oops, forgot one. Remove another curproc/cred dependancy
[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 $
12e4aaff 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.9 2003/07/03 17:24:02 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
MD
83#include <sys/buf2.h>
84
984263bc
MD
85static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
86
87static void insmntque __P((struct vnode *vp, struct mount *mp));
dadab5e9 88static void vclean __P((struct vnode *vp, int flags, struct thread *td));
984263bc
MD
89static unsigned long numvnodes;
90static void vlruvp(struct vnode *vp);
91SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
92
93enum vtype iftovt_tab[16] = {
94 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
95 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
96};
97int vttoif_tab[9] = {
98 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
99 S_IFSOCK, S_IFIFO, S_IFMT,
100};
101
102static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
103
104static u_long wantfreevnodes = 25;
105SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
106static u_long freevnodes = 0;
107SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
108
109static int reassignbufcalls;
110SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
111static int reassignbufloops;
112SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
113static int reassignbufsortgood;
114SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
115static int reassignbufsortbad;
116SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
117static int reassignbufmethod = 1;
118SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
119static int nameileafonly = 0;
120SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
121
122#ifdef ENABLE_VFS_IOOPT
123int vfs_ioopt = 0;
124SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
125#endif
126
127struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
128struct simplelock mountlist_slock;
129struct simplelock mntvnode_slock;
130int nfs_mount_type = -1;
131#ifndef NULL_SIMPLELOCKS
132static struct simplelock mntid_slock;
133static struct simplelock vnode_free_list_slock;
134static struct simplelock spechash_slock;
135#endif
136struct nfs_public nfs_pub; /* publicly exported FS */
137static vm_zone_t vnode_zone;
138
139/*
140 * The workitem queue.
141 */
142#define SYNCER_MAXDELAY 32
143static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
144time_t syncdelay = 30; /* max time to delay syncing data */
145time_t filedelay = 30; /* time to delay syncing files */
146SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
147time_t dirdelay = 29; /* time to delay syncing directories */
148SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
149time_t metadelay = 28; /* time to delay syncing metadata */
150SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
151static int rushjob; /* number of slots to run ASAP */
152static int stat_rush_requests; /* number of times I/O speeded up */
153SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
154
155static int syncer_delayno = 0;
156static long syncer_mask;
157LIST_HEAD(synclist, vnode);
158static struct synclist *syncer_workitem_pending;
159
160int desiredvnodes;
161SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
162 &desiredvnodes, 0, "Maximum number of vnodes");
163static int minvnodes;
164SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
165 &minvnodes, 0, "Minimum number of vnodes");
166static int vnlru_nowhere = 0;
167SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
168 "Number of times the vnlru process ran without success");
169
170static void vfs_free_addrlist __P((struct netexport *nep));
171static int vfs_free_netcred __P((struct radix_node *rn, void *w));
172static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
173 struct export_args *argp));
174
175/*
176 * Initialize the vnode management data structures.
177 */
178void
179vntblinit()
180{
181
12e4aaff 182 desiredvnodes = maxproc + vmstats.v_page_count / 4;
984263bc
MD
183 minvnodes = desiredvnodes / 4;
184 simple_lock_init(&mntvnode_slock);
185 simple_lock_init(&mntid_slock);
186 simple_lock_init(&spechash_slock);
187 TAILQ_INIT(&vnode_free_list);
188 simple_lock_init(&vnode_free_list_slock);
189 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
190 /*
191 * Initialize the filesystem syncer.
192 */
193 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
194 &syncer_mask);
195 syncer_maxdelay = syncer_mask + 1;
196}
197
198/*
199 * Mark a mount point as busy. Used to synchronize access and to delay
200 * unmounting. Interlock is not released on failure.
201 */
202int
dadab5e9
MD
203vfs_busy(struct mount *mp, int flags, struct simplelock *interlkp,
204 struct thread *td)
984263bc
MD
205{
206 int lkflags;
207
208 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
209 if (flags & LK_NOWAIT)
210 return (ENOENT);
211 mp->mnt_kern_flag |= MNTK_MWAIT;
212 if (interlkp) {
213 simple_unlock(interlkp);
214 }
215 /*
216 * Since all busy locks are shared except the exclusive
217 * lock granted when unmounting, the only place that a
218 * wakeup needs to be done is at the release of the
219 * exclusive lock at the end of dounmount.
220 */
221 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
222 if (interlkp) {
223 simple_lock(interlkp);
224 }
225 return (ENOENT);
226 }
227 lkflags = LK_SHARED | LK_NOPAUSE;
228 if (interlkp)
229 lkflags |= LK_INTERLOCK;
dadab5e9 230 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
984263bc
MD
231 panic("vfs_busy: unexpected lock failure");
232 return (0);
233}
234
235/*
236 * Free a busy filesystem.
237 */
238void
dadab5e9 239vfs_unbusy(struct mount *mp, struct thread *td)
984263bc 240{
dadab5e9 241 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
984263bc
MD
242}
243
244/*
245 * Lookup a filesystem type, and if found allocate and initialize
246 * a mount structure for it.
247 *
248 * Devname is usually updated by mount(8) after booting.
249 */
250int
dadab5e9 251vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
984263bc 252{
dadab5e9 253 struct thread *td = curthread; /* XXX */
984263bc
MD
254 struct vfsconf *vfsp;
255 struct mount *mp;
256
257 if (fstypename == NULL)
258 return (ENODEV);
259 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
260 if (!strcmp(vfsp->vfc_name, fstypename))
261 break;
262 if (vfsp == NULL)
263 return (ENODEV);
264 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
265 bzero((char *)mp, (u_long)sizeof(struct mount));
266 lockinit(&mp->mnt_lock, PVFS, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
dadab5e9 267 (void)vfs_busy(mp, LK_NOWAIT, 0, td);
984263bc
MD
268 TAILQ_INIT(&mp->mnt_nvnodelist);
269 TAILQ_INIT(&mp->mnt_reservedvnlist);
270 mp->mnt_nvnodelistsize = 0;
271 mp->mnt_vfc = vfsp;
272 mp->mnt_op = vfsp->vfc_vfsops;
273 mp->mnt_flag = MNT_RDONLY;
274 mp->mnt_vnodecovered = NULLVP;
275 vfsp->vfc_refcount++;
276 mp->mnt_iosize_max = DFLTPHYS;
277 mp->mnt_stat.f_type = vfsp->vfc_typenum;
278 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
279 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
280 mp->mnt_stat.f_mntonname[0] = '/';
281 mp->mnt_stat.f_mntonname[1] = 0;
282 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
283 *mpp = mp;
284 return (0);
285}
286
287/*
288 * Find an appropriate filesystem to use for the root. If a filesystem
289 * has not been preselected, walk through the list of known filesystems
290 * trying those that have mountroot routines, and try them until one
291 * works or we have tried them all.
292 */
293#ifdef notdef /* XXX JH */
294int
295lite2_vfs_mountroot()
296{
297 struct vfsconf *vfsp;
298 extern int (*lite2_mountroot) __P((void));
299 int error;
300
301 if (lite2_mountroot != NULL)
302 return ((*lite2_mountroot)());
303 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
304 if (vfsp->vfc_mountroot == NULL)
305 continue;
306 if ((error = (*vfsp->vfc_mountroot)()) == 0)
307 return (0);
308 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
309 }
310 return (ENODEV);
311}
312#endif
313
314/*
315 * Lookup a mount point by filesystem identifier.
316 */
317struct mount *
318vfs_getvfs(fsid)
319 fsid_t *fsid;
320{
321 register struct mount *mp;
322
323 simple_lock(&mountlist_slock);
324 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
325 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
326 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
327 simple_unlock(&mountlist_slock);
328 return (mp);
329 }
330 }
331 simple_unlock(&mountlist_slock);
332 return ((struct mount *) 0);
333}
334
335/*
336 * Get a new unique fsid. Try to make its val[0] unique, since this value
337 * will be used to create fake device numbers for stat(). Also try (but
338 * not so hard) make its val[0] unique mod 2^16, since some emulators only
339 * support 16-bit device numbers. We end up with unique val[0]'s for the
340 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
341 *
342 * Keep in mind that several mounts may be running in parallel. Starting
343 * the search one past where the previous search terminated is both a
344 * micro-optimization and a defense against returning the same fsid to
345 * different mounts.
346 */
347void
348vfs_getnewfsid(mp)
349 struct mount *mp;
350{
351 static u_int16_t mntid_base;
352 fsid_t tfsid;
353 int mtype;
354
355 simple_lock(&mntid_slock);
356 mtype = mp->mnt_vfc->vfc_typenum;
357 tfsid.val[1] = mtype;
358 mtype = (mtype & 0xFF) << 24;
359 for (;;) {
360 tfsid.val[0] = makeudev(255,
361 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
362 mntid_base++;
363 if (vfs_getvfs(&tfsid) == NULL)
364 break;
365 }
366 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
367 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
368 simple_unlock(&mntid_slock);
369}
370
371/*
372 * Knob to control the precision of file timestamps:
373 *
374 * 0 = seconds only; nanoseconds zeroed.
375 * 1 = seconds and nanoseconds, accurate within 1/HZ.
376 * 2 = seconds and nanoseconds, truncated to microseconds.
377 * >=3 = seconds and nanoseconds, maximum precision.
378 */
379enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
380
381static int timestamp_precision = TSP_SEC;
382SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
383 &timestamp_precision, 0, "");
384
385/*
386 * Get a current timestamp.
387 */
388void
389vfs_timestamp(tsp)
390 struct timespec *tsp;
391{
392 struct timeval tv;
393
394 switch (timestamp_precision) {
395 case TSP_SEC:
396 tsp->tv_sec = time_second;
397 tsp->tv_nsec = 0;
398 break;
399 case TSP_HZ:
400 getnanotime(tsp);
401 break;
402 case TSP_USEC:
403 microtime(&tv);
404 TIMEVAL_TO_TIMESPEC(&tv, tsp);
405 break;
406 case TSP_NSEC:
407 default:
408 nanotime(tsp);
409 break;
410 }
411}
412
413/*
414 * Set vnode attributes to VNOVAL
415 */
416void
417vattr_null(vap)
418 register struct vattr *vap;
419{
420
421 vap->va_type = VNON;
422 vap->va_size = VNOVAL;
423 vap->va_bytes = VNOVAL;
424 vap->va_mode = VNOVAL;
425 vap->va_nlink = VNOVAL;
426 vap->va_uid = VNOVAL;
427 vap->va_gid = VNOVAL;
428 vap->va_fsid = VNOVAL;
429 vap->va_fileid = VNOVAL;
430 vap->va_blocksize = VNOVAL;
431 vap->va_rdev = VNOVAL;
432 vap->va_atime.tv_sec = VNOVAL;
433 vap->va_atime.tv_nsec = VNOVAL;
434 vap->va_mtime.tv_sec = VNOVAL;
435 vap->va_mtime.tv_nsec = VNOVAL;
436 vap->va_ctime.tv_sec = VNOVAL;
437 vap->va_ctime.tv_nsec = VNOVAL;
438 vap->va_flags = VNOVAL;
439 vap->va_gen = VNOVAL;
440 vap->va_vaflags = 0;
441}
442
443/*
444 * This routine is called when we have too many vnodes. It attempts
445 * to free <count> vnodes and will potentially free vnodes that still
446 * have VM backing store (VM backing store is typically the cause
447 * of a vnode blowout so we want to do this). Therefore, this operation
448 * is not considered cheap.
449 *
450 * A number of conditions may prevent a vnode from being reclaimed.
451 * the buffer cache may have references on the vnode, a directory
452 * vnode may still have references due to the namei cache representing
453 * underlying files, or the vnode may be in active use. It is not
454 * desireable to reuse such vnodes. These conditions may cause the
455 * number of vnodes to reach some minimum value regardless of what
456 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
457 */
458static int
459vlrureclaim(struct mount *mp)
460{
461 struct vnode *vp;
462 int done;
463 int trigger;
464 int usevnodes;
465 int count;
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;
480 simple_lock(&mntvnode_slock);
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 */
489 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) &&
490 simple_lock_try(&vp->v_interlock)
491 ) {
492 simple_unlock(&mntvnode_slock);
493 if (VMIGHTFREE(vp)) {
dadab5e9 494 vgonel(vp, curthread);
984263bc
MD
495 done++;
496 } else {
497 simple_unlock(&vp->v_interlock);
498 }
499 simple_lock(&mntvnode_slock);
500 }
501 --count;
502 }
503 simple_unlock(&mntvnode_slock);
504 return done;
505}
506
507/*
508 * Attempt to recycle vnodes in a context that is always safe to block.
509 * Calling vlrurecycle() from the bowels of file system code has some
510 * interesting deadlock problems.
511 */
bc6dffab 512static struct thread *vnlruthread;
984263bc
MD
513static int vnlruproc_sig;
514
515static void
516vnlru_proc(void)
517{
518 struct mount *mp, *nmp;
519 int s;
520 int done;
0cfcada1 521 struct thread *td = curthread;
984263bc 522
bc6dffab 523 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
524 SHUTDOWN_PRI_FIRST);
525
526 s = splbio();
527 for (;;) {
0cfcada1 528 kproc_suspend_loop();
984263bc
MD
529 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
530 vnlruproc_sig = 0;
531 wakeup(&vnlruproc_sig);
0cfcada1 532 tsleep(td, PVFS, "vlruwt", hz);
984263bc
MD
533 continue;
534 }
535 done = 0;
536 simple_lock(&mountlist_slock);
537 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
dadab5e9 538 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, td)) {
984263bc
MD
539 nmp = TAILQ_NEXT(mp, mnt_list);
540 continue;
541 }
542 done += vlrureclaim(mp);
543 simple_lock(&mountlist_slock);
544 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 545 vfs_unbusy(mp, td);
984263bc
MD
546 }
547 simple_unlock(&mountlist_slock);
548 if (done == 0) {
549 vnlru_nowhere++;
0cfcada1 550 tsleep(td, PPAUSE, "vlrup", hz * 3);
984263bc
MD
551 }
552 }
553 splx(s);
554}
555
556static struct kproc_desc vnlru_kp = {
557 "vnlru",
558 vnlru_proc,
bc6dffab 559 &vnlruthread
984263bc
MD
560};
561SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
562
563/*
564 * Routines having to do with the management of the vnode table.
565 */
566extern vop_t **dead_vnodeop_p;
567
568/*
569 * Return the next vnode from the free list.
570 */
571int
572getnewvnode(tag, mp, vops, vpp)
573 enum vtagtype tag;
574 struct mount *mp;
575 vop_t **vops;
576 struct vnode **vpp;
577{
578 int s;
dadab5e9 579 struct thread *td = curthread; /* XXX */
984263bc
MD
580 struct vnode *vp = NULL;
581 vm_object_t object;
582
583 s = splbio();
584
585 /*
586 * Try to reuse vnodes if we hit the max. This situation only
587 * occurs in certain large-memory (2G+) situations. We cannot
588 * attempt to directly reclaim vnodes due to nasty recursion
589 * problems.
590 */
591 while (numvnodes - freevnodes > desiredvnodes) {
592 if (vnlruproc_sig == 0) {
593 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
bc6dffab 594 wakeup(vnlruthread);
984263bc
MD
595 }
596 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
597 }
598
599
600 /*
601 * Attempt to reuse a vnode already on the free list, allocating
602 * a new vnode if we can't find one or if we have not reached a
603 * good minimum for good LRU performance.
604 */
605 simple_lock(&vnode_free_list_slock);
606 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
607 int count;
608
609 for (count = 0; count < freevnodes; count++) {
610 vp = TAILQ_FIRST(&vnode_free_list);
611 if (vp == NULL || vp->v_usecount)
612 panic("getnewvnode: free vnode isn't");
613
614 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
615 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
616 (object->resident_page_count || object->ref_count)) ||
617 !simple_lock_try(&vp->v_interlock)) {
618 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
619 vp = NULL;
620 continue;
621 }
622 if (LIST_FIRST(&vp->v_cache_src)) {
623 /*
624 * note: nameileafonly sysctl is temporary,
625 * for debugging only, and will eventually be
626 * removed.
627 */
628 if (nameileafonly > 0) {
629 /*
630 * Do not reuse namei-cached directory
631 * vnodes that have cached
632 * subdirectories.
633 */
634 if (cache_leaf_test(vp) < 0) {
635 simple_unlock(&vp->v_interlock);
636 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
637 vp = NULL;
638 continue;
639 }
640 } else if (nameileafonly < 0 ||
641 vmiodirenable == 0) {
642 /*
643 * Do not reuse namei-cached directory
644 * vnodes if nameileafonly is -1 or
645 * if VMIO backing for directories is
646 * turned off (otherwise we reuse them
647 * too quickly).
648 */
649 simple_unlock(&vp->v_interlock);
650 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
651 vp = NULL;
652 continue;
653 }
654 }
655 break;
656 }
657 }
658
659 if (vp) {
660 vp->v_flag |= VDOOMED;
661 vp->v_flag &= ~VFREE;
662 freevnodes--;
663 simple_unlock(&vnode_free_list_slock);
664 cache_purge(vp);
665 vp->v_lease = NULL;
666 if (vp->v_type != VBAD) {
dadab5e9 667 vgonel(vp, td);
984263bc
MD
668 } else {
669 simple_unlock(&vp->v_interlock);
670 }
671
672#ifdef INVARIANTS
673 {
674 int s;
675
676 if (vp->v_data)
677 panic("cleaned vnode isn't");
678 s = splbio();
679 if (vp->v_numoutput)
680 panic("Clean vnode has pending I/O's");
681 splx(s);
682 }
683#endif
684 vp->v_flag = 0;
685 vp->v_lastw = 0;
686 vp->v_lasta = 0;
687 vp->v_cstart = 0;
688 vp->v_clen = 0;
689 vp->v_socket = 0;
690 vp->v_writecount = 0; /* XXX */
691 } else {
692 simple_unlock(&vnode_free_list_slock);
693 vp = (struct vnode *) zalloc(vnode_zone);
694 bzero((char *) vp, sizeof *vp);
695 simple_lock_init(&vp->v_interlock);
696 vp->v_dd = vp;
697 cache_purge(vp);
698 LIST_INIT(&vp->v_cache_src);
699 TAILQ_INIT(&vp->v_cache_dst);
700 numvnodes++;
701 }
702
703 TAILQ_INIT(&vp->v_cleanblkhd);
704 TAILQ_INIT(&vp->v_dirtyblkhd);
705 vp->v_type = VNON;
706 vp->v_tag = tag;
707 vp->v_op = vops;
708 insmntque(vp, mp);
709 *vpp = vp;
710 vp->v_usecount = 1;
711 vp->v_data = 0;
712 splx(s);
713
3b568787 714 vfs_object_create(vp, td);
984263bc
MD
715 return (0);
716}
717
718/*
719 * Move a vnode from one mount queue to another.
720 */
721static void
722insmntque(vp, mp)
723 register struct vnode *vp;
724 register struct mount *mp;
725{
726
727 simple_lock(&mntvnode_slock);
728 /*
729 * Delete from old mount point vnode list, if on one.
730 */
731 if (vp->v_mount != NULL) {
732 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
733 ("bad mount point vnode list size"));
734 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
735 vp->v_mount->mnt_nvnodelistsize--;
736 }
737 /*
738 * Insert into list of vnodes for the new mount point, if available.
739 */
740 if ((vp->v_mount = mp) == NULL) {
741 simple_unlock(&mntvnode_slock);
742 return;
743 }
744 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
745 mp->mnt_nvnodelistsize++;
746 simple_unlock(&mntvnode_slock);
747}
748
749/*
750 * Update outstanding I/O count and do wakeup if requested.
751 */
752void
753vwakeup(bp)
754 register struct buf *bp;
755{
756 register struct vnode *vp;
757
758 bp->b_flags &= ~B_WRITEINPROG;
759 if ((vp = bp->b_vp)) {
760 vp->v_numoutput--;
761 if (vp->v_numoutput < 0)
762 panic("vwakeup: neg numoutput");
763 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
764 vp->v_flag &= ~VBWAIT;
765 wakeup((caddr_t) &vp->v_numoutput);
766 }
767 }
768}
769
770/*
771 * Flush out and invalidate all buffers associated with a vnode.
772 * Called with the underlying object locked.
773 */
774int
3b568787
MD
775vinvalbuf(struct vnode *vp, int flags, struct thread *td,
776 int slpflag, int slptimeo)
984263bc
MD
777{
778 register struct buf *bp;
779 struct buf *nbp, *blist;
780 int s, error;
781 vm_object_t object;
782
783 if (flags & V_SAVE) {
784 s = splbio();
785 while (vp->v_numoutput) {
786 vp->v_flag |= VBWAIT;
787 error = tsleep((caddr_t)&vp->v_numoutput,
788 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
789 if (error) {
790 splx(s);
791 return (error);
792 }
793 }
794 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
795 splx(s);
3b568787 796 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
984263bc
MD
797 return (error);
798 s = splbio();
799 if (vp->v_numoutput > 0 ||
800 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
801 panic("vinvalbuf: dirty bufs");
802 }
803 splx(s);
804 }
805 s = splbio();
806 for (;;) {
807 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
808 if (!blist)
809 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
810 if (!blist)
811 break;
812
813 for (bp = blist; bp; bp = nbp) {
814 nbp = TAILQ_NEXT(bp, b_vnbufs);
815 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
816 error = BUF_TIMELOCK(bp,
817 LK_EXCLUSIVE | LK_SLEEPFAIL,
818 "vinvalbuf", slpflag, slptimeo);
819 if (error == ENOLCK)
820 break;
821 splx(s);
822 return (error);
823 }
824 /*
825 * XXX Since there are no node locks for NFS, I
826 * believe there is a slight chance that a delayed
827 * write will occur while sleeping just above, so
828 * check for it. Note that vfs_bio_awrite expects
829 * buffers to reside on a queue, while VOP_BWRITE and
830 * brelse do not.
831 */
832 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
833 (flags & V_SAVE)) {
834
835 if (bp->b_vp == vp) {
836 if (bp->b_flags & B_CLUSTEROK) {
837 BUF_UNLOCK(bp);
838 vfs_bio_awrite(bp);
839 } else {
840 bremfree(bp);
841 bp->b_flags |= B_ASYNC;
842 VOP_BWRITE(bp->b_vp, bp);
843 }
844 } else {
845 bremfree(bp);
846 (void) VOP_BWRITE(bp->b_vp, bp);
847 }
848 break;
849 }
850 bremfree(bp);
851 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
852 bp->b_flags &= ~B_ASYNC;
853 brelse(bp);
854 }
855 }
856
857 /*
858 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
859 * have write I/O in-progress but if there is a VM object then the
860 * VM object can also have read-I/O in-progress.
861 */
862 do {
863 while (vp->v_numoutput > 0) {
864 vp->v_flag |= VBWAIT;
865 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
866 }
867 if (VOP_GETVOBJECT(vp, &object) == 0) {
868 while (object->paging_in_progress)
869 vm_object_pip_sleep(object, "vnvlbx");
870 }
871 } while (vp->v_numoutput > 0);
872
873 splx(s);
874
875 /*
876 * Destroy the copy in the VM cache, too.
877 */
878 simple_lock(&vp->v_interlock);
879 if (VOP_GETVOBJECT(vp, &object) == 0) {
880 vm_object_page_remove(object, 0, 0,
881 (flags & V_SAVE) ? TRUE : FALSE);
882 }
883 simple_unlock(&vp->v_interlock);
884
885 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
886 panic("vinvalbuf: flush failed");
887 return (0);
888}
889
890/*
891 * Truncate a file's buffer and pages to a specified length. This
892 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
893 * sync activity.
894 */
895int
3b568787 896vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
984263bc 897{
dadab5e9 898 struct buf *bp;
984263bc
MD
899 struct buf *nbp;
900 int s, anyfreed;
901 int trunclbn;
902
903 /*
904 * Round up to the *next* lbn.
905 */
906 trunclbn = (length + blksize - 1) / blksize;
907
908 s = splbio();
909restart:
910 anyfreed = 1;
911 for (;anyfreed;) {
912 anyfreed = 0;
913 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
914 nbp = TAILQ_NEXT(bp, b_vnbufs);
915 if (bp->b_lblkno >= trunclbn) {
916 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
917 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
918 goto restart;
919 } else {
920 bremfree(bp);
921 bp->b_flags |= (B_INVAL | B_RELBUF);
922 bp->b_flags &= ~B_ASYNC;
923 brelse(bp);
924 anyfreed = 1;
925 }
926 if (nbp &&
927 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
928 (nbp->b_vp != vp) ||
929 (nbp->b_flags & B_DELWRI))) {
930 goto restart;
931 }
932 }
933 }
934
935 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
936 nbp = TAILQ_NEXT(bp, b_vnbufs);
937 if (bp->b_lblkno >= trunclbn) {
938 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
939 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
940 goto restart;
941 } else {
942 bremfree(bp);
943 bp->b_flags |= (B_INVAL | B_RELBUF);
944 bp->b_flags &= ~B_ASYNC;
945 brelse(bp);
946 anyfreed = 1;
947 }
948 if (nbp &&
949 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
950 (nbp->b_vp != vp) ||
951 (nbp->b_flags & B_DELWRI) == 0)) {
952 goto restart;
953 }
954 }
955 }
956 }
957
958 if (length > 0) {
959restartsync:
960 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
961 nbp = TAILQ_NEXT(bp, b_vnbufs);
962 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
963 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
964 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
965 goto restart;
966 } else {
967 bremfree(bp);
968 if (bp->b_vp == vp) {
969 bp->b_flags |= B_ASYNC;
970 } else {
971 bp->b_flags &= ~B_ASYNC;
972 }
973 VOP_BWRITE(bp->b_vp, bp);
974 }
975 goto restartsync;
976 }
977
978 }
979 }
980
981 while (vp->v_numoutput > 0) {
982 vp->v_flag |= VBWAIT;
983 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
984 }
985
986 splx(s);
987
988 vnode_pager_setsize(vp, length);
989
990 return (0);
991}
992
993/*
994 * Associate a buffer with a vnode.
995 */
996void
997bgetvp(vp, bp)
998 register struct vnode *vp;
999 register struct buf *bp;
1000{
1001 int s;
1002
1003 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1004
1005 vhold(vp);
1006 bp->b_vp = vp;
1007 bp->b_dev = vn_todev(vp);
1008 /*
1009 * Insert onto list for new vnode.
1010 */
1011 s = splbio();
1012 bp->b_xflags |= BX_VNCLEAN;
1013 bp->b_xflags &= ~BX_VNDIRTY;
1014 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1015 splx(s);
1016}
1017
1018/*
1019 * Disassociate a buffer from a vnode.
1020 */
1021void
1022brelvp(bp)
1023 register struct buf *bp;
1024{
1025 struct vnode *vp;
1026 struct buflists *listheadp;
1027 int s;
1028
1029 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1030
1031 /*
1032 * Delete from old vnode list, if on one.
1033 */
1034 vp = bp->b_vp;
1035 s = splbio();
1036 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1037 if (bp->b_xflags & BX_VNDIRTY)
1038 listheadp = &vp->v_dirtyblkhd;
1039 else
1040 listheadp = &vp->v_cleanblkhd;
1041 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1042 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1043 }
1044 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1045 vp->v_flag &= ~VONWORKLST;
1046 LIST_REMOVE(vp, v_synclist);
1047 }
1048 splx(s);
1049 bp->b_vp = (struct vnode *) 0;
1050 vdrop(vp);
1051}
1052
1053/*
1054 * The workitem queue.
1055 *
1056 * It is useful to delay writes of file data and filesystem metadata
1057 * for tens of seconds so that quickly created and deleted files need
1058 * not waste disk bandwidth being created and removed. To realize this,
1059 * we append vnodes to a "workitem" queue. When running with a soft
1060 * updates implementation, most pending metadata dependencies should
1061 * not wait for more than a few seconds. Thus, mounted on block devices
1062 * are delayed only about a half the time that file data is delayed.
1063 * Similarly, directory updates are more critical, so are only delayed
1064 * about a third the time that file data is delayed. Thus, there are
1065 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1066 * one each second (driven off the filesystem syncer process). The
1067 * syncer_delayno variable indicates the next queue that is to be processed.
1068 * Items that need to be processed soon are placed in this queue:
1069 *
1070 * syncer_workitem_pending[syncer_delayno]
1071 *
1072 * A delay of fifteen seconds is done by placing the request fifteen
1073 * entries later in the queue:
1074 *
1075 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1076 *
1077 */
1078
1079/*
1080 * Add an item to the syncer work queue.
1081 */
1082static void
1083vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1084{
1085 int s, slot;
1086
1087 s = splbio();
1088
1089 if (vp->v_flag & VONWORKLST) {
1090 LIST_REMOVE(vp, v_synclist);
1091 }
1092
1093 if (delay > syncer_maxdelay - 2)
1094 delay = syncer_maxdelay - 2;
1095 slot = (syncer_delayno + delay) & syncer_mask;
1096
1097 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1098 vp->v_flag |= VONWORKLST;
1099 splx(s);
1100}
1101
bc6dffab 1102struct thread *updatethread;
984263bc
MD
1103static void sched_sync __P((void));
1104static struct kproc_desc up_kp = {
1105 "syncer",
1106 sched_sync,
bc6dffab 1107 &updatethread
984263bc
MD
1108};
1109SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1110
1111/*
1112 * System filesystem synchronizer daemon.
1113 */
1114void
1115sched_sync(void)
1116{
1117 struct synclist *slp;
1118 struct vnode *vp;
1119 long starttime;
1120 int s;
0cfcada1 1121 struct thread *td = curthread;
984263bc 1122
bc6dffab 1123 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
1124 SHUTDOWN_PRI_LAST);
1125
1126 for (;;) {
0cfcada1 1127 kproc_suspend_loop();
984263bc
MD
1128
1129 starttime = time_second;
1130
1131 /*
1132 * Push files whose dirty time has expired. Be careful
1133 * of interrupt race on slp queue.
1134 */
1135 s = splbio();
1136 slp = &syncer_workitem_pending[syncer_delayno];
1137 syncer_delayno += 1;
1138 if (syncer_delayno == syncer_maxdelay)
1139 syncer_delayno = 0;
1140 splx(s);
1141
1142 while ((vp = LIST_FIRST(slp)) != NULL) {
1143 if (VOP_ISLOCKED(vp, NULL) == 0) {
dadab5e9 1144 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3b568787 1145 (void) VOP_FSYNC(vp, MNT_LAZY, td);
dadab5e9 1146 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1147 }
1148 s = splbio();
1149 if (LIST_FIRST(slp) == vp) {
1150 /*
1151 * Note: v_tag VT_VFS vps can remain on the
1152 * worklist too with no dirty blocks, but
1153 * since sync_fsync() moves it to a different
1154 * slot we are safe.
1155 */
1156 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1157 !vn_isdisk(vp, NULL))
1158 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1159 /*
1160 * Put us back on the worklist. The worklist
1161 * routine will remove us from our current
1162 * position and then add us back in at a later
1163 * position.
1164 */
1165 vn_syncer_add_to_worklist(vp, syncdelay);
1166 }
1167 splx(s);
1168 }
1169
1170 /*
1171 * Do soft update processing.
1172 */
1173 if (bioops.io_sync)
1174 (*bioops.io_sync)(NULL);
1175
1176 /*
1177 * The variable rushjob allows the kernel to speed up the
1178 * processing of the filesystem syncer process. A rushjob
1179 * value of N tells the filesystem syncer to process the next
1180 * N seconds worth of work on its queue ASAP. Currently rushjob
1181 * is used by the soft update code to speed up the filesystem
1182 * syncer process when the incore state is getting so far
1183 * ahead of the disk that the kernel memory pool is being
1184 * threatened with exhaustion.
1185 */
1186 if (rushjob > 0) {
1187 rushjob -= 1;
1188 continue;
1189 }
1190 /*
1191 * If it has taken us less than a second to process the
1192 * current work, then wait. Otherwise start right over
1193 * again. We can still lose time if any single round
1194 * takes more than two seconds, but it does not really
1195 * matter as we are just trying to generally pace the
1196 * filesystem activity.
1197 */
1198 if (time_second == starttime)
1199 tsleep(&lbolt, PPAUSE, "syncer", 0);
1200 }
1201}
1202
1203/*
1204 * Request the syncer daemon to speed up its work.
1205 * We never push it to speed up more than half of its
1206 * normal turn time, otherwise it could take over the cpu.
1207 */
1208int
1209speedup_syncer()
1210{
1211 int s;
1212
1213 s = splhigh();
bc6dffab
MD
1214 if (updatethread->td_proc->p_wchan == &lbolt) /* YYY */
1215 setrunnable(updatethread->td_proc);
984263bc
MD
1216 splx(s);
1217 if (rushjob < syncdelay / 2) {
1218 rushjob += 1;
1219 stat_rush_requests += 1;
1220 return (1);
1221 }
1222 return(0);
1223}
1224
1225/*
1226 * Associate a p-buffer with a vnode.
1227 *
1228 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1229 * with the buffer. i.e. the bp has not been linked into the vnode or
1230 * ref-counted.
1231 */
1232void
1233pbgetvp(vp, bp)
1234 register struct vnode *vp;
1235 register struct buf *bp;
1236{
1237
1238 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1239
1240 bp->b_vp = vp;
1241 bp->b_flags |= B_PAGING;
1242 bp->b_dev = vn_todev(vp);
1243}
1244
1245/*
1246 * Disassociate a p-buffer from a vnode.
1247 */
1248void
1249pbrelvp(bp)
1250 register struct buf *bp;
1251{
1252
1253 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1254
1255 /* XXX REMOVE ME */
1256 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1257 panic(
1258 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1259 bp,
1260 (int)bp->b_flags
1261 );
1262 }
1263 bp->b_vp = (struct vnode *) 0;
1264 bp->b_flags &= ~B_PAGING;
1265}
1266
1267void
1268pbreassignbuf(bp, newvp)
1269 struct buf *bp;
1270 struct vnode *newvp;
1271{
1272 if ((bp->b_flags & B_PAGING) == 0) {
1273 panic(
1274 "pbreassignbuf() on non phys bp %p",
1275 bp
1276 );
1277 }
1278 bp->b_vp = newvp;
1279}
1280
1281/*
1282 * Reassign a buffer from one vnode to another.
1283 * Used to assign file specific control information
1284 * (indirect blocks) to the vnode to which they belong.
1285 */
1286void
1287reassignbuf(bp, newvp)
1288 register struct buf *bp;
1289 register struct vnode *newvp;
1290{
1291 struct buflists *listheadp;
1292 int delay;
1293 int s;
1294
1295 if (newvp == NULL) {
1296 printf("reassignbuf: NULL");
1297 return;
1298 }
1299 ++reassignbufcalls;
1300
1301 /*
1302 * B_PAGING flagged buffers cannot be reassigned because their vp
1303 * is not fully linked in.
1304 */
1305 if (bp->b_flags & B_PAGING)
1306 panic("cannot reassign paging buffer");
1307
1308 s = splbio();
1309 /*
1310 * Delete from old vnode list, if on one.
1311 */
1312 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1313 if (bp->b_xflags & BX_VNDIRTY)
1314 listheadp = &bp->b_vp->v_dirtyblkhd;
1315 else
1316 listheadp = &bp->b_vp->v_cleanblkhd;
1317 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1318 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1319 if (bp->b_vp != newvp) {
1320 vdrop(bp->b_vp);
1321 bp->b_vp = NULL; /* for clarification */
1322 }
1323 }
1324 /*
1325 * If dirty, put on list of dirty buffers; otherwise insert onto list
1326 * of clean buffers.
1327 */
1328 if (bp->b_flags & B_DELWRI) {
1329 struct buf *tbp;
1330
1331 listheadp = &newvp->v_dirtyblkhd;
1332 if ((newvp->v_flag & VONWORKLST) == 0) {
1333 switch (newvp->v_type) {
1334 case VDIR:
1335 delay = dirdelay;
1336 break;
1337 case VCHR:
1338 case VBLK:
1339 if (newvp->v_specmountpoint != NULL) {
1340 delay = metadelay;
1341 break;
1342 }
1343 /* fall through */
1344 default:
1345 delay = filedelay;
1346 }
1347 vn_syncer_add_to_worklist(newvp, delay);
1348 }
1349 bp->b_xflags |= BX_VNDIRTY;
1350 tbp = TAILQ_FIRST(listheadp);
1351 if (tbp == NULL ||
1352 bp->b_lblkno == 0 ||
1353 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1354 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1355 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1356 ++reassignbufsortgood;
1357 } else if (bp->b_lblkno < 0) {
1358 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1359 ++reassignbufsortgood;
1360 } else if (reassignbufmethod == 1) {
1361 /*
1362 * New sorting algorithm, only handle sequential case,
1363 * otherwise append to end (but before metadata)
1364 */
1365 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1366 (tbp->b_xflags & BX_VNDIRTY)) {
1367 /*
1368 * Found the best place to insert the buffer
1369 */
1370 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1371 ++reassignbufsortgood;
1372 } else {
1373 /*
1374 * Missed, append to end, but before meta-data.
1375 * We know that the head buffer in the list is
1376 * not meta-data due to prior conditionals.
1377 *
1378 * Indirect effects: NFS second stage write
1379 * tends to wind up here, giving maximum
1380 * distance between the unstable write and the
1381 * commit rpc.
1382 */
1383 tbp = TAILQ_LAST(listheadp, buflists);
1384 while (tbp && tbp->b_lblkno < 0)
1385 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1386 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1387 ++reassignbufsortbad;
1388 }
1389 } else {
1390 /*
1391 * Old sorting algorithm, scan queue and insert
1392 */
1393 struct buf *ttbp;
1394 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1395 (ttbp->b_lblkno < bp->b_lblkno)) {
1396 ++reassignbufloops;
1397 tbp = ttbp;
1398 }
1399 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1400 }
1401 } else {
1402 bp->b_xflags |= BX_VNCLEAN;
1403 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1404 if ((newvp->v_flag & VONWORKLST) &&
1405 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1406 newvp->v_flag &= ~VONWORKLST;
1407 LIST_REMOVE(newvp, v_synclist);
1408 }
1409 }
1410 if (bp->b_vp != newvp) {
1411 bp->b_vp = newvp;
1412 vhold(bp->b_vp);
1413 }
1414 splx(s);
1415}
1416
1417/*
1418 * Create a vnode for a block device.
1419 * Used for mounting the root file system.
1420 */
1421int
1422bdevvp(dev, vpp)
1423 dev_t dev;
1424 struct vnode **vpp;
1425{
1426 register struct vnode *vp;
1427 struct vnode *nvp;
1428 int error;
1429
1430 if (dev == NODEV) {
1431 *vpp = NULLVP;
1432 return (ENXIO);
1433 }
1434 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1435 if (error) {
1436 *vpp = NULLVP;
1437 return (error);
1438 }
1439 vp = nvp;
1440 vp->v_type = VBLK;
1441 addalias(vp, dev);
1442 *vpp = vp;
1443 return (0);
1444}
1445
1446/*
1447 * Add vnode to the alias list hung off the dev_t.
1448 *
1449 * The reason for this gunk is that multiple vnodes can reference
1450 * the same physical device, so checking vp->v_usecount to see
1451 * how many users there are is inadequate; the v_usecount for
1452 * the vnodes need to be accumulated. vcount() does that.
1453 */
1454void
1455addaliasu(nvp, nvp_rdev)
1456 struct vnode *nvp;
1457 udev_t nvp_rdev;
1458{
1459
1460 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1461 panic("addaliasu on non-special vnode");
1462 addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
1463}
1464
1465void
1466addalias(nvp, dev)
1467 struct vnode *nvp;
1468 dev_t dev;
1469{
1470
1471 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1472 panic("addalias on non-special vnode");
1473
1474 nvp->v_rdev = dev;
1475 simple_lock(&spechash_slock);
1476 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1477 simple_unlock(&spechash_slock);
1478}
1479
1480/*
1481 * Grab a particular vnode from the free list, increment its
1482 * reference count and lock it. The vnode lock bit is set if the
1483 * vnode is being eliminated in vgone. The process is awakened
1484 * when the transition is completed, and an error returned to
1485 * indicate that the vnode is no longer usable (possibly having
1486 * been changed to a new file system type).
1487 */
1488int
dadab5e9
MD
1489vget(vp, flags, td)
1490 struct vnode *vp;
984263bc 1491 int flags;
dadab5e9 1492 struct thread *td;
984263bc
MD
1493{
1494 int error;
1495
1496 /*
1497 * If the vnode is in the process of being cleaned out for
1498 * another use, we wait for the cleaning to finish and then
1499 * return failure. Cleaning is determined by checking that
1500 * the VXLOCK flag is set.
1501 */
1502 if ((flags & LK_INTERLOCK) == 0) {
1503 simple_lock(&vp->v_interlock);
1504 }
1505 if (vp->v_flag & VXLOCK) {
1506 if (vp->v_vxproc == curproc) {
1507#if 0
1508 /* this can now occur in normal operation */
1509 log(LOG_INFO, "VXLOCK interlock avoided\n");
1510#endif
1511 } else {
1512 vp->v_flag |= VXWANT;
1513 simple_unlock(&vp->v_interlock);
1514 tsleep((caddr_t)vp, PINOD, "vget", 0);
1515 return (ENOENT);
1516 }
1517 }
1518
1519 vp->v_usecount++;
1520
1521 if (VSHOULDBUSY(vp))
1522 vbusy(vp);
1523 if (flags & LK_TYPE_MASK) {
dadab5e9 1524 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
984263bc
MD
1525 /*
1526 * must expand vrele here because we do not want
1527 * to call VOP_INACTIVE if the reference count
1528 * drops back to zero since it was never really
1529 * active. We must remove it from the free list
1530 * before sleeping so that multiple processes do
1531 * not try to recycle it.
1532 */
1533 simple_lock(&vp->v_interlock);
1534 vp->v_usecount--;
1535 if (VSHOULDFREE(vp))
1536 vfree(vp);
1537 else
1538 vlruvp(vp);
1539 simple_unlock(&vp->v_interlock);
1540 }
1541 return (error);
1542 }
1543 simple_unlock(&vp->v_interlock);
1544 return (0);
1545}
1546
1547void
1548vref(struct vnode *vp)
1549{
1550 simple_lock(&vp->v_interlock);
1551 vp->v_usecount++;
1552 simple_unlock(&vp->v_interlock);
1553}
1554
1555/*
1556 * Vnode put/release.
1557 * If count drops to zero, call inactive routine and return to freelist.
1558 */
1559void
dadab5e9 1560vrele(struct vnode *vp)
984263bc 1561{
dadab5e9 1562 struct thread *td = curthread; /* XXX */
984263bc
MD
1563
1564 KASSERT(vp != NULL, ("vrele: null vp"));
1565
1566 simple_lock(&vp->v_interlock);
1567
1568 if (vp->v_usecount > 1) {
1569
1570 vp->v_usecount--;
1571 simple_unlock(&vp->v_interlock);
1572
1573 return;
1574 }
1575
1576 if (vp->v_usecount == 1) {
1577 vp->v_usecount--;
1578 /*
1579 * We must call VOP_INACTIVE with the node locked.
1580 * If we are doing a vpu, the node is already locked,
1581 * but, in the case of vrele, we must explicitly lock
1582 * the vnode before calling VOP_INACTIVE
1583 */
1584
dadab5e9
MD
1585 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1586 VOP_INACTIVE(vp, td);
984263bc
MD
1587 if (VSHOULDFREE(vp))
1588 vfree(vp);
1589 else
1590 vlruvp(vp);
1591 } else {
1592#ifdef DIAGNOSTIC
1593 vprint("vrele: negative ref count", vp);
1594 simple_unlock(&vp->v_interlock);
1595#endif
1596 panic("vrele: negative ref cnt");
1597 }
1598}
1599
1600void
dadab5e9 1601vput(struct vnode *vp)
984263bc 1602{
dadab5e9 1603 struct thread *td = curthread; /* XXX */
984263bc
MD
1604
1605 KASSERT(vp != NULL, ("vput: null vp"));
1606
1607 simple_lock(&vp->v_interlock);
1608
1609 if (vp->v_usecount > 1) {
1610 vp->v_usecount--;
dadab5e9 1611 VOP_UNLOCK(vp, LK_INTERLOCK, td);
984263bc
MD
1612 return;
1613 }
1614
1615 if (vp->v_usecount == 1) {
1616 vp->v_usecount--;
1617 /*
1618 * We must call VOP_INACTIVE with the node locked.
1619 * If we are doing a vpu, the node is already locked,
1620 * so we just need to release the vnode mutex.
1621 */
1622 simple_unlock(&vp->v_interlock);
dadab5e9 1623 VOP_INACTIVE(vp, td);
984263bc
MD
1624 if (VSHOULDFREE(vp))
1625 vfree(vp);
1626 else
1627 vlruvp(vp);
1628 } else {
1629#ifdef DIAGNOSTIC
1630 vprint("vput: negative ref count", vp);
1631#endif
1632 panic("vput: negative ref cnt");
1633 }
1634}
1635
1636/*
1637 * Somebody doesn't want the vnode recycled.
1638 */
1639void
1640vhold(vp)
1641 register struct vnode *vp;
1642{
1643 int s;
1644
1645 s = splbio();
1646 vp->v_holdcnt++;
1647 if (VSHOULDBUSY(vp))
1648 vbusy(vp);
1649 splx(s);
1650}
1651
1652/*
1653 * One less who cares about this vnode.
1654 */
1655void
1656vdrop(vp)
1657 register struct vnode *vp;
1658{
1659 int s;
1660
1661 s = splbio();
1662 if (vp->v_holdcnt <= 0)
1663 panic("vdrop: holdcnt");
1664 vp->v_holdcnt--;
1665 if (VSHOULDFREE(vp))
1666 vfree(vp);
1667 splx(s);
1668}
1669
1670/*
1671 * Remove any vnodes in the vnode table belonging to mount point mp.
1672 *
1673 * If FORCECLOSE is not specified, there should not be any active ones,
1674 * return error if any are found (nb: this is a user error, not a
1675 * system error). If FORCECLOSE is specified, detach any active vnodes
1676 * that are found.
1677 *
1678 * If WRITECLOSE is set, only flush out regular file vnodes open for
1679 * writing.
1680 *
1681 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1682 *
1683 * `rootrefs' specifies the base reference count for the root vnode
1684 * of this filesystem. The root vnode is considered busy if its
1685 * v_usecount exceeds this value. On a successful return, vflush()
1686 * will call vrele() on the root vnode exactly rootrefs times.
1687 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1688 * be zero.
1689 */
1690#ifdef DIAGNOSTIC
1691static int busyprt = 0; /* print out busy vnodes */
1692SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1693#endif
1694
1695int
1696vflush(mp, rootrefs, flags)
1697 struct mount *mp;
1698 int rootrefs;
1699 int flags;
1700{
dadab5e9 1701 struct thread *td = curthread; /* XXX */
984263bc
MD
1702 struct vnode *vp, *nvp, *rootvp = NULL;
1703 struct vattr vattr;
1704 int busy = 0, error;
1705
1706 if (rootrefs > 0) {
1707 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1708 ("vflush: bad args"));
1709 /*
1710 * Get the filesystem root vnode. We can vput() it
1711 * immediately, since with rootrefs > 0, it won't go away.
1712 */
1713 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1714 return (error);
1715 vput(rootvp);
1716 }
1717 simple_lock(&mntvnode_slock);
1718loop:
1719 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1720 /*
1721 * Make sure this vnode wasn't reclaimed in getnewvnode().
1722 * Start over if it has (it won't be on the list anymore).
1723 */
1724 if (vp->v_mount != mp)
1725 goto loop;
1726 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1727
1728 simple_lock(&vp->v_interlock);
1729 /*
1730 * Skip over a vnodes marked VSYSTEM.
1731 */
1732 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1733 simple_unlock(&vp->v_interlock);
1734 continue;
1735 }
1736 /*
1737 * If WRITECLOSE is set, flush out unlinked but still open
1738 * files (even if open only for reading) and regular file
1739 * vnodes open for writing.
1740 */
1741 if ((flags & WRITECLOSE) &&
1742 (vp->v_type == VNON ||
3b568787 1743 (VOP_GETATTR(vp, &vattr, td) == 0 &&
984263bc
MD
1744 vattr.va_nlink > 0)) &&
1745 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1746 simple_unlock(&vp->v_interlock);
1747 continue;
1748 }
1749
1750 /*
1751 * With v_usecount == 0, all we need to do is clear out the
1752 * vnode data structures and we are done.
1753 */
1754 if (vp->v_usecount == 0) {
1755 simple_unlock(&mntvnode_slock);
dadab5e9 1756 vgonel(vp, td);
984263bc
MD
1757 simple_lock(&mntvnode_slock);
1758 continue;
1759 }
1760
1761 /*
1762 * If FORCECLOSE is set, forcibly close the vnode. For block
1763 * or character devices, revert to an anonymous device. For
1764 * all other files, just kill them.
1765 */
1766 if (flags & FORCECLOSE) {
1767 simple_unlock(&mntvnode_slock);
1768 if (vp->v_type != VBLK && vp->v_type != VCHR) {
dadab5e9 1769 vgonel(vp, td);
984263bc 1770 } else {
dadab5e9 1771 vclean(vp, 0, td);
984263bc
MD
1772 vp->v_op = spec_vnodeop_p;
1773 insmntque(vp, (struct mount *) 0);
1774 }
1775 simple_lock(&mntvnode_slock);
1776 continue;
1777 }
1778#ifdef DIAGNOSTIC
1779 if (busyprt)
1780 vprint("vflush: busy vnode", vp);
1781#endif
1782 simple_unlock(&vp->v_interlock);
1783 busy++;
1784 }
1785 simple_unlock(&mntvnode_slock);
1786 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1787 /*
1788 * If just the root vnode is busy, and if its refcount
1789 * is equal to `rootrefs', then go ahead and kill it.
1790 */
1791 simple_lock(&rootvp->v_interlock);
1792 KASSERT(busy > 0, ("vflush: not busy"));
1793 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1794 if (busy == 1 && rootvp->v_usecount == rootrefs) {
dadab5e9 1795 vgonel(rootvp, td);
984263bc
MD
1796 busy = 0;
1797 } else
1798 simple_unlock(&rootvp->v_interlock);
1799 }
1800 if (busy)
1801 return (EBUSY);
1802 for (; rootrefs > 0; rootrefs--)
1803 vrele(rootvp);
1804 return (0);
1805}
1806
1807/*
1808 * We do not want to recycle the vnode too quickly.
1809 *
1810 * XXX we can't move vp's around the nvnodelist without really screwing
1811 * up the efficiency of filesystem SYNC and friends. This code is
1812 * disabled until we fix the syncing code's scanning algorithm.
1813 */
1814static void
1815vlruvp(struct vnode *vp)
1816{
1817#if 0
1818 struct mount *mp;
1819
1820 if ((mp = vp->v_mount) != NULL) {
1821 simple_lock(&mntvnode_slock);
1822 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1823 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1824 simple_unlock(&mntvnode_slock);
1825 }
1826#endif
1827}
1828
1829/*
1830 * Disassociate the underlying file system from a vnode.
1831 */
1832static void
dadab5e9 1833vclean(struct vnode *vp, int flags, struct thread *td)
984263bc
MD
1834{
1835 int active;
1836
1837 /*
1838 * Check to see if the vnode is in use. If so we have to reference it
1839 * before we clean it out so that its count cannot fall to zero and
1840 * generate a race against ourselves to recycle it.
1841 */
1842 if ((active = vp->v_usecount))
1843 vp->v_usecount++;
1844
1845 /*
1846 * Prevent the vnode from being recycled or brought into use while we
1847 * clean it out.
1848 */
1849 if (vp->v_flag & VXLOCK)
1850 panic("vclean: deadlock");
1851 vp->v_flag |= VXLOCK;
1852 vp->v_vxproc = curproc;
1853 /*
1854 * Even if the count is zero, the VOP_INACTIVE routine may still
1855 * have the object locked while it cleans it out. The VOP_LOCK
1856 * ensures that the VOP_INACTIVE routine is done with its work.
1857 * For active vnodes, it ensures that no other activity can
1858 * occur while the underlying object is being cleaned out.
1859 */
dadab5e9 1860 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
984263bc
MD
1861
1862 /*
1863 * Clean out any buffers associated with the vnode.
1864 */
3b568787 1865 vinvalbuf(vp, V_SAVE, td, 0, 0);
984263bc
MD
1866
1867 VOP_DESTROYVOBJECT(vp);
1868
1869 /*
1870 * If purging an active vnode, it must be closed and
1871 * deactivated before being reclaimed. Note that the
1872 * VOP_INACTIVE will unlock the vnode.
1873 */
1874 if (active) {
1875 if (flags & DOCLOSE)
3b568787 1876 VOP_CLOSE(vp, FNONBLOCK, td);
dadab5e9 1877 VOP_INACTIVE(vp, td);
984263bc
MD
1878 } else {
1879 /*
1880 * Any other processes trying to obtain this lock must first
1881 * wait for VXLOCK to clear, then call the new lock operation.
1882 */
dadab5e9 1883 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1884 }
1885 /*
1886 * Reclaim the vnode.
1887 */
dadab5e9 1888 if (VOP_RECLAIM(vp, td))
984263bc
MD
1889 panic("vclean: cannot reclaim");
1890
1891 if (active) {
1892 /*
1893 * Inline copy of vrele() since VOP_INACTIVE
1894 * has already been called.
1895 */
1896 simple_lock(&vp->v_interlock);
1897 if (--vp->v_usecount <= 0) {
1898#ifdef DIAGNOSTIC
1899 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1900 vprint("vclean: bad ref count", vp);
1901 panic("vclean: ref cnt");
1902 }
1903#endif
1904 vfree(vp);
1905 }
1906 simple_unlock(&vp->v_interlock);
1907 }
1908
1909 cache_purge(vp);
1910 vp->v_vnlock = NULL;
1911
1912 if (VSHOULDFREE(vp))
1913 vfree(vp);
1914
1915 /*
1916 * Done with purge, notify sleepers of the grim news.
1917 */
1918 vp->v_op = dead_vnodeop_p;
1919 vn_pollgone(vp);
1920 vp->v_tag = VT_NON;
1921 vp->v_flag &= ~VXLOCK;
1922 vp->v_vxproc = NULL;
1923 if (vp->v_flag & VXWANT) {
1924 vp->v_flag &= ~VXWANT;
1925 wakeup((caddr_t) vp);
1926 }
1927}
1928
1929/*
1930 * Eliminate all activity associated with the requested vnode
1931 * and with all vnodes aliased to the requested vnode.
1932 */
1933int
1934vop_revoke(ap)
1935 struct vop_revoke_args /* {
1936 struct vnode *a_vp;
1937 int a_flags;
1938 } */ *ap;
1939{
1940 struct vnode *vp, *vq;
1941 dev_t dev;
1942
1943 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1944
1945 vp = ap->a_vp;
1946 /*
1947 * If a vgone (or vclean) is already in progress,
1948 * wait until it is done and return.
1949 */
1950 if (vp->v_flag & VXLOCK) {
1951 vp->v_flag |= VXWANT;
1952 simple_unlock(&vp->v_interlock);
1953 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1954 return (0);
1955 }
1956 dev = vp->v_rdev;
1957 for (;;) {
1958 simple_lock(&spechash_slock);
1959 vq = SLIST_FIRST(&dev->si_hlist);
1960 simple_unlock(&spechash_slock);
1961 if (!vq)
1962 break;
1963 vgone(vq);
1964 }
1965 return (0);
1966}
1967
1968/*
1969 * Recycle an unused vnode to the front of the free list.
1970 * Release the passed interlock if the vnode will be recycled.
1971 */
1972int
dadab5e9 1973vrecycle(struct vnode *vp, struct simplelock *inter_lkp, struct thread *td)
984263bc 1974{
984263bc
MD
1975 simple_lock(&vp->v_interlock);
1976 if (vp->v_usecount == 0) {
1977 if (inter_lkp) {
1978 simple_unlock(inter_lkp);
1979 }
dadab5e9 1980 vgonel(vp, td);
984263bc
MD
1981 return (1);
1982 }
1983 simple_unlock(&vp->v_interlock);
1984 return (0);
1985}
1986
1987/*
1988 * Eliminate all activity associated with a vnode
1989 * in preparation for reuse.
1990 */
1991void
dadab5e9 1992vgone(struct vnode *vp)
984263bc 1993{
dadab5e9 1994 struct thread *td = curthread; /* XXX */
984263bc
MD
1995
1996 simple_lock(&vp->v_interlock);
dadab5e9 1997 vgonel(vp, td);
984263bc
MD
1998}
1999
2000/*
2001 * vgone, with the vp interlock held.
2002 */
2003void
dadab5e9 2004vgonel(struct vnode *vp, struct thread *td)
984263bc
MD
2005{
2006 int s;
2007
2008 /*
2009 * If a vgone (or vclean) is already in progress,
2010 * wait until it is done and return.
2011 */
2012 if (vp->v_flag & VXLOCK) {
2013 vp->v_flag |= VXWANT;
2014 simple_unlock(&vp->v_interlock);
2015 tsleep((caddr_t)vp, PINOD, "vgone", 0);
2016 return;
2017 }
2018
2019 /*
2020 * Clean out the filesystem specific data.
2021 */
dadab5e9 2022 vclean(vp, DOCLOSE, td);
984263bc
MD
2023 simple_lock(&vp->v_interlock);
2024
2025 /*
2026 * Delete from old mount point vnode list, if on one.
2027 */
2028 if (vp->v_mount != NULL)
2029 insmntque(vp, (struct mount *)0);
2030 /*
2031 * If special device, remove it from special device alias list
2032 * if it is on one.
2033 */
2034 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2035 simple_lock(&spechash_slock);
2036 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2037 freedev(vp->v_rdev);
2038 simple_unlock(&spechash_slock);
2039 vp->v_rdev = NULL;
2040 }
2041
2042 /*
2043 * If it is on the freelist and not already at the head,
2044 * move it to the head of the list. The test of the
2045 * VDOOMED flag and the reference count of zero is because
2046 * it will be removed from the free list by getnewvnode,
2047 * but will not have its reference count incremented until
2048 * after calling vgone. If the reference count were
2049 * incremented first, vgone would (incorrectly) try to
2050 * close the previous instance of the underlying object.
2051 */
2052 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2053 s = splbio();
2054 simple_lock(&vnode_free_list_slock);
2055 if (vp->v_flag & VFREE)
2056 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2057 else
2058 freevnodes++;
2059 vp->v_flag |= VFREE;
2060 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2061 simple_unlock(&vnode_free_list_slock);
2062 splx(s);
2063 }
2064
2065 vp->v_type = VBAD;
2066 simple_unlock(&vp->v_interlock);
2067}
2068
2069/*
2070 * Lookup a vnode by device number.
2071 */
2072int
2073vfinddev(dev, type, vpp)
2074 dev_t dev;
2075 enum vtype type;
2076 struct vnode **vpp;
2077{
2078 struct vnode *vp;
2079
2080 simple_lock(&spechash_slock);
2081 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2082 if (type == vp->v_type) {
2083 *vpp = vp;
2084 simple_unlock(&spechash_slock);
2085 return (1);
2086 }
2087 }
2088 simple_unlock(&spechash_slock);
2089 return (0);
2090}
2091
2092/*
2093 * Calculate the total number of references to a special device.
2094 */
2095int
2096vcount(vp)
2097 struct vnode *vp;
2098{
2099 struct vnode *vq;
2100 int count;
2101
2102 count = 0;
2103 simple_lock(&spechash_slock);
2104 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2105 count += vq->v_usecount;
2106 simple_unlock(&spechash_slock);
2107 return (count);
2108}
2109
2110/*
2111 * Same as above, but using the dev_t as argument
2112 */
2113
2114int
2115count_dev(dev)
2116 dev_t dev;
2117{
2118 struct vnode *vp;
2119
2120 vp = SLIST_FIRST(&dev->si_hlist);
2121 if (vp == NULL)
2122 return (0);
2123 return(vcount(vp));
2124}
2125
2126/*
2127 * Print out a description of a vnode.
2128 */
2129static char *typename[] =
2130{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2131
2132void
2133vprint(label, vp)
2134 char *label;
2135 struct vnode *vp;
2136{
2137 char buf[96];
2138
2139 if (label != NULL)
2140 printf("%s: %p: ", label, (void *)vp);
2141 else
2142 printf("%p: ", (void *)vp);
2143 printf("type %s, usecount %d, writecount %d, refcount %d,",
2144 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2145 vp->v_holdcnt);
2146 buf[0] = '\0';
2147 if (vp->v_flag & VROOT)
2148 strcat(buf, "|VROOT");
2149 if (vp->v_flag & VTEXT)
2150 strcat(buf, "|VTEXT");
2151 if (vp->v_flag & VSYSTEM)
2152 strcat(buf, "|VSYSTEM");
2153 if (vp->v_flag & VXLOCK)
2154 strcat(buf, "|VXLOCK");
2155 if (vp->v_flag & VXWANT)
2156 strcat(buf, "|VXWANT");
2157 if (vp->v_flag & VBWAIT)
2158 strcat(buf, "|VBWAIT");
2159 if (vp->v_flag & VDOOMED)
2160 strcat(buf, "|VDOOMED");
2161 if (vp->v_flag & VFREE)
2162 strcat(buf, "|VFREE");
2163 if (vp->v_flag & VOBJBUF)
2164 strcat(buf, "|VOBJBUF");
2165 if (buf[0] != '\0')
2166 printf(" flags (%s)", &buf[1]);
2167 if (vp->v_data == NULL) {
2168 printf("\n");
2169 } else {
2170 printf("\n\t");
2171 VOP_PRINT(vp);
2172 }
2173}
2174
2175#ifdef DDB
2176#include <ddb/ddb.h>
2177/*
2178 * List all of the locked vnodes in the system.
2179 * Called when debugging the kernel.
2180 */
2181DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2182{
dadab5e9 2183 struct thread *td = curthread; /* XXX */
984263bc
MD
2184 struct mount *mp, *nmp;
2185 struct vnode *vp;
2186
2187 printf("Locked vnodes\n");
2188 simple_lock(&mountlist_slock);
2189 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
dadab5e9 2190 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, td)) {
984263bc
MD
2191 nmp = TAILQ_NEXT(mp, mnt_list);
2192 continue;
2193 }
2194 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2195 if (VOP_ISLOCKED(vp, NULL))
2196 vprint((char *)0, vp);
2197 }
2198 simple_lock(&mountlist_slock);
2199 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 2200 vfs_unbusy(mp, td);
984263bc
MD
2201 }
2202 simple_unlock(&mountlist_slock);
2203}
2204#endif
2205
2206/*
2207 * Top level filesystem related information gathering.
2208 */
2209static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2210
2211static int
2212vfs_sysctl(SYSCTL_HANDLER_ARGS)
2213{
2214 int *name = (int *)arg1 - 1; /* XXX */
2215 u_int namelen = arg2 + 1; /* XXX */
2216 struct vfsconf *vfsp;
2217
2218#if 1 || defined(COMPAT_PRELITE2)
2219 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2220 if (namelen == 1)
2221 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2222#endif
2223
2224#ifdef notyet
2225 /* all sysctl names at this level are at least name and field */
2226 if (namelen < 2)
2227 return (ENOTDIR); /* overloaded */
2228 if (name[0] != VFS_GENERIC) {
2229 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2230 if (vfsp->vfc_typenum == name[0])
2231 break;
2232 if (vfsp == NULL)
2233 return (EOPNOTSUPP);
2234 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2235 oldp, oldlenp, newp, newlen, p));
2236 }
2237#endif
2238 switch (name[1]) {
2239 case VFS_MAXTYPENUM:
2240 if (namelen != 2)
2241 return (ENOTDIR);
2242 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2243 case VFS_CONF:
2244 if (namelen != 3)
2245 return (ENOTDIR); /* overloaded */
2246 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2247 if (vfsp->vfc_typenum == name[2])
2248 break;
2249 if (vfsp == NULL)
2250 return (EOPNOTSUPP);
2251 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2252 }
2253 return (EOPNOTSUPP);
2254}
2255
2256SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2257 "Generic filesystem");
2258
2259#if 1 || defined(COMPAT_PRELITE2)
2260
2261static int
2262sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2263{
2264 int error;
2265 struct vfsconf *vfsp;
2266 struct ovfsconf ovfs;
2267
2268 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2269 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2270 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2271 ovfs.vfc_index = vfsp->vfc_typenum;
2272 ovfs.vfc_refcount = vfsp->vfc_refcount;
2273 ovfs.vfc_flags = vfsp->vfc_flags;
2274 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2275 if (error)
2276 return error;
2277 }
2278 return 0;
2279}
2280
2281#endif /* 1 || COMPAT_PRELITE2 */
2282
2283#if 0
2284#define KINFO_VNODESLOP 10
2285/*
2286 * Dump vnode list (via sysctl).
2287 * Copyout address of vnode followed by vnode.
2288 */
2289/* ARGSUSED */
2290static int
2291sysctl_vnode(SYSCTL_HANDLER_ARGS)
2292{
2293 struct proc *p = curproc; /* XXX */
2294 struct mount *mp, *nmp;
2295 struct vnode *nvp, *vp;
2296 int error;
2297
2298#define VPTRSZ sizeof (struct vnode *)
2299#define VNODESZ sizeof (struct vnode)
2300
2301 req->lock = 0;
2302 if (!req->oldptr) /* Make an estimate */
2303 return (SYSCTL_OUT(req, 0,
2304 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2305
2306 simple_lock(&mountlist_slock);
2307 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2308 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2309 nmp = TAILQ_NEXT(mp, mnt_list);
2310 continue;
2311 }
2312again:
2313 simple_lock(&mntvnode_slock);
2314 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2315 vp != NULL;
2316 vp = nvp) {
2317 /*
2318 * Check that the vp is still associated with
2319 * this filesystem. RACE: could have been
2320 * recycled onto the same filesystem.
2321 */
2322 if (vp->v_mount != mp) {
2323 simple_unlock(&mntvnode_slock);
2324 goto again;
2325 }
2326 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2327 simple_unlock(&mntvnode_slock);
2328 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2329 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2330 return (error);
2331 simple_lock(&mntvnode_slock);
2332 }
2333 simple_unlock(&mntvnode_slock);
2334 simple_lock(&mountlist_slock);
2335 nmp = TAILQ_NEXT(mp, mnt_list);
2336 vfs_unbusy(mp, p);
2337 }
2338 simple_unlock(&mountlist_slock);
2339
2340 return (0);
2341}
2342#endif
2343
2344/*
2345 * XXX
2346 * Exporting the vnode list on large systems causes them to crash.
2347 * Exporting the vnode list on medium systems causes sysctl to coredump.
2348 */
2349#if 0
2350SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2351 0, 0, sysctl_vnode, "S,vnode", "");
2352#endif
2353
2354/*
2355 * Check to see if a filesystem is mounted on a block device.
2356 */
2357int
2358vfs_mountedon(vp)
2359 struct vnode *vp;
2360{
2361
2362 if (vp->v_specmountpoint != NULL)
2363 return (EBUSY);
2364 return (0);
2365}
2366
2367/*
2368 * Unmount all filesystems. The list is traversed in reverse order
2369 * of mounting to avoid dependencies.
2370 */
2371void
2372vfs_unmountall()
2373{
2374 struct mount *mp;
dadab5e9 2375 struct thread *td = curthread;
984263bc
MD
2376 int error;
2377
dadab5e9
MD
2378 if (td->td_proc == NULL)
2379 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2380
984263bc
MD
2381 /*
2382 * Since this only runs when rebooting, it is not interlocked.
2383 */
2384 while(!TAILQ_EMPTY(&mountlist)) {
2385 mp = TAILQ_LAST(&mountlist, mntlist);
dadab5e9 2386 error = dounmount(mp, MNT_FORCE, td);
984263bc
MD
2387 if (error) {
2388 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2389 printf("unmount of %s failed (",
2390 mp->mnt_stat.f_mntonname);
2391 if (error == EBUSY)
2392 printf("BUSY)\n");
2393 else
2394 printf("%d)\n", error);
2395 } else {
2396 /* The unmount has removed mp from the mountlist */
2397 }
2398 }
2399}
2400
2401/*
2402 * Build hash lists of net addresses and hang them off the mount point.
2403 * Called by ufs_mount() to set up the lists of export addresses.
2404 */
2405static int
2406vfs_hang_addrlist(mp, nep, argp)
2407 struct mount *mp;
2408 struct netexport *nep;
2409 struct export_args *argp;
2410{
2411 register struct netcred *np;
2412 register struct radix_node_head *rnh;
2413 register int i;
2414 struct radix_node *rn;
2415 struct sockaddr *saddr, *smask = 0;
2416 struct domain *dom;
2417 int error;
2418
2419 if (argp->ex_addrlen == 0) {
2420 if (mp->mnt_flag & MNT_DEFEXPORTED)
2421 return (EPERM);
2422 np = &nep->ne_defexported;
2423 np->netc_exflags = argp->ex_flags;
2424 np->netc_anon = argp->ex_anon;
2425 np->netc_anon.cr_ref = 1;
2426 mp->mnt_flag |= MNT_DEFEXPORTED;
2427 return (0);
2428 }
2429
2430 if (argp->ex_addrlen > MLEN)
2431 return (EINVAL);
2432
2433 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2434 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2435 bzero((caddr_t) np, i);
2436 saddr = (struct sockaddr *) (np + 1);
2437 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2438 goto out;
2439 if (saddr->sa_len > argp->ex_addrlen)
2440 saddr->sa_len = argp->ex_addrlen;
2441 if (argp->ex_masklen) {
2442 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2443 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2444 if (error)
2445 goto out;
2446 if (smask->sa_len > argp->ex_masklen)
2447 smask->sa_len = argp->ex_masklen;
2448 }
2449 i = saddr->sa_family;
2450 if ((rnh = nep->ne_rtable[i]) == 0) {
2451 /*
2452 * Seems silly to initialize every AF when most are not used,
2453 * do so on demand here
2454 */
2455 for (dom = domains; dom; dom = dom->dom_next)
2456 if (dom->dom_family == i && dom->dom_rtattach) {
2457 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2458 dom->dom_rtoffset);
2459 break;
2460 }
2461 if ((rnh = nep->ne_rtable[i]) == 0) {
2462 error = ENOBUFS;
2463 goto out;
2464 }
2465 }
2466 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2467 np->netc_rnodes);
2468 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2469 error = EPERM;
2470 goto out;
2471 }
2472 np->netc_exflags = argp->ex_flags;
2473 np->netc_anon = argp->ex_anon;
2474 np->netc_anon.cr_ref = 1;
2475 return (0);
2476out:
2477 free(np, M_NETADDR);
2478 return (error);
2479}
2480
2481/* ARGSUSED */
2482static int
2483vfs_free_netcred(rn, w)
2484 struct radix_node *rn;
2485 void *w;
2486{
2487 register struct radix_node_head *rnh = (struct radix_node_head *) w;
2488
2489 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2490 free((caddr_t) rn, M_NETADDR);
2491 return (0);
2492}
2493
2494/*
2495 * Free the net address hash lists that are hanging off the mount points.
2496 */
2497static void
2498vfs_free_addrlist(nep)
2499 struct netexport *nep;
2500{
2501 register int i;
2502 register struct radix_node_head *rnh;
2503
2504 for (i = 0; i <= AF_MAX; i++)
2505 if ((rnh = nep->ne_rtable[i])) {
2506 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2507 (caddr_t) rnh);
2508 free((caddr_t) rnh, M_RTABLE);
2509 nep->ne_rtable[i] = 0;
2510 }
2511}
2512
2513int
2514vfs_export(mp, nep, argp)
2515 struct mount *mp;
2516 struct netexport *nep;
2517 struct export_args *argp;
2518{
2519 int error;
2520
2521 if (argp->ex_flags & MNT_DELEXPORT) {
2522 if (mp->mnt_flag & MNT_EXPUBLIC) {
2523 vfs_setpublicfs(NULL, NULL, NULL);
2524 mp->mnt_flag &= ~MNT_EXPUBLIC;
2525 }
2526 vfs_free_addrlist(nep);
2527 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2528 }
2529 if (argp->ex_flags & MNT_EXPORTED) {
2530 if (argp->ex_flags & MNT_EXPUBLIC) {
2531 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2532 return (error);
2533 mp->mnt_flag |= MNT_EXPUBLIC;
2534 }
2535 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2536 return (error);
2537 mp->mnt_flag |= MNT_EXPORTED;
2538 }
2539 return (0);
2540}
2541
2542
2543/*
2544 * Set the publicly exported filesystem (WebNFS). Currently, only
2545 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2546 */
2547int
2548vfs_setpublicfs(mp, nep, argp)
2549 struct mount *mp;
2550 struct netexport *nep;
2551 struct export_args *argp;
2552{
2553 int error;
2554 struct vnode *rvp;
2555 char *cp;
2556
2557 /*
2558 * mp == NULL -> invalidate the current info, the FS is
2559 * no longer exported. May be called from either vfs_export
2560 * or unmount, so check if it hasn't already been done.
2561 */
2562 if (mp == NULL) {
2563 if (nfs_pub.np_valid) {
2564 nfs_pub.np_valid = 0;
2565 if (nfs_pub.np_index != NULL) {
2566 FREE(nfs_pub.np_index, M_TEMP);
2567 nfs_pub.np_index = NULL;
2568 }
2569 }
2570 return (0);
2571 }
2572
2573 /*
2574 * Only one allowed at a time.
2575 */
2576 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2577 return (EBUSY);
2578
2579 /*
2580 * Get real filehandle for root of exported FS.
2581 */
2582 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2583 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2584
2585 if ((error = VFS_ROOT(mp, &rvp)))
2586 return (error);
2587
2588 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2589 return (error);
2590
2591 vput(rvp);
2592
2593 /*
2594 * If an indexfile was specified, pull it in.
2595 */
2596 if (argp->ex_indexfile != NULL) {
2597 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2598 M_WAITOK);
2599 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2600 MAXNAMLEN, (size_t *)0);
2601 if (!error) {
2602 /*
2603 * Check for illegal filenames.
2604 */
2605 for (cp = nfs_pub.np_index; *cp; cp++) {
2606 if (*cp == '/') {
2607 error = EINVAL;
2608 break;
2609 }
2610 }
2611 }
2612 if (error) {
2613 FREE(nfs_pub.np_index, M_TEMP);
2614 return (error);
2615 }
2616 }
2617
2618 nfs_pub.np_mount = mp;
2619 nfs_pub.np_valid = 1;
2620 return (0);
2621}
2622
2623struct netcred *
2624vfs_export_lookup(mp, nep, nam)
2625 register struct mount *mp;
2626 struct netexport *nep;
2627 struct sockaddr *nam;
2628{
2629 register struct netcred *np;
2630 register struct radix_node_head *rnh;
2631 struct sockaddr *saddr;
2632
2633 np = NULL;
2634 if (mp->mnt_flag & MNT_EXPORTED) {
2635 /*
2636 * Lookup in the export list first.
2637 */
2638 if (nam != NULL) {
2639 saddr = nam;
2640 rnh = nep->ne_rtable[saddr->sa_family];
2641 if (rnh != NULL) {
2642 np = (struct netcred *)
2643 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2644 rnh);
2645 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2646 np = NULL;
2647 }
2648 }
2649 /*
2650 * If no address match, use the default if it exists.
2651 */
2652 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2653 np = &nep->ne_defexported;
2654 }
2655 return (np);
2656}
2657
2658/*
2659 * perform msync on all vnodes under a mount point
2660 * the mount point must be locked.
2661 */
2662void
2663vfs_msync(struct mount *mp, int flags)
2664{
dadab5e9 2665 struct thread *td = curthread; /* XXX */
984263bc
MD
2666 struct vnode *vp, *nvp;
2667 struct vm_object *obj;
2668 int tries;
2669
2670 tries = 5;
2671 simple_lock(&mntvnode_slock);
2672loop:
2673 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2674 if (vp->v_mount != mp) {
2675 if (--tries > 0)
2676 goto loop;
2677 break;
2678 }
2679 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2680
2681 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2682 continue;
2683
2684 /*
2685 * There could be hundreds of thousands of vnodes, we cannot
2686 * afford to do anything heavy-weight until we have a fairly
2687 * good indication that there is something to do.
2688 */
2689 if ((vp->v_flag & VOBJDIRTY) &&
2690 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2691 simple_unlock(&mntvnode_slock);
2692 if (!vget(vp,
dadab5e9 2693 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
984263bc
MD
2694 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2695 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2696 }
2697 vput(vp);
2698 }
2699 simple_lock(&mntvnode_slock);
2700 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2701 if (--tries > 0)
2702 goto loop;
2703 break;
2704 }
2705 }
2706 }
2707 simple_unlock(&mntvnode_slock);
2708}
2709
2710/*
2711 * Create the VM object needed for VMIO and mmap support. This
2712 * is done for all VREG files in the system. Some filesystems might
2713 * afford the additional metadata buffering capability of the
2714 * VMIO code by making the device node be VMIO mode also.
2715 *
2716 * vp must be locked when vfs_object_create is called.
2717 */
2718int
3b568787 2719vfs_object_create(struct vnode *vp, struct thread *td)
984263bc 2720{
3b568787 2721 return (VOP_CREATEVOBJECT(vp, td));
984263bc
MD
2722}
2723
2724void
2725vfree(vp)
2726 struct vnode *vp;
2727{
2728 int s;
2729
2730 s = splbio();
2731 simple_lock(&vnode_free_list_slock);
2732 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2733 if (vp->v_flag & VAGE) {
2734 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2735 } else {
2736 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2737 }
2738 freevnodes++;
2739 simple_unlock(&vnode_free_list_slock);
2740 vp->v_flag &= ~VAGE;
2741 vp->v_flag |= VFREE;
2742 splx(s);
2743}
2744
2745void
2746vbusy(vp)
2747 struct vnode *vp;
2748{
2749 int s;
2750
2751 s = splbio();
2752 simple_lock(&vnode_free_list_slock);
2753 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2754 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2755 freevnodes--;
2756 simple_unlock(&vnode_free_list_slock);
2757 vp->v_flag &= ~(VFREE|VAGE);
2758 splx(s);
2759}
2760
2761/*
2762 * Record a process's interest in events which might happen to
2763 * a vnode. Because poll uses the historic select-style interface
2764 * internally, this routine serves as both the ``check for any
2765 * pending events'' and the ``record my interest in future events''
2766 * functions. (These are done together, while the lock is held,
2767 * to avoid race conditions.)
2768 */
2769int
dadab5e9 2770vn_pollrecord(struct vnode *vp, struct thread *td, int events)
984263bc
MD
2771{
2772 simple_lock(&vp->v_pollinfo.vpi_lock);
2773 if (vp->v_pollinfo.vpi_revents & events) {
2774 /*
2775 * This leaves events we are not interested
2776 * in available for the other process which
2777 * which presumably had requested them
2778 * (otherwise they would never have been
2779 * recorded).
2780 */
2781 events &= vp->v_pollinfo.vpi_revents;
2782 vp->v_pollinfo.vpi_revents &= ~events;
2783
2784 simple_unlock(&vp->v_pollinfo.vpi_lock);
2785 return events;
2786 }
2787 vp->v_pollinfo.vpi_events |= events;
dadab5e9 2788 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
984263bc
MD
2789 simple_unlock(&vp->v_pollinfo.vpi_lock);
2790 return 0;
2791}
2792
2793/*
2794 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2795 * it is possible for us to miss an event due to race conditions, but
2796 * that condition is expected to be rare, so for the moment it is the
2797 * preferred interface.
2798 */
2799void
2800vn_pollevent(vp, events)
2801 struct vnode *vp;
2802 short events;
2803{
2804 simple_lock(&vp->v_pollinfo.vpi_lock);
2805 if (vp->v_pollinfo.vpi_events & events) {
2806 /*
2807 * We clear vpi_events so that we don't
2808 * call selwakeup() twice if two events are
2809 * posted before the polling process(es) is
2810 * awakened. This also ensures that we take at
2811 * most one selwakeup() if the polling process
2812 * is no longer interested. However, it does
2813 * mean that only one event can be noticed at
2814 * a time. (Perhaps we should only clear those
2815 * event bits which we note?) XXX
2816 */
2817 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2818 vp->v_pollinfo.vpi_revents |= events;
2819 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2820 }
2821 simple_unlock(&vp->v_pollinfo.vpi_lock);
2822}
2823
2824/*
2825 * Wake up anyone polling on vp because it is being revoked.
2826 * This depends on dead_poll() returning POLLHUP for correct
2827 * behavior.
2828 */
2829void
2830vn_pollgone(vp)
2831 struct vnode *vp;
2832{
2833 simple_lock(&vp->v_pollinfo.vpi_lock);
2834 if (vp->v_pollinfo.vpi_events) {
2835 vp->v_pollinfo.vpi_events = 0;
2836 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2837 }
2838 simple_unlock(&vp->v_pollinfo.vpi_lock);
2839}
2840
2841
2842
2843/*
2844 * Routine to create and manage a filesystem syncer vnode.
2845 */
2846#define sync_close ((int (*) __P((struct vop_close_args *)))nullop)
2847static int sync_fsync __P((struct vop_fsync_args *));
2848static int sync_inactive __P((struct vop_inactive_args *));
2849static int sync_reclaim __P((struct vop_reclaim_args *));
2850#define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock)
2851#define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock)
2852static int sync_print __P((struct vop_print_args *));
2853#define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2854
2855static vop_t **sync_vnodeop_p;
2856static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2857 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2858 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2859 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2860 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2861 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2862 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2863 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2864 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2865 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2866 { NULL, NULL }
2867};
2868static struct vnodeopv_desc sync_vnodeop_opv_desc =
2869 { &sync_vnodeop_p, sync_vnodeop_entries };
2870
2871VNODEOP_SET(sync_vnodeop_opv_desc);
2872
2873/*
2874 * Create a new filesystem syncer vnode for the specified mount point.
2875 */
2876int
2877vfs_allocate_syncvnode(mp)
2878 struct mount *mp;
2879{
2880 struct vnode *vp;
2881 static long start, incr, next;
2882 int error;
2883
2884 /* Allocate a new vnode */
2885 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2886 mp->mnt_syncer = NULL;
2887 return (error);
2888 }
2889 vp->v_type = VNON;
2890 /*
2891 * Place the vnode onto the syncer worklist. We attempt to
2892 * scatter them about on the list so that they will go off
2893 * at evenly distributed times even if all the filesystems
2894 * are mounted at once.
2895 */
2896 next += incr;
2897 if (next == 0 || next > syncer_maxdelay) {
2898 start /= 2;
2899 incr /= 2;
2900 if (start == 0) {
2901 start = syncer_maxdelay / 2;
2902 incr = syncer_maxdelay;
2903 }
2904 next = start;
2905 }
2906 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2907 mp->mnt_syncer = vp;
2908 return (0);
2909}
2910
2911/*
2912 * Do a lazy sync of the filesystem.
2913 */
2914static int
2915sync_fsync(ap)
2916 struct vop_fsync_args /* {
2917 struct vnode *a_vp;
2918 struct ucred *a_cred;
2919 int a_waitfor;
dadab5e9 2920 struct thread *a_td;
984263bc
MD
2921 } */ *ap;
2922{
2923 struct vnode *syncvp = ap->a_vp;
2924 struct mount *mp = syncvp->v_mount;
dadab5e9 2925 struct thread *td = ap->a_td;
984263bc
MD
2926 int asyncflag;
2927
2928 /*
2929 * We only need to do something if this is a lazy evaluation.
2930 */
2931 if (ap->a_waitfor != MNT_LAZY)
2932 return (0);
2933
2934 /*
2935 * Move ourselves to the back of the sync list.
2936 */
2937 vn_syncer_add_to_worklist(syncvp, syncdelay);
2938
2939 /*
2940 * Walk the list of vnodes pushing all that are dirty and
2941 * not already on the sync list.
2942 */
2943 simple_lock(&mountlist_slock);
dadab5e9 2944 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, td) != 0) {
984263bc
MD
2945 simple_unlock(&mountlist_slock);
2946 return (0);
2947 }
2948 asyncflag = mp->mnt_flag & MNT_ASYNC;
2949 mp->mnt_flag &= ~MNT_ASYNC;
2950 vfs_msync(mp, MNT_NOWAIT);
3b568787 2951 VFS_SYNC(mp, MNT_LAZY, td);
984263bc
MD
2952 if (asyncflag)
2953 mp->mnt_flag |= MNT_ASYNC;
dadab5e9 2954 vfs_unbusy(mp, td);
984263bc
MD
2955 return (0);
2956}
2957
2958/*
2959 * The syncer vnode is no referenced.
2960 */
2961static int
2962sync_inactive(ap)
2963 struct vop_inactive_args /* {
2964 struct vnode *a_vp;
2965 struct proc *a_p;
2966 } */ *ap;
2967{
2968
2969 vgone(ap->a_vp);
2970 return (0);
2971}
2972
2973/*
2974 * The syncer vnode is no longer needed and is being decommissioned.
2975 *
2976 * Modifications to the worklist must be protected at splbio().
2977 */
2978static int
2979sync_reclaim(ap)
2980 struct vop_reclaim_args /* {
2981 struct vnode *a_vp;
2982 } */ *ap;
2983{
2984 struct vnode *vp = ap->a_vp;
2985 int s;
2986
2987 s = splbio();
2988 vp->v_mount->mnt_syncer = NULL;
2989 if (vp->v_flag & VONWORKLST) {
2990 LIST_REMOVE(vp, v_synclist);
2991 vp->v_flag &= ~VONWORKLST;
2992 }
2993 splx(s);
2994
2995 return (0);
2996}
2997
2998/*
2999 * Print out a syncer vnode.
3000 */
3001static int
3002sync_print(ap)
3003 struct vop_print_args /* {
3004 struct vnode *a_vp;
3005 } */ *ap;
3006{
3007 struct vnode *vp = ap->a_vp;
3008
3009 printf("syncer vnode");
3010 if (vp->v_vnlock != NULL)
3011 lockmgr_printinfo(vp->v_vnlock);
3012 printf("\n");
3013 return (0);
3014}
3015
3016/*
3017 * extract the dev_t from a VBLK or VCHR
3018 */
3019dev_t
3020vn_todev(vp)
3021 struct vnode *vp;
3022{
3023 if (vp->v_type != VBLK && vp->v_type != VCHR)
3024 return (NODEV);
3025 return (vp->v_rdev);
3026}
3027
3028/*
3029 * Check if vnode represents a disk device
3030 */
3031int
3032vn_isdisk(vp, errp)
3033 struct vnode *vp;
3034 int *errp;
3035{
3036 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3037 if (errp != NULL)
3038 *errp = ENOTBLK;
3039 return (0);
3040 }
3041 if (vp->v_rdev == NULL) {
3042 if (errp != NULL)
3043 *errp = ENXIO;
3044 return (0);
3045 }
3046 if (!devsw(vp->v_rdev)) {
3047 if (errp != NULL)
3048 *errp = ENXIO;
3049 return (0);
3050 }
3051 if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) {
3052 if (errp != NULL)
3053 *errp = ENOTBLK;
3054 return (0);
3055 }
3056 if (errp != NULL)
3057 *errp = 0;
3058 return (1);
3059}
3060
3061void
3062NDFREE(ndp, flags)
3063 struct nameidata *ndp;
3064 const uint flags;
3065{
3066 if (!(flags & NDF_NO_FREE_PNBUF) &&
3067 (ndp->ni_cnd.cn_flags & HASBUF)) {
3068 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3069 ndp->ni_cnd.cn_flags &= ~HASBUF;
3070 }
3071 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3072 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3073 ndp->ni_dvp != ndp->ni_vp)
dadab5e9 3074 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
984263bc
MD
3075 if (!(flags & NDF_NO_DVP_RELE) &&
3076 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3077 vrele(ndp->ni_dvp);
3078 ndp->ni_dvp = NULL;
3079 }
3080 if (!(flags & NDF_NO_VP_UNLOCK) &&
3081 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
dadab5e9 3082 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
984263bc
MD
3083 if (!(flags & NDF_NO_VP_RELE) &&
3084 ndp->ni_vp) {
3085 vrele(ndp->ni_vp);
3086 ndp->ni_vp = NULL;
3087 }
3088 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3089 (ndp->ni_cnd.cn_flags & SAVESTART)) {
3090 vrele(ndp->ni_startdir);
3091 ndp->ni_startdir = NULL;
3092 }
3093}