MP Implmentation 2A/4: Post commit cleanup, fix missing token releases that
[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 $
8a8d5d85 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.10 2003/07/06 21:23:51 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 */
8a8d5d85
MD
128struct lwkt_token mountlist_token;
129struct lwkt_token mntvnode_token;
984263bc 130int nfs_mount_type = -1;
8a8d5d85
MD
131static struct lwkt_token mntid_token;
132static struct lwkt_token vnode_free_list_token;
133static struct lwkt_token spechash_token;
984263bc
MD
134struct nfs_public nfs_pub; /* publicly exported FS */
135static vm_zone_t vnode_zone;
136
137/*
138 * The workitem queue.
139 */
140#define SYNCER_MAXDELAY 32
141static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
142time_t syncdelay = 30; /* max time to delay syncing data */
143time_t filedelay = 30; /* time to delay syncing files */
144SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
145time_t dirdelay = 29; /* time to delay syncing directories */
146SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
147time_t metadelay = 28; /* time to delay syncing metadata */
148SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
149static int rushjob; /* number of slots to run ASAP */
150static int stat_rush_requests; /* number of times I/O speeded up */
151SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
152
153static int syncer_delayno = 0;
154static long syncer_mask;
155LIST_HEAD(synclist, vnode);
156static struct synclist *syncer_workitem_pending;
157
158int desiredvnodes;
159SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
160 &desiredvnodes, 0, "Maximum number of vnodes");
161static int minvnodes;
162SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
163 &minvnodes, 0, "Minimum number of vnodes");
164static int vnlru_nowhere = 0;
165SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
166 "Number of times the vnlru process ran without success");
167
168static void vfs_free_addrlist __P((struct netexport *nep));
169static int vfs_free_netcred __P((struct radix_node *rn, void *w));
170static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
171 struct export_args *argp));
172
173/*
174 * Initialize the vnode management data structures.
175 */
176void
177vntblinit()
178{
179
12e4aaff 180 desiredvnodes = maxproc + vmstats.v_page_count / 4;
984263bc 181 minvnodes = desiredvnodes / 4;
8a8d5d85
MD
182 lwkt_inittoken(&mntvnode_token);
183 lwkt_inittoken(&mntid_token);
184 lwkt_inittoken(&spechash_token);
984263bc 185 TAILQ_INIT(&vnode_free_list);
8a8d5d85 186 lwkt_inittoken(&vnode_free_list_token);
984263bc
MD
187 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
188 /*
189 * Initialize the filesystem syncer.
190 */
191 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
192 &syncer_mask);
193 syncer_maxdelay = syncer_mask + 1;
194}
195
196/*
197 * Mark a mount point as busy. Used to synchronize access and to delay
198 * unmounting. Interlock is not released on failure.
199 */
200int
8a8d5d85 201vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp,
dadab5e9 202 struct thread *td)
984263bc
MD
203{
204 int lkflags;
205
206 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
207 if (flags & LK_NOWAIT)
208 return (ENOENT);
209 mp->mnt_kern_flag |= MNTK_MWAIT;
210 if (interlkp) {
8a8d5d85 211 lwkt_reltoken(interlkp);
984263bc
MD
212 }
213 /*
214 * Since all busy locks are shared except the exclusive
215 * lock granted when unmounting, the only place that a
216 * wakeup needs to be done is at the release of the
217 * exclusive lock at the end of dounmount.
218 */
219 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
220 if (interlkp) {
8a8d5d85 221 lwkt_gettoken(interlkp);
984263bc
MD
222 }
223 return (ENOENT);
224 }
225 lkflags = LK_SHARED | LK_NOPAUSE;
226 if (interlkp)
227 lkflags |= LK_INTERLOCK;
dadab5e9 228 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
984263bc
MD
229 panic("vfs_busy: unexpected lock failure");
230 return (0);
231}
232
233/*
234 * Free a busy filesystem.
235 */
236void
dadab5e9 237vfs_unbusy(struct mount *mp, struct thread *td)
984263bc 238{
dadab5e9 239 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
984263bc
MD
240}
241
242/*
243 * Lookup a filesystem type, and if found allocate and initialize
244 * a mount structure for it.
245 *
246 * Devname is usually updated by mount(8) after booting.
247 */
248int
dadab5e9 249vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
984263bc 250{
dadab5e9 251 struct thread *td = curthread; /* XXX */
984263bc
MD
252 struct vfsconf *vfsp;
253 struct mount *mp;
254
255 if (fstypename == NULL)
256 return (ENODEV);
257 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
258 if (!strcmp(vfsp->vfc_name, fstypename))
259 break;
260 if (vfsp == NULL)
261 return (ENODEV);
262 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
263 bzero((char *)mp, (u_long)sizeof(struct mount));
264 lockinit(&mp->mnt_lock, PVFS, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
dadab5e9 265 (void)vfs_busy(mp, LK_NOWAIT, 0, td);
984263bc
MD
266 TAILQ_INIT(&mp->mnt_nvnodelist);
267 TAILQ_INIT(&mp->mnt_reservedvnlist);
268 mp->mnt_nvnodelistsize = 0;
269 mp->mnt_vfc = vfsp;
270 mp->mnt_op = vfsp->vfc_vfsops;
271 mp->mnt_flag = MNT_RDONLY;
272 mp->mnt_vnodecovered = NULLVP;
273 vfsp->vfc_refcount++;
274 mp->mnt_iosize_max = DFLTPHYS;
275 mp->mnt_stat.f_type = vfsp->vfc_typenum;
276 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
277 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
278 mp->mnt_stat.f_mntonname[0] = '/';
279 mp->mnt_stat.f_mntonname[1] = 0;
280 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
281 *mpp = mp;
282 return (0);
283}
284
285/*
286 * Find an appropriate filesystem to use for the root. If a filesystem
287 * has not been preselected, walk through the list of known filesystems
288 * trying those that have mountroot routines, and try them until one
289 * works or we have tried them all.
290 */
291#ifdef notdef /* XXX JH */
292int
293lite2_vfs_mountroot()
294{
295 struct vfsconf *vfsp;
296 extern int (*lite2_mountroot) __P((void));
297 int error;
298
299 if (lite2_mountroot != NULL)
300 return ((*lite2_mountroot)());
301 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
302 if (vfsp->vfc_mountroot == NULL)
303 continue;
304 if ((error = (*vfsp->vfc_mountroot)()) == 0)
305 return (0);
306 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
307 }
308 return (ENODEV);
309}
310#endif
311
312/*
313 * Lookup a mount point by filesystem identifier.
314 */
315struct mount *
316vfs_getvfs(fsid)
317 fsid_t *fsid;
318{
319 register struct mount *mp;
320
8a8d5d85 321 lwkt_gettoken(&mountlist_token);
984263bc
MD
322 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
323 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
324 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
8a8d5d85 325 lwkt_reltoken(&mountlist_token);
984263bc
MD
326 return (mp);
327 }
328 }
8a8d5d85 329 lwkt_reltoken(&mountlist_token);
984263bc
MD
330 return ((struct mount *) 0);
331}
332
333/*
334 * Get a new unique fsid. Try to make its val[0] unique, since this value
335 * will be used to create fake device numbers for stat(). Also try (but
336 * not so hard) make its val[0] unique mod 2^16, since some emulators only
337 * support 16-bit device numbers. We end up with unique val[0]'s for the
338 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
339 *
340 * Keep in mind that several mounts may be running in parallel. Starting
341 * the search one past where the previous search terminated is both a
342 * micro-optimization and a defense against returning the same fsid to
343 * different mounts.
344 */
345void
346vfs_getnewfsid(mp)
347 struct mount *mp;
348{
349 static u_int16_t mntid_base;
350 fsid_t tfsid;
351 int mtype;
352
8a8d5d85 353 lwkt_gettoken(&mntid_token);
984263bc
MD
354 mtype = mp->mnt_vfc->vfc_typenum;
355 tfsid.val[1] = mtype;
356 mtype = (mtype & 0xFF) << 24;
357 for (;;) {
358 tfsid.val[0] = makeudev(255,
359 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
360 mntid_base++;
361 if (vfs_getvfs(&tfsid) == NULL)
362 break;
363 }
364 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
365 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
8a8d5d85 366 lwkt_reltoken(&mntid_token);
984263bc
MD
367}
368
369/*
370 * Knob to control the precision of file timestamps:
371 *
372 * 0 = seconds only; nanoseconds zeroed.
373 * 1 = seconds and nanoseconds, accurate within 1/HZ.
374 * 2 = seconds and nanoseconds, truncated to microseconds.
375 * >=3 = seconds and nanoseconds, maximum precision.
376 */
377enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
378
379static int timestamp_precision = TSP_SEC;
380SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
381 &timestamp_precision, 0, "");
382
383/*
384 * Get a current timestamp.
385 */
386void
387vfs_timestamp(tsp)
388 struct timespec *tsp;
389{
390 struct timeval tv;
391
392 switch (timestamp_precision) {
393 case TSP_SEC:
394 tsp->tv_sec = time_second;
395 tsp->tv_nsec = 0;
396 break;
397 case TSP_HZ:
398 getnanotime(tsp);
399 break;
400 case TSP_USEC:
401 microtime(&tv);
402 TIMEVAL_TO_TIMESPEC(&tv, tsp);
403 break;
404 case TSP_NSEC:
405 default:
406 nanotime(tsp);
407 break;
408 }
409}
410
411/*
412 * Set vnode attributes to VNOVAL
413 */
414void
415vattr_null(vap)
416 register struct vattr *vap;
417{
418
419 vap->va_type = VNON;
420 vap->va_size = VNOVAL;
421 vap->va_bytes = VNOVAL;
422 vap->va_mode = VNOVAL;
423 vap->va_nlink = VNOVAL;
424 vap->va_uid = VNOVAL;
425 vap->va_gid = VNOVAL;
426 vap->va_fsid = VNOVAL;
427 vap->va_fileid = VNOVAL;
428 vap->va_blocksize = VNOVAL;
429 vap->va_rdev = VNOVAL;
430 vap->va_atime.tv_sec = VNOVAL;
431 vap->va_atime.tv_nsec = VNOVAL;
432 vap->va_mtime.tv_sec = VNOVAL;
433 vap->va_mtime.tv_nsec = VNOVAL;
434 vap->va_ctime.tv_sec = VNOVAL;
435 vap->va_ctime.tv_nsec = VNOVAL;
436 vap->va_flags = VNOVAL;
437 vap->va_gen = VNOVAL;
438 vap->va_vaflags = 0;
439}
440
441/*
442 * This routine is called when we have too many vnodes. It attempts
443 * to free <count> vnodes and will potentially free vnodes that still
444 * have VM backing store (VM backing store is typically the cause
445 * of a vnode blowout so we want to do this). Therefore, this operation
446 * is not considered cheap.
447 *
448 * A number of conditions may prevent a vnode from being reclaimed.
449 * the buffer cache may have references on the vnode, a directory
450 * vnode may still have references due to the namei cache representing
451 * underlying files, or the vnode may be in active use. It is not
452 * desireable to reuse such vnodes. These conditions may cause the
453 * number of vnodes to reach some minimum value regardless of what
454 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
455 */
456static int
457vlrureclaim(struct mount *mp)
458{
459 struct vnode *vp;
460 int done;
461 int trigger;
462 int usevnodes;
463 int count;
8a8d5d85 464 int gen;
984263bc
MD
465
466 /*
467 * Calculate the trigger point, don't allow user
468 * screwups to blow us up. This prevents us from
469 * recycling vnodes with lots of resident pages. We
470 * aren't trying to free memory, we are trying to
471 * free vnodes.
472 */
473 usevnodes = desiredvnodes;
474 if (usevnodes <= 0)
475 usevnodes = 1;
12e4aaff 476 trigger = vmstats.v_page_count * 2 / usevnodes;
984263bc
MD
477
478 done = 0;
8a8d5d85 479 gen = lwkt_gettoken(&mntvnode_token);
984263bc
MD
480 count = mp->mnt_nvnodelistsize / 10 + 1;
481 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
482 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
483 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
484
485 if (vp->v_type != VNON &&
486 vp->v_type != VBAD &&
487 VMIGHTFREE(vp) && /* critical path opt */
8a8d5d85 488 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger)
984263bc 489 ) {
8a8d5d85
MD
490 lwkt_gettoken(&vp->v_interlock);
491 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) {
492 if (VMIGHTFREE(vp)) {
493 vgonel(vp, curthread);
494 done++;
495 } else {
496 lwkt_reltoken(&vp->v_interlock);
497 }
984263bc 498 } else {
8a8d5d85 499 lwkt_reltoken(&vp->v_interlock);
984263bc 500 }
984263bc
MD
501 }
502 --count;
503 }
8a8d5d85 504 lwkt_reltoken(&mntvnode_token);
984263bc
MD
505 return done;
506}
507
508/*
509 * Attempt to recycle vnodes in a context that is always safe to block.
510 * Calling vlrurecycle() from the bowels of file system code has some
511 * interesting deadlock problems.
512 */
bc6dffab 513static struct thread *vnlruthread;
984263bc
MD
514static int vnlruproc_sig;
515
516static void
517vnlru_proc(void)
518{
519 struct mount *mp, *nmp;
520 int s;
521 int done;
0cfcada1 522 struct thread *td = curthread;
984263bc 523
bc6dffab 524 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
525 SHUTDOWN_PRI_FIRST);
526
527 s = splbio();
528 for (;;) {
0cfcada1 529 kproc_suspend_loop();
984263bc
MD
530 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
531 vnlruproc_sig = 0;
532 wakeup(&vnlruproc_sig);
0cfcada1 533 tsleep(td, PVFS, "vlruwt", hz);
984263bc
MD
534 continue;
535 }
536 done = 0;
8a8d5d85 537 lwkt_gettoken(&mountlist_token);
984263bc 538 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 539 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
984263bc
MD
540 nmp = TAILQ_NEXT(mp, mnt_list);
541 continue;
542 }
543 done += vlrureclaim(mp);
8a8d5d85 544 lwkt_gettoken(&mountlist_token);
984263bc 545 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 546 vfs_unbusy(mp, td);
984263bc 547 }
8a8d5d85 548 lwkt_reltoken(&mountlist_token);
984263bc
MD
549 if (done == 0) {
550 vnlru_nowhere++;
0cfcada1 551 tsleep(td, PPAUSE, "vlrup", hz * 3);
984263bc
MD
552 }
553 }
554 splx(s);
555}
556
557static struct kproc_desc vnlru_kp = {
558 "vnlru",
559 vnlru_proc,
bc6dffab 560 &vnlruthread
984263bc
MD
561};
562SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
563
564/*
565 * Routines having to do with the management of the vnode table.
566 */
567extern vop_t **dead_vnodeop_p;
568
569/*
570 * Return the next vnode from the free list.
571 */
572int
573getnewvnode(tag, mp, vops, vpp)
574 enum vtagtype tag;
575 struct mount *mp;
576 vop_t **vops;
577 struct vnode **vpp;
578{
579 int s;
8a8d5d85
MD
580 int gen;
581 int vgen;
dadab5e9 582 struct thread *td = curthread; /* XXX */
984263bc
MD
583 struct vnode *vp = NULL;
584 vm_object_t object;
585
586 s = splbio();
587
588 /*
589 * Try to reuse vnodes if we hit the max. This situation only
590 * occurs in certain large-memory (2G+) situations. We cannot
591 * attempt to directly reclaim vnodes due to nasty recursion
592 * problems.
593 */
594 while (numvnodes - freevnodes > desiredvnodes) {
595 if (vnlruproc_sig == 0) {
596 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
bc6dffab 597 wakeup(vnlruthread);
984263bc
MD
598 }
599 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
600 }
601
602
603 /*
604 * Attempt to reuse a vnode already on the free list, allocating
605 * a new vnode if we can't find one or if we have not reached a
606 * good minimum for good LRU performance.
607 */
8a8d5d85 608 gen = lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
609 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
610 int count;
611
612 for (count = 0; count < freevnodes; count++) {
613 vp = TAILQ_FIRST(&vnode_free_list);
614 if (vp == NULL || vp->v_usecount)
615 panic("getnewvnode: free vnode isn't");
616
8a8d5d85
MD
617 /*
618 * Get the vnode's interlock, then re-obtain
619 * vnode_free_list_token in case we lost it. If we
620 * did lose it while getting the vnode interlock,
621 * even if we got it back again, then retry.
622 */
623 vgen = lwkt_gettoken(&vp->v_interlock);
624 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
625 --count;
626 lwkt_reltoken(&vp->v_interlock);
627 vp = NULL;
628 continue;
629 }
630
631 /*
632 * Whew! We have both tokens. Since we didn't lose
633 * the free list VFREE had better still be set. But
634 * we aren't out of the woods yet. We have to get
635 * the object (may block). If the vnode is not
636 * suitable then move it to the end of the list
637 * if we can. If we can't move it to the end of the
638 * list retry again.
639 */
984263bc 640 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
8a8d5d85
MD
641 (object->resident_page_count || object->ref_count))
642 ) {
643 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 &&
644 lwkt_gentoken(&vnode_free_list_token, &gen) == 0
645 ) {
646 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
647 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
648 } else {
649 --count;
650 }
651 lwkt_reltoken(&vp->v_interlock);
652 vp = NULL;
653 continue;
654 }
655
656 /*
657 * Still not out of the woods. VOBJECT might have
658 * blocked, if we did not retain our tokens we have
659 * to retry.
660 */
661 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 ||
662 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
663 --count;
984263bc
MD
664 vp = NULL;
665 continue;
666 }
8a8d5d85
MD
667 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
668 KKASSERT(vp->v_flag & VFREE);
669
984263bc
MD
670 if (LIST_FIRST(&vp->v_cache_src)) {
671 /*
672 * note: nameileafonly sysctl is temporary,
673 * for debugging only, and will eventually be
674 * removed.
675 */
676 if (nameileafonly > 0) {
677 /*
678 * Do not reuse namei-cached directory
679 * vnodes that have cached
680 * subdirectories.
681 */
682 if (cache_leaf_test(vp) < 0) {
8a8d5d85 683 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
684 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
685 vp = NULL;
686 continue;
687 }
688 } else if (nameileafonly < 0 ||
689 vmiodirenable == 0) {
690 /*
691 * Do not reuse namei-cached directory
692 * vnodes if nameileafonly is -1 or
693 * if VMIO backing for directories is
694 * turned off (otherwise we reuse them
695 * too quickly).
696 */
8a8d5d85 697 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
698 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
699 vp = NULL;
700 continue;
701 }
702 }
703 break;
704 }
705 }
706
707 if (vp) {
708 vp->v_flag |= VDOOMED;
709 vp->v_flag &= ~VFREE;
710 freevnodes--;
8a8d5d85
MD
711 lwkt_reltoken(&vnode_free_list_token);
712 cache_purge(vp); /* YYY may block */
984263bc
MD
713 vp->v_lease = NULL;
714 if (vp->v_type != VBAD) {
dadab5e9 715 vgonel(vp, td);
984263bc 716 } else {
8a8d5d85 717 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
718 }
719
720#ifdef INVARIANTS
721 {
722 int s;
723
724 if (vp->v_data)
725 panic("cleaned vnode isn't");
726 s = splbio();
727 if (vp->v_numoutput)
728 panic("Clean vnode has pending I/O's");
729 splx(s);
730 }
731#endif
732 vp->v_flag = 0;
733 vp->v_lastw = 0;
734 vp->v_lasta = 0;
735 vp->v_cstart = 0;
736 vp->v_clen = 0;
737 vp->v_socket = 0;
738 vp->v_writecount = 0; /* XXX */
739 } else {
8a8d5d85 740 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
741 vp = (struct vnode *) zalloc(vnode_zone);
742 bzero((char *) vp, sizeof *vp);
8a8d5d85 743 lwkt_inittoken(&vp->v_interlock);
984263bc
MD
744 vp->v_dd = vp;
745 cache_purge(vp);
746 LIST_INIT(&vp->v_cache_src);
747 TAILQ_INIT(&vp->v_cache_dst);
748 numvnodes++;
749 }
750
751 TAILQ_INIT(&vp->v_cleanblkhd);
752 TAILQ_INIT(&vp->v_dirtyblkhd);
753 vp->v_type = VNON;
754 vp->v_tag = tag;
755 vp->v_op = vops;
756 insmntque(vp, mp);
757 *vpp = vp;
758 vp->v_usecount = 1;
759 vp->v_data = 0;
760 splx(s);
761
3b568787 762 vfs_object_create(vp, td);
984263bc
MD
763 return (0);
764}
765
766/*
767 * Move a vnode from one mount queue to another.
768 */
769static void
770insmntque(vp, mp)
771 register struct vnode *vp;
772 register struct mount *mp;
773{
774
8a8d5d85 775 lwkt_gettoken(&mntvnode_token);
984263bc
MD
776 /*
777 * Delete from old mount point vnode list, if on one.
778 */
779 if (vp->v_mount != NULL) {
780 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
781 ("bad mount point vnode list size"));
782 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
783 vp->v_mount->mnt_nvnodelistsize--;
784 }
785 /*
786 * Insert into list of vnodes for the new mount point, if available.
787 */
788 if ((vp->v_mount = mp) == NULL) {
8a8d5d85 789 lwkt_reltoken(&mntvnode_token);
984263bc
MD
790 return;
791 }
792 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
793 mp->mnt_nvnodelistsize++;
8a8d5d85 794 lwkt_reltoken(&mntvnode_token);
984263bc
MD
795}
796
797/*
798 * Update outstanding I/O count and do wakeup if requested.
799 */
800void
801vwakeup(bp)
802 register struct buf *bp;
803{
804 register struct vnode *vp;
805
806 bp->b_flags &= ~B_WRITEINPROG;
807 if ((vp = bp->b_vp)) {
808 vp->v_numoutput--;
809 if (vp->v_numoutput < 0)
810 panic("vwakeup: neg numoutput");
811 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
812 vp->v_flag &= ~VBWAIT;
813 wakeup((caddr_t) &vp->v_numoutput);
814 }
815 }
816}
817
818/*
819 * Flush out and invalidate all buffers associated with a vnode.
820 * Called with the underlying object locked.
821 */
822int
3b568787
MD
823vinvalbuf(struct vnode *vp, int flags, struct thread *td,
824 int slpflag, int slptimeo)
984263bc
MD
825{
826 register struct buf *bp;
827 struct buf *nbp, *blist;
828 int s, error;
829 vm_object_t object;
830
831 if (flags & V_SAVE) {
832 s = splbio();
833 while (vp->v_numoutput) {
834 vp->v_flag |= VBWAIT;
835 error = tsleep((caddr_t)&vp->v_numoutput,
836 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
837 if (error) {
838 splx(s);
839 return (error);
840 }
841 }
842 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
843 splx(s);
3b568787 844 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
984263bc
MD
845 return (error);
846 s = splbio();
847 if (vp->v_numoutput > 0 ||
848 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
849 panic("vinvalbuf: dirty bufs");
850 }
851 splx(s);
852 }
853 s = splbio();
854 for (;;) {
855 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
856 if (!blist)
857 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
858 if (!blist)
859 break;
860
861 for (bp = blist; bp; bp = nbp) {
862 nbp = TAILQ_NEXT(bp, b_vnbufs);
863 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
864 error = BUF_TIMELOCK(bp,
865 LK_EXCLUSIVE | LK_SLEEPFAIL,
866 "vinvalbuf", slpflag, slptimeo);
867 if (error == ENOLCK)
868 break;
869 splx(s);
870 return (error);
871 }
872 /*
873 * XXX Since there are no node locks for NFS, I
874 * believe there is a slight chance that a delayed
875 * write will occur while sleeping just above, so
876 * check for it. Note that vfs_bio_awrite expects
877 * buffers to reside on a queue, while VOP_BWRITE and
878 * brelse do not.
879 */
880 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
881 (flags & V_SAVE)) {
882
883 if (bp->b_vp == vp) {
884 if (bp->b_flags & B_CLUSTEROK) {
885 BUF_UNLOCK(bp);
886 vfs_bio_awrite(bp);
887 } else {
888 bremfree(bp);
889 bp->b_flags |= B_ASYNC;
890 VOP_BWRITE(bp->b_vp, bp);
891 }
892 } else {
893 bremfree(bp);
894 (void) VOP_BWRITE(bp->b_vp, bp);
895 }
896 break;
897 }
898 bremfree(bp);
899 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
900 bp->b_flags &= ~B_ASYNC;
901 brelse(bp);
902 }
903 }
904
905 /*
906 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
907 * have write I/O in-progress but if there is a VM object then the
908 * VM object can also have read-I/O in-progress.
909 */
910 do {
911 while (vp->v_numoutput > 0) {
912 vp->v_flag |= VBWAIT;
913 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
914 }
915 if (VOP_GETVOBJECT(vp, &object) == 0) {
916 while (object->paging_in_progress)
917 vm_object_pip_sleep(object, "vnvlbx");
918 }
919 } while (vp->v_numoutput > 0);
920
921 splx(s);
922
923 /*
924 * Destroy the copy in the VM cache, too.
925 */
8a8d5d85 926 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
927 if (VOP_GETVOBJECT(vp, &object) == 0) {
928 vm_object_page_remove(object, 0, 0,
929 (flags & V_SAVE) ? TRUE : FALSE);
930 }
8a8d5d85 931 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
932
933 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
934 panic("vinvalbuf: flush failed");
935 return (0);
936}
937
938/*
939 * Truncate a file's buffer and pages to a specified length. This
940 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
941 * sync activity.
942 */
943int
3b568787 944vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
984263bc 945{
dadab5e9 946 struct buf *bp;
984263bc
MD
947 struct buf *nbp;
948 int s, anyfreed;
949 int trunclbn;
950
951 /*
952 * Round up to the *next* lbn.
953 */
954 trunclbn = (length + blksize - 1) / blksize;
955
956 s = splbio();
957restart:
958 anyfreed = 1;
959 for (;anyfreed;) {
960 anyfreed = 0;
961 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
962 nbp = TAILQ_NEXT(bp, b_vnbufs);
963 if (bp->b_lblkno >= trunclbn) {
964 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
965 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
966 goto restart;
967 } else {
968 bremfree(bp);
969 bp->b_flags |= (B_INVAL | B_RELBUF);
970 bp->b_flags &= ~B_ASYNC;
971 brelse(bp);
972 anyfreed = 1;
973 }
974 if (nbp &&
975 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
976 (nbp->b_vp != vp) ||
977 (nbp->b_flags & B_DELWRI))) {
978 goto restart;
979 }
980 }
981 }
982
983 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
984 nbp = TAILQ_NEXT(bp, b_vnbufs);
985 if (bp->b_lblkno >= trunclbn) {
986 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
987 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
988 goto restart;
989 } else {
990 bremfree(bp);
991 bp->b_flags |= (B_INVAL | B_RELBUF);
992 bp->b_flags &= ~B_ASYNC;
993 brelse(bp);
994 anyfreed = 1;
995 }
996 if (nbp &&
997 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
998 (nbp->b_vp != vp) ||
999 (nbp->b_flags & B_DELWRI) == 0)) {
1000 goto restart;
1001 }
1002 }
1003 }
1004 }
1005
1006 if (length > 0) {
1007restartsync:
1008 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1009 nbp = TAILQ_NEXT(bp, b_vnbufs);
1010 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1011 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1012 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1013 goto restart;
1014 } else {
1015 bremfree(bp);
1016 if (bp->b_vp == vp) {
1017 bp->b_flags |= B_ASYNC;
1018 } else {
1019 bp->b_flags &= ~B_ASYNC;
1020 }
1021 VOP_BWRITE(bp->b_vp, bp);
1022 }
1023 goto restartsync;
1024 }
1025
1026 }
1027 }
1028
1029 while (vp->v_numoutput > 0) {
1030 vp->v_flag |= VBWAIT;
1031 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
1032 }
1033
1034 splx(s);
1035
1036 vnode_pager_setsize(vp, length);
1037
1038 return (0);
1039}
1040
1041/*
1042 * Associate a buffer with a vnode.
1043 */
1044void
1045bgetvp(vp, bp)
1046 register struct vnode *vp;
1047 register struct buf *bp;
1048{
1049 int s;
1050
1051 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1052
1053 vhold(vp);
1054 bp->b_vp = vp;
1055 bp->b_dev = vn_todev(vp);
1056 /*
1057 * Insert onto list for new vnode.
1058 */
1059 s = splbio();
1060 bp->b_xflags |= BX_VNCLEAN;
1061 bp->b_xflags &= ~BX_VNDIRTY;
1062 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1063 splx(s);
1064}
1065
1066/*
1067 * Disassociate a buffer from a vnode.
1068 */
1069void
1070brelvp(bp)
1071 register struct buf *bp;
1072{
1073 struct vnode *vp;
1074 struct buflists *listheadp;
1075 int s;
1076
1077 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1078
1079 /*
1080 * Delete from old vnode list, if on one.
1081 */
1082 vp = bp->b_vp;
1083 s = splbio();
1084 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1085 if (bp->b_xflags & BX_VNDIRTY)
1086 listheadp = &vp->v_dirtyblkhd;
1087 else
1088 listheadp = &vp->v_cleanblkhd;
1089 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1090 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1091 }
1092 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1093 vp->v_flag &= ~VONWORKLST;
1094 LIST_REMOVE(vp, v_synclist);
1095 }
1096 splx(s);
1097 bp->b_vp = (struct vnode *) 0;
1098 vdrop(vp);
1099}
1100
1101/*
1102 * The workitem queue.
1103 *
1104 * It is useful to delay writes of file data and filesystem metadata
1105 * for tens of seconds so that quickly created and deleted files need
1106 * not waste disk bandwidth being created and removed. To realize this,
1107 * we append vnodes to a "workitem" queue. When running with a soft
1108 * updates implementation, most pending metadata dependencies should
1109 * not wait for more than a few seconds. Thus, mounted on block devices
1110 * are delayed only about a half the time that file data is delayed.
1111 * Similarly, directory updates are more critical, so are only delayed
1112 * about a third the time that file data is delayed. Thus, there are
1113 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1114 * one each second (driven off the filesystem syncer process). The
1115 * syncer_delayno variable indicates the next queue that is to be processed.
1116 * Items that need to be processed soon are placed in this queue:
1117 *
1118 * syncer_workitem_pending[syncer_delayno]
1119 *
1120 * A delay of fifteen seconds is done by placing the request fifteen
1121 * entries later in the queue:
1122 *
1123 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1124 *
1125 */
1126
1127/*
1128 * Add an item to the syncer work queue.
1129 */
1130static void
1131vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1132{
1133 int s, slot;
1134
1135 s = splbio();
1136
1137 if (vp->v_flag & VONWORKLST) {
1138 LIST_REMOVE(vp, v_synclist);
1139 }
1140
1141 if (delay > syncer_maxdelay - 2)
1142 delay = syncer_maxdelay - 2;
1143 slot = (syncer_delayno + delay) & syncer_mask;
1144
1145 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1146 vp->v_flag |= VONWORKLST;
1147 splx(s);
1148}
1149
bc6dffab 1150struct thread *updatethread;
984263bc
MD
1151static void sched_sync __P((void));
1152static struct kproc_desc up_kp = {
1153 "syncer",
1154 sched_sync,
bc6dffab 1155 &updatethread
984263bc
MD
1156};
1157SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1158
1159/*
1160 * System filesystem synchronizer daemon.
1161 */
1162void
1163sched_sync(void)
1164{
1165 struct synclist *slp;
1166 struct vnode *vp;
1167 long starttime;
1168 int s;
0cfcada1 1169 struct thread *td = curthread;
984263bc 1170
bc6dffab 1171 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
1172 SHUTDOWN_PRI_LAST);
1173
1174 for (;;) {
0cfcada1 1175 kproc_suspend_loop();
984263bc
MD
1176
1177 starttime = time_second;
1178
1179 /*
1180 * Push files whose dirty time has expired. Be careful
1181 * of interrupt race on slp queue.
1182 */
1183 s = splbio();
1184 slp = &syncer_workitem_pending[syncer_delayno];
1185 syncer_delayno += 1;
1186 if (syncer_delayno == syncer_maxdelay)
1187 syncer_delayno = 0;
1188 splx(s);
1189
1190 while ((vp = LIST_FIRST(slp)) != NULL) {
1191 if (VOP_ISLOCKED(vp, NULL) == 0) {
dadab5e9 1192 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3b568787 1193 (void) VOP_FSYNC(vp, MNT_LAZY, td);
dadab5e9 1194 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1195 }
1196 s = splbio();
1197 if (LIST_FIRST(slp) == vp) {
1198 /*
1199 * Note: v_tag VT_VFS vps can remain on the
1200 * worklist too with no dirty blocks, but
1201 * since sync_fsync() moves it to a different
1202 * slot we are safe.
1203 */
1204 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1205 !vn_isdisk(vp, NULL))
1206 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1207 /*
1208 * Put us back on the worklist. The worklist
1209 * routine will remove us from our current
1210 * position and then add us back in at a later
1211 * position.
1212 */
1213 vn_syncer_add_to_worklist(vp, syncdelay);
1214 }
1215 splx(s);
1216 }
1217
1218 /*
1219 * Do soft update processing.
1220 */
1221 if (bioops.io_sync)
1222 (*bioops.io_sync)(NULL);
1223
1224 /*
1225 * The variable rushjob allows the kernel to speed up the
1226 * processing of the filesystem syncer process. A rushjob
1227 * value of N tells the filesystem syncer to process the next
1228 * N seconds worth of work on its queue ASAP. Currently rushjob
1229 * is used by the soft update code to speed up the filesystem
1230 * syncer process when the incore state is getting so far
1231 * ahead of the disk that the kernel memory pool is being
1232 * threatened with exhaustion.
1233 */
1234 if (rushjob > 0) {
1235 rushjob -= 1;
1236 continue;
1237 }
1238 /*
1239 * If it has taken us less than a second to process the
1240 * current work, then wait. Otherwise start right over
1241 * again. We can still lose time if any single round
1242 * takes more than two seconds, but it does not really
1243 * matter as we are just trying to generally pace the
1244 * filesystem activity.
1245 */
1246 if (time_second == starttime)
1247 tsleep(&lbolt, PPAUSE, "syncer", 0);
1248 }
1249}
1250
1251/*
1252 * Request the syncer daemon to speed up its work.
1253 * We never push it to speed up more than half of its
1254 * normal turn time, otherwise it could take over the cpu.
1255 */
1256int
1257speedup_syncer()
1258{
1259 int s;
1260
1261 s = splhigh();
bc6dffab
MD
1262 if (updatethread->td_proc->p_wchan == &lbolt) /* YYY */
1263 setrunnable(updatethread->td_proc);
984263bc
MD
1264 splx(s);
1265 if (rushjob < syncdelay / 2) {
1266 rushjob += 1;
1267 stat_rush_requests += 1;
1268 return (1);
1269 }
1270 return(0);
1271}
1272
1273/*
1274 * Associate a p-buffer with a vnode.
1275 *
1276 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1277 * with the buffer. i.e. the bp has not been linked into the vnode or
1278 * ref-counted.
1279 */
1280void
1281pbgetvp(vp, bp)
1282 register struct vnode *vp;
1283 register struct buf *bp;
1284{
1285
1286 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1287
1288 bp->b_vp = vp;
1289 bp->b_flags |= B_PAGING;
1290 bp->b_dev = vn_todev(vp);
1291}
1292
1293/*
1294 * Disassociate a p-buffer from a vnode.
1295 */
1296void
1297pbrelvp(bp)
1298 register struct buf *bp;
1299{
1300
1301 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1302
1303 /* XXX REMOVE ME */
1304 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1305 panic(
1306 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1307 bp,
1308 (int)bp->b_flags
1309 );
1310 }
1311 bp->b_vp = (struct vnode *) 0;
1312 bp->b_flags &= ~B_PAGING;
1313}
1314
1315void
1316pbreassignbuf(bp, newvp)
1317 struct buf *bp;
1318 struct vnode *newvp;
1319{
1320 if ((bp->b_flags & B_PAGING) == 0) {
1321 panic(
1322 "pbreassignbuf() on non phys bp %p",
1323 bp
1324 );
1325 }
1326 bp->b_vp = newvp;
1327}
1328
1329/*
1330 * Reassign a buffer from one vnode to another.
1331 * Used to assign file specific control information
1332 * (indirect blocks) to the vnode to which they belong.
1333 */
1334void
1335reassignbuf(bp, newvp)
1336 register struct buf *bp;
1337 register struct vnode *newvp;
1338{
1339 struct buflists *listheadp;
1340 int delay;
1341 int s;
1342
1343 if (newvp == NULL) {
1344 printf("reassignbuf: NULL");
1345 return;
1346 }
1347 ++reassignbufcalls;
1348
1349 /*
1350 * B_PAGING flagged buffers cannot be reassigned because their vp
1351 * is not fully linked in.
1352 */
1353 if (bp->b_flags & B_PAGING)
1354 panic("cannot reassign paging buffer");
1355
1356 s = splbio();
1357 /*
1358 * Delete from old vnode list, if on one.
1359 */
1360 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1361 if (bp->b_xflags & BX_VNDIRTY)
1362 listheadp = &bp->b_vp->v_dirtyblkhd;
1363 else
1364 listheadp = &bp->b_vp->v_cleanblkhd;
1365 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1366 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1367 if (bp->b_vp != newvp) {
1368 vdrop(bp->b_vp);
1369 bp->b_vp = NULL; /* for clarification */
1370 }
1371 }
1372 /*
1373 * If dirty, put on list of dirty buffers; otherwise insert onto list
1374 * of clean buffers.
1375 */
1376 if (bp->b_flags & B_DELWRI) {
1377 struct buf *tbp;
1378
1379 listheadp = &newvp->v_dirtyblkhd;
1380 if ((newvp->v_flag & VONWORKLST) == 0) {
1381 switch (newvp->v_type) {
1382 case VDIR:
1383 delay = dirdelay;
1384 break;
1385 case VCHR:
1386 case VBLK:
1387 if (newvp->v_specmountpoint != NULL) {
1388 delay = metadelay;
1389 break;
1390 }
1391 /* fall through */
1392 default:
1393 delay = filedelay;
1394 }
1395 vn_syncer_add_to_worklist(newvp, delay);
1396 }
1397 bp->b_xflags |= BX_VNDIRTY;
1398 tbp = TAILQ_FIRST(listheadp);
1399 if (tbp == NULL ||
1400 bp->b_lblkno == 0 ||
1401 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1402 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1403 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1404 ++reassignbufsortgood;
1405 } else if (bp->b_lblkno < 0) {
1406 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1407 ++reassignbufsortgood;
1408 } else if (reassignbufmethod == 1) {
1409 /*
1410 * New sorting algorithm, only handle sequential case,
1411 * otherwise append to end (but before metadata)
1412 */
1413 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1414 (tbp->b_xflags & BX_VNDIRTY)) {
1415 /*
1416 * Found the best place to insert the buffer
1417 */
1418 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1419 ++reassignbufsortgood;
1420 } else {
1421 /*
1422 * Missed, append to end, but before meta-data.
1423 * We know that the head buffer in the list is
1424 * not meta-data due to prior conditionals.
1425 *
1426 * Indirect effects: NFS second stage write
1427 * tends to wind up here, giving maximum
1428 * distance between the unstable write and the
1429 * commit rpc.
1430 */
1431 tbp = TAILQ_LAST(listheadp, buflists);
1432 while (tbp && tbp->b_lblkno < 0)
1433 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1434 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1435 ++reassignbufsortbad;
1436 }
1437 } else {
1438 /*
1439 * Old sorting algorithm, scan queue and insert
1440 */
1441 struct buf *ttbp;
1442 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1443 (ttbp->b_lblkno < bp->b_lblkno)) {
1444 ++reassignbufloops;
1445 tbp = ttbp;
1446 }
1447 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1448 }
1449 } else {
1450 bp->b_xflags |= BX_VNCLEAN;
1451 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1452 if ((newvp->v_flag & VONWORKLST) &&
1453 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1454 newvp->v_flag &= ~VONWORKLST;
1455 LIST_REMOVE(newvp, v_synclist);
1456 }
1457 }
1458 if (bp->b_vp != newvp) {
1459 bp->b_vp = newvp;
1460 vhold(bp->b_vp);
1461 }
1462 splx(s);
1463}
1464
1465/*
1466 * Create a vnode for a block device.
1467 * Used for mounting the root file system.
1468 */
1469int
1470bdevvp(dev, vpp)
1471 dev_t dev;
1472 struct vnode **vpp;
1473{
1474 register struct vnode *vp;
1475 struct vnode *nvp;
1476 int error;
1477
1478 if (dev == NODEV) {
1479 *vpp = NULLVP;
1480 return (ENXIO);
1481 }
1482 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1483 if (error) {
1484 *vpp = NULLVP;
1485 return (error);
1486 }
1487 vp = nvp;
1488 vp->v_type = VBLK;
1489 addalias(vp, dev);
1490 *vpp = vp;
1491 return (0);
1492}
1493
1494/*
1495 * Add vnode to the alias list hung off the dev_t.
1496 *
1497 * The reason for this gunk is that multiple vnodes can reference
1498 * the same physical device, so checking vp->v_usecount to see
1499 * how many users there are is inadequate; the v_usecount for
1500 * the vnodes need to be accumulated. vcount() does that.
1501 */
1502void
1503addaliasu(nvp, nvp_rdev)
1504 struct vnode *nvp;
1505 udev_t nvp_rdev;
1506{
1507
1508 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1509 panic("addaliasu on non-special vnode");
1510 addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
1511}
1512
1513void
1514addalias(nvp, dev)
1515 struct vnode *nvp;
1516 dev_t dev;
1517{
1518
1519 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1520 panic("addalias on non-special vnode");
1521
1522 nvp->v_rdev = dev;
8a8d5d85 1523 lwkt_gettoken(&spechash_token);
984263bc 1524 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
8a8d5d85 1525 lwkt_reltoken(&spechash_token);
984263bc
MD
1526}
1527
1528/*
1529 * Grab a particular vnode from the free list, increment its
1530 * reference count and lock it. The vnode lock bit is set if the
1531 * vnode is being eliminated in vgone. The process is awakened
1532 * when the transition is completed, and an error returned to
1533 * indicate that the vnode is no longer usable (possibly having
1534 * been changed to a new file system type).
1535 */
1536int
dadab5e9
MD
1537vget(vp, flags, td)
1538 struct vnode *vp;
984263bc 1539 int flags;
dadab5e9 1540 struct thread *td;
984263bc
MD
1541{
1542 int error;
1543
1544 /*
1545 * If the vnode is in the process of being cleaned out for
1546 * another use, we wait for the cleaning to finish and then
1547 * return failure. Cleaning is determined by checking that
1548 * the VXLOCK flag is set.
1549 */
1550 if ((flags & LK_INTERLOCK) == 0) {
8a8d5d85 1551 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1552 }
1553 if (vp->v_flag & VXLOCK) {
1554 if (vp->v_vxproc == curproc) {
1555#if 0
1556 /* this can now occur in normal operation */
1557 log(LOG_INFO, "VXLOCK interlock avoided\n");
1558#endif
1559 } else {
1560 vp->v_flag |= VXWANT;
8a8d5d85 1561 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1562 tsleep((caddr_t)vp, PINOD, "vget", 0);
1563 return (ENOENT);
1564 }
1565 }
1566
1567 vp->v_usecount++;
1568
1569 if (VSHOULDBUSY(vp))
1570 vbusy(vp);
1571 if (flags & LK_TYPE_MASK) {
dadab5e9 1572 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
984263bc
MD
1573 /*
1574 * must expand vrele here because we do not want
1575 * to call VOP_INACTIVE if the reference count
1576 * drops back to zero since it was never really
1577 * active. We must remove it from the free list
1578 * before sleeping so that multiple processes do
1579 * not try to recycle it.
1580 */
8a8d5d85 1581 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1582 vp->v_usecount--;
1583 if (VSHOULDFREE(vp))
1584 vfree(vp);
1585 else
1586 vlruvp(vp);
8a8d5d85 1587 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1588 }
1589 return (error);
1590 }
8a8d5d85 1591 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1592 return (0);
1593}
1594
1595void
1596vref(struct vnode *vp)
1597{
8a8d5d85 1598 lwkt_gettoken(&vp->v_interlock);
984263bc 1599 vp->v_usecount++;
8a8d5d85 1600 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1601}
1602
1603/*
1604 * Vnode put/release.
1605 * If count drops to zero, call inactive routine and return to freelist.
1606 */
1607void
dadab5e9 1608vrele(struct vnode *vp)
984263bc 1609{
dadab5e9 1610 struct thread *td = curthread; /* XXX */
984263bc
MD
1611
1612 KASSERT(vp != NULL, ("vrele: null vp"));
1613
8a8d5d85 1614 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1615
1616 if (vp->v_usecount > 1) {
1617
1618 vp->v_usecount--;
8a8d5d85 1619 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1620
1621 return;
1622 }
1623
1624 if (vp->v_usecount == 1) {
1625 vp->v_usecount--;
1626 /*
1627 * We must call VOP_INACTIVE with the node locked.
1628 * If we are doing a vpu, the node is already locked,
1629 * but, in the case of vrele, we must explicitly lock
1630 * the vnode before calling VOP_INACTIVE
1631 */
1632
dadab5e9
MD
1633 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1634 VOP_INACTIVE(vp, td);
984263bc
MD
1635 if (VSHOULDFREE(vp))
1636 vfree(vp);
1637 else
1638 vlruvp(vp);
1639 } else {
1640#ifdef DIAGNOSTIC
1641 vprint("vrele: negative ref count", vp);
8a8d5d85 1642 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1643#endif
1644 panic("vrele: negative ref cnt");
1645 }
1646}
1647
1648void
dadab5e9 1649vput(struct vnode *vp)
984263bc 1650{
dadab5e9 1651 struct thread *td = curthread; /* XXX */
984263bc
MD
1652
1653 KASSERT(vp != NULL, ("vput: null vp"));
1654
8a8d5d85 1655 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1656
1657 if (vp->v_usecount > 1) {
1658 vp->v_usecount--;
dadab5e9 1659 VOP_UNLOCK(vp, LK_INTERLOCK, td);
984263bc
MD
1660 return;
1661 }
1662
1663 if (vp->v_usecount == 1) {
1664 vp->v_usecount--;
1665 /*
1666 * We must call VOP_INACTIVE with the node locked.
1667 * If we are doing a vpu, the node is already locked,
1668 * so we just need to release the vnode mutex.
1669 */
8a8d5d85 1670 lwkt_reltoken(&vp->v_interlock);
dadab5e9 1671 VOP_INACTIVE(vp, td);
984263bc
MD
1672 if (VSHOULDFREE(vp))
1673 vfree(vp);
1674 else
1675 vlruvp(vp);
1676 } else {
1677#ifdef DIAGNOSTIC
1678 vprint("vput: negative ref count", vp);
1679#endif
1680 panic("vput: negative ref cnt");
1681 }
1682}
1683
1684/*
1685 * Somebody doesn't want the vnode recycled.
1686 */
1687void
1688vhold(vp)
1689 register struct vnode *vp;
1690{
1691 int s;
1692
1693 s = splbio();
1694 vp->v_holdcnt++;
1695 if (VSHOULDBUSY(vp))
1696 vbusy(vp);
1697 splx(s);
1698}
1699
1700/*
1701 * One less who cares about this vnode.
1702 */
1703void
1704vdrop(vp)
1705 register struct vnode *vp;
1706{
1707 int s;
1708
1709 s = splbio();
1710 if (vp->v_holdcnt <= 0)
1711 panic("vdrop: holdcnt");
1712 vp->v_holdcnt--;
1713 if (VSHOULDFREE(vp))
1714 vfree(vp);
1715 splx(s);
1716}
1717
1718/*
1719 * Remove any vnodes in the vnode table belonging to mount point mp.
1720 *
1721 * If FORCECLOSE is not specified, there should not be any active ones,
1722 * return error if any are found (nb: this is a user error, not a
1723 * system error). If FORCECLOSE is specified, detach any active vnodes
1724 * that are found.
1725 *
1726 * If WRITECLOSE is set, only flush out regular file vnodes open for
1727 * writing.
1728 *
1729 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1730 *
1731 * `rootrefs' specifies the base reference count for the root vnode
1732 * of this filesystem. The root vnode is considered busy if its
1733 * v_usecount exceeds this value. On a successful return, vflush()
1734 * will call vrele() on the root vnode exactly rootrefs times.
1735 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1736 * be zero.
1737 */
1738#ifdef DIAGNOSTIC
1739static int busyprt = 0; /* print out busy vnodes */
1740SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1741#endif
1742
1743int
1744vflush(mp, rootrefs, flags)
1745 struct mount *mp;
1746 int rootrefs;
1747 int flags;
1748{
dadab5e9 1749 struct thread *td = curthread; /* XXX */
984263bc
MD
1750 struct vnode *vp, *nvp, *rootvp = NULL;
1751 struct vattr vattr;
1752 int busy = 0, error;
1753
1754 if (rootrefs > 0) {
1755 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1756 ("vflush: bad args"));
1757 /*
1758 * Get the filesystem root vnode. We can vput() it
1759 * immediately, since with rootrefs > 0, it won't go away.
1760 */
1761 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1762 return (error);
1763 vput(rootvp);
1764 }
8a8d5d85 1765 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1766loop:
1767 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1768 /*
1769 * Make sure this vnode wasn't reclaimed in getnewvnode().
1770 * Start over if it has (it won't be on the list anymore).
1771 */
1772 if (vp->v_mount != mp)
1773 goto loop;
1774 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1775
8a8d5d85 1776 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1777 /*
1778 * Skip over a vnodes marked VSYSTEM.
1779 */
1780 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
8a8d5d85 1781 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1782 continue;
1783 }
1784 /*
1785 * If WRITECLOSE is set, flush out unlinked but still open
1786 * files (even if open only for reading) and regular file
1787 * vnodes open for writing.
1788 */
1789 if ((flags & WRITECLOSE) &&
1790 (vp->v_type == VNON ||
3b568787 1791 (VOP_GETATTR(vp, &vattr, td) == 0 &&
984263bc
MD
1792 vattr.va_nlink > 0)) &&
1793 (vp->v_writecount == 0 || vp->v_type != VREG)) {
8a8d5d85 1794 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1795 continue;
1796 }
1797
1798 /*
1799 * With v_usecount == 0, all we need to do is clear out the
1800 * vnode data structures and we are done.
1801 */
1802 if (vp->v_usecount == 0) {
8a8d5d85 1803 lwkt_reltoken(&mntvnode_token);
dadab5e9 1804 vgonel(vp, td);
8a8d5d85 1805 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1806 continue;
1807 }
1808
1809 /*
1810 * If FORCECLOSE is set, forcibly close the vnode. For block
1811 * or character devices, revert to an anonymous device. For
1812 * all other files, just kill them.
1813 */
1814 if (flags & FORCECLOSE) {
8a8d5d85 1815 lwkt_reltoken(&mntvnode_token);
984263bc 1816 if (vp->v_type != VBLK && vp->v_type != VCHR) {
dadab5e9 1817 vgonel(vp, td);
984263bc 1818 } else {
dadab5e9 1819 vclean(vp, 0, td);
984263bc
MD
1820 vp->v_op = spec_vnodeop_p;
1821 insmntque(vp, (struct mount *) 0);
1822 }
8a8d5d85 1823 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1824 continue;
1825 }
1826#ifdef DIAGNOSTIC
1827 if (busyprt)
1828 vprint("vflush: busy vnode", vp);
1829#endif
8a8d5d85 1830 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1831 busy++;
1832 }
8a8d5d85 1833 lwkt_reltoken(&mntvnode_token);
984263bc
MD
1834 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1835 /*
1836 * If just the root vnode is busy, and if its refcount
1837 * is equal to `rootrefs', then go ahead and kill it.
1838 */
8a8d5d85 1839 lwkt_gettoken(&rootvp->v_interlock);
984263bc
MD
1840 KASSERT(busy > 0, ("vflush: not busy"));
1841 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1842 if (busy == 1 && rootvp->v_usecount == rootrefs) {
dadab5e9 1843 vgonel(rootvp, td);
984263bc
MD
1844 busy = 0;
1845 } else
8a8d5d85 1846 lwkt_reltoken(&rootvp->v_interlock);
984263bc
MD
1847 }
1848 if (busy)
1849 return (EBUSY);
1850 for (; rootrefs > 0; rootrefs--)
1851 vrele(rootvp);
1852 return (0);
1853}
1854
1855/*
1856 * We do not want to recycle the vnode too quickly.
1857 *
1858 * XXX we can't move vp's around the nvnodelist without really screwing
1859 * up the efficiency of filesystem SYNC and friends. This code is
1860 * disabled until we fix the syncing code's scanning algorithm.
1861 */
1862static void
1863vlruvp(struct vnode *vp)
1864{
1865#if 0
1866 struct mount *mp;
1867
1868 if ((mp = vp->v_mount) != NULL) {
8a8d5d85 1869 lwkt_gettoken(&mntvnode_token);
984263bc
MD
1870 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1871 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
8a8d5d85 1872 lwkt_reltoken(&mntvnode_token);
984263bc
MD
1873 }
1874#endif
1875}
1876
1877/*
1878 * Disassociate the underlying file system from a vnode.
1879 */
1880static void
dadab5e9 1881vclean(struct vnode *vp, int flags, struct thread *td)
984263bc
MD
1882{
1883 int active;
1884
1885 /*
1886 * Check to see if the vnode is in use. If so we have to reference it
1887 * before we clean it out so that its count cannot fall to zero and
1888 * generate a race against ourselves to recycle it.
1889 */
1890 if ((active = vp->v_usecount))
1891 vp->v_usecount++;
1892
1893 /*
1894 * Prevent the vnode from being recycled or brought into use while we
1895 * clean it out.
1896 */
1897 if (vp->v_flag & VXLOCK)
1898 panic("vclean: deadlock");
1899 vp->v_flag |= VXLOCK;
1900 vp->v_vxproc = curproc;
1901 /*
1902 * Even if the count is zero, the VOP_INACTIVE routine may still
1903 * have the object locked while it cleans it out. The VOP_LOCK
1904 * ensures that the VOP_INACTIVE routine is done with its work.
1905 * For active vnodes, it ensures that no other activity can
1906 * occur while the underlying object is being cleaned out.
1907 */
dadab5e9 1908 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
984263bc
MD
1909
1910 /*
1911 * Clean out any buffers associated with the vnode.
1912 */
3b568787 1913 vinvalbuf(vp, V_SAVE, td, 0, 0);
984263bc
MD
1914
1915 VOP_DESTROYVOBJECT(vp);
1916
1917 /*
1918 * If purging an active vnode, it must be closed and
1919 * deactivated before being reclaimed. Note that the
1920 * VOP_INACTIVE will unlock the vnode.
1921 */
1922 if (active) {
1923 if (flags & DOCLOSE)
3b568787 1924 VOP_CLOSE(vp, FNONBLOCK, td);
dadab5e9 1925 VOP_INACTIVE(vp, td);
984263bc
MD
1926 } else {
1927 /*
1928 * Any other processes trying to obtain this lock must first
1929 * wait for VXLOCK to clear, then call the new lock operation.
1930 */
dadab5e9 1931 VOP_UNLOCK(vp, 0, td);
984263bc
MD
1932 }
1933 /*
1934 * Reclaim the vnode.
1935 */
dadab5e9 1936 if (VOP_RECLAIM(vp, td))
984263bc
MD
1937 panic("vclean: cannot reclaim");
1938
1939 if (active) {
1940 /*
1941 * Inline copy of vrele() since VOP_INACTIVE
1942 * has already been called.
1943 */
8a8d5d85 1944 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
1945 if (--vp->v_usecount <= 0) {
1946#ifdef DIAGNOSTIC
1947 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1948 vprint("vclean: bad ref count", vp);
1949 panic("vclean: ref cnt");
1950 }
1951#endif
1952 vfree(vp);
1953 }
8a8d5d85 1954 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
1955 }
1956
1957 cache_purge(vp);
1958 vp->v_vnlock = NULL;
1959
1960 if (VSHOULDFREE(vp))
1961 vfree(vp);
1962
1963 /*
1964 * Done with purge, notify sleepers of the grim news.
1965 */
1966 vp->v_op = dead_vnodeop_p;
1967 vn_pollgone(vp);
1968 vp->v_tag = VT_NON;
1969 vp->v_flag &= ~VXLOCK;
1970 vp->v_vxproc = NULL;
1971 if (vp->v_flag & VXWANT) {
1972 vp->v_flag &= ~VXWANT;
1973 wakeup((caddr_t) vp);
1974 }
1975}
1976
1977/*
1978 * Eliminate all activity associated with the requested vnode
1979 * and with all vnodes aliased to the requested vnode.
1980 */
1981int
1982vop_revoke(ap)
1983 struct vop_revoke_args /* {
1984 struct vnode *a_vp;
1985 int a_flags;
1986 } */ *ap;
1987{
1988 struct vnode *vp, *vq;
1989 dev_t dev;
1990
1991 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1992
1993 vp = ap->a_vp;
1994 /*
1995 * If a vgone (or vclean) is already in progress,
1996 * wait until it is done and return.
1997 */
1998 if (vp->v_flag & VXLOCK) {
1999 vp->v_flag |= VXWANT;
8a8d5d85 2000 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2001 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
2002 return (0);
2003 }
2004 dev = vp->v_rdev;
2005 for (;;) {
8a8d5d85 2006 lwkt_gettoken(&spechash_token);
984263bc 2007 vq = SLIST_FIRST(&dev->si_hlist);
8a8d5d85 2008 lwkt_reltoken(&spechash_token);
984263bc
MD
2009 if (!vq)
2010 break;
2011 vgone(vq);
2012 }
2013 return (0);
2014}
2015
2016/*
2017 * Recycle an unused vnode to the front of the free list.
2018 * Release the passed interlock if the vnode will be recycled.
2019 */
2020int
8a8d5d85 2021vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td)
984263bc 2022{
8a8d5d85 2023 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
2024 if (vp->v_usecount == 0) {
2025 if (inter_lkp) {
8a8d5d85 2026 lwkt_reltoken(inter_lkp);
984263bc 2027 }
dadab5e9 2028 vgonel(vp, td);
984263bc
MD
2029 return (1);
2030 }
8a8d5d85 2031 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2032 return (0);
2033}
2034
2035/*
2036 * Eliminate all activity associated with a vnode
2037 * in preparation for reuse.
2038 */
2039void
dadab5e9 2040vgone(struct vnode *vp)
984263bc 2041{
dadab5e9 2042 struct thread *td = curthread; /* XXX */
984263bc 2043
8a8d5d85 2044 lwkt_gettoken(&vp->v_interlock);
dadab5e9 2045 vgonel(vp, td);
984263bc
MD
2046}
2047
2048/*
2049 * vgone, with the vp interlock held.
2050 */
2051void
dadab5e9 2052vgonel(struct vnode *vp, struct thread *td)
984263bc
MD
2053{
2054 int s;
2055
2056 /*
2057 * If a vgone (or vclean) is already in progress,
2058 * wait until it is done and return.
2059 */
2060 if (vp->v_flag & VXLOCK) {
2061 vp->v_flag |= VXWANT;
8a8d5d85 2062 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2063 tsleep((caddr_t)vp, PINOD, "vgone", 0);
2064 return;
2065 }
2066
2067 /*
2068 * Clean out the filesystem specific data.
2069 */
dadab5e9 2070 vclean(vp, DOCLOSE, td);
8a8d5d85 2071 lwkt_gettoken(&vp->v_interlock);
984263bc
MD
2072
2073 /*
2074 * Delete from old mount point vnode list, if on one.
2075 */
2076 if (vp->v_mount != NULL)
2077 insmntque(vp, (struct mount *)0);
2078 /*
2079 * If special device, remove it from special device alias list
2080 * if it is on one.
2081 */
2082 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
8a8d5d85 2083 lwkt_gettoken(&spechash_token);
984263bc
MD
2084 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2085 freedev(vp->v_rdev);
8a8d5d85 2086 lwkt_reltoken(&spechash_token);
984263bc
MD
2087 vp->v_rdev = NULL;
2088 }
2089
2090 /*
2091 * If it is on the freelist and not already at the head,
2092 * move it to the head of the list. The test of the
2093 * VDOOMED flag and the reference count of zero is because
2094 * it will be removed from the free list by getnewvnode,
2095 * but will not have its reference count incremented until
2096 * after calling vgone. If the reference count were
2097 * incremented first, vgone would (incorrectly) try to
2098 * close the previous instance of the underlying object.
2099 */
2100 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2101 s = splbio();
8a8d5d85 2102 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2103 if (vp->v_flag & VFREE)
2104 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2105 else
2106 freevnodes++;
2107 vp->v_flag |= VFREE;
2108 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
8a8d5d85 2109 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2110 splx(s);
2111 }
2112
2113 vp->v_type = VBAD;
8a8d5d85 2114 lwkt_reltoken(&vp->v_interlock);
984263bc
MD
2115}
2116
2117/*
2118 * Lookup a vnode by device number.
2119 */
2120int
2121vfinddev(dev, type, vpp)
2122 dev_t dev;
2123 enum vtype type;
2124 struct vnode **vpp;
2125{
2126 struct vnode *vp;
2127
8a8d5d85 2128 lwkt_gettoken(&spechash_token);
984263bc
MD
2129 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2130 if (type == vp->v_type) {
2131 *vpp = vp;
8a8d5d85 2132 lwkt_reltoken(&spechash_token);
984263bc
MD
2133 return (1);
2134 }
2135 }
8a8d5d85 2136 lwkt_reltoken(&spechash_token);
984263bc
MD
2137 return (0);
2138}
2139
2140/*
2141 * Calculate the total number of references to a special device.
2142 */
2143int
2144vcount(vp)
2145 struct vnode *vp;
2146{
2147 struct vnode *vq;
2148 int count;
2149
2150 count = 0;
8a8d5d85 2151 lwkt_gettoken(&spechash_token);
984263bc
MD
2152 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2153 count += vq->v_usecount;
8a8d5d85 2154 lwkt_reltoken(&spechash_token);
984263bc
MD
2155 return (count);
2156}
2157
2158/*
2159 * Same as above, but using the dev_t as argument
2160 */
2161
2162int
2163count_dev(dev)
2164 dev_t dev;
2165{
2166 struct vnode *vp;
2167
2168 vp = SLIST_FIRST(&dev->si_hlist);
2169 if (vp == NULL)
2170 return (0);
2171 return(vcount(vp));
2172}
2173
2174/*
2175 * Print out a description of a vnode.
2176 */
2177static char *typename[] =
2178{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2179
2180void
2181vprint(label, vp)
2182 char *label;
2183 struct vnode *vp;
2184{
2185 char buf[96];
2186
2187 if (label != NULL)
2188 printf("%s: %p: ", label, (void *)vp);
2189 else
2190 printf("%p: ", (void *)vp);
2191 printf("type %s, usecount %d, writecount %d, refcount %d,",
2192 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2193 vp->v_holdcnt);
2194 buf[0] = '\0';
2195 if (vp->v_flag & VROOT)
2196 strcat(buf, "|VROOT");
2197 if (vp->v_flag & VTEXT)
2198 strcat(buf, "|VTEXT");
2199 if (vp->v_flag & VSYSTEM)
2200 strcat(buf, "|VSYSTEM");
2201 if (vp->v_flag & VXLOCK)
2202 strcat(buf, "|VXLOCK");
2203 if (vp->v_flag & VXWANT)
2204 strcat(buf, "|VXWANT");
2205 if (vp->v_flag & VBWAIT)
2206 strcat(buf, "|VBWAIT");
2207 if (vp->v_flag & VDOOMED)
2208 strcat(buf, "|VDOOMED");
2209 if (vp->v_flag & VFREE)
2210 strcat(buf, "|VFREE");
2211 if (vp->v_flag & VOBJBUF)
2212 strcat(buf, "|VOBJBUF");
2213 if (buf[0] != '\0')
2214 printf(" flags (%s)", &buf[1]);
2215 if (vp->v_data == NULL) {
2216 printf("\n");
2217 } else {
2218 printf("\n\t");
2219 VOP_PRINT(vp);
2220 }
2221}
2222
2223#ifdef DDB
2224#include <ddb/ddb.h>
2225/*
2226 * List all of the locked vnodes in the system.
2227 * Called when debugging the kernel.
2228 */
2229DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2230{
dadab5e9 2231 struct thread *td = curthread; /* XXX */
984263bc
MD
2232 struct mount *mp, *nmp;
2233 struct vnode *vp;
2234
2235 printf("Locked vnodes\n");
8a8d5d85 2236 lwkt_gettoken(&mountlist_token);
984263bc 2237 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 2238 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
984263bc
MD
2239 nmp = TAILQ_NEXT(mp, mnt_list);
2240 continue;
2241 }
2242 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2243 if (VOP_ISLOCKED(vp, NULL))
2244 vprint((char *)0, vp);
2245 }
8a8d5d85 2246 lwkt_gettoken(&mountlist_token);
984263bc 2247 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 2248 vfs_unbusy(mp, td);
984263bc 2249 }
8a8d5d85 2250 lwkt_reltoken(&mountlist_token);
984263bc
MD
2251}
2252#endif
2253
2254/*
2255 * Top level filesystem related information gathering.
2256 */
2257static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2258
2259static int
2260vfs_sysctl(SYSCTL_HANDLER_ARGS)
2261{
2262 int *name = (int *)arg1 - 1; /* XXX */
2263 u_int namelen = arg2 + 1; /* XXX */
2264 struct vfsconf *vfsp;
2265
2266#if 1 || defined(COMPAT_PRELITE2)
2267 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2268 if (namelen == 1)
2269 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2270#endif
2271
2272#ifdef notyet
2273 /* all sysctl names at this level are at least name and field */
2274 if (namelen < 2)
2275 return (ENOTDIR); /* overloaded */
2276 if (name[0] != VFS_GENERIC) {
2277 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2278 if (vfsp->vfc_typenum == name[0])
2279 break;
2280 if (vfsp == NULL)
2281 return (EOPNOTSUPP);
2282 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2283 oldp, oldlenp, newp, newlen, p));
2284 }
2285#endif
2286 switch (name[1]) {
2287 case VFS_MAXTYPENUM:
2288 if (namelen != 2)
2289 return (ENOTDIR);
2290 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2291 case VFS_CONF:
2292 if (namelen != 3)
2293 return (ENOTDIR); /* overloaded */
2294 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2295 if (vfsp->vfc_typenum == name[2])
2296 break;
2297 if (vfsp == NULL)
2298 return (EOPNOTSUPP);
2299 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2300 }
2301 return (EOPNOTSUPP);
2302}
2303
2304SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2305 "Generic filesystem");
2306
2307#if 1 || defined(COMPAT_PRELITE2)
2308
2309static int
2310sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2311{
2312 int error;
2313 struct vfsconf *vfsp;
2314 struct ovfsconf ovfs;
2315
2316 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2317 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2318 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2319 ovfs.vfc_index = vfsp->vfc_typenum;
2320 ovfs.vfc_refcount = vfsp->vfc_refcount;
2321 ovfs.vfc_flags = vfsp->vfc_flags;
2322 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2323 if (error)
2324 return error;
2325 }
2326 return 0;
2327}
2328
2329#endif /* 1 || COMPAT_PRELITE2 */
2330
2331#if 0
2332#define KINFO_VNODESLOP 10
2333/*
2334 * Dump vnode list (via sysctl).
2335 * Copyout address of vnode followed by vnode.
2336 */
2337/* ARGSUSED */
2338static int
2339sysctl_vnode(SYSCTL_HANDLER_ARGS)
2340{
2341 struct proc *p = curproc; /* XXX */
2342 struct mount *mp, *nmp;
2343 struct vnode *nvp, *vp;
2344 int error;
2345
2346#define VPTRSZ sizeof (struct vnode *)
2347#define VNODESZ sizeof (struct vnode)
2348
2349 req->lock = 0;
2350 if (!req->oldptr) /* Make an estimate */
2351 return (SYSCTL_OUT(req, 0,
2352 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2353
8a8d5d85 2354 lwkt_gettoken(&mountlist_token);
984263bc 2355 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
8a8d5d85 2356 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) {
984263bc
MD
2357 nmp = TAILQ_NEXT(mp, mnt_list);
2358 continue;
2359 }
2360again:
8a8d5d85 2361 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2362 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2363 vp != NULL;
2364 vp = nvp) {
2365 /*
2366 * Check that the vp is still associated with
2367 * this filesystem. RACE: could have been
2368 * recycled onto the same filesystem.
2369 */
2370 if (vp->v_mount != mp) {
8a8d5d85 2371 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2372 goto again;
2373 }
2374 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
8a8d5d85 2375 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2376 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2377 (error = SYSCTL_OUT(req, vp, VNODESZ)))
2378 return (error);
8a8d5d85 2379 lwkt_gettoken(&mntvnode_token);
984263bc 2380 }
8a8d5d85
MD
2381 lwkt_reltoken(&mntvnode_token);
2382 lwkt_gettoken(&mountlist_token);
984263bc
MD
2383 nmp = TAILQ_NEXT(mp, mnt_list);
2384 vfs_unbusy(mp, p);
2385 }
8a8d5d85 2386 lwkt_reltoken(&mountlist_token);
984263bc
MD
2387
2388 return (0);
2389}
2390#endif
2391
2392/*
2393 * XXX
2394 * Exporting the vnode list on large systems causes them to crash.
2395 * Exporting the vnode list on medium systems causes sysctl to coredump.
2396 */
2397#if 0
2398SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2399 0, 0, sysctl_vnode, "S,vnode", "");
2400#endif
2401
2402/*
2403 * Check to see if a filesystem is mounted on a block device.
2404 */
2405int
2406vfs_mountedon(vp)
2407 struct vnode *vp;
2408{
2409
2410 if (vp->v_specmountpoint != NULL)
2411 return (EBUSY);
2412 return (0);
2413}
2414
2415/*
2416 * Unmount all filesystems. The list is traversed in reverse order
2417 * of mounting to avoid dependencies.
2418 */
2419void
2420vfs_unmountall()
2421{
2422 struct mount *mp;
dadab5e9 2423 struct thread *td = curthread;
984263bc
MD
2424 int error;
2425
dadab5e9
MD
2426 if (td->td_proc == NULL)
2427 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2428
984263bc
MD
2429 /*
2430 * Since this only runs when rebooting, it is not interlocked.
2431 */
2432 while(!TAILQ_EMPTY(&mountlist)) {
2433 mp = TAILQ_LAST(&mountlist, mntlist);
dadab5e9 2434 error = dounmount(mp, MNT_FORCE, td);
984263bc
MD
2435 if (error) {
2436 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2437 printf("unmount of %s failed (",
2438 mp->mnt_stat.f_mntonname);
2439 if (error == EBUSY)
2440 printf("BUSY)\n");
2441 else
2442 printf("%d)\n", error);
2443 } else {
2444 /* The unmount has removed mp from the mountlist */
2445 }
2446 }
2447}
2448
2449/*
2450 * Build hash lists of net addresses and hang them off the mount point.
2451 * Called by ufs_mount() to set up the lists of export addresses.
2452 */
2453static int
2454vfs_hang_addrlist(mp, nep, argp)
2455 struct mount *mp;
2456 struct netexport *nep;
2457 struct export_args *argp;
2458{
2459 register struct netcred *np;
2460 register struct radix_node_head *rnh;
2461 register int i;
2462 struct radix_node *rn;
2463 struct sockaddr *saddr, *smask = 0;
2464 struct domain *dom;
2465 int error;
2466
2467 if (argp->ex_addrlen == 0) {
2468 if (mp->mnt_flag & MNT_DEFEXPORTED)
2469 return (EPERM);
2470 np = &nep->ne_defexported;
2471 np->netc_exflags = argp->ex_flags;
2472 np->netc_anon = argp->ex_anon;
2473 np->netc_anon.cr_ref = 1;
2474 mp->mnt_flag |= MNT_DEFEXPORTED;
2475 return (0);
2476 }
2477
2478 if (argp->ex_addrlen > MLEN)
2479 return (EINVAL);
2480
2481 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2482 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2483 bzero((caddr_t) np, i);
2484 saddr = (struct sockaddr *) (np + 1);
2485 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2486 goto out;
2487 if (saddr->sa_len > argp->ex_addrlen)
2488 saddr->sa_len = argp->ex_addrlen;
2489 if (argp->ex_masklen) {
2490 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2491 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2492 if (error)
2493 goto out;
2494 if (smask->sa_len > argp->ex_masklen)
2495 smask->sa_len = argp->ex_masklen;
2496 }
2497 i = saddr->sa_family;
2498 if ((rnh = nep->ne_rtable[i]) == 0) {
2499 /*
2500 * Seems silly to initialize every AF when most are not used,
2501 * do so on demand here
2502 */
2503 for (dom = domains; dom; dom = dom->dom_next)
2504 if (dom->dom_family == i && dom->dom_rtattach) {
2505 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2506 dom->dom_rtoffset);
2507 break;
2508 }
2509 if ((rnh = nep->ne_rtable[i]) == 0) {
2510 error = ENOBUFS;
2511 goto out;
2512 }
2513 }
2514 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2515 np->netc_rnodes);
2516 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2517 error = EPERM;
2518 goto out;
2519 }
2520 np->netc_exflags = argp->ex_flags;
2521 np->netc_anon = argp->ex_anon;
2522 np->netc_anon.cr_ref = 1;
2523 return (0);
2524out:
2525 free(np, M_NETADDR);
2526 return (error);
2527}
2528
2529/* ARGSUSED */
2530static int
2531vfs_free_netcred(rn, w)
2532 struct radix_node *rn;
2533 void *w;
2534{
2535 register struct radix_node_head *rnh = (struct radix_node_head *) w;
2536
2537 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2538 free((caddr_t) rn, M_NETADDR);
2539 return (0);
2540}
2541
2542/*
2543 * Free the net address hash lists that are hanging off the mount points.
2544 */
2545static void
2546vfs_free_addrlist(nep)
2547 struct netexport *nep;
2548{
2549 register int i;
2550 register struct radix_node_head *rnh;
2551
2552 for (i = 0; i <= AF_MAX; i++)
2553 if ((rnh = nep->ne_rtable[i])) {
2554 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2555 (caddr_t) rnh);
2556 free((caddr_t) rnh, M_RTABLE);
2557 nep->ne_rtable[i] = 0;
2558 }
2559}
2560
2561int
2562vfs_export(mp, nep, argp)
2563 struct mount *mp;
2564 struct netexport *nep;
2565 struct export_args *argp;
2566{
2567 int error;
2568
2569 if (argp->ex_flags & MNT_DELEXPORT) {
2570 if (mp->mnt_flag & MNT_EXPUBLIC) {
2571 vfs_setpublicfs(NULL, NULL, NULL);
2572 mp->mnt_flag &= ~MNT_EXPUBLIC;
2573 }
2574 vfs_free_addrlist(nep);
2575 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2576 }
2577 if (argp->ex_flags & MNT_EXPORTED) {
2578 if (argp->ex_flags & MNT_EXPUBLIC) {
2579 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2580 return (error);
2581 mp->mnt_flag |= MNT_EXPUBLIC;
2582 }
2583 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2584 return (error);
2585 mp->mnt_flag |= MNT_EXPORTED;
2586 }
2587 return (0);
2588}
2589
2590
2591/*
2592 * Set the publicly exported filesystem (WebNFS). Currently, only
2593 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2594 */
2595int
2596vfs_setpublicfs(mp, nep, argp)
2597 struct mount *mp;
2598 struct netexport *nep;
2599 struct export_args *argp;
2600{
2601 int error;
2602 struct vnode *rvp;
2603 char *cp;
2604
2605 /*
2606 * mp == NULL -> invalidate the current info, the FS is
2607 * no longer exported. May be called from either vfs_export
2608 * or unmount, so check if it hasn't already been done.
2609 */
2610 if (mp == NULL) {
2611 if (nfs_pub.np_valid) {
2612 nfs_pub.np_valid = 0;
2613 if (nfs_pub.np_index != NULL) {
2614 FREE(nfs_pub.np_index, M_TEMP);
2615 nfs_pub.np_index = NULL;
2616 }
2617 }
2618 return (0);
2619 }
2620
2621 /*
2622 * Only one allowed at a time.
2623 */
2624 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2625 return (EBUSY);
2626
2627 /*
2628 * Get real filehandle for root of exported FS.
2629 */
2630 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2631 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2632
2633 if ((error = VFS_ROOT(mp, &rvp)))
2634 return (error);
2635
2636 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2637 return (error);
2638
2639 vput(rvp);
2640
2641 /*
2642 * If an indexfile was specified, pull it in.
2643 */
2644 if (argp->ex_indexfile != NULL) {
2645 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2646 M_WAITOK);
2647 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2648 MAXNAMLEN, (size_t *)0);
2649 if (!error) {
2650 /*
2651 * Check for illegal filenames.
2652 */
2653 for (cp = nfs_pub.np_index; *cp; cp++) {
2654 if (*cp == '/') {
2655 error = EINVAL;
2656 break;
2657 }
2658 }
2659 }
2660 if (error) {
2661 FREE(nfs_pub.np_index, M_TEMP);
2662 return (error);
2663 }
2664 }
2665
2666 nfs_pub.np_mount = mp;
2667 nfs_pub.np_valid = 1;
2668 return (0);
2669}
2670
2671struct netcred *
2672vfs_export_lookup(mp, nep, nam)
2673 register struct mount *mp;
2674 struct netexport *nep;
2675 struct sockaddr *nam;
2676{
2677 register struct netcred *np;
2678 register struct radix_node_head *rnh;
2679 struct sockaddr *saddr;
2680
2681 np = NULL;
2682 if (mp->mnt_flag & MNT_EXPORTED) {
2683 /*
2684 * Lookup in the export list first.
2685 */
2686 if (nam != NULL) {
2687 saddr = nam;
2688 rnh = nep->ne_rtable[saddr->sa_family];
2689 if (rnh != NULL) {
2690 np = (struct netcred *)
2691 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2692 rnh);
2693 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2694 np = NULL;
2695 }
2696 }
2697 /*
2698 * If no address match, use the default if it exists.
2699 */
2700 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2701 np = &nep->ne_defexported;
2702 }
2703 return (np);
2704}
2705
2706/*
2707 * perform msync on all vnodes under a mount point
2708 * the mount point must be locked.
2709 */
2710void
2711vfs_msync(struct mount *mp, int flags)
2712{
dadab5e9 2713 struct thread *td = curthread; /* XXX */
984263bc
MD
2714 struct vnode *vp, *nvp;
2715 struct vm_object *obj;
2716 int tries;
2717
2718 tries = 5;
8a8d5d85 2719 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2720loop:
2721 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2722 if (vp->v_mount != mp) {
2723 if (--tries > 0)
2724 goto loop;
2725 break;
2726 }
2727 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2728
2729 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */
2730 continue;
2731
2732 /*
2733 * There could be hundreds of thousands of vnodes, we cannot
2734 * afford to do anything heavy-weight until we have a fairly
2735 * good indication that there is something to do.
2736 */
2737 if ((vp->v_flag & VOBJDIRTY) &&
2738 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
8a8d5d85 2739 lwkt_reltoken(&mntvnode_token);
984263bc 2740 if (!vget(vp,
dadab5e9 2741 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
984263bc
MD
2742 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2743 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2744 }
2745 vput(vp);
2746 }
8a8d5d85 2747 lwkt_gettoken(&mntvnode_token);
984263bc
MD
2748 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2749 if (--tries > 0)
2750 goto loop;
2751 break;
2752 }
2753 }
2754 }
8a8d5d85 2755 lwkt_reltoken(&mntvnode_token);
984263bc
MD
2756}
2757
2758/*
2759 * Create the VM object needed for VMIO and mmap support. This
2760 * is done for all VREG files in the system. Some filesystems might
2761 * afford the additional metadata buffering capability of the
2762 * VMIO code by making the device node be VMIO mode also.
2763 *
2764 * vp must be locked when vfs_object_create is called.
2765 */
2766int
3b568787 2767vfs_object_create(struct vnode *vp, struct thread *td)
984263bc 2768{
3b568787 2769 return (VOP_CREATEVOBJECT(vp, td));
984263bc
MD
2770}
2771
2772void
2773vfree(vp)
2774 struct vnode *vp;
2775{
2776 int s;
2777
2778 s = splbio();
8a8d5d85 2779 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2780 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2781 if (vp->v_flag & VAGE) {
2782 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2783 } else {
2784 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2785 }
2786 freevnodes++;
8a8d5d85 2787 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2788 vp->v_flag &= ~VAGE;
2789 vp->v_flag |= VFREE;
2790 splx(s);
2791}
2792
2793void
2794vbusy(vp)
2795 struct vnode *vp;
2796{
2797 int s;
2798
2799 s = splbio();
8a8d5d85 2800 lwkt_gettoken(&vnode_free_list_token);
984263bc
MD
2801 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2802 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2803 freevnodes--;
8a8d5d85 2804 lwkt_reltoken(&vnode_free_list_token);
984263bc
MD
2805 vp->v_flag &= ~(VFREE|VAGE);
2806 splx(s);
2807}
2808
2809/*
2810 * Record a process's interest in events which might happen to
2811 * a vnode. Because poll uses the historic select-style interface
2812 * internally, this routine serves as both the ``check for any
2813 * pending events'' and the ``record my interest in future events''
2814 * functions. (These are done together, while the lock is held,
2815 * to avoid race conditions.)
2816 */
2817int
dadab5e9 2818vn_pollrecord(struct vnode *vp, struct thread *td, int events)
984263bc 2819{
8a8d5d85 2820 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2821 if (vp->v_pollinfo.vpi_revents & events) {
2822 /*
2823 * This leaves events we are not interested
2824 * in available for the other process which
2825 * which presumably had requested them
2826 * (otherwise they would never have been
2827 * recorded).
2828 */
2829 events &= vp->v_pollinfo.vpi_revents;
2830 vp->v_pollinfo.vpi_revents &= ~events;
2831
8a8d5d85 2832 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2833 return events;
2834 }
2835 vp->v_pollinfo.vpi_events |= events;
dadab5e9 2836 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
8a8d5d85 2837 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2838 return 0;
2839}
2840
2841/*
2842 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2843 * it is possible for us to miss an event due to race conditions, but
2844 * that condition is expected to be rare, so for the moment it is the
2845 * preferred interface.
2846 */
2847void
2848vn_pollevent(vp, events)
2849 struct vnode *vp;
2850 short events;
2851{
8a8d5d85 2852 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2853 if (vp->v_pollinfo.vpi_events & events) {
2854 /*
2855 * We clear vpi_events so that we don't
2856 * call selwakeup() twice if two events are
2857 * posted before the polling process(es) is
2858 * awakened. This also ensures that we take at
2859 * most one selwakeup() if the polling process
2860 * is no longer interested. However, it does
2861 * mean that only one event can be noticed at
2862 * a time. (Perhaps we should only clear those
2863 * event bits which we note?) XXX
2864 */
2865 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2866 vp->v_pollinfo.vpi_revents |= events;
2867 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2868 }
8a8d5d85 2869 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2870}
2871
2872/*
2873 * Wake up anyone polling on vp because it is being revoked.
2874 * This depends on dead_poll() returning POLLHUP for correct
2875 * behavior.
2876 */
2877void
2878vn_pollgone(vp)
2879 struct vnode *vp;
2880{
8a8d5d85 2881 lwkt_gettoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2882 if (vp->v_pollinfo.vpi_events) {
2883 vp->v_pollinfo.vpi_events = 0;
2884 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2885 }
8a8d5d85 2886 lwkt_reltoken(&vp->v_pollinfo.vpi_token);
984263bc
MD
2887}
2888
2889
2890
2891/*
2892 * Routine to create and manage a filesystem syncer vnode.
2893 */
2894#define sync_close ((int (*) __P((struct vop_close_args *)))nullop)
2895static int sync_fsync __P((struct vop_fsync_args *));
2896static int sync_inactive __P((struct vop_inactive_args *));
2897static int sync_reclaim __P((struct vop_reclaim_args *));
2898#define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock)
2899#define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock)
2900static int sync_print __P((struct vop_print_args *));
2901#define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2902
2903static vop_t **sync_vnodeop_p;
2904static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2905 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
2906 { &vop_close_desc, (vop_t *) sync_close }, /* close */
2907 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
2908 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
2909 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
2910 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
2911 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
2912 { &vop_print_desc, (vop_t *) sync_print }, /* print */
2913 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
2914 { NULL, NULL }
2915};
2916static struct vnodeopv_desc sync_vnodeop_opv_desc =
2917 { &sync_vnodeop_p, sync_vnodeop_entries };
2918
2919VNODEOP_SET(sync_vnodeop_opv_desc);
2920
2921/*
2922 * Create a new filesystem syncer vnode for the specified mount point.
2923 */
2924int
2925vfs_allocate_syncvnode(mp)
2926 struct mount *mp;
2927{
2928 struct vnode *vp;
2929 static long start, incr, next;
2930 int error;
2931
2932 /* Allocate a new vnode */
2933 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2934 mp->mnt_syncer = NULL;
2935 return (error);
2936 }
2937 vp->v_type = VNON;
2938 /*
2939 * Place the vnode onto the syncer worklist. We attempt to
2940 * scatter them about on the list so that they will go off
2941 * at evenly distributed times even if all the filesystems
2942 * are mounted at once.
2943 */
2944 next += incr;
2945 if (next == 0 || next > syncer_maxdelay) {
2946 start /= 2;
2947 incr /= 2;
2948 if (start == 0) {
2949 start = syncer_maxdelay / 2;
2950 incr = syncer_maxdelay;
2951 }
2952 next = start;
2953 }
2954 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2955 mp->mnt_syncer = vp;
2956 return (0);
2957}
2958
2959/*
2960 * Do a lazy sync of the filesystem.
2961 */
2962static int
2963sync_fsync(ap)
2964 struct vop_fsync_args /* {
2965 struct vnode *a_vp;
2966 struct ucred *a_cred;
2967 int a_waitfor;
dadab5e9 2968 struct thread *a_td;
984263bc
MD
2969 } */ *ap;
2970{
2971 struct vnode *syncvp = ap->a_vp;
2972 struct mount *mp = syncvp->v_mount;
dadab5e9 2973 struct thread *td = ap->a_td;
984263bc
MD
2974 int asyncflag;
2975
2976 /*
2977 * We only need to do something if this is a lazy evaluation.
2978 */
2979 if (ap->a_waitfor != MNT_LAZY)
2980 return (0);
2981
2982 /*
2983 * Move ourselves to the back of the sync list.
2984 */
2985 vn_syncer_add_to_worklist(syncvp, syncdelay);
2986
2987 /*
2988 * Walk the list of vnodes pushing all that are dirty and
2989 * not already on the sync list.
2990 */
8a8d5d85
MD
2991 lwkt_gettoken(&mountlist_token);
2992 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) {
2993 lwkt_reltoken(&mountlist_token);
984263bc
MD
2994 return (0);
2995 }
2996 asyncflag = mp->mnt_flag & MNT_ASYNC;
2997 mp->mnt_flag &= ~MNT_ASYNC;
2998 vfs_msync(mp, MNT_NOWAIT);
3b568787 2999 VFS_SYNC(mp, MNT_LAZY, td);
984263bc
MD
3000 if (asyncflag)
3001 mp->mnt_flag |= MNT_ASYNC;
dadab5e9 3002 vfs_unbusy(mp, td);
984263bc
MD
3003 return (0);
3004}
3005
3006/*
3007 * The syncer vnode is no referenced.
3008 */
3009static int
3010sync_inactive(ap)
3011 struct vop_inactive_args /* {
3012 struct vnode *a_vp;
3013 struct proc *a_p;
3014 } */ *ap;
3015{
3016
3017 vgone(ap->a_vp);
3018 return (0);
3019}
3020
3021/*
3022 * The syncer vnode is no longer needed and is being decommissioned.
3023 *
3024 * Modifications to the worklist must be protected at splbio().
3025 */
3026static int
3027sync_reclaim(ap)
3028 struct vop_reclaim_args /* {
3029 struct vnode *a_vp;
3030 } */ *ap;
3031{
3032 struct vnode *vp = ap->a_vp;
3033 int s;
3034
3035 s = splbio();
3036 vp->v_mount->mnt_syncer = NULL;
3037 if (vp->v_flag & VONWORKLST) {
3038 LIST_REMOVE(vp, v_synclist);
3039 vp->v_flag &= ~VONWORKLST;
3040 }
3041 splx(s);
3042
3043 return (0);
3044}
3045
3046/*
3047 * Print out a syncer vnode.
3048 */
3049static int
3050sync_print(ap)
3051 struct vop_print_args /* {
3052 struct vnode *a_vp;
3053 } */ *ap;
3054{
3055 struct vnode *vp = ap->a_vp;
3056
3057 printf("syncer vnode");
3058 if (vp->v_vnlock != NULL)
3059 lockmgr_printinfo(vp->v_vnlock);
3060 printf("\n");
3061 return (0);
3062}
3063
3064/*
3065 * extract the dev_t from a VBLK or VCHR
3066 */
3067dev_t
3068vn_todev(vp)
3069 struct vnode *vp;
3070{
3071 if (vp->v_type != VBLK && vp->v_type != VCHR)
3072 return (NODEV);
3073 return (vp->v_rdev);
3074}
3075
3076/*
3077 * Check if vnode represents a disk device
3078 */
3079int
3080vn_isdisk(vp, errp)
3081 struct vnode *vp;
3082 int *errp;
3083{
3084 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3085 if (errp != NULL)
3086 *errp = ENOTBLK;
3087 return (0);
3088 }
3089 if (vp->v_rdev == NULL) {
3090 if (errp != NULL)
3091 *errp = ENXIO;
3092 return (0);
3093 }
3094 if (!devsw(vp->v_rdev)) {
3095 if (errp != NULL)
3096 *errp = ENXIO;
3097 return (0);
3098 }
3099 if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) {
3100 if (errp != NULL)
3101 *errp = ENOTBLK;
3102 return (0);
3103 }
3104 if (errp != NULL)
3105 *errp = 0;
3106 return (1);
3107}
3108
3109void
3110NDFREE(ndp, flags)
3111 struct nameidata *ndp;
3112 const uint flags;
3113{
3114 if (!(flags & NDF_NO_FREE_PNBUF) &&
3115 (ndp->ni_cnd.cn_flags & HASBUF)) {
3116 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3117 ndp->ni_cnd.cn_flags &= ~HASBUF;
3118 }
3119 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3120 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3121 ndp->ni_dvp != ndp->ni_vp)
dadab5e9 3122 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
984263bc
MD
3123 if (!(flags & NDF_NO_DVP_RELE) &&
3124 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3125 vrele(ndp->ni_dvp);
3126 ndp->ni_dvp = NULL;
3127 }
3128 if (!(flags & NDF_NO_VP_UNLOCK) &&
3129 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
dadab5e9 3130 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
984263bc
MD
3131 if (!(flags & NDF_NO_VP_RELE) &&
3132 ndp->ni_vp) {
3133 vrele(ndp->ni_vp);
3134 ndp->ni_vp = NULL;
3135 }
3136 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3137 (ndp->ni_cnd.cn_flags & SAVESTART)) {
3138 vrele(ndp->ni_startdir);
3139 ndp->ni_startdir = NULL;
3140 }
3141}