Split wait4(), setrlimit(), getrlimit(), statfs(), fstatfs(), chdir(),
[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 $
bc0c094e 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.22 2003/10/09 22:27:19 dillon 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, "");
984263bc
MD
120
121#ifdef ENABLE_VFS_IOOPT
122int vfs_ioopt = 0;
123SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
124#endif
125
126struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
8a8d5d85
MD
127struct lwkt_token mountlist_token;
128struct lwkt_token mntvnode_token;
984263bc 129int nfs_mount_type = -1;
8a8d5d85
MD
130static struct lwkt_token mntid_token;
131static struct lwkt_token vnode_free_list_token;
132static struct lwkt_token spechash_token;
984263bc
MD
133struct nfs_public nfs_pub; /* publicly exported FS */
134static vm_zone_t vnode_zone;
135
136/*
137 * The workitem queue.
138 */
139#define SYNCER_MAXDELAY 32
140static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
141time_t syncdelay = 30; /* max time to delay syncing data */
99138c2f
HP
142SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
143 "VFS data synchronization delay");
984263bc 144time_t filedelay = 30; /* time to delay syncing files */
99138c2f
HP
145SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
146 "File synchronization delay");
984263bc 147time_t dirdelay = 29; /* time to delay syncing directories */
99138c2f
HP
148SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
149 "Directory synchronization delay");
984263bc 150time_t metadelay = 28; /* time to delay syncing metadata */
99138c2f
HP
151SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
152 "VFS metadata synchronization delay");
984263bc
MD
153static int rushjob; /* number of slots to run ASAP */
154static int stat_rush_requests; /* number of times I/O speeded up */
155SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
156
157static int syncer_delayno = 0;
158static long syncer_mask;
159LIST_HEAD(synclist, vnode);
160static struct synclist *syncer_workitem_pending;
161
162int desiredvnodes;
163SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
164 &desiredvnodes, 0, "Maximum number of vnodes");
165static int minvnodes;
166SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
167 &minvnodes, 0, "Minimum number of vnodes");
168static int vnlru_nowhere = 0;
169SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
170 "Number of times the vnlru process ran without success");
171
402ed7e1
RG
172static void vfs_free_addrlist (struct netexport *nep);
173static int vfs_free_netcred (struct radix_node *rn, void *w);
174static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
175 struct export_args *argp);
984263bc
MD
176
177/*
178 * Initialize the vnode management data structures.
179 */
180void
181vntblinit()
182{
183
12e4aaff 184 desiredvnodes = maxproc + vmstats.v_page_count / 4;
984263bc 185 minvnodes = desiredvnodes / 4;
8a8d5d85
MD
186 lwkt_inittoken(&mntvnode_token);
187 lwkt_inittoken(&mntid_token);
188 lwkt_inittoken(&spechash_token);
984263bc 189 TAILQ_INIT(&vnode_free_list);
8a8d5d85 190 lwkt_inittoken(&vnode_free_list_token);
984263bc
MD
191 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
192 /*
193 * Initialize the filesystem syncer.
194 */
195 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
196 &syncer_mask);
197 syncer_maxdelay = syncer_mask + 1;
198}
199
200/*
201 * Mark a mount point as busy. Used to synchronize access and to delay
202 * unmounting. Interlock is not released on failure.
203 */
204int
8a8d5d85 205vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp,
dadab5e9 206 struct thread *td)
984263bc
MD
207{
208 int lkflags;
209
210 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
211 if (flags & LK_NOWAIT)
212 return (ENOENT);
213 mp->mnt_kern_flag |= MNTK_MWAIT;
214 if (interlkp) {
8a8d5d85 215 lwkt_reltoken(interlkp);
984263bc
MD
216 }
217 /*
218 * Since all busy locks are shared except the exclusive
219 * lock granted when unmounting, the only place that a
220 * wakeup needs to be done is at the release of the
221 * exclusive lock at the end of dounmount.
222 */
377d4740 223 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
984263bc 224 if (interlkp) {
8a8d5d85 225 lwkt_gettoken(interlkp);
984263bc
MD
226 }
227 return (ENOENT);
228 }
229 lkflags = LK_SHARED | LK_NOPAUSE;
230 if (interlkp)
231 lkflags |= LK_INTERLOCK;
dadab5e9 232 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
984263bc
MD
233 panic("vfs_busy: unexpected lock failure");
234 return (0);
235}
236
237/*
238 * Free a busy filesystem.
239 */
240void
dadab5e9 241vfs_unbusy(struct mount *mp, struct thread *td)
984263bc 242{
dadab5e9 243 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
984263bc
MD
244}
245
246/*
247 * Lookup a filesystem type, and if found allocate and initialize
248 * a mount structure for it.
249 *
250 * Devname is usually updated by mount(8) after booting.
251 */
252int
dadab5e9 253vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
984263bc 254{
dadab5e9 255 struct thread *td = curthread; /* XXX */
984263bc
MD
256 struct vfsconf *vfsp;
257 struct mount *mp;
258
259 if (fstypename == NULL)
260 return (ENODEV);
261 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
262 if (!strcmp(vfsp->vfc_name, fstypename))
263 break;
264 if (vfsp == NULL)
265 return (ENODEV);
266 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
267 bzero((char *)mp, (u_long)sizeof(struct mount));
377d4740 268 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
dadab5e9 269 (void)vfs_busy(mp, LK_NOWAIT, 0, td);
984263bc
MD
270 TAILQ_INIT(&mp->mnt_nvnodelist);
271 TAILQ_INIT(&mp->mnt_reservedvnlist);
272 mp->mnt_nvnodelistsize = 0;
273 mp->mnt_vfc = vfsp;
274 mp->mnt_op = vfsp->vfc_vfsops;
275 mp->mnt_flag = MNT_RDONLY;
276 mp->mnt_vnodecovered = NULLVP;
277 vfsp->vfc_refcount++;
278 mp->mnt_iosize_max = DFLTPHYS;
279 mp->mnt_stat.f_type = vfsp->vfc_typenum;
280 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
281 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
282 mp->mnt_stat.f_mntonname[0] = '/';
283 mp->mnt_stat.f_mntonname[1] = 0;
284 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
285 *mpp = mp;
286 return (0);
287}
288
289/*
290 * Find an appropriate filesystem to use for the root. If a filesystem
291 * has not been preselected, walk through the list of known filesystems
292 * trying those that have mountroot routines, and try them until one
293 * works or we have tried them all.
294 */
295#ifdef notdef /* XXX JH */
296int
297lite2_vfs_mountroot()
298{
299 struct vfsconf *vfsp;
402ed7e1 300 extern int (*lite2_mountroot) (void);
984263bc
MD
301 int error;
302
303 if (lite2_mountroot != NULL)
304 return ((*lite2_mountroot)());
305 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
306 if (vfsp->vfc_mountroot == NULL)
307 continue;
308 if ((error = (*vfsp->vfc_mountroot)()) == 0)
309 return (0);
310 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
311 }
312 return (ENODEV);
313}
314#endif
315
316/*
317 * Lookup a mount point by filesystem identifier.
318 */
319struct mount *
320vfs_getvfs(fsid)
321 fsid_t *fsid;
322{
1fd87d54 323 struct mount *mp;
984263bc 324
8a8d5d85 325 lwkt_gettoken(&mountlist_token);
984263bc
MD
326 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
327 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
328 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
8a8d5d85 329 lwkt_reltoken(&mountlist_token);
984263bc
MD
330 return (mp);
331 }
332 }
8a8d5d85 333 lwkt_reltoken(&mountlist_token);
984263bc
MD
334 return ((struct mount *) 0);
335}
336
337/*
338 * Get a new unique fsid. Try to make its val[0] unique, since this value
339 * will be used to create fake device numbers for stat(). Also try (but
340 * not so hard) make its val[0] unique mod 2^16, since some emulators only
341 * support 16-bit device numbers. We end up with unique val[0]'s for the
342 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
343 *
344 * Keep in mind that several mounts may be running in parallel. Starting
345 * the search one past where the previous search terminated is both a
346 * micro-optimization and a defense against returning the same fsid to
347 * different mounts.
348 */
349void
350vfs_getnewfsid(mp)
351 struct mount *mp;
352{
353 static u_int16_t mntid_base;
354 fsid_t tfsid;
355 int mtype;
356
8a8d5d85 357 lwkt_gettoken(&mntid_token);
984263bc
MD
358 mtype = mp->mnt_vfc->vfc_typenum;
359 tfsid.val[1] = mtype;
360 mtype = (mtype & 0xFF) << 24;
361 for (;;) {
362 tfsid.val[0] = makeudev(255,
363 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
364 mntid_base++;
365 if (vfs_getvfs(&tfsid) == NULL)
366 break;
367 }
368 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
369 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
8a8d5d85 370 lwkt_reltoken(&mntid_token);
984263bc
MD
371}
372
373/*
374 * Knob to control the precision of file timestamps:
375 *
376 * 0 = seconds only; nanoseconds zeroed.
377 * 1 = seconds and nanoseconds, accurate within 1/HZ.
378 * 2 = seconds and nanoseconds, truncated to microseconds.
379 * >=3 = seconds and nanoseconds, maximum precision.
380 */
381enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
382
383static int timestamp_precision = TSP_SEC;
384SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
385 &timestamp_precision, 0, "");
386
387/*
388 * Get a current timestamp.
389 */
390void
391vfs_timestamp(tsp)
392 struct timespec *tsp;
393{
394 struct timeval tv;
395
396 switch (timestamp_precision) {
397 case TSP_SEC:
398 tsp->tv_sec = time_second;
399 tsp->tv_nsec = 0;
400 break;
401 case TSP_HZ:
402 getnanotime(tsp);
403 break;
404 case TSP_USEC:
405 microtime(&tv);
406 TIMEVAL_TO_TIMESPEC(&tv, tsp);
407 break;
408 case TSP_NSEC:
409 default:
410 nanotime(tsp);
411 break;
412 }
413}
414
415/*
416 * Set vnode attributes to VNOVAL
417 */
418void
419vattr_null(vap)
1fd87d54 420 struct vattr *vap;
984263bc
MD
421{
422
423 vap->va_type = VNON;
424 vap->va_size = VNOVAL;
425 vap->va_bytes = VNOVAL;
426 vap->va_mode = VNOVAL;
427 vap->va_nlink = VNOVAL;
428 vap->va_uid = VNOVAL;
429 vap->va_gid = VNOVAL;
430 vap->va_fsid = VNOVAL;
431 vap->va_fileid = VNOVAL;
432 vap->va_blocksize = VNOVAL;
433 vap->va_rdev = VNOVAL;
434 vap->va_atime.tv_sec = VNOVAL;
435 vap->va_atime.tv_nsec = VNOVAL;
436 vap->va_mtime.tv_sec = VNOVAL;
437 vap->va_mtime.tv_nsec = VNOVAL;
438 vap->va_ctime.tv_sec = VNOVAL;
439 vap->va_ctime.tv_nsec = VNOVAL;
440 vap->va_flags = VNOVAL;
441 vap->va_gen = VNOVAL;
442 vap->va_vaflags = 0;
443}
444
445/*
446 * This routine is called when we have too many vnodes. It attempts
447 * to free <count> vnodes and will potentially free vnodes that still
448 * have VM backing store (VM backing store is typically the cause
449 * of a vnode blowout so we want to do this). Therefore, this operation
450 * is not considered cheap.
451 *
452 * A number of conditions may prevent a vnode from being reclaimed.
453 * the buffer cache may have references on the vnode, a directory
454 * vnode may still have references due to the namei cache representing
455 * underlying files, or the vnode may be in active use. It is not
456 * desireable to reuse such vnodes. These conditions may cause the
457 * number of vnodes to reach some minimum value regardless of what
458 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
459 */
460static int
461vlrureclaim(struct mount *mp)
462{
463 struct vnode *vp;
464 int done;
465 int trigger;
466 int usevnodes;
467 int count;
8a8d5d85 468 int gen;
984263bc
MD
469
470 /*
471 * Calculate the trigger point, don't allow user
472 * screwups to blow us up. This prevents us from
473 * recycling vnodes with lots of resident pages. We
474 * aren't trying to free memory, we are trying to
475 * free vnodes.
476 */
477 usevnodes = desiredvnodes;
478 if (usevnodes <= 0)
479 usevnodes = 1;
12e4aaff 480 trigger = vmstats.v_page_count * 2 / usevnodes;
984263bc
MD
481
482 done = 0;
8a8d5d85 483 gen = lwkt_gettoken(&mntvnode_token);
984263bc
MD
484 count = mp->mnt_nvnodelistsize / 10 + 1;
485 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
486 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
487 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
488
489 if (vp->v_type != VNON &&
490 vp->v_type != VBAD &&
491 VMIGHTFREE(vp) && /* critical path opt */
8a8d5d85 492 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger)
984263bc 493 ) {
8a8d5d85
MD
494 lwkt_gettoken(&vp->v_interlock);
495 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) {
496 if (VMIGHTFREE(vp)) {
497 vgonel(vp, curthread);
498 done++;
499 } else {
500 lwkt_reltoken(&vp->v_interlock);
501 }
984263bc 502 } else {
8a8d5d85 503 lwkt_reltoken(&vp->v_interlock);
984263bc 504 }
984263bc
MD
505 }
506 --count;
507 }
8a8d5d85 508 lwkt_reltoken(&mntvnode_token);
984263bc
MD
509 return done;
510}
511
512/*
513 * Attempt to recycle vnodes in a context that is always safe to block.
514 * Calling vlrurecycle() from the bowels of file system code has some
515 * interesting deadlock problems.
516 */
bc6dffab 517static struct thread *vnlruthread;
984263bc
MD
518static int vnlruproc_sig;
519
520static void
521vnlru_proc(void)
522{
523 struct mount *mp, *nmp;
524 int s;
525 int done;
0cfcada1 526 struct thread *td = curthread;
984263bc 527
bc6dffab 528 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
529 SHUTDOWN_PRI_FIRST);
530
531 s = splbio();
532 for (;;) {
0cfcada1 533 kproc_suspend_loop();
984263bc
MD
534 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
535 vnlruproc_sig = 0;
536 wakeup(&vnlruproc_sig);
377d4740 537 tsleep(td, 0, "vlruwt", hz);
984263bc
MD
538 continue;
539 }
540 done = 0;
8a8d5d85 541 lwkt_gettoken(&mountlist_token);
984263bc 542 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 543 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
984263bc
MD
544 nmp = TAILQ_NEXT(mp, mnt_list);
545 continue;
546 }
547 done += vlrureclaim(mp);
8a8d5d85 548 lwkt_gettoken(&mountlist_token);
984263bc 549 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 550 vfs_unbusy(mp, td);
984263bc 551 }
8a8d5d85 552 lwkt_reltoken(&mountlist_token);
984263bc
MD
553 if (done == 0) {
554 vnlru_nowhere++;
377d4740 555 tsleep(td, 0, "vlrup", hz * 3);
984263bc
MD
556 }
557 }
558 splx(s);
559}
560
561static struct kproc_desc vnlru_kp = {
562 "vnlru",
563 vnlru_proc,
bc6dffab 564 &vnlruthread
984263bc
MD
565};
566SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
567
568/*
569 * Routines having to do with the management of the vnode table.
570 */
571extern vop_t **dead_vnodeop_p;
572
573/*
574 * Return the next vnode from the free list.
575 */
576int
577getnewvnode(tag, mp, vops, vpp)
578 enum vtagtype tag;
579 struct mount *mp;
580 vop_t **vops;
581 struct vnode **vpp;
582{
583 int s;
8a8d5d85
MD
584 int gen;
585 int vgen;
dadab5e9 586 struct thread *td = curthread; /* XXX */
984263bc
MD
587 struct vnode *vp = NULL;
588 vm_object_t object;
589
590 s = splbio();
591
592 /*
593 * Try to reuse vnodes if we hit the max. This situation only
594 * occurs in certain large-memory (2G+) situations. We cannot
595 * attempt to directly reclaim vnodes due to nasty recursion
596 * problems.
597 */
598 while (numvnodes - freevnodes > desiredvnodes) {
599 if (vnlruproc_sig == 0) {
600 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
bc6dffab 601 wakeup(vnlruthread);
984263bc 602 }
377d4740 603 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
984263bc
MD
604 }
605
606
607 /*
608 * Attempt to reuse a vnode already on the free list, allocating
609 * a new vnode if we can't find one or if we have not reached a
610 * good minimum for good LRU performance.
611 */
8a8d5d85 612 gen = lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
613 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
614 int count;
615
616 for (count = 0; count < freevnodes; count++) {
617 vp = TAILQ_FIRST(&vnode_free_list);
618 if (vp == NULL || vp->v_usecount)
619 panic("getnewvnode: free vnode isn't");
620
8a8d5d85
MD
621 /*
622 * Get the vnode's interlock, then re-obtain
623 * vnode_free_list_token in case we lost it. If we
624 * did lose it while getting the vnode interlock,
625 * even if we got it back again, then retry.
626 */
627 vgen = lwkt_gettoken(&vp->v_interlock);
628 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
629 --count;
630 lwkt_reltoken(&vp->v_interlock);
631 vp = NULL;
632 continue;
633 }
634
635 /*
636 * Whew! We have both tokens. Since we didn't lose
637 * the free list VFREE had better still be set. But
638 * we aren't out of the woods yet. We have to get
639 * the object (may block). If the vnode is not
640 * suitable then move it to the end of the list
641 * if we can. If we can't move it to the end of the
642 * list retry again.
643 */
984263bc 644 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
8a8d5d85
MD
645 (object->resident_page_count || object->ref_count))
646 ) {
647 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 &&
648 lwkt_gentoken(&vnode_free_list_token, &gen) == 0
649 ) {
650 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
651 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
652 } else {
653 --count;
654 }
655 lwkt_reltoken(&vp->v_interlock);
656 vp = NULL;
657 continue;
658 }
659
660 /*
661 * Still not out of the woods. VOBJECT might have
662 * blocked, if we did not retain our tokens we have
663 * to retry.
664 */
665 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 ||
666 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
667 --count;
984263bc
MD
668 vp = NULL;
669 continue;
670 }
8a8d5d85
MD
671 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
672 KKASSERT(vp->v_flag & VFREE);
673
7ea21ed1
MD
674 /*
675 * If we have children in the namecache we cannot
676 * reuse the vnode yet because it will break the
677 * namecache chain (YYY use nc_refs for the check?)
678 */
679 if (TAILQ_FIRST(&vp->v_namecache)) {
680 if (cache_leaf_test(vp) < 0) {
8a8d5d85 681 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
682 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
683 vp = NULL;
684 continue;
685 }
686 }
687 break;
688 }
689 }
690
691 if (vp) {
692 vp->v_flag |= VDOOMED;
693 vp->v_flag &= ~VFREE;
694 freevnodes--;
8a8d5d85
MD
695 lwkt_reltoken(&vnode_free_list_token);
696 cache_purge(vp); /* YYY may block */
984263bc
MD
697 vp->v_lease = NULL;
698 if (vp->v_type != VBAD) {
dadab5e9 699 vgonel(vp, td);
984263bc 700 } else {
8a8d5d85 701 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
702 }
703
704#ifdef INVARIANTS
705 {
706 int s;
707
708 if (vp->v_data)
709 panic("cleaned vnode isn't");
710 s = splbio();
711 if (vp->v_numoutput)
712 panic("Clean vnode has pending I/O's");
713 splx(s);
714 }
715#endif
716 vp->v_flag = 0;
717 vp->v_lastw = 0;
718 vp->v_lasta = 0;
719 vp->v_cstart = 0;
720 vp->v_clen = 0;
721 vp->v_socket = 0;
722 vp->v_writecount = 0; /* XXX */
723 } else {
8a8d5d85 724 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
725 vp = (struct vnode *) zalloc(vnode_zone);
726 bzero((char *) vp, sizeof *vp);
8a8d5d85 727 lwkt_inittoken(&vp->v_interlock);
984263bc
MD
728 vp->v_dd = vp;
729 cache_purge(vp);
7ea21ed1 730 TAILQ_INIT(&vp->v_namecache);
984263bc
MD
731 numvnodes++;
732 }
733
734 TAILQ_INIT(&vp->v_cleanblkhd);
735 TAILQ_INIT(&vp->v_dirtyblkhd);
736 vp->v_type = VNON;
737 vp->v_tag = tag;
738 vp->v_op = vops;
739 insmntque(vp, mp);
740 *vpp = vp;
741 vp->v_usecount = 1;
742 vp->v_data = 0;
743 splx(s);
744
3b568787 745 vfs_object_create(vp, td);
984263bc
MD
746 return (0);
747}
748
749/*
750 * Move a vnode from one mount queue to another.
751 */
752static void
753insmntque(vp, mp)
1fd87d54
RG
754 struct vnode *vp;
755 struct mount *mp;
984263bc
MD
756{
757
8a8d5d85 758 lwkt_gettoken(&mntvnode_token);
984263bc
MD
759 /*
760 * Delete from old mount point vnode list, if on one.
761 */
762 if (vp->v_mount != NULL) {
763 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
764 ("bad mount point vnode list size"));
765 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
766 vp->v_mount->mnt_nvnodelistsize--;
767 }
768 /*
769 * Insert into list of vnodes for the new mount point, if available.
770 */
771 if ((vp->v_mount = mp) == NULL) {
8a8d5d85 772 lwkt_reltoken(&mntvnode_token);
984263bc
MD
773 return;
774 }
775 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
776 mp->mnt_nvnodelistsize++;
8a8d5d85 777 lwkt_reltoken(&mntvnode_token);
984263bc
MD
778}
779
780/*
781 * Update outstanding I/O count and do wakeup if requested.
782 */
783void
784vwakeup(bp)
1fd87d54 785 struct buf *bp;
984263bc 786{
1fd87d54 787 struct vnode *vp;
984263bc
MD
788
789 bp->b_flags &= ~B_WRITEINPROG;
790 if ((vp = bp->b_vp)) {
791 vp->v_numoutput--;
792 if (vp->v_numoutput < 0)
793 panic("vwakeup: neg numoutput");
794 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
795 vp->v_flag &= ~VBWAIT;
796 wakeup((caddr_t) &vp->v_numoutput);
797 }
798 }
799}
800
801/*
802 * Flush out and invalidate all buffers associated with a vnode.
803 * Called with the underlying object locked.
804 */
805int
3b568787
MD
806vinvalbuf(struct vnode *vp, int flags, struct thread *td,
807 int slpflag, int slptimeo)
984263bc 808{
1fd87d54 809 struct buf *bp;
984263bc
MD
810 struct buf *nbp, *blist;
811 int s, error;
812 vm_object_t object;
813
814 if (flags & V_SAVE) {
815 s = splbio();
816 while (vp->v_numoutput) {
817 vp->v_flag |= VBWAIT;
818 error = tsleep((caddr_t)&vp->v_numoutput,
377d4740 819 slpflag, "vinvlbuf", slptimeo);
984263bc
MD
820 if (error) {
821 splx(s);
822 return (error);
823 }
824 }
825 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
826 splx(s);
3b568787 827 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
984263bc
MD
828 return (error);
829 s = splbio();
830 if (vp->v_numoutput > 0 ||
831 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
832 panic("vinvalbuf: dirty bufs");
833 }
834 splx(s);
835 }
836 s = splbio();
837 for (;;) {
838 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
839 if (!blist)
840 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
841 if (!blist)
842 break;
843
844 for (bp = blist; bp; bp = nbp) {
845 nbp = TAILQ_NEXT(bp, b_vnbufs);
846 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
847 error = BUF_TIMELOCK(bp,
848 LK_EXCLUSIVE | LK_SLEEPFAIL,
849 "vinvalbuf", slpflag, slptimeo);
850 if (error == ENOLCK)
851 break;
852 splx(s);
853 return (error);
854 }
855 /*
856 * XXX Since there are no node locks for NFS, I
857 * believe there is a slight chance that a delayed
858 * write will occur while sleeping just above, so
859 * check for it. Note that vfs_bio_awrite expects
860 * buffers to reside on a queue, while VOP_BWRITE and
861 * brelse do not.
862 */
863 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
864 (flags & V_SAVE)) {
865
866 if (bp->b_vp == vp) {
867 if (bp->b_flags & B_CLUSTEROK) {
868 BUF_UNLOCK(bp);
869 vfs_bio_awrite(bp);
870 } else {
871 bremfree(bp);
872 bp->b_flags |= B_ASYNC;
873 VOP_BWRITE(bp->b_vp, bp);
874 }
875 } else {
876 bremfree(bp);
877 (void) VOP_BWRITE(bp->b_vp, bp);
878 }
879 break;
880 }
881 bremfree(bp);
882 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
883 bp->b_flags &= ~B_ASYNC;
884 brelse(bp);
885 }
886 }
887
888 /*
889 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
890 * have write I/O in-progress but if there is a VM object then the
891 * VM object can also have read-I/O in-progress.
892 */
893 do {
894 while (vp->v_numoutput > 0) {
895 vp->v_flag |= VBWAIT;
377d4740 896 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
984263bc
MD
897 }
898 if (VOP_GETVOBJECT(vp, &object) == 0) {
899 while (object->paging_in_progress)
900 vm_object_pip_sleep(object, "vnvlbx");
901 }
902 } while (vp->v_numoutput > 0);
903
904 splx(s);
905
906 /*
907 * Destroy the copy in the VM cache, too.
908 */
8a8d5d85 909 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
910 if (VOP_GETVOBJECT(vp, &object) == 0) {
911 vm_object_page_remove(object, 0, 0,
912 (flags & V_SAVE) ? TRUE : FALSE);
913 }
8a8d5d85 914 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
915
916 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
917 panic("vinvalbuf: flush failed");
918 return (0);
919}
920
921/*
922 * Truncate a file's buffer and pages to a specified length. This
923 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
924 * sync activity.
925 */
926int
3b568787 927vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
984263bc 928{
dadab5e9 929 struct buf *bp;
984263bc
MD
930 struct buf *nbp;
931 int s, anyfreed;
932 int trunclbn;
933
934 /*
935 * Round up to the *next* lbn.
936 */
937 trunclbn = (length + blksize - 1) / blksize;
938
939 s = splbio();
940restart:
941 anyfreed = 1;
942 for (;anyfreed;) {
943 anyfreed = 0;
944 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
945 nbp = TAILQ_NEXT(bp, b_vnbufs);
946 if (bp->b_lblkno >= trunclbn) {
947 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
948 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
949 goto restart;
950 } else {
951 bremfree(bp);
952 bp->b_flags |= (B_INVAL | B_RELBUF);
953 bp->b_flags &= ~B_ASYNC;
954 brelse(bp);
955 anyfreed = 1;
956 }
957 if (nbp &&
958 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
959 (nbp->b_vp != vp) ||
960 (nbp->b_flags & B_DELWRI))) {
961 goto restart;
962 }
963 }
964 }
965
966 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
967 nbp = TAILQ_NEXT(bp, b_vnbufs);
968 if (bp->b_lblkno >= trunclbn) {
969 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
970 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
971 goto restart;
972 } else {
973 bremfree(bp);
974 bp->b_flags |= (B_INVAL | B_RELBUF);
975 bp->b_flags &= ~B_ASYNC;
976 brelse(bp);
977 anyfreed = 1;
978 }
979 if (nbp &&
980 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
981 (nbp->b_vp != vp) ||
982 (nbp->b_flags & B_DELWRI) == 0)) {
983 goto restart;
984 }
985 }
986 }
987 }
988
989 if (length > 0) {
990restartsync:
991 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
992 nbp = TAILQ_NEXT(bp, b_vnbufs);
993 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
994 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
995 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
996 goto restart;
997 } else {
998 bremfree(bp);
999 if (bp->b_vp == vp) {
1000 bp->b_flags |= B_ASYNC;
1001 } else {
1002 bp->b_flags &= ~B_ASYNC;
1003 }
1004 VOP_BWRITE(bp->b_vp, bp);
1005 }
1006 goto restartsync;
1007 }
1008
1009 }
1010 }
1011
1012 while (vp->v_numoutput > 0) {
1013 vp->v_flag |= VBWAIT;
377d4740 1014 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
984263bc
MD
1015 }
1016
1017 splx(s);
1018
1019 vnode_pager_setsize(vp, length);
1020
1021 return (0);
1022}
1023
1024/*
1025 * Associate a buffer with a vnode.
1026 */
1027void
1028bgetvp(vp, bp)
1fd87d54
RG
1029 struct vnode *vp;
1030 struct buf *bp;
984263bc
MD
1031{
1032 int s;
1033
1034 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1035
1036 vhold(vp);
1037 bp->b_vp = vp;
1038 bp->b_dev = vn_todev(vp);
1039 /*
1040 * Insert onto list for new vnode.
1041 */
1042 s = splbio();
1043 bp->b_xflags |= BX_VNCLEAN;
1044 bp->b_xflags &= ~BX_VNDIRTY;
1045 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1046 splx(s);
1047}
1048
1049/*
1050 * Disassociate a buffer from a vnode.
1051 */
1052void
1053brelvp(bp)
1fd87d54 1054 struct buf *bp;
984263bc
MD
1055{
1056 struct vnode *vp;
1057 struct buflists *listheadp;
1058 int s;
1059
1060 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1061
1062 /*
1063 * Delete from old vnode list, if on one.
1064 */
1065 vp = bp->b_vp;
1066 s = splbio();
1067 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1068 if (bp->b_xflags & BX_VNDIRTY)
1069 listheadp = &vp->v_dirtyblkhd;
1070 else
1071 listheadp = &vp->v_cleanblkhd;
1072 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1073 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1074 }
1075 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1076 vp->v_flag &= ~VONWORKLST;
1077 LIST_REMOVE(vp, v_synclist);
1078 }
1079 splx(s);
1080 bp->b_vp = (struct vnode *) 0;
1081 vdrop(vp);
1082}
1083
1084/*
1085 * The workitem queue.
1086 *
1087 * It is useful to delay writes of file data and filesystem metadata
1088 * for tens of seconds so that quickly created and deleted files need
1089 * not waste disk bandwidth being created and removed. To realize this,
1090 * we append vnodes to a "workitem" queue. When running with a soft
1091 * updates implementation, most pending metadata dependencies should
1092 * not wait for more than a few seconds. Thus, mounted on block devices
1093 * are delayed only about a half the time that file data is delayed.
1094 * Similarly, directory updates are more critical, so are only delayed
1095 * about a third the time that file data is delayed. Thus, there are
1096 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1097 * one each second (driven off the filesystem syncer process). The
1098 * syncer_delayno variable indicates the next queue that is to be processed.
1099 * Items that need to be processed soon are placed in this queue:
1100 *
1101 * syncer_workitem_pending[syncer_delayno]
1102 *
1103 * A delay of fifteen seconds is done by placing the request fifteen
1104 * entries later in the queue:
1105 *
1106 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1107 *
1108 */
1109
1110/*
1111 * Add an item to the syncer work queue.
1112 */
1113static void
1114vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1115{
1116 int s, slot;
1117
1118 s = splbio();
1119
1120 if (vp->v_flag & VONWORKLST) {
1121 LIST_REMOVE(vp, v_synclist);
1122 }
1123
1124 if (delay > syncer_maxdelay - 2)
1125 delay = syncer_maxdelay - 2;
1126 slot = (syncer_delayno + delay) & syncer_mask;
1127
1128 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1129 vp->v_flag |= VONWORKLST;
1130 splx(s);
1131}
1132
bc6dffab 1133struct thread *updatethread;
402ed7e1 1134static void sched_sync (void);
984263bc
MD
1135static struct kproc_desc up_kp = {
1136 "syncer",
1137 sched_sync,
bc6dffab 1138 &updatethread
984263bc
MD
1139};
1140SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1141
1142/*
1143 * System filesystem synchronizer daemon.
1144 */
1145void
1146sched_sync(void)
1147{
1148 struct synclist *slp;
1149 struct vnode *vp;
1150 long starttime;
1151 int s;
0cfcada1 1152 struct thread *td = curthread;
984263bc 1153
bc6dffab 1154 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
1155 SHUTDOWN_PRI_LAST);
1156
1157 for (;;) {
0cfcada1 1158 kproc_suspend_loop();
984263bc
MD
1159
1160 starttime = time_second;
1161
1162 /*
1163 * Push files whose dirty time has expired. Be careful
1164 * of interrupt race on slp queue.
1165 */
1166 s = splbio();
1167 slp = &syncer_workitem_pending[syncer_delayno];
1168 syncer_delayno += 1;
1169 if (syncer_delayno == syncer_maxdelay)
1170 syncer_delayno = 0;
1171 splx(s);
1172
1173 while ((vp = LIST_FIRST(slp)) != NULL) {
1174 if (VOP_ISLOCKED(vp, NULL) == 0) {
dadab5e9 1175 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3b568787 1176 (void) VOP_FSYNC(vp, MNT_LAZY, td);
dadab5e9 1177 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1178 }
1179 s = splbio();
1180 if (LIST_FIRST(slp) == vp) {
1181 /*
1182 * Note: v_tag VT_VFS vps can remain on the
1183 * worklist too with no dirty blocks, but
1184 * since sync_fsync() moves it to a different
1185 * slot we are safe.
1186 */
1187 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1188 !vn_isdisk(vp, NULL))
1189 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1190 /*
1191 * Put us back on the worklist. The worklist
1192 * routine will remove us from our current
1193 * position and then add us back in at a later
1194 * position.
1195 */
1196 vn_syncer_add_to_worklist(vp, syncdelay);
1197 }
1198 splx(s);
1199 }
1200
1201 /*
1202 * Do soft update processing.
1203 */
1204 if (bioops.io_sync)
1205 (*bioops.io_sync)(NULL);
1206
1207 /*
1208 * The variable rushjob allows the kernel to speed up the
1209 * processing of the filesystem syncer process. A rushjob
1210 * value of N tells the filesystem syncer to process the next
1211 * N seconds worth of work on its queue ASAP. Currently rushjob
1212 * is used by the soft update code to speed up the filesystem
1213 * syncer process when the incore state is getting so far
1214 * ahead of the disk that the kernel memory pool is being
1215 * threatened with exhaustion.
1216 */
1217 if (rushjob > 0) {
1218 rushjob -= 1;
1219 continue;
1220 }
1221 /*
1222 * If it has taken us less than a second to process the
1223 * current work, then wait. Otherwise start right over
1224 * again. We can still lose time if any single round
1225 * takes more than two seconds, but it does not really
1226 * matter as we are just trying to generally pace the
1227 * filesystem activity.
1228 */
1229 if (time_second == starttime)
377d4740 1230 tsleep(&lbolt, 0, "syncer", 0);
984263bc
MD
1231 }
1232}
1233
1234/*
1235 * Request the syncer daemon to speed up its work.
1236 * We never push it to speed up more than half of its
1237 * normal turn time, otherwise it could take over the cpu.
f5865223
MD
1238 *
1239 * YYY wchan field protected by the BGL.
984263bc
MD
1240 */
1241int
1242speedup_syncer()
1243{
f5865223
MD
1244 crit_enter();
1245 if (updatethread->td_wchan == &lbolt) { /* YYY */
1246 unsleep(updatethread);
1247 lwkt_schedule(updatethread);
1248 }
1249 crit_exit();
984263bc
MD
1250 if (rushjob < syncdelay / 2) {
1251 rushjob += 1;
1252 stat_rush_requests += 1;
1253 return (1);
1254 }
1255 return(0);
1256}
1257
1258/*
1259 * Associate a p-buffer with a vnode.
1260 *
1261 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1262 * with the buffer. i.e. the bp has not been linked into the vnode or
1263 * ref-counted.
1264 */
1265void
1266pbgetvp(vp, bp)
1fd87d54
RG
1267 struct vnode *vp;
1268 struct buf *bp;
984263bc
MD
1269{
1270
1271 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1272
1273 bp->b_vp = vp;
1274 bp->b_flags |= B_PAGING;
1275 bp->b_dev = vn_todev(vp);
1276}
1277
1278/*
1279 * Disassociate a p-buffer from a vnode.
1280 */
1281void
1282pbrelvp(bp)
1fd87d54 1283 struct buf *bp;
984263bc
MD
1284{
1285
1286 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1287
1288 /* XXX REMOVE ME */
1289 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1290 panic(
1291 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1292 bp,
1293 (int)bp->b_flags
1294 );
1295 }
1296 bp->b_vp = (struct vnode *) 0;
1297 bp->b_flags &= ~B_PAGING;
1298}
1299
1300void
1301pbreassignbuf(bp, newvp)
1302 struct buf *bp;
1303 struct vnode *newvp;
1304{
1305 if ((bp->b_flags & B_PAGING) == 0) {
1306 panic(
1307 "pbreassignbuf() on non phys bp %p",
1308 bp
1309 );
1310 }
1311 bp->b_vp = newvp;
1312}
1313
1314/*
1315 * Reassign a buffer from one vnode to another.
1316 * Used to assign file specific control information
1317 * (indirect blocks) to the vnode to which they belong.
1318 */
1319void
1320reassignbuf(bp, newvp)
1fd87d54
RG
1321 struct buf *bp;
1322 struct vnode *newvp;
984263bc
MD
1323{
1324 struct buflists *listheadp;
1325 int delay;
1326 int s;
1327
1328 if (newvp == NULL) {
1329 printf("reassignbuf: NULL");
1330 return;
1331 }
1332 ++reassignbufcalls;
1333
1334 /*
1335 * B_PAGING flagged buffers cannot be reassigned because their vp
1336 * is not fully linked in.
1337 */
1338 if (bp->b_flags & B_PAGING)
1339 panic("cannot reassign paging buffer");
1340
1341 s = splbio();
1342 /*
1343 * Delete from old vnode list, if on one.
1344 */
1345 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1346 if (bp->b_xflags & BX_VNDIRTY)
1347 listheadp = &bp->b_vp->v_dirtyblkhd;
1348 else
1349 listheadp = &bp->b_vp->v_cleanblkhd;
1350 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1351 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1352 if (bp->b_vp != newvp) {
1353 vdrop(bp->b_vp);
1354 bp->b_vp = NULL; /* for clarification */
1355 }
1356 }
1357 /*
1358 * If dirty, put on list of dirty buffers; otherwise insert onto list
1359 * of clean buffers.
1360 */
1361 if (bp->b_flags & B_DELWRI) {
1362 struct buf *tbp;
1363
1364 listheadp = &newvp->v_dirtyblkhd;
1365 if ((newvp->v_flag & VONWORKLST) == 0) {
1366 switch (newvp->v_type) {
1367 case VDIR:
1368 delay = dirdelay;
1369 break;
1370 case VCHR:
1371 case VBLK:
1372 if (newvp->v_specmountpoint != NULL) {
1373 delay = metadelay;
1374 break;
1375 }
1376 /* fall through */
1377 default:
1378 delay = filedelay;
1379 }
1380 vn_syncer_add_to_worklist(newvp, delay);
1381 }
1382 bp->b_xflags |= BX_VNDIRTY;
1383 tbp = TAILQ_FIRST(listheadp);
1384 if (tbp == NULL ||
1385 bp->b_lblkno == 0 ||
1386 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1387 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1388 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1389 ++reassignbufsortgood;
1390 } else if (bp->b_lblkno < 0) {
1391 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1392 ++reassignbufsortgood;
1393 } else if (reassignbufmethod == 1) {
1394 /*
1395 * New sorting algorithm, only handle sequential case,
1396 * otherwise append to end (but before metadata)
1397 */
1398 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1399 (tbp->b_xflags & BX_VNDIRTY)) {
1400 /*
1401 * Found the best place to insert the buffer
1402 */
1403 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1404 ++reassignbufsortgood;
1405 } else {
1406 /*
1407 * Missed, append to end, but before meta-data.
1408 * We know that the head buffer in the list is
1409 * not meta-data due to prior conditionals.
1410 *
1411 * Indirect effects: NFS second stage write
1412 * tends to wind up here, giving maximum
1413 * distance between the unstable write and the
1414 * commit rpc.
1415 */
1416 tbp = TAILQ_LAST(listheadp, buflists);
1417 while (tbp && tbp->b_lblkno < 0)
1418 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1419 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1420 ++reassignbufsortbad;
1421 }
1422 } else {
1423 /*
1424 * Old sorting algorithm, scan queue and insert
1425 */
1426 struct buf *ttbp;
1427 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1428 (ttbp->b_lblkno < bp->b_lblkno)) {
1429 ++reassignbufloops;
1430 tbp = ttbp;
1431 }
1432 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1433 }
1434 } else {
1435 bp->b_xflags |= BX_VNCLEAN;
1436 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1437 if ((newvp->v_flag & VONWORKLST) &&
1438 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1439 newvp->v_flag &= ~VONWORKLST;
1440 LIST_REMOVE(newvp, v_synclist);
1441 }
1442 }
1443 if (bp->b_vp != newvp) {
1444 bp->b_vp = newvp;
1445 vhold(bp->b_vp);
1446 }
1447 splx(s);
1448}
1449
1450/*
1451 * Create a vnode for a block device.
1452 * Used for mounting the root file system.
1453 */
1454int
1455bdevvp(dev, vpp)
1456 dev_t dev;
1457 struct vnode **vpp;
1458{
1fd87d54 1459 struct vnode *vp;
984263bc
MD
1460 struct vnode *nvp;
1461 int error;
1462
1463 if (dev == NODEV) {
1464 *vpp = NULLVP;
1465 return (ENXIO);
1466 }
1467 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1468 if (error) {
1469 *vpp = NULLVP;
1470 return (error);
1471 }
1472 vp = nvp;
1473 vp->v_type = VBLK;
1474 addalias(vp, dev);
1475 *vpp = vp;
1476 return (0);
1477}
1478
1479/*
7732c8cd 1480 * Add a vnode to the alias list hung off the dev_t.
984263bc
MD
1481 *
1482 * The reason for this gunk is that multiple vnodes can reference
1483 * the same physical device, so checking vp->v_usecount to see
1484 * how many users there are is inadequate; the v_usecount for
1485 * the vnodes need to be accumulated. vcount() does that.
1486 */
1487void
7732c8cd 1488addaliasu(struct vnode *nvp, udev_t nvp_rdev)
984263bc 1489{
7732c8cd 1490 dev_t dev;
984263bc
MD
1491
1492 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1493 panic("addaliasu on non-special vnode");
7732c8cd 1494 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
4908c260
JH
1495 if (dev != NODEV) {
1496 nvp->v_rdev = dev;
7732c8cd 1497 addalias(nvp, dev);
4908c260
JH
1498 } else
1499 nvp->v_rdev = NULL;
984263bc
MD
1500}
1501
1502void
7732c8cd 1503addalias(struct vnode *nvp, dev_t dev)
984263bc
MD
1504{
1505
1506 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1507 panic("addalias on non-special vnode");
1508
1509 nvp->v_rdev = dev;
8a8d5d85 1510 lwkt_gettoken(&spechash_token);
984263bc 1511 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
8a8d5d85 1512 lwkt_reltoken(&spechash_token);
984263bc
MD
1513}
1514
1515/*
1516 * Grab a particular vnode from the free list, increment its
1517 * reference count and lock it. The vnode lock bit is set if the
1518 * vnode is being eliminated in vgone. The process is awakened
1519 * when the transition is completed, and an error returned to
1520 * indicate that the vnode is no longer usable (possibly having
1521 * been changed to a new file system type).
1522 */
1523int
dadab5e9
MD
1524vget(vp, flags, td)
1525 struct vnode *vp;
984263bc 1526 int flags;
dadab5e9 1527 struct thread *td;
984263bc
MD
1528{
1529 int error;
1530
1531 /*
1532 * If the vnode is in the process of being cleaned out for
1533 * another use, we wait for the cleaning to finish and then
1534 * return failure. Cleaning is determined by checking that
1535 * the VXLOCK flag is set.
1536 */
1537 if ((flags & LK_INTERLOCK) == 0) {
8a8d5d85 1538 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1539 }
1540 if (vp->v_flag & VXLOCK) {
1541 if (vp->v_vxproc == curproc) {
1542#if 0
1543 /* this can now occur in normal operation */
1544 log(LOG_INFO, "VXLOCK interlock avoided\n");
1545#endif
1546 } else {
1547 vp->v_flag |= VXWANT;
8a8d5d85 1548 lwkt_reltoken(&vp->v_interlock);
377d4740 1549 tsleep((caddr_t)vp, 0, "vget", 0);
984263bc
MD
1550 return (ENOENT);
1551 }
1552 }
1553
1554 vp->v_usecount++;
1555
1556 if (VSHOULDBUSY(vp))
1557 vbusy(vp);
1558 if (flags & LK_TYPE_MASK) {
dadab5e9 1559 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
984263bc
MD
1560 /*
1561 * must expand vrele here because we do not want
1562 * to call VOP_INACTIVE if the reference count
1563 * drops back to zero since it was never really
1564 * active. We must remove it from the free list
1565 * before sleeping so that multiple processes do
1566 * not try to recycle it.
1567 */
8a8d5d85 1568 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1569 vp->v_usecount--;
1570 if (VSHOULDFREE(vp))
1571 vfree(vp);
1572 else
1573 vlruvp(vp);
8a8d5d85 1574 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1575 }
1576 return (error);
1577 }
8a8d5d85 1578 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1579 return (0);
1580}
1581
1582void
1583vref(struct vnode *vp)
1584{
8a8d5d85 1585 lwkt_gettoken(&vp->v_interlock);
984263bc 1586 vp->v_usecount++;
8a8d5d85 1587 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1588}
1589
1590/*
1591 * Vnode put/release.
1592 * If count drops to zero, call inactive routine and return to freelist.
1593 */
1594void
dadab5e9 1595vrele(struct vnode *vp)
984263bc 1596{
dadab5e9 1597 struct thread *td = curthread; /* XXX */
984263bc
MD
1598
1599 KASSERT(vp != NULL, ("vrele: null vp"));
1600
8a8d5d85 1601 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1602
1603 if (vp->v_usecount > 1) {
1604
1605 vp->v_usecount--;
8a8d5d85 1606 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1607
1608 return;
1609 }
1610
1611 if (vp->v_usecount == 1) {
1612 vp->v_usecount--;
1613 /*
1614 * We must call VOP_INACTIVE with the node locked.
1615 * If we are doing a vpu, the node is already locked,
1616 * but, in the case of vrele, we must explicitly lock
1617 * the vnode before calling VOP_INACTIVE
1618 */
1619
dadab5e9
MD
1620 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1621 VOP_INACTIVE(vp, td);
984263bc
MD
1622 if (VSHOULDFREE(vp))
1623 vfree(vp);
1624 else
1625 vlruvp(vp);
1626 } else {
1627#ifdef DIAGNOSTIC
1628 vprint("vrele: negative ref count", vp);
8a8d5d85 1629 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1630#endif
1631 panic("vrele: negative ref cnt");
1632 }
1633}
1634
1635void
dadab5e9 1636vput(struct vnode *vp)
984263bc 1637{
dadab5e9 1638 struct thread *td = curthread; /* XXX */
984263bc
MD
1639
1640 KASSERT(vp != NULL, ("vput: null vp"));
1641
8a8d5d85 1642 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1643
1644 if (vp->v_usecount > 1) {
1645 vp->v_usecount--;
dadab5e9 1646 VOP_UNLOCK(vp, LK_INTERLOCK, td);
984263bc
MD
1647 return;
1648 }
1649
1650 if (vp->v_usecount == 1) {
1651 vp->v_usecount--;
1652 /*
1653 * We must call VOP_INACTIVE with the node locked.
1654 * If we are doing a vpu, the node is already locked,
1655 * so we just need to release the vnode mutex.
1656 */
8a8d5d85 1657 lwkt_reltoken(&vp->v_interlock);
dadab5e9 1658 VOP_INACTIVE(vp, td);
984263bc
MD
1659 if (VSHOULDFREE(vp))
1660 vfree(vp);
1661 else
1662 vlruvp(vp);
1663 } else {
1664#ifdef DIAGNOSTIC
1665 vprint("vput: negative ref count", vp);
1666#endif
1667 panic("vput: negative ref cnt");
1668 }
1669}
1670
1671/*
1672 * Somebody doesn't want the vnode recycled.
1673 */
1674void
1675vhold(vp)
1fd87d54 1676 struct vnode *vp;
984263bc
MD
1677{
1678 int s;
1679
1680 s = splbio();
1681 vp->v_holdcnt++;
1682 if (VSHOULDBUSY(vp))
1683 vbusy(vp);
1684 splx(s);
1685}
1686
1687/*
1688 * One less who cares about this vnode.
1689 */
1690void
1691vdrop(vp)
1fd87d54 1692 struct vnode *vp;
984263bc
MD
1693{
1694 int s;
1695
1696 s = splbio();
1697 if (vp->v_holdcnt <= 0)
1698 panic("vdrop: holdcnt");
1699 vp->v_holdcnt--;
1700 if (VSHOULDFREE(vp))
1701 vfree(vp);
1702 splx(s);
1703}
1704
1705/*
1706 * Remove any vnodes in the vnode table belonging to mount point mp.
1707 *
1708 * If FORCECLOSE is not specified, there should not be any active ones,
1709 * return error if any are found (nb: this is a user error, not a
1710 * system error). If FORCECLOSE is specified, detach any active vnodes
1711 * that are found.
1712 *
1713 * If WRITECLOSE is set, only flush out regular file vnodes open for
1714 * writing.
1715 *
1716 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1717 *
1718 * `rootrefs' specifies the base reference count for the root vnode
1719 * of this filesystem. The root vnode is considered busy if its
1720 * v_usecount exceeds this value. On a successful return, vflush()
1721 * will call vrele() on the root vnode exactly rootrefs times.
1722 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1723 * be zero.
1724 */
1725#ifdef DIAGNOSTIC
1726static int busyprt = 0; /* print out busy vnodes */
1727SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1728#endif
1729
1730int
1731vflush(mp, rootrefs, flags)
1732 struct mount *mp;
1733 int rootrefs;
1734 int flags;
1735{
dadab5e9 1736 struct thread *td = curthread; /* XXX */
984263bc
MD
1737 struct vnode *vp, *nvp, *rootvp = NULL;
1738 struct vattr vattr;
1739 int busy = 0, error;
1740
1741 if (rootrefs > 0) {
1742 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1743 ("vflush: bad args"));
1744 /*
1745 * Get the filesystem root vnode. We can vput() it
1746 * immediately, since with rootrefs > 0, it won't go away.
1747 */
1748 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1749 return (error);
1750 vput(rootvp);
1751 }
8a8d5d85 1752 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1753loop:
1754 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1755 /*
1756 * Make sure this vnode wasn't reclaimed in getnewvnode().
1757 * Start over if it has (it won't be on the list anymore).
1758 */
1759 if (vp->v_mount != mp)
1760 goto loop;
1761 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1762
8a8d5d85 1763 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1764 /*
1765 * Skip over a vnodes marked VSYSTEM.
1766 */
1767 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
8a8d5d85 1768 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1769 continue;
1770 }
1771 /*
1772 * If WRITECLOSE is set, flush out unlinked but still open
1773 * files (even if open only for reading) and regular file
1774 * vnodes open for writing.
1775 */
1776 if ((flags & WRITECLOSE) &&
1777 (vp->v_type == VNON ||
3b568787 1778 (VOP_GETATTR(vp, &vattr, td) == 0 &&
984263bc
MD
1779 vattr.va_nlink > 0)) &&
1780 (vp->v_writecount == 0 || vp->v_type != VREG)) {
8a8d5d85 1781 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1782 continue;
1783 }
1784
1785 /*
1786 * With v_usecount == 0, all we need to do is clear out the
1787 * vnode data structures and we are done.
1788 */
1789 if (vp->v_usecount == 0) {
8a8d5d85 1790 lwkt_reltoken(&mntvnode_token);
dadab5e9 1791 vgonel(vp, td);
8a8d5d85 1792 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1793 continue;
1794 }
1795
1796 /*
1797 * If FORCECLOSE is set, forcibly close the vnode. For block
1798 * or character devices, revert to an anonymous device. For
1799 * all other files, just kill them.
1800 */
1801 if (flags & FORCECLOSE) {
8a8d5d85 1802 lwkt_reltoken(&mntvnode_token);
984263bc 1803 if (vp->v_type != VBLK && vp->v_type != VCHR) {
dadab5e9 1804 vgonel(vp, td);
984263bc 1805 } else {
dadab5e9 1806 vclean(vp, 0, td);
984263bc
MD
1807 vp->v_op = spec_vnodeop_p;
1808 insmntque(vp, (struct mount *) 0);
1809 }
8a8d5d85 1810 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1811 continue;
1812 }
1813#ifdef DIAGNOSTIC
1814 if (busyprt)
1815 vprint("vflush: busy vnode", vp);
1816#endif
8a8d5d85 1817 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1818 busy++;
1819 }
8a8d5d85 1820 lwkt_reltoken(&mntvnode_token);
984263bc
MD
1821 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1822 /*
1823 * If just the root vnode is busy, and if its refcount
1824 * is equal to `rootrefs', then go ahead and kill it.
1825 */
8a8d5d85 1826 lwkt_gettoken(&rootvp->v_interlock);
984263bc
MD
1827 KASSERT(busy > 0, ("vflush: not busy"));
1828 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1829 if (busy == 1 && rootvp->v_usecount == rootrefs) {
dadab5e9 1830 vgonel(rootvp, td);
984263bc
MD
1831 busy = 0;
1832 } else
8a8d5d85 1833 lwkt_reltoken(&rootvp->v_interlock);
984263bc
MD
1834 }
1835 if (busy)
1836 return (EBUSY);
1837 for (; rootrefs > 0; rootrefs--)
1838 vrele(rootvp);
1839 return (0);
1840}
1841
1842/*
1843 * We do not want to recycle the vnode too quickly.
1844 *
1845 * XXX we can't move vp's around the nvnodelist without really screwing
1846 * up the efficiency of filesystem SYNC and friends. This code is
1847 * disabled until we fix the syncing code's scanning algorithm.
1848 */
1849static void
1850vlruvp(struct vnode *vp)
1851{
1852#if 0
1853 struct mount *mp;
1854
1855 if ((mp = vp->v_mount) != NULL) {
8a8d5d85 1856 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1857 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1858 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
8a8d5d85 1859 lwkt_reltoken(&mntvnode_token);
984263bc
MD
1860 }
1861#endif
1862}
1863
1864/*
1865 * Disassociate the underlying file system from a vnode.
1866 */
1867static void
dadab5e9 1868vclean(struct vnode *vp, int flags, struct thread *td)
984263bc
MD
1869{
1870 int active;
1871
1872 /*
1873 * Check to see if the vnode is in use. If so we have to reference it
1874 * before we clean it out so that its count cannot fall to zero and
1875 * generate a race against ourselves to recycle it.
1876 */
1877 if ((active = vp->v_usecount))
1878 vp->v_usecount++;
1879
1880 /*
1881 * Prevent the vnode from being recycled or brought into use while we
1882 * clean it out.
1883 */
1884 if (vp->v_flag & VXLOCK)
1885 panic("vclean: deadlock");
1886 vp->v_flag |= VXLOCK;
1887 vp->v_vxproc = curproc;
1888 /*
1889 * Even if the count is zero, the VOP_INACTIVE routine may still
1890 * have the object locked while it cleans it out. The VOP_LOCK
1891 * ensures that the VOP_INACTIVE routine is done with its work.
1892 * For active vnodes, it ensures that no other activity can
1893 * occur while the underlying object is being cleaned out.
1894 */
dadab5e9 1895 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
984263bc
MD
1896
1897 /*
1898 * Clean out any buffers associated with the vnode.
1899 */
3b568787 1900 vinvalbuf(vp, V_SAVE, td, 0, 0);
984263bc
MD
1901
1902 VOP_DESTROYVOBJECT(vp);
1903
1904 /*
1905 * If purging an active vnode, it must be closed and
1906 * deactivated before being reclaimed. Note that the
1907 * VOP_INACTIVE will unlock the vnode.
1908 */
1909 if (active) {
1910 if (flags & DOCLOSE)
3b568787 1911 VOP_CLOSE(vp, FNONBLOCK, td);
dadab5e9 1912 VOP_INACTIVE(vp, td);
984263bc
MD
1913 } else {
1914 /*
1915 * Any other processes trying to obtain this lock must first
1916 * wait for VXLOCK to clear, then call the new lock operation.
1917 */
dadab5e9 1918 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1919 }
1920 /*
1921 * Reclaim the vnode.
1922 */
dadab5e9 1923 if (VOP_RECLAIM(vp, td))
984263bc
MD
1924 panic("vclean: cannot reclaim");
1925
1926 if (active) {
1927 /*
1928 * Inline copy of vrele() since VOP_INACTIVE
1929 * has already been called.
1930 */
8a8d5d85 1931 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1932 if (--vp->v_usecount <= 0) {
1933#ifdef DIAGNOSTIC
1934 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1935 vprint("vclean: bad ref count", vp);
1936 panic("vclean: ref cnt");
1937 }
1938#endif
1939 vfree(vp);
1940 }
8a8d5d85 1941 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1942 }
1943
1944 cache_purge(vp);
1945 vp->v_vnlock = NULL;
1946
1947 if (VSHOULDFREE(vp))
1948 vfree(vp);
1949
1950 /*
1951 * Done with purge, notify sleepers of the grim news.
1952 */
1953 vp->v_op = dead_vnodeop_p;
1954 vn_pollgone(vp);
1955 vp->v_tag = VT_NON;
1956 vp->v_flag &= ~VXLOCK;
1957 vp->v_vxproc = NULL;
1958 if (vp->v_flag & VXWANT) {
1959 vp->v_flag &= ~VXWANT;
1960 wakeup((caddr_t) vp);
1961 }
1962}
1963
1964/*
1965 * Eliminate all activity associated with the requested vnode
1966 * and with all vnodes aliased to the requested vnode.
1967 */
1968int
1969vop_revoke(ap)
1970 struct vop_revoke_args /* {
1971 struct vnode *a_vp;
1972 int a_flags;
1973 } */ *ap;
1974{
1975 struct vnode *vp, *vq;
1976 dev_t dev;
1977
1978 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1979
1980 vp = ap->a_vp;
1981 /*
1982 * If a vgone (or vclean) is already in progress,
1983 * wait until it is done and return.
1984 */
1985 if (vp->v_flag & VXLOCK) {
1986 vp->v_flag |= VXWANT;
8a8d5d85 1987 lwkt_reltoken(&vp->v_interlock);
377d4740 1988 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
984263bc
MD
1989 return (0);
1990 }
1991 dev = vp->v_rdev;
1992 for (;;) {
8a8d5d85 1993 lwkt_gettoken(&spechash_token);
984263bc 1994 vq = SLIST_FIRST(&dev->si_hlist);
8a8d5d85 1995 lwkt_reltoken(&spechash_token);
984263bc
MD
1996 if (!vq)
1997 break;
1998 vgone(vq);
1999 }
2000 return (0);
2001}
2002
2003/*
2004 * Recycle an unused vnode to the front of the free list.
2005 * Release the passed interlock if the vnode will be recycled.
2006 */
2007int
8a8d5d85 2008vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td)
984263bc 2009{
8a8d5d85 2010 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
2011 if (vp->v_usecount == 0) {
2012 if (inter_lkp) {
8a8d5d85 2013 lwkt_reltoken(inter_lkp);
984263bc 2014 }
dadab5e9 2015 vgonel(vp, td);
984263bc
MD
2016 return (1);
2017 }
8a8d5d85 2018 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2019 return (0);
2020}
2021
2022/*
2023 * Eliminate all activity associated with a vnode
2024 * in preparation for reuse.
2025 */
2026void
dadab5e9 2027vgone(struct vnode *vp)
984263bc 2028{
dadab5e9 2029 struct thread *td = curthread; /* XXX */
984263bc 2030
8a8d5d85 2031 lwkt_gettoken(&vp->v_interlock);
dadab5e9 2032 vgonel(vp, td);
984263bc
MD
2033}
2034
2035/*
2036 * vgone, with the vp interlock held.
2037 */
2038void
dadab5e9 2039vgonel(struct vnode *vp, struct thread *td)
984263bc
MD
2040{
2041 int s;
2042
2043 /*
2044 * If a vgone (or vclean) is already in progress,
2045 * wait until it is done and return.
2046 */
2047 if (vp->v_flag & VXLOCK) {
2048 vp->v_flag |= VXWANT;
8a8d5d85 2049 lwkt_reltoken(&vp->v_interlock);
377d4740 2050 tsleep((caddr_t)vp, 0, "vgone", 0);
984263bc
MD
2051 return;
2052 }
2053
2054 /*
2055 * Clean out the filesystem specific data.
2056 */
dadab5e9 2057 vclean(vp, DOCLOSE, td);
8a8d5d85 2058 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
2059
2060 /*
2061 * Delete from old mount point vnode list, if on one.
2062 */
2063 if (vp->v_mount != NULL)
2064 insmntque(vp, (struct mount *)0);
2065 /*
2066 * If special device, remove it from special device alias list
2067 * if it is on one.
2068 */
2069 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
8a8d5d85 2070 lwkt_gettoken(&spechash_token);
984263bc
MD
2071 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2072 freedev(vp->v_rdev);
8a8d5d85 2073 lwkt_reltoken(&spechash_token);
984263bc
MD
2074 vp->v_rdev = NULL;
2075 }
2076
2077 /*
2078 * If it is on the freelist and not already at the head,
2079 * move it to the head of the list. The test of the
2080 * VDOOMED flag and the reference count of zero is because
2081 * it will be removed from the free list by getnewvnode,
2082 * but will not have its reference count incremented until
2083 * after calling vgone. If the reference count were
2084 * incremented first, vgone would (incorrectly) try to
2085 * close the previous instance of the underlying object.
2086 */
2087 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2088 s = splbio();
8a8d5d85 2089 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2090 if (vp->v_flag & VFREE)
2091 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2092 else
2093 freevnodes++;
2094 vp->v_flag |= VFREE;
2095 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
8a8d5d85 2096 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2097 splx(s);
2098 }
2099
2100 vp->v_type = VBAD;
8a8d5d85 2101 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2102}
2103
2104/*
2105 * Lookup a vnode by device number.
2106 */
2107int
2108vfinddev(dev, type, vpp)
2109 dev_t dev;
2110 enum vtype type;
2111 struct vnode **vpp;
2112{
2113 struct vnode *vp;
2114
8a8d5d85 2115 lwkt_gettoken(&spechash_token);
984263bc
MD
2116 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2117 if (type == vp->v_type) {
2118 *vpp = vp;
8a8d5d85 2119 lwkt_reltoken(&spechash_token);
984263bc
MD
2120 return (1);
2121 }
2122 }
8a8d5d85 2123 lwkt_reltoken(&spechash_token);
984263bc
MD
2124 return (0);
2125}
2126
2127/*
2128 * Calculate the total number of references to a special device.
2129 */
2130int
2131vcount(vp)
2132 struct vnode *vp;
2133{
2134 struct vnode *vq;
2135 int count;
2136
2137 count = 0;
8a8d5d85 2138 lwkt_gettoken(&spechash_token);
984263bc
MD
2139 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2140 count += vq->v_usecount;
8a8d5d85 2141 lwkt_reltoken(&spechash_token);
984263bc
MD
2142 return (count);
2143}
2144
2145/*
2146 * Same as above, but using the dev_t as argument
2147 */
2148
2149int
2150count_dev(dev)
2151 dev_t dev;
2152{
2153 struct vnode *vp;
2154
2155 vp = SLIST_FIRST(&dev->si_hlist);
2156 if (vp == NULL)
2157 return (0);
2158 return(vcount(vp));
2159}
2160
2161/*
2162 * Print out a description of a vnode.
2163 */
2164static char *typename[] =
2165{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2166
2167void
2168vprint(label, vp)
2169 char *label;
2170 struct vnode *vp;
2171{
2172 char buf[96];
2173
2174 if (label != NULL)
2175 printf("%s: %p: ", label, (void *)vp);
2176 else
2177 printf("%p: ", (void *)vp);
2178 printf("type %s, usecount %d, writecount %d, refcount %d,",
2179 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2180 vp->v_holdcnt);
2181 buf[0] = '\0';
2182 if (vp->v_flag & VROOT)
2183 strcat(buf, "|VROOT");
2184 if (vp->v_flag & VTEXT)
2185 strcat(buf, "|VTEXT");
2186 if (vp->v_flag & VSYSTEM)
2187 strcat(buf, "|VSYSTEM");
2188 if (vp->v_flag & VXLOCK)
2189 strcat(buf, "|VXLOCK");
2190 if (vp->v_flag & VXWANT)
2191 strcat(buf, "|VXWANT");
2192 if (vp->v_flag & VBWAIT)
2193 strcat(buf, "|VBWAIT");
2194 if (vp->v_flag & VDOOMED)
2195 strcat(buf, "|VDOOMED");
2196 if (vp->v_flag & VFREE)
2197 strcat(buf, "|VFREE");
2198 if (vp->v_flag & VOBJBUF)
2199 strcat(buf, "|VOBJBUF");
2200 if (buf[0] != '\0')
2201 printf(" flags (%s)", &buf[1]);
2202 if (vp->v_data == NULL) {
2203 printf("\n");
2204 } else {
2205 printf("\n\t");
2206 VOP_PRINT(vp);
2207 }
2208}
2209
2210#ifdef DDB
2211#include <ddb/ddb.h>
2212/*
2213 * List all of the locked vnodes in the system.
2214 * Called when debugging the kernel.
2215 */
2216DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2217{
dadab5e9 2218 struct thread *td = curthread; /* XXX */
984263bc
MD
2219 struct mount *mp, *nmp;
2220 struct vnode *vp;
2221
2222 printf("Locked vnodes\n");
8a8d5d85 2223 lwkt_gettoken(&mountlist_token);
984263bc 2224 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 2225 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
984263bc
MD
2226 nmp = TAILQ_NEXT(mp, mnt_list);
2227 continue;
2228 }
2229 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2230 if (VOP_ISLOCKED(vp, NULL))
2231 vprint((char *)0, vp);
2232 }
8a8d5d85 2233 lwkt_gettoken(&mountlist_token);
984263bc 2234 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 2235 vfs_unbusy(mp, td);
984263bc 2236 }
8a8d5d85 2237 lwkt_reltoken(&mountlist_token);
984263bc
MD
2238}
2239#endif
2240
2241/*
2242 * Top level filesystem related information gathering.
2243 */
402ed7e1 2244static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
984263bc
MD
2245
2246static int
2247vfs_sysctl(SYSCTL_HANDLER_ARGS)
2248{
2249 int *name = (int *)arg1 - 1; /* XXX */
2250 u_int namelen = arg2 + 1; /* XXX */
2251 struct vfsconf *vfsp;
2252
2253#if 1 || defined(COMPAT_PRELITE2)
2254 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2255 if (namelen == 1)
2256 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2257#endif
2258
2259#ifdef notyet
2260 /* all sysctl names at this level are at least name and field */
2261 if (namelen < 2)
2262 return (ENOTDIR); /* overloaded */
2263 if (name[0] != VFS_GENERIC) {
2264 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2265 if (vfsp->vfc_typenum == name[0])
2266 break;
2267 if (vfsp == NULL)
2268 return (EOPNOTSUPP);
2269 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2270 oldp, oldlenp, newp, newlen, p));
2271 }
2272#endif
2273 switch (name[1]) {
2274 case VFS_MAXTYPENUM:
2275 if (namelen != 2)
2276 return (ENOTDIR);
2277 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2278 case VFS_CONF:
2279 if (namelen != 3)
2280 return (ENOTDIR); /* overloaded */
2281 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2282 if (vfsp->vfc_typenum == name[2])
2283 break;
2284 if (vfsp == NULL)
2285 return (EOPNOTSUPP);
2286 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2287 }
2288 return (EOPNOTSUPP);
2289}
2290
2291SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2292 "Generic filesystem");
2293
2294#if 1 || defined(COMPAT_PRELITE2)
2295
2296static int
2297sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2298{
2299 int error;
2300 struct vfsconf *vfsp;
2301 struct ovfsconf ovfs;
2302
2303 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2304 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2305 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2306 ovfs.vfc_index = vfsp->vfc_typenum;
2307 ovfs.vfc_refcount = vfsp->vfc_refcount;
2308 ovfs.vfc_flags = vfsp->vfc_flags;
2309 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2310 if (error)
2311 return error;
2312 }
2313 return 0;
2314}
2315
2316#endif /* 1 || COMPAT_PRELITE2 */
2317
2318#if 0
2319#define KINFO_VNODESLOP 10
2320/*
2321 * Dump vnode list (via sysctl).
2322 * Copyout address of vnode followed by vnode.
2323 */
2324/* ARGSUSED */
2325static int
2326sysctl_vnode(SYSCTL_HANDLER_ARGS)
2327{
2328 struct proc *p = curproc; /* XXX */
2329 struct mount *mp, *nmp;
2330 struct vnode *nvp, *vp;
2331 int error;
2332
2333#define VPTRSZ sizeof (struct vnode *)
2334#define VNODESZ sizeof (struct vnode)
2335
2336 req->lock = 0;
2337 if (!req->oldptr) /* Make an estimate */
2338 return (SYSCTL_OUT(req, 0,
2339 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2340
8a8d5d85 2341 lwkt_gettoken(&mountlist_token);
984263bc 2342 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 2343 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) {
984263bc
MD
2344 nmp = TAILQ_NEXT(mp, mnt_list);
2345 continue;
2346 }
2347again:
8a8d5d85 2348 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2349 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2350 vp != NULL;
2351 vp = nvp) {
2352 /*
2353 * Check that the vp is still associated with
2354 * this filesystem. RACE: could have been
2355 * recycled onto the same filesystem.
2356 */
2357 if (vp->v_mount != mp) {
8a8d5d85 2358 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2359 goto again;
2360 }
2361 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
8a8d5d85 2362 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2363 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2364 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2365 return (error);
8a8d5d85 2366 lwkt_gettoken(&mntvnode_token);
984263bc 2367 }
8a8d5d85
MD
2368 lwkt_reltoken(&mntvnode_token);
2369 lwkt_gettoken(&mountlist_token);
984263bc
MD
2370 nmp = TAILQ_NEXT(mp, mnt_list);
2371 vfs_unbusy(mp, p);
2372 }
8a8d5d85 2373 lwkt_reltoken(&mountlist_token);
984263bc
MD
2374
2375 return (0);
2376}
2377#endif
2378
2379/*
2380 * XXX
2381 * Exporting the vnode list on large systems causes them to crash.
2382 * Exporting the vnode list on medium systems causes sysctl to coredump.
2383 */
2384#if 0
2385SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2386 0, 0, sysctl_vnode, "S,vnode", "");
2387#endif
2388
2389/*
2390 * Check to see if a filesystem is mounted on a block device.
2391 */
2392int
2393vfs_mountedon(vp)
2394 struct vnode *vp;
2395{
2396
2397 if (vp->v_specmountpoint != NULL)
2398 return (EBUSY);
2399 return (0);
2400}
2401
2402/*
2403 * Unmount all filesystems. The list is traversed in reverse order
2404 * of mounting to avoid dependencies.
2405 */
2406void
2407vfs_unmountall()
2408{
2409 struct mount *mp;
dadab5e9 2410 struct thread *td = curthread;
984263bc
MD
2411 int error;
2412
dadab5e9
MD
2413 if (td->td_proc == NULL)
2414 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2415
984263bc
MD
2416 /*
2417 * Since this only runs when rebooting, it is not interlocked.
2418 */
2419 while(!TAILQ_EMPTY(&mountlist)) {
2420 mp = TAILQ_LAST(&mountlist, mntlist);
dadab5e9 2421 error = dounmount(mp, MNT_FORCE, td);
984263bc
MD
2422 if (error) {
2423 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2424 printf("unmount of %s failed (",
2425 mp->mnt_stat.f_mntonname);
2426 if (error == EBUSY)
2427 printf("BUSY)\n");
2428 else
2429 printf("%d)\n", error);
2430 } else {
2431 /* The unmount has removed mp from the mountlist */
2432 }
2433 }
2434}
2435
2436/*
2437 * Build hash lists of net addresses and hang them off the mount point.
2438 * Called by ufs_mount() to set up the lists of export addresses.
2439 */
2440static int
2441vfs_hang_addrlist(mp, nep, argp)
2442 struct mount *mp;
2443 struct netexport *nep;
2444 struct export_args *argp;
2445{
1fd87d54
RG
2446 struct netcred *np;
2447 struct radix_node_head *rnh;
2448 int i;
984263bc
MD
2449 struct radix_node *rn;
2450 struct sockaddr *saddr, *smask = 0;
2451 struct domain *dom;
2452 int error;
2453
2454 if (argp->ex_addrlen == 0) {
2455 if (mp->mnt_flag & MNT_DEFEXPORTED)
2456 return (EPERM);
2457 np = &nep->ne_defexported;
2458 np->netc_exflags = argp->ex_flags;
2459 np->netc_anon = argp->ex_anon;
2460 np->netc_anon.cr_ref = 1;
2461 mp->mnt_flag |= MNT_DEFEXPORTED;
2462 return (0);
2463 }
2464
0260ddf9
MD
2465 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2466 return (EINVAL);
2467 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
984263bc
MD
2468 return (EINVAL);
2469
2470 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2471 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2472 bzero((caddr_t) np, i);
2473 saddr = (struct sockaddr *) (np + 1);
2474 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2475 goto out;
2476 if (saddr->sa_len > argp->ex_addrlen)
2477 saddr->sa_len = argp->ex_addrlen;
2478 if (argp->ex_masklen) {
2479 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2480 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2481 if (error)
2482 goto out;
2483 if (smask->sa_len > argp->ex_masklen)
2484 smask->sa_len = argp->ex_masklen;
2485 }
2486 i = saddr->sa_family;
2487 if ((rnh = nep->ne_rtable[i]) == 0) {
2488 /*
2489 * Seems silly to initialize every AF when most are not used,
2490 * do so on demand here
2491 */
2492 for (dom = domains; dom; dom = dom->dom_next)
2493 if (dom->dom_family == i && dom->dom_rtattach) {
2494 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2495 dom->dom_rtoffset);
2496 break;
2497 }
2498 if ((rnh = nep->ne_rtable[i]) == 0) {
2499 error = ENOBUFS;
2500 goto out;
2501 }
2502 }
2503 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2504 np->netc_rnodes);
2505 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2506 error = EPERM;
2507 goto out;
2508 }
2509 np->netc_exflags = argp->ex_flags;
2510 np->netc_anon = argp->ex_anon;
2511 np->netc_anon.cr_ref = 1;
2512 return (0);
2513out:
2514 free(np, M_NETADDR);
2515 return (error);
2516}
2517
2518/* ARGSUSED */
2519static int
2520vfs_free_netcred(rn, w)
2521 struct radix_node *rn;
2522 void *w;
2523{
1fd87d54 2524 struct radix_node_head *rnh = (struct radix_node_head *) w;
984263bc
MD
2525
2526 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2527 free((caddr_t) rn, M_NETADDR);
2528 return (0);
2529}
2530
2531/*
2532 * Free the net address hash lists that are hanging off the mount points.
2533 */
2534static void
2535vfs_free_addrlist(nep)
2536 struct netexport *nep;
2537{
1fd87d54
RG
2538 int i;
2539 struct radix_node_head *rnh;
984263bc
MD
2540
2541 for (i = 0; i <= AF_MAX; i++)
2542 if ((rnh = nep->ne_rtable[i])) {
2543 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2544 (caddr_t) rnh);
2545 free((caddr_t) rnh, M_RTABLE);
2546 nep->ne_rtable[i] = 0;
2547 }
2548}
2549
2550int
2551vfs_export(mp, nep, argp)
2552 struct mount *mp;
2553 struct netexport *nep;
2554 struct export_args *argp;
2555{
2556 int error;
2557
2558 if (argp->ex_flags & MNT_DELEXPORT) {
2559 if (mp->mnt_flag & MNT_EXPUBLIC) {
2560 vfs_setpublicfs(NULL, NULL, NULL);
2561 mp->mnt_flag &= ~MNT_EXPUBLIC;
2562 }
2563 vfs_free_addrlist(nep);
2564 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2565 }
2566 if (argp->ex_flags & MNT_EXPORTED) {
2567 if (argp->ex_flags & MNT_EXPUBLIC) {
2568 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2569 return (error);
2570 mp->mnt_flag |= MNT_EXPUBLIC;
2571 }
2572 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2573 return (error);
2574 mp->mnt_flag |= MNT_EXPORTED;
2575 }
2576 return (0);
2577}
2578
2579
2580/*
2581 * Set the publicly exported filesystem (WebNFS). Currently, only
2582 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2583 */
2584int
2585vfs_setpublicfs(mp, nep, argp)
2586 struct mount *mp;
2587 struct netexport *nep;
2588 struct export_args *argp;
2589{
2590 int error;
2591 struct vnode *rvp;
2592 char *cp;
2593
2594 /*
2595 * mp == NULL -> invalidate the current info, the FS is
2596 * no longer exported. May be called from either vfs_export
2597 * or unmount, so check if it hasn't already been done.
2598 */
2599 if (mp == NULL) {
2600 if (nfs_pub.np_valid) {
2601 nfs_pub.np_valid = 0;
2602 if (nfs_pub.np_index != NULL) {
2603 FREE(nfs_pub.np_index, M_TEMP);
2604 nfs_pub.np_index = NULL;
2605 }
2606 }
2607 return (0);
2608 }
2609
2610 /*
2611 * Only one allowed at a time.
2612 */
2613 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2614 return (EBUSY);
2615
2616 /*
2617 * Get real filehandle for root of exported FS.
2618 */
2619 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2620 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2621
2622 if ((error = VFS_ROOT(mp, &rvp)))
2623 return (error);
2624
2625 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2626 return (error);
2627
2628 vput(rvp);
2629
2630 /*
2631 * If an indexfile was specified, pull it in.
2632 */
2633 if (argp->ex_indexfile != NULL) {
2634 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2635 M_WAITOK);
2636 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2637 MAXNAMLEN, (size_t *)0);
2638 if (!error) {
2639 /*
2640 * Check for illegal filenames.
2641 */
2642 for (cp = nfs_pub.np_index; *cp; cp++) {
2643 if (*cp == '/') {
2644 error = EINVAL;
2645 break;
2646 }
2647 }
2648 }
2649 if (error) {
2650 FREE(nfs_pub.np_index, M_TEMP);
2651 return (error);
2652 }
2653 }
2654
2655 nfs_pub.np_mount = mp;
2656 nfs_pub.np_valid = 1;
2657 return (0);
2658}
2659
2660struct netcred *
2661vfs_export_lookup(mp, nep, nam)
1fd87d54 2662 struct mount *mp;
984263bc
MD
2663 struct netexport *nep;
2664 struct sockaddr *nam;
2665{
1fd87d54
RG
2666 struct netcred *np;
2667 struct radix_node_head *rnh;
984263bc
MD
2668 struct sockaddr *saddr;
2669
2670 np = NULL;
2671 if (mp->mnt_flag & MNT_EXPORTED) {
2672 /*
2673 * Lookup in the export list first.
2674 */
2675 if (nam != NULL) {
2676 saddr = nam;
2677 rnh = nep->ne_rtable[saddr->sa_family];
2678 if (rnh != NULL) {
2679 np = (struct netcred *)
2680 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2681 rnh);
2682 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2683 np = NULL;
2684 }
2685 }
2686 /*
2687 * If no address match, use the default if it exists.
2688 */
2689 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2690 np = &nep->ne_defexported;
2691 }
2692 return (np);
2693}
2694
2695/*
2696 * perform msync on all vnodes under a mount point
2697 * the mount point must be locked.
2698 */
2699void
2700vfs_msync(struct mount *mp, int flags)
2701{
dadab5e9 2702 struct thread *td = curthread; /* XXX */
984263bc
MD
2703 struct vnode *vp, *nvp;
2704 struct vm_object *obj;
2705 int tries;
2706
2707 tries = 5;
8a8d5d85 2708 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2709loop:
2710 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2711 if (vp->v_mount != mp) {
2712 if (--tries > 0)
2713 goto loop;
2714 break;
2715 }
2716 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2717
2718 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2719 continue;
2720
2721 /*
2722 * There could be hundreds of thousands of vnodes, we cannot
2723 * afford to do anything heavy-weight until we have a fairly
2724 * good indication that there is something to do.
2725 */
2726 if ((vp->v_flag & VOBJDIRTY) &&
2727 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
8a8d5d85 2728 lwkt_reltoken(&mntvnode_token);
984263bc 2729 if (!vget(vp,
dadab5e9 2730 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
984263bc
MD
2731 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2732 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2733 }
2734 vput(vp);
2735 }
8a8d5d85 2736 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2737 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2738 if (--tries > 0)
2739 goto loop;
2740 break;
2741 }
2742 }
2743 }
8a8d5d85 2744 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2745}
2746
2747/*
2748 * Create the VM object needed for VMIO and mmap support. This
2749 * is done for all VREG files in the system. Some filesystems might
2750 * afford the additional metadata buffering capability of the
2751 * VMIO code by making the device node be VMIO mode also.
2752 *
2753 * vp must be locked when vfs_object_create is called.
2754 */
2755int
3b568787 2756vfs_object_create(struct vnode *vp, struct thread *td)
984263bc 2757{
3b568787 2758 return (VOP_CREATEVOBJECT(vp, td));
984263bc
MD
2759}
2760
2761void
2762vfree(vp)
2763 struct vnode *vp;
2764{
2765 int s;
2766
2767 s = splbio();
8a8d5d85 2768 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2769 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2770 if (vp->v_flag & VAGE) {
2771 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2772 } else {
2773 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2774 }
2775 freevnodes++;
8a8d5d85 2776 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2777 vp->v_flag &= ~VAGE;
2778 vp->v_flag |= VFREE;
2779 splx(s);
2780}
2781
2782void
2783vbusy(vp)
2784 struct vnode *vp;
2785{
2786 int s;
2787
2788 s = splbio();
8a8d5d85 2789 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2790 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2791 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2792 freevnodes--;
8a8d5d85 2793 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2794 vp->v_flag &= ~(VFREE|VAGE);
2795 splx(s);
2796}
2797
2798/*
2799 * Record a process's interest in events which might happen to
2800 * a vnode. Because poll uses the historic select-style interface
2801 * internally, this routine serves as both the ``check for any
2802 * pending events'' and the ``record my interest in future events''
2803 * functions. (These are done together, while the lock is held,
2804 * to avoid race conditions.)
2805 */
2806int
dadab5e9 2807vn_pollrecord(struct vnode *vp, struct thread *td, int events)
984263bc 2808{
8a8d5d85 2809 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2810 if (vp->v_pollinfo.vpi_revents & events) {
2811 /*
2812 * This leaves events we are not interested
2813 * in available for the other process which
2814 * which presumably had requested them
2815 * (otherwise they would never have been
2816 * recorded).
2817 */
2818 events &= vp->v_pollinfo.vpi_revents;
2819 vp->v_pollinfo.vpi_revents &= ~events;
2820
8a8d5d85 2821 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2822 return events;
2823 }
2824 vp->v_pollinfo.vpi_events |= events;
dadab5e9 2825 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
8a8d5d85 2826 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2827 return 0;
2828}
2829
2830/*
2831 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2832 * it is possible for us to miss an event due to race conditions, but
2833 * that condition is expected to be rare, so for the moment it is the
2834 * preferred interface.
2835 */
2836void
2837vn_pollevent(vp, events)
2838 struct vnode *vp;
2839 short events;
2840{
8a8d5d85 2841 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2842 if (vp->v_pollinfo.vpi_events & events) {
2843 /*
2844 * We clear vpi_events so that we don't
2845 * call selwakeup() twice if two events are
2846 * posted before the polling process(es) is
2847 * awakened. This also ensures that we take at
2848 * most one selwakeup() if the polling process
2849 * is no longer interested. However, it does
2850 * mean that only one event can be noticed at
2851 * a time. (Perhaps we should only clear those
2852 * event bits which we note?) XXX
2853 */
2854 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2855 vp->v_pollinfo.vpi_revents |= events;
2856 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2857 }
8a8d5d85 2858 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2859}
2860
2861/*
2862 * Wake up anyone polling on vp because it is being revoked.
2863 * This depends on dead_poll() returning POLLHUP for correct
2864 * behavior.
2865 */
2866void
2867vn_pollgone(vp)
2868 struct vnode *vp;
2869{
8a8d5d85 2870 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2871 if (vp->v_pollinfo.vpi_events) {
2872 vp->v_pollinfo.vpi_events = 0;
2873 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2874 }
8a8d5d85 2875 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2876}
2877
2878
2879
2880/*
2881 * Routine to create and manage a filesystem syncer vnode.
2882 */
402ed7e1
RG
2883#define sync_close ((int (*) (struct vop_close_args *))nullop)
2884static int sync_fsync (struct vop_fsync_args *);
2885static int sync_inactive (struct vop_inactive_args *);
2886static int sync_reclaim (struct vop_reclaim_args *);
2887#define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
2888#define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
2889static int sync_print (struct vop_print_args *);
2890#define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
984263bc
MD
2891
2892static vop_t **sync_vnodeop_p;
2893static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2894 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2895 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2896 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2897 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2898 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2899 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2900 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2901 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2902 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2903 { NULL, NULL }
2904};
2905static struct vnodeopv_desc sync_vnodeop_opv_desc =
2906 { &sync_vnodeop_p, sync_vnodeop_entries };
2907
2908VNODEOP_SET(sync_vnodeop_opv_desc);
2909
2910/*
2911 * Create a new filesystem syncer vnode for the specified mount point.
2912 */
2913int
2914vfs_allocate_syncvnode(mp)
2915 struct mount *mp;
2916{
2917 struct vnode *vp;
2918 static long start, incr, next;
2919 int error;
2920
2921 /* Allocate a new vnode */
2922 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2923 mp->mnt_syncer = NULL;
2924 return (error);
2925 }
2926 vp->v_type = VNON;
2927 /*
2928 * Place the vnode onto the syncer worklist. We attempt to
2929 * scatter them about on the list so that they will go off
2930 * at evenly distributed times even if all the filesystems
2931 * are mounted at once.
2932 */
2933 next += incr;
2934 if (next == 0 || next > syncer_maxdelay) {
2935 start /= 2;
2936 incr /= 2;
2937 if (start == 0) {
2938 start = syncer_maxdelay / 2;
2939 incr = syncer_maxdelay;
2940 }
2941 next = start;
2942 }
2943 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2944 mp->mnt_syncer = vp;
2945 return (0);
2946}
2947
2948/*
2949 * Do a lazy sync of the filesystem.
2950 */
2951static int
2952sync_fsync(ap)
2953 struct vop_fsync_args /* {
2954 struct vnode *a_vp;
2955 struct ucred *a_cred;
2956 int a_waitfor;
dadab5e9 2957 struct thread *a_td;
984263bc
MD
2958 } */ *ap;
2959{
2960 struct vnode *syncvp = ap->a_vp;
2961 struct mount *mp = syncvp->v_mount;
dadab5e9 2962 struct thread *td = ap->a_td;
984263bc
MD
2963 int asyncflag;
2964
2965 /*
2966 * We only need to do something if this is a lazy evaluation.
2967 */
2968 if (ap->a_waitfor != MNT_LAZY)
2969 return (0);
2970
2971 /*
2972 * Move ourselves to the back of the sync list.
2973 */
2974 vn_syncer_add_to_worklist(syncvp, syncdelay);
2975
2976 /*
2977 * Walk the list of vnodes pushing all that are dirty and
2978 * not already on the sync list.
2979 */
8a8d5d85
MD
2980 lwkt_gettoken(&mountlist_token);
2981 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) {
2982 lwkt_reltoken(&mountlist_token);
984263bc
MD
2983 return (0);
2984 }
2985 asyncflag = mp->mnt_flag & MNT_ASYNC;
2986 mp->mnt_flag &= ~MNT_ASYNC;
2987 vfs_msync(mp, MNT_NOWAIT);
3b568787 2988 VFS_SYNC(mp, MNT_LAZY, td);
984263bc
MD
2989 if (asyncflag)
2990 mp->mnt_flag |= MNT_ASYNC;
dadab5e9 2991 vfs_unbusy(mp, td);
984263bc
MD
2992 return (0);
2993}
2994
2995/*
2996 * The syncer vnode is no referenced.
2997 */
2998static int
2999sync_inactive(ap)
3000 struct vop_inactive_args /* {
3001 struct vnode *a_vp;
3002 struct proc *a_p;
3003 } */ *ap;
3004{
3005
3006 vgone(ap->a_vp);
3007 return (0);
3008}
3009
3010/*
3011 * The syncer vnode is no longer needed and is being decommissioned.
3012 *
3013 * Modifications to the worklist must be protected at splbio().
3014 */
3015static int
3016sync_reclaim(ap)
3017 struct vop_reclaim_args /* {
3018 struct vnode *a_vp;
3019 } */ *ap;
3020{
3021 struct vnode *vp = ap->a_vp;
3022 int s;
3023
3024 s = splbio();
3025 vp->v_mount->mnt_syncer = NULL;
3026 if (vp->v_flag & VONWORKLST) {
3027 LIST_REMOVE(vp, v_synclist);
3028 vp->v_flag &= ~VONWORKLST;
3029 }
3030 splx(s);
3031
3032 return (0);
3033}
3034
3035/*
3036 * Print out a syncer vnode.
3037 */
3038static int
3039sync_print(ap)
3040 struct vop_print_args /* {
3041 struct vnode *a_vp;
3042 } */ *ap;
3043{
3044 struct vnode *vp = ap->a_vp;
3045
3046 printf("syncer vnode");
3047 if (vp->v_vnlock != NULL)
3048 lockmgr_printinfo(vp->v_vnlock);
3049 printf("\n");
3050 return (0);
3051}
3052
3053/*
3054 * extract the dev_t from a VBLK or VCHR
3055 */
3056dev_t
3057vn_todev(vp)
3058 struct vnode *vp;
3059{
3060 if (vp->v_type != VBLK && vp->v_type != VCHR)
3061 return (NODEV);
3062 return (vp->v_rdev);
3063}
3064
3065/*
3066 * Check if vnode represents a disk device
3067 */
3068int
3069vn_isdisk(vp, errp)
3070 struct vnode *vp;
3071 int *errp;
3072{
3073 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3074 if (errp != NULL)
3075 *errp = ENOTBLK;
3076 return (0);
3077 }
3078 if (vp->v_rdev == NULL) {
3079 if (errp != NULL)
3080 *errp = ENXIO;
3081 return (0);
3082 }
335dda38 3083 if (!dev_dport(vp->v_rdev)) {
984263bc
MD
3084 if (errp != NULL)
3085 *errp = ENXIO;
3086 return (0);
3087 }
335dda38 3088 if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
984263bc
MD
3089 if (errp != NULL)
3090 *errp = ENOTBLK;
3091 return (0);
3092 }
3093 if (errp != NULL)
3094 *errp = 0;
3095 return (1);
3096}
3097
3098void
3099NDFREE(ndp, flags)
3100 struct nameidata *ndp;
3101 const uint flags;
3102{
3103 if (!(flags & NDF_NO_FREE_PNBUF) &&
2b69e610 3104 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
984263bc 3105 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
2b69e610 3106 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
984263bc 3107 }
bc0c094e
MD
3108 if (!(flags & NDF_NO_DNCP_RELE) &&
3109 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3110 ndp->ni_dncp) {
3111 cache_drop(ndp->ni_dncp);
3112 ndp->ni_dncp = NULL;
3113 }
3114 if (!(flags & NDF_NO_NCP_RELE) &&
3115 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3116 ndp->ni_ncp) {
3117 cache_drop(ndp->ni_ncp);
3118 ndp->ni_ncp = NULL;
3119 }
984263bc 3120 if (!(flags & NDF_NO_DVP_UNLOCK) &&
2b69e610 3121 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
bc0c094e 3122 ndp->ni_dvp != ndp->ni_vp) {
dadab5e9 3123 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
bc0c094e 3124 }
984263bc 3125 if (!(flags & NDF_NO_DVP_RELE) &&
2b69e610 3126 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
984263bc
MD
3127 vrele(ndp->ni_dvp);
3128 ndp->ni_dvp = NULL;
3129 }
3130 if (!(flags & NDF_NO_VP_UNLOCK) &&
bc0c094e 3131 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
dadab5e9 3132 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
bc0c094e 3133 }
984263bc
MD
3134 if (!(flags & NDF_NO_VP_RELE) &&
3135 ndp->ni_vp) {
3136 vrele(ndp->ni_vp);
3137 ndp->ni_vp = NULL;
3138 }
3139 if (!(flags & NDF_NO_STARTDIR_RELE) &&
2b69e610 3140 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
984263bc
MD
3141 vrele(ndp->ni_startdir);
3142 ndp->ni_startdir = NULL;
3143 }
3144}
bc0c094e 3145