnodes <= 0)
476 usevnodes = 1;
12e4aaff 477 trigger = vmstats.v_page_count * 2 / usevnodes;
984263bc
MD
478
479 done = 0;
8a8d5d85 480 gen = lwkt_gettoken(&mntvnode_token);
984263bc
MD
481 count = mp->mnt_nvnodelistsize / 10 + 1;
482 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
483 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
484 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
485
486 if (vp->v_type != VNON &&
487 vp->v_type != VBAD &&
488 VMIGHTFREE(vp) && /* critical path opt */
8a8d5d85 489 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger)
984263bc 490 ) {
8a8d5d85
MD
491 lwkt_gettoken(&vp->v_interlock);
492 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) {
493 if (VMIGHTFREE(vp)) {
494 vgonel(vp, curthread);
495 done++;
496 } else {
497 lwkt_reltoken(&vp->v_interlock);
498 }
984263bc 499 } else {
8a8d5d85 500 lwkt_reltoken(&vp->v_interlock);
984263bc 501 }
984263bc
MD
502 }
503 --count;
504 }
8a8d5d85 505 lwkt_reltoken(&mntvnode_token);
984263bc
MD
506 return done;
507}
508
509/*
510 * Attempt to recycle vnodes in a context that is always safe to block.
511 * Calling vlrurecycle() from the bowels of file system code has some
512 * interesting deadlock problems.
513 */
bc6dffab 514static struct thread *vnlruthread;
984263bc
MD
515static int vnlruproc_sig;
516
517static void
518vnlru_proc(void)
519{
520 struct mount *mp, *nmp;
521 int s;
522 int done;
0cfcada1 523 struct thread *td = curthread;
984263bc 524
bc6dffab 525 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
526 SHUTDOWN_PRI_FIRST);
527
528 s = splbio();
529 for (;;) {
0cfcada1 530 kproc_suspend_loop();
984263bc
MD
531 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
532 vnlruproc_sig = 0;
533 wakeup(&vnlruproc_sig);
377d4740 534 tsleep(td, 0, "vlruwt", hz);
984263bc
MD
535 continue;
536 }
537 done = 0;
8a8d5d85 538 lwkt_gettoken(&mountlist_token);
984263bc 539 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 540 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
984263bc
MD
541 nmp = TAILQ_NEXT(mp, mnt_list);
542 continue;
543 }
544 done += vlrureclaim(mp);
8a8d5d85 545 lwkt_gettoken(&mountlist_token);
984263bc 546 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 547 vfs_unbusy(mp, td);
984263bc 548 }
8a8d5d85 549 lwkt_reltoken(&mountlist_token);
984263bc
MD
550 if (done == 0) {
551 vnlru_nowhere++;
377d4740 552 tsleep(td, 0, "vlrup", hz * 3);
984263bc
MD
553 }
554 }
555 splx(s);
556}
557
558static struct kproc_desc vnlru_kp = {
559 "vnlru",
560 vnlru_proc,
bc6dffab 561 &vnlruthread
984263bc
MD
562};
563SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
564
565/*
566 * Routines having to do with the management of the vnode table.
567 */
568extern vop_t **dead_vnodeop_p;
569
570/*
571 * Return the next vnode from the free list.
572 */
573int
574getnewvnode(tag, mp, vops, vpp)
575 enum vtagtype tag;
576 struct mount *mp;
577 vop_t **vops;
578 struct vnode **vpp;
579{
580 int s;
8a8d5d85
MD
581 int gen;
582 int vgen;
dadab5e9 583 struct thread *td = curthread; /* XXX */
984263bc
MD
584 struct vnode *vp = NULL;
585 vm_object_t object;
586
587 s = splbio();
588
589 /*
590 * Try to reuse vnodes if we hit the max. This situation only
591 * occurs in certain large-memory (2G+) situations. We cannot
592 * attempt to directly reclaim vnodes due to nasty recursion
593 * problems.
594 */
595 while (numvnodes - freevnodes > desiredvnodes) {
596 if (vnlruproc_sig == 0) {
597 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
bc6dffab 598 wakeup(vnlruthread);
984263bc 599 }
377d4740 600 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
984263bc
MD
601 }
602
603
604 /*
605 * Attempt to reuse a vnode already on the free list, allocating
606 * a new vnode if we can't find one or if we have not reached a
607 * good minimum for good LRU performance.
608 */
8a8d5d85 609 gen = lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
610 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
611 int count;
612
613 for (count = 0; count < freevnodes; count++) {
614 vp = TAILQ_FIRST(&vnode_free_list);
615 if (vp == NULL || vp->v_usecount)
616 panic("getnewvnode: free vnode isn't");
617
8a8d5d85
MD
618 /*
619 * Get the vnode's interlock, then re-obtain
620 * vnode_free_list_token in case we lost it. If we
621 * did lose it while getting the vnode interlock,
622 * even if we got it back again, then retry.
623 */
624 vgen = lwkt_gettoken(&vp->v_interlock);
625 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
626 --count;
627 lwkt_reltoken(&vp->v_interlock);
628 vp = NULL;
629 continue;
630 }
631
632 /*
633 * Whew! We have both tokens. Since we didn't lose
634 * the free list VFREE had better still be set. But
635 * we aren't out of the woods yet. We have to get
636 * the object (may block). If the vnode is not
637 * suitable then move it to the end of the list
638 * if we can. If we can't move it to the end of the
639 * list retry again.
640 */
984263bc 641 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
8a8d5d85
MD
642 (object->resident_page_count || object->ref_count))
643 ) {
644 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 &&
645 lwkt_gentoken(&vnode_free_list_token, &gen) == 0
646 ) {
647 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
648 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
649 } else {
650 --count;
651 }
652 lwkt_reltoken(&vp->v_interlock);
653 vp = NULL;
654 continue;
655 }
656
657 /*
658 * Still not out of the woods. VOBJECT might have
659 * blocked, if we did not retain our tokens we have
660 * to retry.
661 */
662 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 ||
663 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
664 --count;
984263bc
MD
665 vp = NULL;
666 continue;
667 }
8a8d5d85
MD
668 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
669 KKASSERT(vp->v_flag & VFREE);
670
984263bc
MD
671 if (LIST_FIRST(&vp->v_cache_src)) {
672 /*
673 * note: nameileafonly sysctl is temporary,
674 * for debugging only, and will eventually be
675 * removed.
676 */
677 if (nameileafonly > 0) {
678 /*
679 * Do not reuse namei-cached directory
680 * vnodes that have cached
681 * subdirectories.
682 */
683 if (cache_leaf_test(vp) < 0) {
8a8d5d85 684 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
685 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
686 vp = NULL;
687 continue;
688 }
689 } else if (nameileafonly < 0 ||
690 vmiodirenable == 0) {
691 /*
692 * Do not reuse namei-cached directory
693 * vnodes if nameileafonly is -1 or
694 * if VMIO backing for directories is
695 * turned off (otherwise we reuse them
696 * too quickly).
697 */
8a8d5d85 698 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
699 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
700 vp = NULL;
701 continue;
702 }
703 }
704 break;
705 }
706 }
707
708 if (vp) {
709 vp->v_flag |= VDOOMED;
710 vp->v_flag &= ~VFREE;
711 freevnodes--;
8a8d5d85
MD
712 lwkt_reltoken(&vnode_free_list_token);
713 cache_purge(vp); /* YYY may block */
984263bc
MD
714 vp->v_lease = NULL;
715 if (vp->v_type != VBAD) {
dadab5e9 716 vgonel(vp, td);
984263bc 717 } else {
8a8d5d85 718 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
719 }
720
721#ifdef INVARIANTS
722 {
723 int s;
724
725 if (vp->v_data)
726 panic("cleaned vnode isn't");
727 s = splbio();
728 if (vp->v_numoutput)
729 panic("Clean vnode has pending I/O's");
730 splx(s);
731 }
732#endif
733 vp->v_flag = 0;
734 vp->v_lastw = 0;
735 vp->v_lasta = 0;
736 vp->v_cstart = 0;
737 vp->v_clen = 0;
738 vp->v_socket = 0;
739 vp->v_writecount = 0; /* XXX */
740 } else {
8a8d5d85 741 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
742 vp = (struct vnode *) zalloc(vnode_zone);
743 bzero((char *) vp, sizeof *vp);
8a8d5d85 744 lwkt_inittoken(&vp->v_interlock);
984263bc
MD
745 vp->v_dd = vp;
746 cache_purge(vp);
747 LIST_INIT(&vp->v_cache_src);
748 TAILQ_INIT(&vp->v_cache_dst);
749 numvnodes++;
750 }
751
752 TAILQ_INIT(&vp->v_cleanblkhd);
753 TAILQ_INIT(&vp->v_dirtyblkhd);
754 vp->v_type = VNON;
755 vp->v_tag = tag;
756 vp->v_op = vops;
757 insmntque(vp, mp);
758 *vpp = vp;
759 vp->v_usecount = 1;
760 vp->v_data = 0;
761 splx(s);
762
3b568787 763 vfs_object_create(vp, td);
984263bc
MD
764 return (0);
765}
766
767/*
768 * Move a vnode from one mount queue to another.
769 */
770static void
771insmntque(vp, mp)
1fd87d54
RG
772 struct vnode *vp;
773 struct mount *mp;
984263bc
MD
774{
775
8a8d5d85 776 lwkt_gettoken(&mntvnode_token);
984263bc
MD
777 /*
778 * Delete from old mount point vnode list, if on one.
779 */
780 if (vp->v_mount != NULL) {
781 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
782 ("bad mount point vnode list size"));
783 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
784 vp->v_mount->mnt_nvnodelistsize--;
785 }
786 /*
787 * Insert into list of vnodes for the new mount point, if available.
788 */
789 if ((vp->v_mount = mp) == NULL) {
8a8d5d85 790 lwkt_reltoken(&mntvnode_token);
984263bc
MD
791 return;
792 }
793 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
794 mp->mnt_nvnodelistsize++;
8a8d5d85 795 lwkt_reltoken(&mntvnode_token);
984263bc
MD
796}
797
798/*
799 * Update outstanding I/O count and do wakeup if requested.
800 */
801void
802vwakeup(bp)
1fd87d54 803 struct buf *bp;
984263bc 804{
1fd87d54 805 struct vnode *vp;
984263bc
MD
806
807 bp->b_flags &= ~B_WRITEINPROG;
808 if ((vp = bp->b_vp)) {
809 vp->v_numoutput--;
810 if (vp->v_numoutput < 0)
811 panic("vwakeup: neg numoutput");
812 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
813 vp->v_flag &= ~VBWAIT;
814 wakeup((caddr_t) &vp->v_numoutput);
815 }
816 }
817}
818
819/*
820 * Flush out and invalidate all buffers associated with a vnode.
821 * Called with the underlying object locked.
822 */
823int
3b568787
MD
824vinvalbuf(struct vnode *vp, int flags, struct thread *td,
825 int slpflag, int slptimeo)
984263bc 826{
1fd87d54 827 struct buf *bp;
984263bc
MD
828 struct buf *nbp, *blist;
829 int s, error;
830 vm_object_t object;
831
832 if (flags & V_SAVE) {
833 s = splbio();
834 while (vp->v_numoutput) {
835 vp->v_flag |= VBWAIT;
836 error = tsleep((caddr_t)&vp->v_numoutput,
377d4740 837 slpflag, "vinvlbuf", slptimeo);
984263bc
MD
838 if (error) {
839 splx(s);
840 return (error);
841 }
842 }
843 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
844 splx(s);
3b568787 845 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
984263bc
MD
846 return (error);
847 s = splbio();
848 if (vp->v_numoutput > 0 ||
849 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
850 panic("vinvalbuf: dirty bufs");
851 }
852 splx(s);
853 }
854 s = splbio();
855 for (;;) {
856 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
857 if (!blist)
858 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
859 if (!blist)
860 break;
861
862 for (bp = blist; bp; bp = nbp) {
863 nbp = TAILQ_NEXT(bp, b_vnbufs);
864 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
865 error = BUF_TIMELOCK(bp,
866 LK_EXCLUSIVE | LK_SLEEPFAIL,
867 "vinvalbuf", slpflag, slptimeo);
868 if (error == ENOLCK)
869 break;
870 splx(s);
871 return (error);
872 }
873 /*
874 * XXX Since there are no node locks for NFS, I
875 * believe there is a slight chance that a delayed
876 * write will occur while sleeping just above, so
877 * check for it. Note that vfs_bio_awrite expects
878 * buffers to reside on a queue, while VOP_BWRITE and
879 * brelse do not.
880 */
881 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
882 (flags & V_SAVE)) {
883
884 if (bp->b_vp == vp) {
885 if (bp->b_flags & B_CLUSTEROK) {
886 BUF_UNLOCK(bp);
887 vfs_bio_awrite(bp);
888 } else {
889 bremfree(bp);
890 bp->b_flags |= B_ASYNC;
891 VOP_BWRITE(bp->b_vp, bp);
892 }
893 } else {
894 bremfree(bp);
895 (void) VOP_BWRITE(bp->b_vp, bp);
896 }
897 break;
898 }
899 bremfree(bp);
900 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
901 bp->b_flags &= ~B_ASYNC;
902 brelse(bp);
903 }
904 }
905
906 /*
907 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
908 * have write I/O in-progress but if there is a VM object then the
909 * VM object can also have read-I/O in-progress.
910 */
911 do {
912 while (vp->v_numoutput > 0) {
913 vp->v_flag |= VBWAIT;
377d4740 914 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
984263bc
MD
915 }
916 if (VOP_GETVOBJECT(vp, &object) == 0) {
917 while (object->paging_in_progress)
918 vm_object_pip_sleep(object, "vnvlbx");
919 }
920 } while (vp->v_numoutput > 0);
921
922 splx(s);
923
924 /*
925 * Destroy the copy in the VM cache, too.
926 */
8a8d5d85 927 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
928 if (VOP_GETVOBJECT(vp, &object) == 0) {
929 vm_object_page_remove(object, 0, 0,
930 (flags & V_SAVE) ? TRUE : FALSE);
931 }
8a8d5d85 932 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
933
934 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
935 panic("vinvalbuf: flush failed");
936 return (0);
937}
938
939/*
940 * Truncate a file's buffer and pages to a specified length. This
941 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
942 * sync activity.
943 */
944int
3b568787 945vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
984263bc 946{
dadab5e9 947 struct buf *bp;
984263bc
MD
948 struct buf *nbp;
949 int s, anyfreed;
950 int trunclbn;
951
952 /*
953 * Round up to the *next* lbn.
954 */
955 trunclbn = (length + blksize - 1) / blksize;
956
957 s = splbio();
958restart:
959 anyfreed = 1;
960 for (;anyfreed;) {
961 anyfreed = 0;
962 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
963 nbp = TAILQ_NEXT(bp, b_vnbufs);
964 if (bp->b_lblkno >= trunclbn) {
965 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
966 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
967 goto restart;
968 } else {
969 bremfree(bp);
970 bp->b_flags |= (B_INVAL | B_RELBUF);
971 bp->b_flags &= ~B_ASYNC;
972 brelse(bp);
973 anyfreed = 1;
974 }
975 if (nbp &&
976 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
977 (nbp->b_vp != vp) ||
978 (nbp->b_flags & B_DELWRI))) {
979 goto restart;
980 }
981 }
982 }
983
984 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
985 nbp = TAILQ_NEXT(bp, b_vnbufs);
986 if (bp->b_lblkno >= trunclbn) {
987 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
988 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
989 goto restart;
990 } else {
991 bremfree(bp);
992 bp->b_flags |= (B_INVAL | B_RELBUF);
993 bp->b_flags &= ~B_ASYNC;
994 brelse(bp);
995 anyfreed = 1;
996 }
997 if (nbp &&
998 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
999 (nbp->b_vp != vp) ||
1000 (nbp->b_flags & B_DELWRI) == 0)) {
1001 goto restart;
1002 }
1003 }
1004 }
1005 }
1006
1007 if (length > 0) {
1008restartsync:
1009 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1010 nbp = TAILQ_NEXT(bp, b_vnbufs);
1011 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1012 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1013 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1014 goto restart;
1015 } else {
1016 bremfree(bp);
1017 if (bp->b_vp == vp) {
1018 bp->b_flags |= B_ASYNC;
1019 } else {
1020 bp->b_flags &= ~B_ASYNC;
1021 }
1022 VOP_BWRITE(bp->b_vp, bp);
1023 }
1024 goto restartsync;
1025 }
1026
1027 }
1028 }
1029
1030 while (vp->v_numoutput > 0) {
1031 vp->v_flag |= VBWAIT;
377d4740 1032 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
984263bc
MD
1033 }
1034
1035 splx(s);
1036
1037 vnode_pager_setsize(vp, length);
1038
1039 return (0);
1040}
1041
1042/*
1043 * Associate a buffer with a vnode.
1044 */
1045void
1046bgetvp(vp, bp)
1fd87d54
RG
1047 struct vnode *vp;
1048 struct buf *bp;
984263bc
MD
1049{
1050 int s;
1051
1052 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1053
1054 vhold(vp);
1055 bp->b_vp = vp;
1056 bp->b_dev = vn_todev(vp);
1057 /*
1058 * Insert onto list for new vnode.
1059 */
1060 s = splbio();
1061 bp->b_xflags |= BX_VNCLEAN;
1062 bp->b_xflags &= ~BX_VNDIRTY;
1063 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1064 splx(s);
1065}
1066
1067/*
1068 * Disassociate a buffer from a vnode.
1069 */
1070void
1071brelvp(bp)
1fd87d54 1072 struct buf *bp;
984263bc
MD
1073{
1074 struct vnode *vp;
1075 struct buflists *listheadp;
1076 int s;
1077
1078 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1079
1080 /*
1081 * Delete from old vnode list, if on one.
1082 */
1083 vp = bp->b_vp;
1084 s = splbio();
1085 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1086 if (bp->b_xflags & BX_VNDIRTY)
1087 listheadp = &vp->v_dirtyblkhd;
1088 else
1089 listheadp = &vp->v_cleanblkhd;
1090 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1091 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1092 }
1093 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1094 vp->v_flag &= ~VONWORKLST;
1095 LIST_REMOVE(vp, v_synclist);
1096 }
1097 splx(s);
1098 bp->b_vp = (struct vnode *) 0;
1099 vdrop(vp);
1100}
1101
1102/*
1103 * The workitem queue.
1104 *
1105 * It is useful to delay writes of file data and filesystem metadata
1106 * for tens of seconds so that quickly created and deleted files need
1107 * not waste disk bandwidth being created and removed. To realize this,
1108 * we append vnodes to a "workitem" queue. When running with a soft
1109 * updates implementation, most pending metadata dependencies should
1110 * not wait for more than a few seconds. Thus, mounted on block devices
1111 * are delayed only about a half the time that file data is delayed.
1112 * Similarly, directory updates are more critical, so are only delayed
1113 * about a third the time that file data is delayed. Thus, there are
1114 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1115 * one each second (driven off the filesystem syncer process). The
1116 * syncer_delayno variable indicates the next queue that is to be processed.
1117 * Items that need to be processed soon are placed in this queue:
1118 *
1119 * syncer_workitem_pending[syncer_delayno]
1120 *
1121 * A delay of fifteen seconds is done by placing the request fifteen
1122 * entries later in the queue:
1123 *
1124 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1125 *
1126 */
1127
1128/*
1129 * Add an item to the syncer work queue.
1130 */
1131static void
1132vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1133{
1134 int s, slot;
1135
1136 s = splbio();
1137
1138 if (vp->v_flag & VONWORKLST) {
1139 LIST_REMOVE(vp, v_synclist);
1140 }
1141
1142 if (delay > syncer_maxdelay - 2)
1143 delay = syncer_maxdelay - 2;
1144 slot = (syncer_delayno + delay) & syncer_mask;
1145
1146 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1147 vp->v_flag |= VONWORKLST;
1148 splx(s);
1149}
1150
bc6dffab 1151struct thread *updatethread;
402ed7e1 1152static void sched_sync (void);
984263bc
MD
1153static struct kproc_desc up_kp = {
1154 "syncer",
1155 sched_sync,
bc6dffab 1156 &updatethread
984263bc
MD
1157};
1158SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1159
1160/*
1161 * System filesystem synchronizer daemon.
1162 */
1163void
1164sched_sync(void)
1165{
1166 struct synclist *slp;
1167 struct vnode *vp;
1168 long starttime;
1169 int s;
0cfcada1 1170 struct thread *td = curthread;
984263bc 1171
bc6dffab 1172 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
1173 SHUTDOWN_PRI_LAST);
1174
1175 for (;;) {
0cfcada1 1176 kproc_suspend_loop();
984263bc
MD
1177
1178 starttime = time_second;
1179
1180 /*
1181 * Push files whose dirty time has expired. Be careful
1182 * of interrupt race on slp queue.
1183 */
1184 s = splbio();
1185 slp = &syncer_workitem_pending[syncer_delayno];
1186 syncer_delayno += 1;
1187 if (syncer_delayno == syncer_maxdelay)
1188 syncer_delayno = 0;
1189 splx(s);
1190
1191 while ((vp = LIST_FIRST(slp)) != NULL) {
1192 if (VOP_ISLOCKED(vp, NULL) == 0) {
dadab5e9 1193 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3b568787 1194 (void) VOP_FSYNC(vp, MNT_LAZY, td);
dadab5e9 1195 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1196 }
1197 s = splbio();
1198 if (LIST_FIRST(slp) == vp) {
1199 /*
1200 * Note: v_tag VT_VFS vps can remain on the
1201 * worklist too with no dirty blocks, but
1202 * since sync_fsync() moves it to a different
1203 * slot we are safe.
1204 */
1205 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1206 !vn_isdisk(vp, NULL))
1207 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1208 /*
1209 * Put us back on the worklist. The worklist
1210 * routine will remove us from our current
1211 * position and then add us back in at a later
1212 * position.
1213 */
1214 vn_syncer_add_to_worklist(vp, syncdelay);
1215 }
1216 splx(s);
1217 }
1218
1219 /*
1220 * Do soft update processing.
1221 */
1222 if (bioops.io_sync)
1223 (*bioops.io_sync)(NULL);
1224
1225 /*
1226 * The variable rushjob allows the kernel to speed up the
1227 * processing of the filesystem syncer process. A rushjob
1228 * value of N tells the filesystem syncer to process the next
1229 * N seconds worth of work on its queue ASAP. Currently rushjob
1230 * is used by the soft update code to speed up the filesystem
1231 * syncer process when the incore state is getting so far
1232 * ahead of the disk that the kernel memory pool is being
1233 * threatened with exhaustion.
1234 */
1235 if (rushjob > 0) {
1236 rushjob -= 1;
1237 continue;
1238 }
1239 /*
1240 * If it has taken us less than a second to process the
1241 * current work, then wait. Otherwise start right over
1242 * again. We can still lose time if any single round
1243 * takes more than two seconds, but it does not really
1244 * matter as we are just trying to generally pace the
1245 * filesystem activity.
1246 */
1247 if (time_second == starttime)
377d4740 1248 tsleep(&lbolt, 0, "syncer", 0);
984263bc
MD
1249 }
1250}
1251
1252/*
1253 * Request the syncer daemon to speed up its work.
1254 * We never push it to speed up more than half of its
1255 * normal turn time, otherwise it could take over the cpu.
f5865223
MD
1256 *
1257 * YYY wchan field protected by the BGL.
984263bc
MD
1258 */
1259int
1260speedup_syncer()
1261{
f5865223
MD
1262 crit_enter();
1263 if (updatethread->td_wchan == &lbolt) { /* YYY */
1264 unsleep(updatethread);
1265 lwkt_schedule(updatethread);
1266 }
1267 crit_exit();
984263bc
MD
1268 if (rushjob < syncdelay / 2) {
1269 rushjob += 1;
1270 stat_rush_requests += 1;
1271 return (1);
1272 }
1273 return(0);
1274}
1275
1276/*
1277 * Associate a p-buffer with a vnode.
1278 *
1279 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1280 * with the buffer. i.e. the bp has not been linked into the vnode or
1281 * ref-counted.
1282 */
1283void
1284pbgetvp(vp, bp)
1fd87d54
RG
1285 struct vnode *vp;
1286 struct buf *bp;
984263bc
MD
1287{
1288
1289 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1290
1291 bp->b_vp = vp;
1292 bp->b_flags |= B_PAGING;
1293 bp->b_dev = vn_todev(vp);
1294}
1295
1296/*
1297 * Disassociate a p-buffer from a vnode.
1298 */
1299void
1300pbrelvp(bp)
1fd87d54 1301 struct buf *bp;
984263bc
MD
1302{
1303
1304 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1305
1306 /* XXX REMOVE ME */
1307 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1308 panic(
1309 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1310 bp,
1311 (int)bp->b_flags
1312 );
1313 }
1314 bp->b_vp = (struct vnode *) 0;
1315 bp->b_flags &= ~B_PAGING;
1316}
1317
1318void
1319pbreassignbuf(bp, newvp)
1320 struct buf *bp;
1321 struct vnode *newvp;
1322{
1323 if ((bp->b_flags & B_PAGING) == 0) {
1324 panic(
1325 "pbreassignbuf() on non phys bp %p",
1326 bp
1327 );
1328 }
1329 bp->b_vp = newvp;
1330}
1331
1332/*
1333 * Reassign a buffer from one vnode to another.
1334 * Used to assign file specific control information
1335 * (indirect blocks) to the vnode to which they belong.
1336 */
1337void
1338reassignbuf(bp, newvp)
1fd87d54
RG
1339 struct buf *bp;
1340 struct vnode *newvp;
984263bc
MD
1341{
1342 struct buflists *listheadp;
1343 int delay;
1344 int s;
1345
1346 if (newvp == NULL) {
1347 printf("reassignbuf: NULL");
1348 return;
1349 }
1350 ++reassignbufcalls;
1351
1352 /*
1353 * B_PAGING flagged buffers cannot be reassigned because their vp
1354 * is not fully linked in.
1355 */
1356 if (bp->b_flags & B_PAGING)
1357 panic("cannot reassign paging buffer");
1358
1359 s = splbio();
1360 /*
1361 * Delete from old vnode list, if on one.
1362 */
1363 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1364 if (bp->b_xflags & BX_VNDIRTY)
1365 listheadp = &bp->b_vp->v_dirtyblkhd;
1366 else
1367 listheadp = &bp->b_vp->v_cleanblkhd;
1368 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1369 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1370 if (bp->b_vp != newvp) {
1371 vdrop(bp->b_vp);
1372 bp->b_vp = NULL; /* for clarification */
1373 }
1374 }
1375 /*
1376 * If dirty, put on list of dirty buffers; otherwise insert onto list
1377 * of clean buffers.
1378 */
1379 if (bp->b_flags & B_DELWRI) {
1380 struct buf *tbp;
1381
1382 listheadp = &newvp->v_dirtyblkhd;
1383 if ((newvp->v_flag & VONWORKLST) == 0) {
1384 switch (newvp->v_type) {
1385 case VDIR:
1386 delay = dirdelay;
1387 break;
1388 case VCHR:
1389 case VBLK:
1390 if (newvp->v_specmountpoint != NULL) {
1391 delay = metadelay;
1392 break;
1393 }
1394 /* fall through */
1395 default:
1396 delay = filedelay;
1397 }
1398 vn_syncer_add_to_worklist(newvp, delay);
1399 }
1400 bp->b_xflags |= BX_VNDIRTY;
1401 tbp = TAILQ_FIRST(listheadp);
1402 if (tbp == NULL ||
1403 bp->b_lblkno == 0 ||
1404 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1405 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1406 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1407 ++reassignbufsortgood;
1408 } else if (bp->b_lblkno < 0) {
1409 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1410 ++reassignbufsortgood;
1411 } else if (reassignbufmethod == 1) {
1412 /*
1413 * New sorting algorithm, only handle sequential case,
1414 * otherwise append to end (but before metadata)
1415 */
1416 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1417 (tbp->b_xflags & BX_VNDIRTY)) {
1418 /*
1419 * Found the best place to insert the buffer
1420 */
1421 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1422 ++reassignbufsortgood;
1423 } else {
1424 /*
1425 * Missed, append to end, but before meta-data.
1426 * We know that the head buffer in the list is
1427 * not meta-data due to prior conditionals.
1428 *
1429 * Indirect effects: NFS second stage write
1430 * tends to wind up here, giving maximum
1431 * distance between the unstable write and the
1432 * commit rpc.
1433 */
1434 tbp = TAILQ_LAST(listheadp, buflists);
1435 while (tbp && tbp->b_lblkno < 0)
1436 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1437 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1438 ++reassignbufsortbad;
1439 }
1440 } else {
1441 /*
1442 * Old sorting algorithm, scan queue and insert
1443 */
1444 struct buf *ttbp;
1445 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1446 (ttbp->b_lblkno < bp->b_lblkno)) {
1447 ++reassignbufloops;
1448 tbp = ttbp;
1449 }
1450 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1451 }
1452 } else {
1453 bp->b_xflags |= BX_VNCLEAN;
1454 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1455 if ((newvp->v_flag & VONWORKLST) &&
1456 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1457 newvp->v_flag &= ~VONWORKLST;
1458 LIST_REMOVE(newvp, v_synclist);
1459 }
1460 }
1461 if (bp->b_vp != newvp) {
1462 bp->b_vp = newvp;
1463 vhold(bp->b_vp);
1464 }
1465 splx(s);
1466}
1467
1468/*
1469 * Create a vnode for a block device.
1470 * Used for mounting the root file system.
1471 */
1472int
1473bdevvp(dev, vpp)
1474 dev_t dev;
1475 struct vnode **vpp;
1476{
1fd87d54 1477 struct vnode *vp;
984263bc
MD
1478 struct vnode *nvp;
1479 int error;
1480
1481 if (dev == NODEV) {
1482 *vpp = NULLVP;
1483 return (ENXIO);
1484 }
1485 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1486 if (error) {
1487 *vpp = NULLVP;
1488 return (error);
1489 }
1490 vp = nvp;
1491 vp->v_type = VBLK;
1492 addalias(vp, dev);
1493 *vpp = vp;
1494 return (0);
1495}
1496
1497/*
7732c8cd 1498 * Add a vnode to the alias list hung off the dev_t.
984263bc
MD
1499 *
1500 * The reason for this gunk is that multiple vnodes can reference
1501 * the same physical device, so checking vp->v_usecount to see
1502 * how many users there are is inadequate; the v_usecount for
1503 * the vnodes need to be accumulated. vcount() does that.
1504 */
1505void
7732c8cd 1506addaliasu(struct vnode *nvp, udev_t nvp_rdev)
984263bc 1507{
7732c8cd 1508 dev_t dev;
984263bc
MD
1509
1510 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1511 panic("addaliasu on non-special vnode");
7732c8cd 1512 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
4908c260
JH
1513 if (dev != NODEV) {
1514 nvp->v_rdev = dev;
7732c8cd 1515 addalias(nvp, dev);
4908c260
JH
1516 } else
1517 nvp->v_rdev = NULL;
984263bc
MD
1518}
1519
1520void
7732c8cd 1521addalias(struct vnode *nvp, dev_t dev)
984263bc
MD
1522{
1523
1524 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1525 panic("addalias on non-special vnode");
1526
1527 nvp->v_rdev = dev;
8a8d5d85 1528 lwkt_gettoken(&spechash_token);
984263bc 1529 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
8a8d5d85 1530 lwkt_reltoken(&spechash_token);
984263bc
MD
1531}
1532
1533/*
1534 * Grab a particular vnode from the free list, increment its
1535 * reference count and lock it. The vnode lock bit is set if the
1536 * vnode is being eliminated in vgone. The process is awakened
1537 * when the transition is completed, and an error returned to
1538 * indicate that the vnode is no longer usable (possibly having
1539 * been changed to a new file system type).
1540 */
1541int
dadab5e9
MD
1542vget(vp, flags, td)
1543 struct vnode *vp;
984263bc 1544 int flags;
dadab5e9 1545 struct thread *td;
984263bc
MD
1546{
1547 int error;
1548
1549 /*
1550 * If the vnode is in the process of being cleaned out for
1551 * another use, we wait for the cleaning to finish and then
1552 * return failure. Cleaning is determined by checking that
1553 * the VXLOCK flag is set.
1554 */
1555 if ((flags & LK_INTERLOCK) == 0) {
8a8d5d85 1556 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1557 }
1558 if (vp->v_flag & VXLOCK) {
1559 if (vp->v_vxproc == curproc) {
1560#if 0
1561 /* this can now occur in normal operation */
1562 log(LOG_INFO, "VXLOCK interlock avoided\n");
1563#endif
1564 } else {
1565 vp->v_flag |= VXWANT;
8a8d5d85 1566 lwkt_reltoken(&vp->v_interlock);
377d4740 1567 tsleep((caddr_t)vp, 0, "vget", 0);
984263bc
MD
1568 return (ENOENT);
1569 }
1570 }
1571
1572 vp->v_usecount++;
1573
1574 if (VSHOULDBUSY(vp))
1575 vbusy(vp);
1576 if (flags & LK_TYPE_MASK) {
dadab5e9 1577 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
984263bc
MD
1578 /*
1579 * must expand vrele here because we do not want
1580 * to call VOP_INACTIVE if the reference count
1581 * drops back to zero since it was never really
1582 * active. We must remove it from the free list
1583 * before sleeping so that multiple processes do
1584 * not try to recycle it.
1585 */
8a8d5d85 1586 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1587 vp->v_usecount--;
1588 if (VSHOULDFREE(vp))
1589 vfree(vp);
1590 else
1591 vlruvp(vp);
8a8d5d85 1592 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1593 }
1594 return (error);
1595 }
8a8d5d85 1596 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1597 return (0);
1598}
1599
1600void
1601vref(struct vnode *vp)
1602{
8a8d5d85 1603 lwkt_gettoken(&vp->v_interlock);
984263bc 1604 vp->v_usecount++;
8a8d5d85 1605 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1606}
1607
1608/*
1609 * Vnode put/release.
1610 * If count drops to zero, call inactive routine and return to freelist.
1611 */
1612void
dadab5e9 1613vrele(struct vnode *vp)
984263bc 1614{
dadab5e9 1615 struct thread *td = curthread; /* XXX */
984263bc
MD
1616
1617 KASSERT(vp != NULL, ("vrele: null vp"));
1618
8a8d5d85 1619 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1620
1621 if (vp->v_usecount > 1) {
1622
1623 vp->v_usecount--;
8a8d5d85 1624 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1625
1626 return;
1627 }
1628
1629 if (vp->v_usecount == 1) {
1630 vp->v_usecount--;
1631 /*
1632 * We must call VOP_INACTIVE with the node locked.
1633 * If we are doing a vpu, the node is already locked,
1634 * but, in the case of vrele, we must explicitly lock
1635 * the vnode before calling VOP_INACTIVE
1636 */
1637
dadab5e9
MD
1638 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1639 VOP_INACTIVE(vp, td);
984263bc
MD
1640 if (VSHOULDFREE(vp))
1641 vfree(vp);
1642 else
1643 vlruvp(vp);
1644 } else {
1645#ifdef DIAGNOSTIC
1646 vprint("vrele: negative ref count", vp);
8a8d5d85 1647 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1648#endif
1649 panic("vrele: negative ref cnt");
1650 }
1651}
1652
1653void
dadab5e9 1654vput(struct vnode *vp)
984263bc 1655{
dadab5e9 1656 struct thread *td = curthread; /* XXX */
984263bc
MD
1657
1658 KASSERT(vp != NULL, ("vput: null vp"));
1659
8a8d5d85 1660 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1661
1662 if (vp->v_usecount > 1) {
1663 vp->v_usecount--;
dadab5e9 1664 VOP_UNLOCK(vp, LK_INTERLOCK, td);
984263bc
MD
1665 return;
1666 }
1667
1668 if (vp->v_usecount == 1) {
1669 vp->v_usecount--;
1670 /*
1671 * We must call VOP_INACTIVE with the node locked.
1672 * If we are doing a vpu, the node is already locked,
1673 * so we just need to release the vnode mutex.
1674 */
8a8d5d85 1675 lwkt_reltoken(&vp->v_interlock);
dadab5e9 1676 VOP_INACTIVE(vp, td);
984263bc
MD
1677 if (VSHOULDFREE(vp))
1678 vfree(vp);
1679 else
1680 vlruvp(vp);
1681 } else {
1682#ifdef DIAGNOSTIC
1683 vprint("vput: negative ref count", vp);
1684#endif
1685 panic("vput: negative ref cnt");
1686 }
1687}
1688
1689/*
1690 * Somebody doesn't want the vnode recycled.
1691 */
1692void
1693vhold(vp)
1fd87d54 1694 struct vnode *vp;
984263bc
MD
1695{
1696 int s;
1697
1698 s = splbio();
1699 vp->v_holdcnt++;
1700 if (VSHOULDBUSY(vp))
1701 vbusy(vp);
1702 splx(s);
1703}
1704
1705/*
1706 * One less who cares about this vnode.
1707 */
1708void
1709vdrop(vp)
1fd87d54 1710 struct vnode *vp;
984263bc
MD
1711{
1712 int s;
1713
1714 s = splbio();
1715 if (vp->v_holdcnt <= 0)
1716 panic("vdrop: holdcnt");
1717 vp->v_holdcnt--;
1718 if (VSHOULDFREE(vp))
1719 vfree(vp);
1720 splx(s);
1721}
1722
1723/*
1724 * Remove any vnodes in the vnode table belonging to mount point mp.
1725 *
1726 * If FORCECLOSE is not specified, there should not be any active ones,
1727 * return error if any are found (nb: this is a user error, not a
1728 * system error). If FORCECLOSE is specified, detach any active vnodes
1729 * that are found.
1730 *
1731 * If WRITECLOSE is set, only flush out regular file vnodes open for
1732 * writing.
1733 *
1734 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1735 *
1736 * `rootrefs' specifies the base reference count for the root vnode
1737 * of this filesystem. The root vnode is considered busy if its
1738 * v_usecount exceeds this value. On a successful return, vflush()
1739 * will call vrele() on the root vnode exactly rootrefs times.
1740 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1741 * be zero.
1742 */
1743#ifdef DIAGNOSTIC
1744static int busyprt = 0; /* print out busy vnodes */
1745SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1746#endif
1747
1748int
1749vflush(mp, rootrefs, flags)
1750 struct mount *mp;
1751 int rootrefs;
1752 int flags;
1753{
dadab5e9 1754 struct thread *td = curthread; /* XXX */
984263bc
MD
1755 struct vnode *vp, *nvp, *rootvp = NULL;
1756 struct vattr vattr;
1757 int busy = 0, error;
1758
1759 if (rootrefs > 0) {
1760 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1761 ("vflush: bad args"));
1762 /*
1763 * Get the filesystem root vnode. We can vput() it
1764 * immediately, since with rootrefs > 0, it won't go away.
1765 */
1766 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1767 return (error);
1768 vput(rootvp);
1769 }
8a8d5d85 1770 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1771loop:
1772 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1773 /*
1774 * Make sure this vnode wasn't reclaimed in getnewvnode().
1775 * Start over if it has (it won't be on the list anymore).
1776 */
1777 if (vp->v_mount != mp)
1778 goto loop;
1779 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1780
8a8d5d85 1781 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1782 /*
1783 * Skip over a vnodes marked VSYSTEM.
1784 */
1785 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
8a8d5d85 1786 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1787 continue;
1788 }
1789 /*
1790 * If WRITECLOSE is set, flush out unlinked but still open
1791 * files (even if open only for reading) and regular file
1792 * vnodes open for writing.
1793 */
1794 if ((flags & WRITECLOSE) &&
1795 (vp->v_type == VNON ||
3b568787 1796 (VOP_GETATTR(vp, &vattr, td) == 0 &&
984263bc
MD
1797 vattr.va_nlink > 0)) &&
1798 (vp->v_writecount == 0 || vp->v_type != VREG)) {
8a8d5d85 1799 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1800 continue;
1801 }
1802
1803 /*
1804 * With v_usecount == 0, all we need to do is clear out the
1805 * vnode data structures and we are done.
1806 */
1807 if (vp->v_usecount == 0) {
8a8d5d85 1808 lwkt_reltoken(&mntvnode_token);
dadab5e9 1809 vgonel(vp, td);
8a8d5d85 1810 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1811 continue;
1812 }
1813
1814 /*
1815 * If FORCECLOSE is set, forcibly close the vnode. For block
1816 * or character devices, revert to an anonymous device. For
1817 * all other files, just kill them.
1818 */
1819 if (flags & FORCECLOSE) {
8a8d5d85 1820 lwkt_reltoken(&mntvnode_token);
984263bc 1821 if (vp->v_type != VBLK && vp->v_type != VCHR) {
dadab5e9 1822 vgonel(vp, td);
984263bc 1823 } else {
dadab5e9 1824 vclean(vp, 0, td);
984263bc
MD
1825 vp->v_op = spec_vnodeop_p;
1826 insmntque(vp, (struct mount *) 0);
1827 }
8a8d5d85 1828 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1829 continue;
1830 }
1831#ifdef DIAGNOSTIC
1832 if (busyprt)
1833 vprint("vflush: busy vnode", vp);
1834#endif
8a8d5d85 1835 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1836 busy++;
1837 }
8a8d5d85 1838 lwkt_reltoken(&mntvnode_token);
984263bc
MD
1839 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1840 /*
1841 * If just the root vnode is busy, and if its refcount
1842 * is equal to `rootrefs', then go ahead and kill it.
1843 */
8a8d5d85 1844 lwkt_gettoken(&rootvp->v_interlock);
984263bc
MD
1845 KASSERT(busy > 0, ("vflush: not busy"));
1846 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1847 if (busy == 1 && rootvp->v_usecount == rootrefs) {
dadab5e9 1848 vgonel(rootvp, td);
984263bc
MD
1849 busy = 0;
1850 } else
8a8d5d85 1851 lwkt_reltoken(&rootvp->v_interlock);
984263bc
MD
1852 }
1853 if (busy)
1854 return (EBUSY);
1855 for (; rootrefs > 0; rootrefs--)
1856 vrele(rootvp);
1857 return (0);
1858}
1859
1860/*
1861 * We do not want to recycle the vnode too quickly.
1862 *
1863 * XXX we can't move vp's around the nvnodelist without really screwing
1864 * up the efficiency of filesystem SYNC and friends. This code is
1865 * disabled until we fix the syncing code's scanning algorithm.
1866 */
1867static void
1868vlruvp(struct vnode *vp)
1869{
1870#if 0
1871 struct mount *mp;
1872
1873 if ((mp = vp->v_mount) != NULL) {
8a8d5d85 1874 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1875 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1876 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
8a8d5d85 1877 lwkt_reltoken(&mntvnode_token);
984263bc
MD
1878 }
1879#endif
1880}
1881
1882/*
1883 * Disassociate the underlying file system from a vnode.
1884 */
1885static void
dadab5e9 1886vclean(struct vnode *vp, int flags, struct thread *td)
984263bc
MD
1887{
1888 int active;
1889
1890 /*
1891 * Check to see if the vnode is in use. If so we have to reference it
1892 * before we clean it out so that its count cannot fall to zero and
1893 * generate a race against ourselves to recycle it.
1894 */
1895 if ((active = vp->v_usecount))
1896 vp->v_usecount++;
1897
1898 /*
1899 * Prevent the vnode from being recycled or brought into use while we
1900 * clean it out.
1901 */
1902 if (vp->v_flag & VXLOCK)
1903 panic("vclean: deadlock");
1904 vp->v_flag |= VXLOCK;
1905 vp->v_vxproc = curproc;
1906 /*
1907 * Even if the count is zero, the VOP_INACTIVE routine may still
1908 * have the object locked while it cleans it out. The VOP_LOCK
1909 * ensures that the VOP_INACTIVE routine is done with its work.
1910 * For active vnodes, it ensures that no other activity can
1911 * occur while the underlying object is being cleaned out.
1912 */
dadab5e9 1913 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
984263bc
MD
1914
1915 /*
1916 * Clean out any buffers associated with the vnode.
1917 */
3b568787 1918 vinvalbuf(vp, V_SAVE, td, 0, 0);
984263bc
MD
1919
1920 VOP_DESTROYVOBJECT(vp);
1921
1922 /*
1923 * If purging an active vnode, it must be closed and
1924 * deactivated before being reclaimed. Note that the
1925 * VOP_INACTIVE will unlock the vnode.
1926 */
1927 if (active) {
1928 if (flags & DOCLOSE)
3b568787 1929 VOP_CLOSE(vp, FNONBLOCK, td);
dadab5e9 1930 VOP_INACTIVE(vp, td);
984263bc
MD
1931 } else {
1932 /*
1933 * Any other processes trying to obtain this lock must first
1934 * wait for VXLOCK to clear, then call the new lock operation.
1935 */
dadab5e9 1936 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1937 }
1938 /*
1939 * Reclaim the vnode.
1940 */
dadab5e9 1941 if (VOP_RECLAIM(vp, td))
984263bc
MD
1942 panic("vclean: cannot reclaim");
1943
1944 if (active) {
1945 /*
1946 * Inline copy of vrele() since VOP_INACTIVE
1947 * has already been called.
1948 */
8a8d5d85 1949 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1950 if (--vp->v_usecount <= 0) {
1951#ifdef DIAGNOSTIC
1952 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1953 vprint("vclean: bad ref count", vp);
1954 panic("vclean: ref cnt");
1955 }
1956#endif
1957 vfree(vp);
1958 }
8a8d5d85 1959 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1960 }
1961
1962 cache_purge(vp);
1963 vp->v_vnlock = NULL;
1964
1965 if (VSHOULDFREE(vp))
1966 vfree(vp);
1967
1968 /*
1969 * Done with purge, notify sleepers of the grim news.
1970 */
1971 vp->v_op = dead_vnodeop_p;
1972 vn_pollgone(vp);
1973 vp->v_tag = VT_NON;
1974 vp->v_flag &= ~VXLOCK;
1975 vp->v_vxproc = NULL;
1976 if (vp->v_flag & VXWANT) {
1977 vp->v_flag &= ~VXWANT;
1978 wakeup((caddr_t) vp);
1979 }
1980}
1981
1982/*
1983 * Eliminate all activity associated with the requested vnode
1984 * and with all vnodes aliased to the requested vnode.
1985 */
1986int
1987vop_revoke(ap)
1988 struct vop_revoke_args /* {
1989 struct vnode *a_vp;
1990 int a_flags;
1991 } */ *ap;
1992{
1993 struct vnode *vp, *vq;
1994 dev_t dev;
1995
1996 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1997
1998 vp = ap->a_vp;
1999 /*
2000 * If a vgone (or vclean) is already in progress,
2001 * wait until it is done and return.
2002 */
2003 if (vp->v_flag & VXLOCK) {
2004 vp->v_flag |= VXWANT;
8a8d5d85 2005 lwkt_reltoken(&vp->v_interlock);
377d4740 2006 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
984263bc
MD
2007 return (0);
2008 }
2009 dev = vp->v_rdev;
2010 for (;;) {
8a8d5d85 2011 lwkt_gettoken(&spechash_token);
984263bc 2012 vq = SLIST_FIRST(&dev->si_hlist);
8a8d5d85 2013 lwkt_reltoken(&spechash_token);
984263bc
MD
2014 if (!vq)
2015 break;
2016 vgone(vq);
2017 }
2018 return (0);
2019}
2020
2021/*
2022 * Recycle an unused vnode to the front of the free list.
2023 * Release the passed interlock if the vnode will be recycled.
2024 */
2025int
8a8d5d85 2026vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td)
984263bc 2027{
8a8d5d85 2028 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
2029 if (vp->v_usecount == 0) {
2030 if (inter_lkp) {
8a8d5d85 2031 lwkt_reltoken(inter_lkp);
984263bc 2032 }
dadab5e9 2033 vgonel(vp, td);
984263bc
MD
2034 return (1);
2035 }
8a8d5d85 2036 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2037 return (0);
2038}
2039
2040/*
2041 * Eliminate all activity associated with a vnode
2042 * in preparation for reuse.
2043 */
2044void
dadab5e9 2045vgone(struct vnode *vp)
984263bc 2046{
dadab5e9 2047 struct thread *td = curthread; /* XXX */
984263bc 2048
8a8d5d85 2049 lwkt_gettoken(&vp->v_interlock);
dadab5e9 2050 vgonel(vp, td);
984263bc
MD
2051}
2052
2053/*
2054 * vgone, with the vp interlock held.
2055 */
2056void
dadab5e9 2057vgonel(struct vnode *vp, struct thread *td)
984263bc
MD
2058{
2059 int s;
2060
2061 /*
2062 * If a vgone (or vclean) is already in progress,
2063 * wait until it is done and return.
2064 */
2065 if (vp->v_flag & VXLOCK) {
2066 vp->v_flag |= VXWANT;
8a8d5d85 2067 lwkt_reltoken(&vp->v_interlock);
377d4740 2068 tsleep((caddr_t)vp, 0, "vgone", 0);
984263bc
MD
2069 return;
2070 }
2071
2072 /*
2073 * Clean out the filesystem specific data.
2074 */
dadab5e9 2075 vclean(vp, DOCLOSE, td);
8a8d5d85 2076 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
2077
2078 /*
2079 * Delete from old mount point vnode list, if on one.
2080 */
2081 if (vp->v_mount != NULL)
2082 insmntque(vp, (struct mount *)0);
2083 /*
2084 * If special device, remove it from special device alias list
2085 * if it is on one.
2086 */
2087 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
8a8d5d85 2088 lwkt_gettoken(&spechash_token);
984263bc
MD
2089 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2090 freedev(vp->v_rdev);
8a8d5d85 2091 lwkt_reltoken(&spechash_token);
984263bc
MD
2092 vp->v_rdev = NULL;
2093 }
2094
2095 /*
2096 * If it is on the freelist and not already at the head,
2097 * move it to the head of the list. The test of the
2098 * VDOOMED flag and the reference count of zero is because
2099 * it will be removed from the free list by getnewvnode,
2100 * but will not have its reference count incremented until
2101 * after calling vgone. If the reference count were
2102 * incremented first, vgone would (incorrectly) try to
2103 * close the previous instance of the underlying object.
2104 */
2105 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2106 s = splbio();
8a8d5d85 2107 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2108 if (vp->v_flag & VFREE)
2109 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2110 else
2111 freevnodes++;
2112 vp->v_flag |= VFREE;
2113 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
8a8d5d85 2114 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2115 splx(s);
2116 }
2117
2118 vp->v_type = VBAD;
8a8d5d85 2119 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2120}
2121
2122/*
2123 * Lookup a vnode by device number.
2124 */
2125int
2126vfinddev(dev, type, vpp)
2127 dev_t dev;
2128 enum vtype type;
2129 struct vnode **vpp;
2130{
2131 struct vnode *vp;
2132
8a8d5d85 2133 lwkt_gettoken(&spechash_token);
984263bc
MD
2134 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2135 if (type == vp->v_type) {
2136 *vpp = vp;
8a8d5d85 2137 lwkt_reltoken(&spechash_token);
984263bc
MD
2138 return (1);
2139 }
2140 }
8a8d5d85 2141 lwkt_reltoken(&spechash_token);
984263bc
MD
2142 return (0);
2143}
2144
2145/*
2146 * Calculate the total number of references to a special device.
2147 */
2148int
2149vcount(vp)
2150 struct vnode *vp;
2151{
2152 struct vnode *vq;
2153 int count;
2154
2155 count = 0;
8a8d5d85 2156 lwkt_gettoken(&spechash_token);
984263bc
MD
2157 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2158 count += vq->v_usecount;
8a8d5d85 2159 lwkt_reltoken(&spechash_token);
984263bc
MD
2160 return (count);
2161}
2162
2163/*
2164 * Same as above, but using the dev_t as argument
2165 */
2166
2167int
2168count_dev(dev)
2169 dev_t dev;
2170{
2171 struct vnode *vp;
2172
2173 vp = SLIST_FIRST(&dev->si_hlist);
2174 if (vp == NULL)
2175 return (0);
2176 return(vcount(vp));
2177}
2178
2179/*
2180 * Print out a description of a vnode.
2181 */
2182static char *typename[] =
2183{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2184
2185void
2186vprint(label, vp)
2187 char *label;
2188 struct vnode *vp;
2189{
2190 char buf[96];
2191
2192 if (label != NULL)
2193 printf("%s: %p: ", label, (void *)vp);
2194 else
2195 printf("%p: ", (void *)vp);
2196 printf("type %s, usecount %d, writecount %d, refcount %d,",
2197 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2198 vp->v_holdcnt);
2199 buf[0] = '\0';
2200 if (vp->v_flag & VROOT)
2201 strcat(buf, "|VROOT");
2202 if (vp->v_flag & VTEXT)
2203 strcat(buf, "|VTEXT");
2204 if (vp->v_flag & VSYSTEM)
2205 strcat(buf, "|VSYSTEM");
2206 if (vp->v_flag & VXLOCK)
2207 strcat(buf, "|VXLOCK");
2208 if (vp->v_flag & VXWANT)
2209 strcat(buf, "|VXWANT");
2210 if (vp->v_flag & VBWAIT)
2211 strcat(buf, "|VBWAIT");
2212 if (vp->v_flag & VDOOMED)
2213 strcat(buf, "|VDOOMED");
2214 if (vp->v_flag & VFREE)
2215 strcat(buf, "|VFREE");
2216 if (vp->v_flag & VOBJBUF)
2217 strcat(buf, "|VOBJBUF");
2218 if (buf[0] != '\0')
2219 printf(" flags (%s)", &buf[1]);
2220 if (vp->v_data == NULL) {
2221 printf("\n");
2222 } else {
2223 printf("\n\t");
2224 VOP_PRINT(vp);
2225 }
2226}
2227
2228#ifdef DDB
2229#include <ddb/ddb.h>
2230/*
2231 * List all of the locked vnodes in the system.
2232 * Called when debugging the kernel.
2233 */
2234DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2235{
dadab5e9 2236 struct thread *td = curthread; /* XXX */
984263bc
MD
2237 struct mount *mp, *nmp;
2238 struct vnode *vp;
2239
2240 printf("Locked vnodes\n");
8a8d5d85 2241 lwkt_gettoken(&mountlist_token);
984263bc 2242 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 2243 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
984263bc
MD
2244 nmp = TAILQ_NEXT(mp, mnt_list);
2245 continue;
2246 }
2247 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2248 if (VOP_ISLOCKED(vp, NULL))
2249 vprint((char *)0, vp);
2250 }
8a8d5d85 2251 lwkt_gettoken(&mountlist_token);
984263bc 2252 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 2253 vfs_unbusy(mp, td);
984263bc 2254 }
8a8d5d85 2255 lwkt_reltoken(&mountlist_token);
984263bc
MD
2256}
2257#endif
2258
2259/*
2260 * Top level filesystem related information gathering.
2261 */
402ed7e1 2262static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
984263bc
MD
2263
2264static int
2265vfs_sysctl(SYSCTL_HANDLER_ARGS)
2266{
2267 int *name = (int *)arg1 - 1; /* XXX */
2268 u_int namelen = arg2 + 1; /* XXX */
2269 struct vfsconf *vfsp;
2270
2271#if 1 || defined(COMPAT_PRELITE2)
2272 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2273 if (namelen == 1)
2274 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2275#endif
2276
2277#ifdef notyet
2278 /* all sysctl names at this level are at least name and field */
2279 if (namelen < 2)
2280 return (ENOTDIR); /* overloaded */
2281 if (name[0] != VFS_GENERIC) {
2282 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2283 if (vfsp->vfc_typenum == name[0])
2284 break;
2285 if (vfsp == NULL)
2286 return (EOPNOTSUPP);
2287 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2288 oldp, oldlenp, newp, newlen, p));
2289 }
2290#endif
2291 switch (name[1]) {
2292 case VFS_MAXTYPENUM:
2293 if (namelen != 2)
2294 return (ENOTDIR);
2295 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2296 case VFS_CONF:
2297 if (namelen != 3)
2298 return (ENOTDIR); /* overloaded */
2299 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2300 if (vfsp->vfc_typenum == name[2])
2301 break;
2302 if (vfsp == NULL)
2303 return (EOPNOTSUPP);
2304 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2305 }
2306 return (EOPNOTSUPP);
2307}
2308
2309SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2310 "Generic filesystem");
2311
2312#if 1 || defined(COMPAT_PRELITE2)
2313
2314static int
2315sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2316{
2317 int error;
2318 struct vfsconf *vfsp;
2319 struct ovfsconf ovfs;
2320
2321 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2322 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2323 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2324 ovfs.vfc_index = vfsp->vfc_typenum;
2325 ovfs.vfc_refcount = vfsp->vfc_refcount;
2326 ovfs.vfc_flags = vfsp->vfc_flags;
2327 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2328 if (error)
2329 return error;
2330 }
2331 return 0;
2332}
2333
2334#endif /* 1 || COMPAT_PRELITE2 */
2335
2336#if 0
2337#define KINFO_VNODESLOP 10
2338/*
2339 * Dump vnode list (via sysctl).
2340 * Copyout address of vnode followed by vnode.
2341 */
2342/* ARGSUSED */
2343static int
2344sysctl_vnode(SYSCTL_HANDLER_ARGS)
2345{
2346 struct proc *p = curproc; /* XXX */
2347 struct mount *mp, *nmp;
2348 struct vnode *nvp, *vp;
2349 int error;
2350
2351#define VPTRSZ sizeof (struct vnode *)
2352#define VNODESZ sizeof (struct vnode)
2353
2354 req->lock = 0;
2355 if (!req->oldptr) /* Make an estimate */
2356 return (SYSCTL_OUT(req, 0,
2357 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2358
8a8d5d85 2359 lwkt_gettoken(&mountlist_token);
984263bc 2360 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 2361 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) {
984263bc
MD
2362 nmp = TAILQ_NEXT(mp, mnt_list);
2363 continue;
2364 }
2365again:
8a8d5d85 2366 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2367 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2368 vp != NULL;
2369 vp = nvp) {
2370 /*
2371 * Check that the vp is still associated with
2372 * this filesystem. RACE: could have been
2373 * recycled onto the same filesystem.
2374 */
2375 if (vp->v_mount != mp) {
8a8d5d85 2376 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2377 goto again;
2378 }
2379 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
8a8d5d85 2380 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2381 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2382 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2383 return (error);
8a8d5d85 2384 lwkt_gettoken(&mntvnode_token);
984263bc 2385 }
8a8d5d85
MD
2386 lwkt_reltoken(&mntvnode_token);
2387 lwkt_gettoken(&mountlist_token);
984263bc
MD
2388 nmp = TAILQ_NEXT(mp, mnt_list);
2389 vfs_unbusy(mp, p);
2390 }
8a8d5d85 2391 lwkt_reltoken(&mountlist_token);
984263bc
MD
2392
2393 return (0);
2394}
2395#endif
2396
2397/*
2398 * XXX
2399 * Exporting the vnode list on large systems causes them to crash.
2400 * Exporting the vnode list on medium systems causes sysctl to coredump.
2401 */
2402#if 0
2403SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2404 0, 0, sysctl_vnode, "S,vnode", "");
2405#endif
2406
2407/*
2408 * Check to see if a filesystem is mounted on a block device.
2409 */
2410int
2411vfs_mountedon(vp)
2412 struct vnode *vp;
2413{
2414
2415 if (vp->v_specmountpoint != NULL)
2416 return (EBUSY);
2417 return (0);
2418}
2419
2420/*
2421 * Unmount all filesystems. The list is traversed in reverse order
2422 * of mounting to avoid dependencies.
2423 */
2424void
2425vfs_unmountall()
2426{
2427 struct mount *mp;
dadab5e9 2428 struct thread *td = curthread;
984263bc
MD
2429 int error;
2430
dadab5e9
MD
2431 if (td->td_proc == NULL)
2432 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2433
984263bc
MD
2434 /*
2435 * Since this only runs when rebooting, it is not interlocked.
2436 */
2437 while(!TAILQ_EMPTY(&mountlist)) {
2438 mp = TAILQ_LAST(&mountlist, mntlist);
dadab5e9 2439 error = dounmount(mp, MNT_FORCE, td);
984263bc
MD
2440 if (error) {
2441 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2442 printf("unmount of %s failed (",
2443 mp->mnt_stat.f_mntonname);
2444 if (error == EBUSY)
2445 printf("BUSY)\n");
2446 else
2447 printf("%d)\n", error);
2448 } else {
2449 /* The unmount has removed mp from the mountlist */
2450 }
2451 }
2452}
2453
2454/*
2455 * Build hash lists of net addresses and hang them off the mount point.
2456 * Called by ufs_mount() to set up the lists of export addresses.
2457 */
2458static int
2459vfs_hang_addrlist(mp, nep, argp)
2460 struct mount *mp;
2461 struct netexport *nep;
2462 struct export_args *argp;
2463{
1fd87d54
RG
2464 struct netcred *np;
2465 struct radix_node_head *rnh;
2466 int i;
984263bc
MD
2467 struct radix_node *rn;
2468 struct sockaddr *saddr, *smask = 0;
2469 struct domain *dom;
2470 int error;
2471
2472 if (argp->ex_addrlen == 0) {
2473 if (mp->mnt_flag & MNT_DEFEXPORTED)
2474 return (EPERM);
2475 np = &nep->ne_defexported;
2476 np->netc_exflags = argp->ex_flags;
2477 np->netc_anon = argp->ex_anon;
2478 np->netc_anon.cr_ref = 1;
2479 mp->mnt_flag |= MNT_DEFEXPORTED;
2480 return (0);
2481 }
2482
0260ddf9
MD
2483 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2484 return (EINVAL);
2485 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
984263bc
MD
2486 return (EINVAL);
2487
2488 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2489 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2490 bzero((caddr_t) np, i);
2491 saddr = (struct sockaddr *) (np + 1);
2492 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2493 goto out;
2494 if (saddr->sa_len > argp->ex_addrlen)
2495 saddr->sa_len = argp->ex_addrlen;
2496 if (argp->ex_masklen) {
2497 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2498 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2499 if (error)
2500 goto out;
2501 if (smask->sa_len > argp->ex_masklen)
2502 smask->sa_len = argp->ex_masklen;
2503 }
2504 i = saddr->sa_family;
2505 if ((rnh = nep->ne_rtable[i]) == 0) {
2506 /*
2507 * Seems silly to initialize every AF when most are not used,
2508 * do so on demand here
2509 */
2510 for (dom = domains; dom; dom = dom->dom_next)
2511 if (dom->dom_family == i && dom->dom_rtattach) {
2512 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2513 dom->dom_rtoffset);
2514 break;
2515 }
2516 if ((rnh = nep->ne_rtable[i]) == 0) {
2517 error = ENOBUFS;
2518 goto out;
2519 }
2520 }
2521 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2522 np->netc_rnodes);
2523 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2524 error = EPERM;
2525 goto out;
2526 }
2527 np->netc_exflags = argp->ex_flags;
2528 np->netc_anon = argp->ex_anon;
2529 np->netc_anon.cr_ref = 1;
2530 return (0);
2531out:
2532 free(np, M_NETADDR);
2533 return (error);
2534}
2535
2536/* ARGSUSED */
2537static int
2538vfs_free_netcred(rn, w)
2539 struct radix_node *rn;
2540 void *w;
2541{
1fd87d54 2542 struct radix_node_head *rnh = (struct radix_node_head *) w;
984263bc
MD
2543
2544 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2545 free((caddr_t) rn, M_NETADDR);
2546 return (0);
2547}
2548
2549/*
2550 * Free the net address hash lists that are hanging off the mount points.
2551 */
2552static void
2553vfs_free_addrlist(nep)
2554 struct netexport *nep;
2555{
1fd87d54
RG
2556 int i;
2557 struct radix_node_head *rnh;
984263bc
MD
2558
2559 for (i = 0; i <= AF_MAX; i++)
2560 if ((rnh = nep->ne_rtable[i])) {
2561 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2562 (caddr_t) rnh);
2563 free((caddr_t) rnh, M_RTABLE);
2564 nep->ne_rtable[i] = 0;
2565 }
2566}
2567
2568int
2569vfs_export(mp, nep, argp)
2570 struct mount *mp;
2571 struct netexport *nep;
2572 struct export_args *argp;
2573{
2574 int error;
2575
2576 if (argp->ex_flags & MNT_DELEXPORT) {
2577 if (mp->mnt_flag & MNT_EXPUBLIC) {
2578 vfs_setpublicfs(NULL, NULL, NULL);
2579 mp->mnt_flag &= ~MNT_EXPUBLIC;
2580 }
2581 vfs_free_addrlist(nep);
2582 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2583 }
2584 if (argp->ex_flags & MNT_EXPORTED) {
2585 if (argp->ex_flags & MNT_EXPUBLIC) {
2586 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2587 return (error);
2588 mp->mnt_flag |= MNT_EXPUBLIC;
2589 }
2590 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2591 return (error);
2592 mp->mnt_flag |= MNT_EXPORTED;
2593 }
2594 return (0);
2595}
2596
2597
2598/*
2599 * Set the publicly exported filesystem (WebNFS). Currently, only
2600 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2601 */
2602int
2603vfs_setpublicfs(mp, nep, argp)
2604 struct mount *mp;
2605 struct netexport *nep;
2606 struct export_args *argp;
2607{
2608 int error;
2609 struct vnode *rvp;
2610 char *cp;
2611
2612 /*
2613 * mp == NULL -> invalidate the current info, the FS is
2614 * no longer exported. May be called from either vfs_export
2615 * or unmount, so check if it hasn't already been done.
2616 */
2617 if (mp == NULL) {
2618 if (nfs_pub.np_valid) {
2619 nfs_pub.np_valid = 0;
2620 if (nfs_pub.np_index != NULL) {
2621 FREE(nfs_pub.np_index, M_TEMP);
2622 nfs_pub.np_index = NULL;
2623 }
2624 }
2625 return (0);
2626 }
2627
2628 /*
2629 * Only one allowed at a time.
2630 */
2631 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2632 return (EBUSY);
2633
2634 /*
2635 * Get real filehandle for root of exported FS.
2636 */
2637 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2638 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2639
2640 if ((error = VFS_ROOT(mp, &rvp)))
2641 return (error);
2642
2643 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2644 return (error);
2645
2646 vput(rvp);
2647
2648 /*
2649 * If an indexfile was specified, pull it in.
2650 */
2651 if (argp->ex_indexfile != NULL) {
2652 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2653 M_WAITOK);
2654 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2655 MAXNAMLEN, (size_t *)0);
2656 if (!error) {
2657 /*
2658 * Check for illegal filenames.
2659 */
2660 for (cp = nfs_pub.np_index; *cp; cp++) {
2661 if (*cp == '/') {
2662 error = EINVAL;
2663 break;
2664 }
2665 }
2666 }
2667 if (error) {
2668 FREE(nfs_pub.np_index, M_TEMP);
2669 return (error);
2670 }
2671 }
2672
2673 nfs_pub.np_mount = mp;
2674 nfs_pub.np_valid = 1;
2675 return (0);
2676}
2677
2678struct netcred *
2679vfs_export_lookup(mp, nep, nam)
1fd87d54 2680 struct mount *mp;
984263bc
MD
2681 struct netexport *nep;
2682 struct sockaddr *nam;
2683{
1fd87d54
RG
2684 struct netcred *np;
2685 struct radix_node_head *rnh;
984263bc
MD
2686 struct sockaddr *saddr;
2687
2688 np = NULL;
2689 if (mp->mnt_flag & MNT_EXPORTED) {
2690 /*
2691 * Lookup in the export list first.
2692 */
2693 if (nam != NULL) {
2694 saddr = nam;
2695 rnh = nep->ne_rtable[saddr->sa_family];
2696 if (rnh != NULL) {
2697 np = (struct netcred *)
2698 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2699 rnh);
2700 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2701 np = NULL;
2702 }
2703 }
2704 /*
2705 * If no address match, use the default if it exists.
2706 */
2707 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2708 np = &nep->ne_defexported;
2709 }
2710 return (np);
2711}
2712
2713/*
2714 * perform msync on all vnodes under a mount point
2715 * the mount point must be locked.
2716 */
2717void
2718vfs_msync(struct mount *mp, int flags)
2719{
dadab5e9 2720 struct thread *td = curthread; /* XXX */
984263bc
MD
2721 struct vnode *vp, *nvp;
2722 struct vm_object *obj;
2723 int tries;
2724
2725 tries = 5;
8a8d5d85 2726 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2727loop:
2728 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2729 if (vp->v_mount != mp) {
2730 if (--tries > 0)
2731 goto loop;
2732 break;
2733 }
2734 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2735
2736 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2737 continue;
2738
2739 /*
2740 * There could be hundreds of thousands of vnodes, we cannot
2741 * afford to do anything heavy-weight until we have a fairly
2742 * good indication that there is something to do.
2743 */
2744 if ((vp->v_flag & VOBJDIRTY) &&
2745 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
8a8d5d85 2746 lwkt_reltoken(&mntvnode_token);
984263bc 2747 if (!vget(vp,
dadab5e9 2748 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
984263bc
MD
2749 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2750 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2751 }
2752 vput(vp);
2753 }
8a8d5d85 2754 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2755 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2756 if (--tries > 0)
2757 goto loop;
2758 break;
2759 }
2760 }
2761 }
8a8d5d85 2762 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2763}
2764
2765/*
2766 * Create the VM object needed for VMIO and mmap support. This
2767 * is done for all VREG files in the system. Some filesystems might
2768 * afford the additional metadata buffering capability of the
2769 * VMIO code by making the device node be VMIO mode also.
2770 *
2771 * vp must be locked when vfs_object_create is called.
2772 */
2773int
3b568787 2774vfs_object_create(struct vnode *vp, struct thread *td)
984263bc 2775{
3b568787 2776 return (VOP_CREATEVOBJECT(vp, td));
984263bc
MD
2777}
2778
2779void
2780vfree(vp)
2781 struct vnode *vp;
2782{
2783 int s;
2784
2785 s = splbio();
8a8d5d85 2786 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2787 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2788 if (vp->v_flag & VAGE) {
2789 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2790 } else {
2791 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2792 }
2793 freevnodes++;
8a8d5d85 2794 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2795 vp->v_flag &= ~VAGE;
2796 vp->v_flag |= VFREE;
2797 splx(s);
2798}
2799
2800void
2801vbusy(vp)
2802 struct vnode *vp;
2803{
2804 int s;
2805
2806 s = splbio();
8a8d5d85 2807 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2808 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2809 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2810 freevnodes--;
8a8d5d85 2811 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2812 vp->v_flag &= ~(VFREE|VAGE);
2813 splx(s);
2814}
2815
2816/*
2817 * Record a process's interest in events which might happen to
2818 * a vnode. Because poll uses the historic select-style interface
2819 * internally, this routine serves as both the ``check for any
2820 * pending events'' and the ``record my interest in future events''
2821 * functions. (These are done together, while the lock is held,
2822 * to avoid race conditions.)
2823 */
2824int
dadab5e9 2825vn_pollrecord(struct vnode *vp, struct thread *td, int events)
984263bc 2826{
8a8d5d85 2827 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2828 if (vp->v_pollinfo.vpi_revents & events) {
2829 /*
2830 * This leaves events we are not interested
2831 * in available for the other process which
2832 * which presumably had requested them
2833 * (otherwise they would never have been
2834 * recorded).
2835 */
2836 events &= vp->v_pollinfo.vpi_revents;
2837 vp->v_pollinfo.vpi_revents &= ~events;
2838
8a8d5d85 2839 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2840 return events;
2841 }
2842 vp->v_pollinfo.vpi_events |= events;
dadab5e9 2843 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
8a8d5d85 2844 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2845 return 0;
2846}
2847
2848/*
2849 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2850 * it is possible for us to miss an event due to race conditions, but
2851 * that condition is expected to be rare, so for the moment it is the
2852 * preferred interface.
2853 */
2854void
2855vn_pollevent(vp, events)
2856 struct vnode *vp;
2857 short events;
2858{
8a8d5d85 2859 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2860 if (vp->v_pollinfo.vpi_events & events) {
2861 /*
2862 * We clear vpi_events so that we don't
2863 * call selwakeup() twice if two events are
2864 * posted before the polling process(es) is
2865 * awakened. This also ensures that we take at
2866 * most one selwakeup() if the polling process
2867 * is no longer interested. However, it does
2868 * mean that only one event can be noticed at
2869 * a time. (Perhaps we should only clear those
2870 * event bits which we note?) XXX
2871 */
2872 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2873 vp->v_pollinfo.vpi_revents |= events;
2874 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2875 }
8a8d5d85 2876 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2877}
2878
2879/*
2880 * Wake up anyone polling on vp because it is being revoked.
2881 * This depends on dead_poll() returning POLLHUP for correct
2882 * behavior.
2883 */
2884void
2885vn_pollgone(vp)
2886 struct vnode *vp;
2887{
8a8d5d85 2888 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2889 if (vp->v_pollinfo.vpi_events) {
2890 vp->v_pollinfo.vpi_events = 0;
2891 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2892 }
8a8d5d85 2893 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2894}
2895
2896
2897
2898/*
2899 * Routine to create and manage a filesystem syncer vnode.
2900 */
402ed7e1
RG
2901#define sync_close ((int (*) (struct vop_close_args *))nullop)
2902static int sync_fsync (struct vop_fsync_args *);
2903static int sync_inactive (struct vop_inactive_args *);
2904static int sync_reclaim (struct vop_reclaim_args *);
2905#define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
2906#define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
2907static int sync_print (struct vop_print_args *);
2908#define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
984263bc
MD
2909
2910static vop_t **sync_vnodeop_p;
2911static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2912 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2913 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2914 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2915 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2916 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2917 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2918 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2919 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2920 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2921 { NULL, NULL }
2922};
2923static struct vnodeopv_desc sync_vnodeop_opv_desc =
2924 { &sync_vnodeop_p, sync_vnodeop_entries };
2925
2926VNODEOP_SET(sync_vnodeop_opv_desc);
2927
2928/*
2929 * Create a new filesystem syncer vnode for the specified mount point.
2930 */
2931int
2932vfs_allocate_syncvnode(mp)
2933 struct mount *mp;
2934{
2935 struct vnode *vp;
2936 static long start, incr, next;
2937 int error;
2938
2939 /* Allocate a new vnode */
2940 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2941 mp->mnt_syncer = NULL;
2942 return (error);
2943 }
2944 vp->v_type = VNON;
2945 /*
2946 * Place the vnode onto the syncer worklist. We attempt to
2947 * scatter them about on the list so that they will go off
2948 * at evenly distributed times even if all the filesystems
2949 * are mounted at once.
2950 */
2951 next += incr;
2952 if (next == 0 || next > syncer_maxdelay) {
2953 start /= 2;
2954 incr /= 2;
2955 if (start == 0) {
2956 start = syncer_maxdelay / 2;
2957 incr = syncer_maxdelay;
2958 }
2959 next = start;
2960 }
2961 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2962 mp->mnt_syncer = vp;
2963 return (0);
2964}
2965
2966/*
2967 * Do a lazy sync of the filesystem.
2968 */
2969static int
2970sync_fsync(ap)
2971 struct vop_fsync_args /* {
2972 struct vnode *a_vp;
2973 struct ucred *a_cred;
2974 int a_waitfor;
dadab5e9 2975 struct thread *a_td;
984263bc
MD
2976 } */ *ap;
2977{
2978 struct vnode *syncvp = ap->a_vp;
2979 struct mount *mp = syncvp->v_mount;
dadab5e9 2980 struct thread *td = ap->a_td;
984263bc
MD
2981 int asyncflag;
2982
2983 /*
2984 * We only need to do something if this is a lazy evaluation.
2985 */
2986 if (ap->a_waitfor != MNT_LAZY)
2987 return (0);
2988
2989 /*
2990 * Move ourselves to the back of the sync list.
2991 */
2992 vn_syncer_add_to_worklist(syncvp, syncdelay);
2993
2994 /*
2995 * Walk the list of vnodes pushing all that are dirty and
2996 * not already on the sync list.
2997 */
8a8d5d85
MD
2998 lwkt_gettoken(&mountlist_token);
2999 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) {
3000 lwkt_reltoken(&mountlist_token);
984263bc
MD
3001 return (0);
3002 }
3003 asyncflag = mp->mnt_flag & MNT_ASYNC;
3004 mp->mnt_flag &= ~MNT_ASYNC;
3005 vfs_msync(mp, MNT_NOWAIT);
3b568787 3006 VFS_SYNC(mp, MNT_LAZY, td);
984263bc
MD
3007 if (asyncflag)
3008 mp->mnt_flag |= MNT_ASYNC;
dadab5e9 3009 vfs_unbusy(mp, td);
984263bc
MD
3010 return (0);
3011}
3012
3013/*
3014 * The syncer vnode is no referenced.
3015 */
3016static int
3017sync_inactive(ap)
3018 struct vop_inactive_args /* {
3019 struct vnode *a_vp;
3020 struct proc *a_p;
3021 } */ *ap;
3022{
3023
3024 vgone(ap->a_vp);
3025 return (0);
3026}
3027
3028/*
3029 * The syncer vnode is no longer needed and is being decommissioned.
3030 *
3031 * Modifications to the worklist must be protected at splbio().
3032 */
3033static int
3034sync_reclaim(ap)
3035 struct vop_reclaim_args /* {
3036 struct vnode *a_vp;
3037 } */ *ap;
3038{
3039 struct vnode *vp = ap->a_vp;
3040 int s;
3041
3042 s = splbio();
3043 vp->v_mount->mnt_syncer = NULL;
3044 if (vp->v_flag & VONWORKLST) {
3045 LIST_REMOVE(vp, v_synclist);
3046 vp->v_flag &= ~VONWORKLST;
3047 }
3048 splx(s);
3049
3050 return (0);
3051}
3052
3053/*
3054 * Print out a syncer vnode.
3055 */
3056static int
3057sync_print(ap)
3058 struct vop_print_args /* {
3059 struct vnode *a_vp;
3060 } */ *ap;
3061{
3062 struct vnode *vp = ap->a_vp;
3063
3064 printf("syncer vnode");
3065 if (vp->v_vnlock != NULL)
3066 lockmgr_printinfo(vp->v_vnlock);
3067 printf("\n");
3068 return (0);
3069}
3070
3071/*
3072 * extract the dev_t from a VBLK or VCHR
3073 */
3074dev_t
3075vn_todev(vp)
3076 struct vnode *vp;
3077{
3078 if (vp->v_type != VBLK && vp->v_type != VCHR)
3079 return (NODEV);
3080 return (vp->v_rdev);
3081}
3082
3083/*
3084 * Check if vnode represents a disk device
3085 */
3086int
3087vn_isdisk(vp, errp)
3088 struct vnode *vp;
3089 int *errp;
3090{
3091 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3092 if (errp != NULL)
3093 *errp = ENOTBLK;
3094 return (0);
3095 }
3096 if (vp->v_rdev == NULL) {
3097 if (errp != NULL)
3098 *errp = ENXIO;
3099 return (0);
3100 }
335dda38 3101 if (!dev_dport(vp->v_rdev)) {
984263bc
MD
3102 if (errp != NULL)
3103 *errp = ENXIO;
3104 return (0);
3105 }
335dda38 3106 if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
984263bc
MD
3107 if (errp != NULL)
3108 *errp = ENOTBLK;
3109 return (0);
3110 }
3111 if (errp != NULL)
3112 *errp = 0;
3113 return (1);
3114}
3115
3116void
3117NDFREE(ndp, flags)
3118 struct nameidata *ndp;
3119 const uint flags;
3120{
3121 if (!(flags & NDF_NO_FREE_PNBUF) &&
3122 (ndp->ni_cnd.cn_flags & HASBUF)) {
3123 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3124 ndp->ni_cnd.cn_flags &= ~HASBUF;
3125 }
3126 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3127 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3128 ndp->ni_dvp != ndp->ni_vp)
dadab5e9 3129 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
984263bc
MD
3130 if (!(flags & NDF_NO_DVP_RELE) &&
3131 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3132 vrele(ndp->ni_dvp);
3133 ndp->ni_dvp = NULL;
3134 }
3135 if (!(flags & NDF_NO_VP_UNLOCK) &&
3136 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
dadab5e9 3137 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
984263bc
MD
3138 if (!(flags & NDF_NO_VP_RELE) &&
3139 ndp->ni_vp) {
3140 vrele(ndp->ni_vp);
3141 ndp->ni_vp = NULL;
3142 }
3143 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3144 (ndp->ni_cnd.cn_flags & SAVESTART)) {
3145 vrele(ndp->ni_startdir);
3146 ndp->ni_startdir = NULL;
3147 }
3148}