Mark it as ALTQ ready too.
[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 $
75ffff0d 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.51 2005/02/02 21:34:18 joerg 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
MD
61#include <sys/proc.h>
62#include <sys/reboot.h>
63#include <sys/socket.h>
64#include <sys/stat.h>
65#include <sys/sysctl.h>
66#include <sys/syslog.h>
67#include <sys/vmmeter.h>
68#include <sys/vnode.h>
69
70#include <machine/limits.h>
71
72#include <vm/vm.h>
73#include <vm/vm_object.h>
74#include <vm/vm_extern.h>
6ef943a3 75#include <vm/vm_kern.h>
984263bc
MD
76#include <vm/pmap.h>
77#include <vm/vm_map.h>
78#include <vm/vm_page.h>
79#include <vm/vm_pager.h>
80#include <vm/vnode_pager.h>
81#include <vm/vm_zone.h>
82
3020e3be 83#include <sys/buf2.h>
f5865223 84#include <sys/thread2.h>
3020e3be 85
984263bc
MD
86static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
87
5fd012e0 88int numvnodes;
984263bc 89SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
36dbf6bd 90int vfs_fastdev = 1;
fad57d0e 91SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
984263bc
MD
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
984263bc 102static int reassignbufcalls;
dd98570a
MD
103SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
104 &reassignbufcalls, 0, "");
984263bc 105static int reassignbufloops;
dd98570a
MD
106SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
107 &reassignbufloops, 0, "");
984263bc 108static int reassignbufsortgood;
dd98570a
MD
109SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
110 &reassignbufsortgood, 0, "");
984263bc 111static int reassignbufsortbad;
dd98570a
MD
112SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
113 &reassignbufsortbad, 0, "");
984263bc 114static int reassignbufmethod = 1;
dd98570a
MD
115SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
116 &reassignbufmethod, 0, "");
984263bc 117
984263bc 118int nfs_mount_type = -1;
8a8d5d85 119static struct lwkt_token spechash_token;
984263bc 120struct nfs_public nfs_pub; /* publicly exported FS */
984263bc
MD
121
122int desiredvnodes;
123SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
dd98570a 124 &desiredvnodes, 0, "Maximum number of vnodes");
984263bc 125
402ed7e1
RG
126static void vfs_free_addrlist (struct netexport *nep);
127static int vfs_free_netcred (struct radix_node *rn, void *w);
128static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
129 struct export_args *argp);
984263bc 130
e4c9c0c8 131extern int dev_ref_debug;
0961aa92 132extern struct vnodeopv_entry_desc spec_vnodeop_entries[];
e4c9c0c8 133
41a01a4d 134/*
5fd012e0
MD
135 * Return 0 if the vnode is already on the free list or cannot be placed
136 * on the free list. Return 1 if the vnode can be placed on the free list.
41a01a4d 137 */
5fd012e0
MD
138static __inline int
139vshouldfree(struct vnode *vp, int usecount)
41a01a4d 140{
5fd012e0
MD
141 if (vp->v_flag & VFREE)
142 return (0); /* already free */
143 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount)
144 return (0); /* other holderse */
145 if (vp->v_object &&
146 (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
147 return (0);
148 }
149 return (1);
41a01a4d 150}
5fd012e0 151
984263bc 152/*
5fd012e0
MD
153 * Initialize the vnode management data structures.
154 *
155 * Called from vfsinit()
984263bc
MD
156 */
157void
5fd012e0 158vfs_subr_init(void)
984263bc 159{
6ef943a3
MD
160 /*
161 * Desired vnodes is a result of the physical page count
162 * and the size of kernel's heap. It scales in proportion
163 * to the amount of available physical memory. This can
164 * cause trouble on 64-bit and large memory platforms.
165 */
166 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
167 desiredvnodes =
168 min(maxproc + vmstats.v_page_count /4,
169 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
170 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
171
41a01a4d 172 lwkt_token_init(&spechash_token);
984263bc
MD
173}
174
175/*
176 * Knob to control the precision of file timestamps:
177 *
178 * 0 = seconds only; nanoseconds zeroed.
179 * 1 = seconds and nanoseconds, accurate within 1/HZ.
180 * 2 = seconds and nanoseconds, truncated to microseconds.
181 * >=3 = seconds and nanoseconds, maximum precision.
182 */
183enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
184
185static int timestamp_precision = TSP_SEC;
186SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
dd98570a 187 &timestamp_precision, 0, "");
984263bc
MD
188
189/*
190 * Get a current timestamp.
191 */
192void
dd98570a 193vfs_timestamp(struct timespec *tsp)
984263bc
MD
194{
195 struct timeval tv;
196
197 switch (timestamp_precision) {
198 case TSP_SEC:
199 tsp->tv_sec = time_second;
200 tsp->tv_nsec = 0;
201 break;
202 case TSP_HZ:
203 getnanotime(tsp);
204 break;
205 case TSP_USEC:
206 microtime(&tv);
207 TIMEVAL_TO_TIMESPEC(&tv, tsp);
208 break;
209 case TSP_NSEC:
210 default:
211 nanotime(tsp);
212 break;
213 }
214}
215
216/*
217 * Set vnode attributes to VNOVAL
218 */
219void
dd98570a 220vattr_null(struct vattr *vap)
984263bc 221{
984263bc
MD
222 vap->va_type = VNON;
223 vap->va_size = VNOVAL;
224 vap->va_bytes = VNOVAL;
225 vap->va_mode = VNOVAL;
226 vap->va_nlink = VNOVAL;
227 vap->va_uid = VNOVAL;
228 vap->va_gid = VNOVAL;
229 vap->va_fsid = VNOVAL;
230 vap->va_fileid = VNOVAL;
231 vap->va_blocksize = VNOVAL;
232 vap->va_rdev = VNOVAL;
233 vap->va_atime.tv_sec = VNOVAL;
234 vap->va_atime.tv_nsec = VNOVAL;
235 vap->va_mtime.tv_sec = VNOVAL;
236 vap->va_mtime.tv_nsec = VNOVAL;
237 vap->va_ctime.tv_sec = VNOVAL;
238 vap->va_ctime.tv_nsec = VNOVAL;
239 vap->va_flags = VNOVAL;
240 vap->va_gen = VNOVAL;
241 vap->va_vaflags = 0;
242}
243
984263bc
MD
244/*
245 * Update outstanding I/O count and do wakeup if requested.
246 */
247void
dd98570a 248vwakeup(struct buf *bp)
984263bc 249{
1fd87d54 250 struct vnode *vp;
984263bc
MD
251
252 bp->b_flags &= ~B_WRITEINPROG;
253 if ((vp = bp->b_vp)) {
254 vp->v_numoutput--;
255 if (vp->v_numoutput < 0)
256 panic("vwakeup: neg numoutput");
257 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
258 vp->v_flag &= ~VBWAIT;
259 wakeup((caddr_t) &vp->v_numoutput);
260 }
261 }
262}
263
264/*
265 * Flush out and invalidate all buffers associated with a vnode.
5fd012e0
MD
266 *
267 * vp must be locked.
984263bc
MD
268 */
269int
3b568787
MD
270vinvalbuf(struct vnode *vp, int flags, struct thread *td,
271 int slpflag, int slptimeo)
984263bc 272{
1fd87d54 273 struct buf *bp;
984263bc
MD
274 struct buf *nbp, *blist;
275 int s, error;
276 vm_object_t object;
277
278 if (flags & V_SAVE) {
279 s = splbio();
280 while (vp->v_numoutput) {
281 vp->v_flag |= VBWAIT;
282 error = tsleep((caddr_t)&vp->v_numoutput,
377d4740 283 slpflag, "vinvlbuf", slptimeo);
984263bc
MD
284 if (error) {
285 splx(s);
286 return (error);
287 }
288 }
289 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
290 splx(s);
3b568787 291 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
984263bc
MD
292 return (error);
293 s = splbio();
294 if (vp->v_numoutput > 0 ||
295 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
296 panic("vinvalbuf: dirty bufs");
297 }
298 splx(s);
299 }
300 s = splbio();
301 for (;;) {
302 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
303 if (!blist)
304 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
305 if (!blist)
306 break;
307
308 for (bp = blist; bp; bp = nbp) {
309 nbp = TAILQ_NEXT(bp, b_vnbufs);
310 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
311 error = BUF_TIMELOCK(bp,
312 LK_EXCLUSIVE | LK_SLEEPFAIL,
313 "vinvalbuf", slpflag, slptimeo);
314 if (error == ENOLCK)
315 break;
316 splx(s);
317 return (error);
318 }
319 /*
320 * XXX Since there are no node locks for NFS, I
321 * believe there is a slight chance that a delayed
322 * write will occur while sleeping just above, so
323 * check for it. Note that vfs_bio_awrite expects
324 * buffers to reside on a queue, while VOP_BWRITE and
325 * brelse do not.
326 */
327 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
328 (flags & V_SAVE)) {
329
330 if (bp->b_vp == vp) {
331 if (bp->b_flags & B_CLUSTEROK) {
332 BUF_UNLOCK(bp);
333 vfs_bio_awrite(bp);
334 } else {
335 bremfree(bp);
336 bp->b_flags |= B_ASYNC;
337 VOP_BWRITE(bp->b_vp, bp);
338 }
339 } else {
340 bremfree(bp);
341 (void) VOP_BWRITE(bp->b_vp, bp);
342 }
343 break;
344 }
345 bremfree(bp);
346 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
347 bp->b_flags &= ~B_ASYNC;
348 brelse(bp);
349 }
350 }
351
352 /*
353 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
354 * have write I/O in-progress but if there is a VM object then the
355 * VM object can also have read-I/O in-progress.
356 */
357 do {
358 while (vp->v_numoutput > 0) {
359 vp->v_flag |= VBWAIT;
377d4740 360 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
984263bc
MD
361 }
362 if (VOP_GETVOBJECT(vp, &object) == 0) {
363 while (object->paging_in_progress)
364 vm_object_pip_sleep(object, "vnvlbx");
365 }
366 } while (vp->v_numoutput > 0);
367
368 splx(s);
369
370 /*
371 * Destroy the copy in the VM cache, too.
372 */
984263bc
MD
373 if (VOP_GETVOBJECT(vp, &object) == 0) {
374 vm_object_page_remove(object, 0, 0,
375 (flags & V_SAVE) ? TRUE : FALSE);
376 }
984263bc
MD
377
378 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
379 panic("vinvalbuf: flush failed");
380 return (0);
381}
382
383/*
384 * Truncate a file's buffer and pages to a specified length. This
385 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
386 * sync activity.
5fd012e0
MD
387 *
388 * The vnode must be locked.
984263bc
MD
389 */
390int
3b568787 391vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
984263bc 392{
dadab5e9 393 struct buf *bp;
984263bc
MD
394 struct buf *nbp;
395 int s, anyfreed;
396 int trunclbn;
397
398 /*
399 * Round up to the *next* lbn.
400 */
401 trunclbn = (length + blksize - 1) / blksize;
402
403 s = splbio();
404restart:
405 anyfreed = 1;
406 for (;anyfreed;) {
407 anyfreed = 0;
408 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
409 nbp = TAILQ_NEXT(bp, b_vnbufs);
410 if (bp->b_lblkno >= trunclbn) {
411 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
412 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
413 goto restart;
414 } else {
415 bremfree(bp);
416 bp->b_flags |= (B_INVAL | B_RELBUF);
417 bp->b_flags &= ~B_ASYNC;
418 brelse(bp);
419 anyfreed = 1;
420 }
421 if (nbp &&
422 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
423 (nbp->b_vp != vp) ||
424 (nbp->b_flags & B_DELWRI))) {
425 goto restart;
426 }
427 }
428 }
429
430 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
431 nbp = TAILQ_NEXT(bp, b_vnbufs);
432 if (bp->b_lblkno >= trunclbn) {
433 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
434 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
435 goto restart;
436 } else {
437 bremfree(bp);
438 bp->b_flags |= (B_INVAL | B_RELBUF);
439 bp->b_flags &= ~B_ASYNC;
440 brelse(bp);
441 anyfreed = 1;
442 }
443 if (nbp &&
444 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
445 (nbp->b_vp != vp) ||
446 (nbp->b_flags & B_DELWRI) == 0)) {
447 goto restart;
448 }
449 }
450 }
451 }
452
453 if (length > 0) {
454restartsync:
455 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
456 nbp = TAILQ_NEXT(bp, b_vnbufs);
457 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
458 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
459 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
460 goto restart;
461 } else {
462 bremfree(bp);
463 if (bp->b_vp == vp) {
464 bp->b_flags |= B_ASYNC;
465 } else {
466 bp->b_flags &= ~B_ASYNC;
467 }
468 VOP_BWRITE(bp->b_vp, bp);
469 }
470 goto restartsync;
471 }
472
473 }
474 }
475
476 while (vp->v_numoutput > 0) {
477 vp->v_flag |= VBWAIT;
377d4740 478 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
984263bc
MD
479 }
480
481 splx(s);
482
483 vnode_pager_setsize(vp, length);
484
485 return (0);
486}
487
488/*
489 * Associate a buffer with a vnode.
490 */
491void
dd98570a 492bgetvp(struct vnode *vp, struct buf *bp)
984263bc 493{
984263bc
MD
494 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
495
496 vhold(vp);
497 bp->b_vp = vp;
498 bp->b_dev = vn_todev(vp);
499 /*
500 * Insert onto list for new vnode.
501 */
5fd012e0 502 crit_enter();
984263bc
MD
503 bp->b_xflags |= BX_VNCLEAN;
504 bp->b_xflags &= ~BX_VNDIRTY;
505 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
5fd012e0 506 crit_exit();
984263bc
MD
507}
508
509/*
510 * Disassociate a buffer from a vnode.
511 */
512void
dd98570a 513brelvp(struct buf *bp)
984263bc
MD
514{
515 struct vnode *vp;
516 struct buflists *listheadp;
984263bc
MD
517
518 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
519
520 /*
521 * Delete from old vnode list, if on one.
522 */
523 vp = bp->b_vp;
5fd012e0 524 crit_enter();
984263bc
MD
525 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
526 if (bp->b_xflags & BX_VNDIRTY)
527 listheadp = &vp->v_dirtyblkhd;
528 else
529 listheadp = &vp->v_cleanblkhd;
530 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
531 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
532 }
533 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
534 vp->v_flag &= ~VONWORKLST;
535 LIST_REMOVE(vp, v_synclist);
536 }
f5865223 537 crit_exit();
5fd012e0
MD
538 bp->b_vp = NULL;
539 vdrop(vp);
984263bc
MD
540}
541
542/*
543 * Associate a p-buffer with a vnode.
544 *
545 * Also sets B_PAGING flag to indicate that vnode is not fully associated
546 * with the buffer. i.e. the bp has not been linked into the vnode or
547 * ref-counted.
548 */
549void
dd98570a 550pbgetvp(struct vnode *vp, struct buf *bp)
984263bc 551{
984263bc
MD
552 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
553
554 bp->b_vp = vp;
555 bp->b_flags |= B_PAGING;
556 bp->b_dev = vn_todev(vp);
557}
558
559/*
560 * Disassociate a p-buffer from a vnode.
561 */
562void
dd98570a 563pbrelvp(struct buf *bp)
984263bc 564{
984263bc
MD
565 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
566
567 /* XXX REMOVE ME */
568 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
569 panic(
570 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
571 bp,
572 (int)bp->b_flags
573 );
574 }
575 bp->b_vp = (struct vnode *) 0;
576 bp->b_flags &= ~B_PAGING;
577}
578
579void
dd98570a 580pbreassignbuf(struct buf *bp, struct vnode *newvp)
984263bc
MD
581{
582 if ((bp->b_flags & B_PAGING) == 0) {
583 panic(
584 "pbreassignbuf() on non phys bp %p",
585 bp
586 );
587 }
588 bp->b_vp = newvp;
589}
590
591/*
592 * Reassign a buffer from one vnode to another.
593 * Used to assign file specific control information
594 * (indirect blocks) to the vnode to which they belong.
595 */
596void
dd98570a 597reassignbuf(struct buf *bp, struct vnode *newvp)
984263bc
MD
598{
599 struct buflists *listheadp;
600 int delay;
984263bc
MD
601
602 if (newvp == NULL) {
603 printf("reassignbuf: NULL");
604 return;
605 }
606 ++reassignbufcalls;
607
608 /*
609 * B_PAGING flagged buffers cannot be reassigned because their vp
610 * is not fully linked in.
611 */
612 if (bp->b_flags & B_PAGING)
613 panic("cannot reassign paging buffer");
614
5fd012e0 615 crit_enter();
984263bc
MD
616 /*
617 * Delete from old vnode list, if on one.
618 */
619 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
620 if (bp->b_xflags & BX_VNDIRTY)
621 listheadp = &bp->b_vp->v_dirtyblkhd;
622 else
623 listheadp = &bp->b_vp->v_cleanblkhd;
624 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
625 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
626 if (bp->b_vp != newvp) {
627 vdrop(bp->b_vp);
628 bp->b_vp = NULL; /* for clarification */
629 }
630 }
631 /*
632 * If dirty, put on list of dirty buffers; otherwise insert onto list
633 * of clean buffers.
634 */
635 if (bp->b_flags & B_DELWRI) {
636 struct buf *tbp;
637
638 listheadp = &newvp->v_dirtyblkhd;
639 if ((newvp->v_flag & VONWORKLST) == 0) {
640 switch (newvp->v_type) {
641 case VDIR:
642 delay = dirdelay;
643 break;
644 case VCHR:
645 case VBLK:
e4c9c0c8
MD
646 if (newvp->v_rdev &&
647 newvp->v_rdev->si_mountpoint != NULL) {
984263bc
MD
648 delay = metadelay;
649 break;
650 }
651 /* fall through */
652 default:
653 delay = filedelay;
654 }
655 vn_syncer_add_to_worklist(newvp, delay);
656 }
657 bp->b_xflags |= BX_VNDIRTY;
658 tbp = TAILQ_FIRST(listheadp);
659 if (tbp == NULL ||
660 bp->b_lblkno == 0 ||
661 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
662 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
663 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
664 ++reassignbufsortgood;
665 } else if (bp->b_lblkno < 0) {
666 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
667 ++reassignbufsortgood;
668 } else if (reassignbufmethod == 1) {
669 /*
670 * New sorting algorithm, only handle sequential case,
671 * otherwise append to end (but before metadata)
672 */
673 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
674 (tbp->b_xflags & BX_VNDIRTY)) {
675 /*
676 * Found the best place to insert the buffer
677 */
678 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
679 ++reassignbufsortgood;
680 } else {
681 /*
682 * Missed, append to end, but before meta-data.
683 * We know that the head buffer in the list is
684 * not meta-data due to prior conditionals.
685 *
686 * Indirect effects: NFS second stage write
687 * tends to wind up here, giving maximum
688 * distance between the unstable write and the
689 * commit rpc.
690 */
691 tbp = TAILQ_LAST(listheadp, buflists);
692 while (tbp && tbp->b_lblkno < 0)
693 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
694 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
695 ++reassignbufsortbad;
696 }
697 } else {
698 /*
699 * Old sorting algorithm, scan queue and insert
700 */
701 struct buf *ttbp;
702 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
703 (ttbp->b_lblkno < bp->b_lblkno)) {
704 ++reassignbufloops;
705 tbp = ttbp;
706 }
707 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
708 }
709 } else {
710 bp->b_xflags |= BX_VNCLEAN;
711 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
712 if ((newvp->v_flag & VONWORKLST) &&
713 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
714 newvp->v_flag &= ~VONWORKLST;
715 LIST_REMOVE(newvp, v_synclist);
716 }
717 }
718 if (bp->b_vp != newvp) {
719 bp->b_vp = newvp;
720 vhold(bp->b_vp);
721 }
5fd012e0 722 crit_exit();
984263bc
MD
723}
724
725/*
726 * Create a vnode for a block device.
727 * Used for mounting the root file system.
728 */
729int
e4c9c0c8 730bdevvp(dev_t dev, struct vnode **vpp)
984263bc 731{
1fd87d54 732 struct vnode *vp;
984263bc
MD
733 struct vnode *nvp;
734 int error;
735
736 if (dev == NODEV) {
737 *vpp = NULLVP;
738 return (ENXIO);
739 }
6ddb7618 740 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0);
984263bc
MD
741 if (error) {
742 *vpp = NULLVP;
743 return (error);
744 }
745 vp = nvp;
e4c9c0c8
MD
746 vp->v_type = VCHR;
747 vp->v_udev = dev->si_udev;
5fd012e0 748 vx_unlock(vp);
984263bc
MD
749 *vpp = vp;
750 return (0);
5fd012e0 751}
41a01a4d 752
984263bc 753int
5fd012e0 754v_associate_rdev(struct vnode *vp, dev_t dev)
984263bc 755{
5fd012e0 756 lwkt_tokref ilock;
984263bc 757
5fd012e0
MD
758 if (dev == NULL || dev == NODEV)
759 return(ENXIO);
760 if (dev_is_good(dev) == 0)
761 return(ENXIO);
762 KKASSERT(vp->v_rdev == NULL);
763 if (dev_ref_debug)
764 printf("Z1");
765 vp->v_rdev = reference_dev(dev);
766 lwkt_gettoken(&ilock, &spechash_token);
767 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
768 lwkt_reltoken(&ilock);
769 return(0);
770}
984263bc 771
5fd012e0
MD
772void
773v_release_rdev(struct vnode *vp)
774{
775 lwkt_tokref ilock;
776 dev_t dev;
984263bc 777
5fd012e0
MD
778 if ((dev = vp->v_rdev) != NULL) {
779 lwkt_gettoken(&ilock, &spechash_token);
780 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
781 if (dev_ref_debug && vp->v_opencount != 0) {
782 printf("releasing rdev with non-0 "
783 "v_opencount(%d) (revoked?)\n",
784 vp->v_opencount);
41a01a4d 785 }
5fd012e0
MD
786 vp->v_rdev = NULL;
787 vp->v_opencount = 0;
788 release_dev(dev);
789 lwkt_reltoken(&ilock);
984263bc 790 }
984263bc
MD
791}
792
793/*
5fd012e0
MD
794 * Add a vnode to the alias list hung off the dev_t. We only associate
795 * the device number with the vnode. The actual device is not associated
796 * until the vnode is opened (usually in spec_open()), and will be
797 * disassociated on last close.
984263bc 798 */
5fd012e0
MD
799void
800addaliasu(struct vnode *nvp, udev_t nvp_udev)
984263bc 801{
5fd012e0
MD
802 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
803 panic("addaliasu on non-special vnode");
804 nvp->v_udev = nvp_udev;
984263bc
MD
805}
806
807/*
5fd012e0
MD
808 * Disassociate a vnode from its underlying filesystem.
809 *
810 * The vnode must be VX locked and refd
811 *
812 * If there are v_usecount references to the vnode other then ours we have
813 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
984263bc 814 */
5fd012e0
MD
815void
816vclean(struct vnode *vp, int flags, struct thread *td)
984263bc
MD
817{
818 int active;
819
820 /*
5fd012e0 821 * If the vnode has already been reclaimed we have nothing to do.
984263bc 822 */
5fd012e0
MD
823 if (vp->v_flag & VRECLAIMED)
824 return;
825 vp->v_flag |= VRECLAIMED;
984263bc
MD
826
827 /*
5fd012e0 828 * Scrap the vfs cache
984263bc 829 */
e09206ba 830 cache_inval_vp(vp, 0);
41a01a4d 831
984263bc 832 /*
5fd012e0
MD
833 * Check to see if the vnode is in use. If so we have to reference it
834 * before we clean it out so that its count cannot fall to zero and
835 * generate a race against ourselves to recycle it.
984263bc 836 */
5fd012e0 837 active = (vp->v_usecount > 1);
984263bc
MD
838
839 /*
5fd012e0
MD
840 * Clean out any buffers associated with the vnode and destroy its
841 * object, if it has one.
984263bc 842 */
3b568787 843 vinvalbuf(vp, V_SAVE, td, 0, 0);
984263bc
MD
844 VOP_DESTROYVOBJECT(vp);
845
846 /*
847 * If purging an active vnode, it must be closed and
5fd012e0
MD
848 * deactivated before being reclaimed. XXX
849 *
850 * Note that neither of these routines unlocks the vnode.
984263bc
MD
851 */
852 if (active) {
853 if (flags & DOCLOSE)
3b568787 854 VOP_CLOSE(vp, FNONBLOCK, td);
5fd012e0
MD
855 }
856
857 /*
858 * If the vnode has not be deactivated, deactivated it.
859 */
860 if ((vp->v_flag & VINACTIVE) == 0) {
861 vp->v_flag |= VINACTIVE;
dadab5e9 862 VOP_INACTIVE(vp, td);
984263bc 863 }
5fd012e0 864
984263bc
MD
865 /*
866 * Reclaim the vnode.
867 */
dadab5e9 868 if (VOP_RECLAIM(vp, td))
984263bc
MD
869 panic("vclean: cannot reclaim");
870
984263bc
MD
871 /*
872 * Done with purge, notify sleepers of the grim news.
873 */
6ddb7618 874 vp->v_ops = &dead_vnode_vops;
984263bc
MD
875 vn_pollgone(vp);
876 vp->v_tag = VT_NON;
984263bc
MD
877}
878
879/*
880 * Eliminate all activity associated with the requested vnode
881 * and with all vnodes aliased to the requested vnode.
dd98570a 882 *
5fd012e0
MD
883 * The vnode must be referenced and vx_lock()'d
884 *
dd98570a 885 * revoke { struct vnode *a_vp, int a_flags }
984263bc
MD
886 */
887int
2d3e977e 888vop_stdrevoke(struct vop_revoke_args *ap)
984263bc
MD
889{
890 struct vnode *vp, *vq;
41a01a4d 891 lwkt_tokref ilock;
984263bc
MD
892 dev_t dev;
893
894 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
895
896 vp = ap->a_vp;
5fd012e0 897
984263bc 898 /*
5fd012e0 899 * If the vnode is already dead don't try to revoke it
984263bc 900 */
5fd012e0 901 if (vp->v_flag & VRECLAIMED)
984263bc 902 return (0);
e4c9c0c8
MD
903
904 /*
905 * If the vnode has a device association, scrap all vnodes associated
906 * with the device. Don't let the device disappear on us while we
907 * are scrapping the vnodes.
5fd012e0
MD
908 *
909 * The passed vp will probably show up in the list, do not VX lock
910 * it twice!
e4c9c0c8
MD
911 */
912 if (vp->v_type != VCHR && vp->v_type != VBLK)
913 return(0);
914 if ((dev = vp->v_rdev) == NULL) {
915 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
916 return(0);
917 }
918 reference_dev(dev);
5fd012e0
MD
919 lwkt_gettoken(&ilock, &spechash_token);
920 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
921 if (vp == vq || vx_get(vq) == 0) {
922 if (vq == SLIST_FIRST(&dev->si_hlist))
923 vgone(vq);
924 if (vp != vq)
925 vx_put(vq);
926 }
984263bc 927 }
5fd012e0 928 lwkt_reltoken(&ilock);
e4c9c0c8 929 release_dev(dev);
984263bc
MD
930 return (0);
931}
932
933/*
934 * Recycle an unused vnode to the front of the free list.
5fd012e0
MD
935 *
936 * Returns 1 if we were successfully able to recycle the vnode,
937 * 0 otherwise.
984263bc
MD
938 */
939int
5fd012e0 940vrecycle(struct vnode *vp, struct thread *td)
984263bc 941{
5fd012e0
MD
942 if (vp->v_usecount == 1) {
943 vgone(vp);
984263bc
MD
944 return (1);
945 }
984263bc
MD
946 return (0);
947}
948
949/*
5fd012e0
MD
950 * Eliminate all activity associated with a vnode in preparation for reuse.
951 *
57ac0c99
MD
952 * The vnode must be VX locked and refd and will remain VX locked and refd
953 * on return. This routine may be called with the vnode in any state, as
954 * long as it is VX locked. The vnode will be cleaned out and marked
955 * VRECLAIMED but will not actually be reused until all existing refs and
956 * holds go away.
5fd012e0
MD
957 *
958 * NOTE: This routine may be called on a vnode which has not yet been
959 * already been deactivated (VOP_INACTIVE), or on a vnode which has
960 * already been reclaimed.
961 *
962 * This routine is not responsible for placing us back on the freelist.
963 * Instead, it happens automatically when the caller releases the VX lock
964 * (assuming there aren't any other references).
984263bc
MD
965 */
966void
dadab5e9 967vgone(struct vnode *vp)
984263bc 968{
984263bc 969 /*
5fd012e0
MD
970 * assert that the VX lock is held. This is an absolute requirement
971 * now for vgone() to be called.
984263bc 972 */
5fd012e0 973 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
984263bc
MD
974
975 /*
5fd012e0
MD
976 * Clean out the filesystem specific data and set the VRECLAIMED
977 * bit. Also deactivate the vnode if necessary.
984263bc 978 */
5fd012e0 979 vclean(vp, DOCLOSE, curthread);
984263bc
MD
980
981 /*
982 * Delete from old mount point vnode list, if on one.
983 */
984 if (vp->v_mount != NULL)
5fd012e0 985 insmntque(vp, NULL);
dd98570a 986
984263bc
MD
987 /*
988 * If special device, remove it from special device alias list
1fbb5fc0
MD
989 * if it is on one. This should normally only occur if a vnode is
990 * being revoked as the device should otherwise have been released
991 * naturally.
984263bc
MD
992 */
993 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
e4c9c0c8 994 v_release_rdev(vp);
984263bc
MD
995 }
996
997 /*
5fd012e0 998 * Set us to VBAD
984263bc 999 */
984263bc 1000 vp->v_type = VBAD;
984263bc
MD
1001}
1002
1003/*
1004 * Lookup a vnode by device number.
1005 */
1006int
dd98570a 1007vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
984263bc 1008{
41a01a4d 1009 lwkt_tokref ilock;
984263bc
MD
1010 struct vnode *vp;
1011
41a01a4d 1012 lwkt_gettoken(&ilock, &spechash_token);
984263bc
MD
1013 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1014 if (type == vp->v_type) {
1015 *vpp = vp;
41a01a4d 1016 lwkt_reltoken(&ilock);
984263bc
MD
1017 return (1);
1018 }
1019 }
41a01a4d 1020 lwkt_reltoken(&ilock);
984263bc
MD
1021 return (0);
1022}
1023
1024/*
e4c9c0c8
MD
1025 * Calculate the total number of references to a special device. This
1026 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1027 * an overloaded field. Since udev2dev can now return NODEV, we have
1028 * to check for a NULL v_rdev.
984263bc
MD
1029 */
1030int
e4c9c0c8 1031count_dev(dev_t dev)
984263bc 1032{
41a01a4d 1033 lwkt_tokref ilock;
e4c9c0c8
MD
1034 struct vnode *vp;
1035 int count = 0;
984263bc 1036
e4c9c0c8
MD
1037 if (SLIST_FIRST(&dev->si_hlist)) {
1038 lwkt_gettoken(&ilock, &spechash_token);
1039 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1040 count += vp->v_usecount;
1041 }
1042 lwkt_reltoken(&ilock);
1043 }
1044 return(count);
984263bc
MD
1045}
1046
984263bc 1047int
e4c9c0c8 1048count_udev(udev_t udev)
984263bc 1049{
e4c9c0c8 1050 dev_t dev;
984263bc 1051
e4c9c0c8
MD
1052 if ((dev = udev2dev(udev, 0)) == NODEV)
1053 return(0);
1054 return(count_dev(dev));
1055}
1056
1057int
1058vcount(struct vnode *vp)
1059{
1060 if (vp->v_rdev == NULL)
1061 return(0);
1062 return(count_dev(vp->v_rdev));
984263bc
MD
1063}
1064
1065/*
1066 * Print out a description of a vnode.
1067 */
1068static char *typename[] =
1069{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1070
1071void
dd98570a 1072vprint(char *label, struct vnode *vp)
984263bc
MD
1073{
1074 char buf[96];
1075
1076 if (label != NULL)
1077 printf("%s: %p: ", label, (void *)vp);
1078 else
1079 printf("%p: ", (void *)vp);
1080 printf("type %s, usecount %d, writecount %d, refcount %d,",
1081 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1082 vp->v_holdcnt);
1083 buf[0] = '\0';
1084 if (vp->v_flag & VROOT)
1085 strcat(buf, "|VROOT");
1086 if (vp->v_flag & VTEXT)
1087 strcat(buf, "|VTEXT");
1088 if (vp->v_flag & VSYSTEM)
1089 strcat(buf, "|VSYSTEM");
984263bc
MD
1090 if (vp->v_flag & VBWAIT)
1091 strcat(buf, "|VBWAIT");
984263bc
MD
1092 if (vp->v_flag & VFREE)
1093 strcat(buf, "|VFREE");
1094 if (vp->v_flag & VOBJBUF)
1095 strcat(buf, "|VOBJBUF");
1096 if (buf[0] != '\0')
1097 printf(" flags (%s)", &buf[1]);
1098 if (vp->v_data == NULL) {
1099 printf("\n");
1100 } else {
1101 printf("\n\t");
1102 VOP_PRINT(vp);
1103 }
1104}
1105
1106#ifdef DDB
1107#include <ddb/ddb.h>
1108/*
1109 * List all of the locked vnodes in the system.
1110 * Called when debugging the kernel.
1111 */
1112DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1113{
dadab5e9 1114 struct thread *td = curthread; /* XXX */
41a01a4d 1115 lwkt_tokref ilock;
984263bc
MD
1116 struct mount *mp, *nmp;
1117 struct vnode *vp;
1118
1119 printf("Locked vnodes\n");
41a01a4d 1120 lwkt_gettoken(&ilock, &mountlist_token);
984263bc 1121 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
41a01a4d 1122 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
984263bc
MD
1123 nmp = TAILQ_NEXT(mp, mnt_list);
1124 continue;
1125 }
1126 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1127 if (VOP_ISLOCKED(vp, NULL))
1128 vprint((char *)0, vp);
1129 }
41a01a4d 1130 lwkt_gettokref(&ilock);
984263bc 1131 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 1132 vfs_unbusy(mp, td);
984263bc 1133 }
41a01a4d 1134 lwkt_reltoken(&ilock);
984263bc
MD
1135}
1136#endif
1137
1138/*
1139 * Top level filesystem related information gathering.
1140 */
402ed7e1 1141static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
984263bc
MD
1142
1143static int
1144vfs_sysctl(SYSCTL_HANDLER_ARGS)
1145{
1146 int *name = (int *)arg1 - 1; /* XXX */
1147 u_int namelen = arg2 + 1; /* XXX */
1148 struct vfsconf *vfsp;
1149
1150#if 1 || defined(COMPAT_PRELITE2)
1151 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1152 if (namelen == 1)
1153 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1154#endif
1155
1156#ifdef notyet
1157 /* all sysctl names at this level are at least name and field */
1158 if (namelen < 2)
1159 return (ENOTDIR); /* overloaded */
1160 if (name[0] != VFS_GENERIC) {
1161 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1162 if (vfsp->vfc_typenum == name[0])
1163 break;
1164 if (vfsp == NULL)
1165 return (EOPNOTSUPP);
1166 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1167 oldp, oldlenp, newp, newlen, p));
1168 }
1169#endif
1170 switch (name[1]) {
1171 case VFS_MAXTYPENUM:
1172 if (namelen != 2)
1173 return (ENOTDIR);
1174 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1175 case VFS_CONF:
1176 if (namelen != 3)
1177 return (ENOTDIR); /* overloaded */
1178 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1179 if (vfsp->vfc_typenum == name[2])
1180 break;
1181 if (vfsp == NULL)
1182 return (EOPNOTSUPP);
1183 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1184 }
1185 return (EOPNOTSUPP);
1186}
1187
1188SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1189 "Generic filesystem");
1190
1191#if 1 || defined(COMPAT_PRELITE2)
1192
1193static int
1194sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1195{
1196 int error;
1197 struct vfsconf *vfsp;
1198 struct ovfsconf ovfs;
1199
1200 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1201 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1202 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1203 ovfs.vfc_index = vfsp->vfc_typenum;
1204 ovfs.vfc_refcount = vfsp->vfc_refcount;
1205 ovfs.vfc_flags = vfsp->vfc_flags;
1206 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1207 if (error)
1208 return error;
1209 }
1210 return 0;
1211}
1212
1213#endif /* 1 || COMPAT_PRELITE2 */
1214
1215#if 0
1216#define KINFO_VNODESLOP 10
1217/*
1218 * Dump vnode list (via sysctl).
1219 * Copyout address of vnode followed by vnode.
1220 */
1221/* ARGSUSED */
1222static int
1223sysctl_vnode(SYSCTL_HANDLER_ARGS)
1224{
1225 struct proc *p = curproc; /* XXX */
1226 struct mount *mp, *nmp;
1227 struct vnode *nvp, *vp;
41a01a4d
MD
1228 lwkt_tokref ilock;
1229 lwkt_tokref jlock;
984263bc
MD
1230 int error;
1231
1232#define VPTRSZ sizeof (struct vnode *)
1233#define VNODESZ sizeof (struct vnode)
1234
1235 req->lock = 0;
1236 if (!req->oldptr) /* Make an estimate */
1237 return (SYSCTL_OUT(req, 0,
1238 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
1239
41a01a4d 1240 lwkt_gettoken(&ilock, &mountlist_token);
984263bc 1241 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
41a01a4d 1242 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
984263bc
MD
1243 nmp = TAILQ_NEXT(mp, mnt_list);
1244 continue;
1245 }
41a01a4d 1246 lwkt_gettoken(&jlock, &mntvnode_token);
984263bc 1247again:
984263bc
MD
1248 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1249 vp != NULL;
1250 vp = nvp) {
1251 /*
1252 * Check that the vp is still associated with
1253 * this filesystem. RACE: could have been
1254 * recycled onto the same filesystem.
1255 */
41a01a4d 1256 if (vp->v_mount != mp)
984263bc 1257 goto again;
984263bc 1258 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
984263bc 1259 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
41a01a4d
MD
1260 (error = SYSCTL_OUT(req, vp, VNODESZ))) {
1261 lwkt_reltoken(&jlock);
984263bc 1262 return (error);
41a01a4d 1263 }
984263bc 1264 }
41a01a4d
MD
1265 lwkt_reltoken(&jlock);
1266 lwkt_gettokref(&ilock);
1267 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */
984263bc
MD
1268 vfs_unbusy(mp, p);
1269 }
41a01a4d 1270 lwkt_reltoken(&ilock);
984263bc
MD
1271
1272 return (0);
1273}
1274#endif
1275
1276/*
1277 * XXX
1278 * Exporting the vnode list on large systems causes them to crash.
1279 * Exporting the vnode list on medium systems causes sysctl to coredump.
1280 */
1281#if 0
1282SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
1283 0, 0, sysctl_vnode, "S,vnode", "");
1284#endif
1285
1286/*
1287 * Check to see if a filesystem is mounted on a block device.
1288 */
1289int
e4c9c0c8 1290vfs_mountedon(struct vnode *vp)
984263bc 1291{
e4c9c0c8 1292 dev_t dev;
984263bc 1293
e4c9c0c8
MD
1294 if ((dev = vp->v_rdev) == NULL)
1295 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1296 if (dev != NODEV && dev->si_mountpoint)
984263bc
MD
1297 return (EBUSY);
1298 return (0);
1299}
1300
1301/*
1302 * Unmount all filesystems. The list is traversed in reverse order
1303 * of mounting to avoid dependencies.
1304 */
1305void
dd98570a 1306vfs_unmountall(void)
984263bc
MD
1307{
1308 struct mount *mp;
dadab5e9 1309 struct thread *td = curthread;
984263bc
MD
1310 int error;
1311
dadab5e9
MD
1312 if (td->td_proc == NULL)
1313 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
1314
984263bc
MD
1315 /*
1316 * Since this only runs when rebooting, it is not interlocked.
1317 */
1318 while(!TAILQ_EMPTY(&mountlist)) {
1319 mp = TAILQ_LAST(&mountlist, mntlist);
dadab5e9 1320 error = dounmount(mp, MNT_FORCE, td);
984263bc
MD
1321 if (error) {
1322 TAILQ_REMOVE(&mountlist, mp, mnt_list);
75ffff0d 1323 printf("unmount of filesystem mounted from %s failed (", mp->mnt_stat.f_mntfromname);
984263bc
MD
1324 if (error == EBUSY)
1325 printf("BUSY)\n");
1326 else
1327 printf("%d)\n", error);
1328 } else {
1329 /* The unmount has removed mp from the mountlist */
1330 }
1331 }
1332}
1333
1334/*
1335 * Build hash lists of net addresses and hang them off the mount point.
1336 * Called by ufs_mount() to set up the lists of export addresses.
1337 */
1338static int
dd98570a
MD
1339vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1340 struct export_args *argp)
984263bc 1341{
1fd87d54
RG
1342 struct netcred *np;
1343 struct radix_node_head *rnh;
1344 int i;
984263bc
MD
1345 struct radix_node *rn;
1346 struct sockaddr *saddr, *smask = 0;
1347 struct domain *dom;
1348 int error;
1349
1350 if (argp->ex_addrlen == 0) {
1351 if (mp->mnt_flag & MNT_DEFEXPORTED)
1352 return (EPERM);
1353 np = &nep->ne_defexported;
1354 np->netc_exflags = argp->ex_flags;
1355 np->netc_anon = argp->ex_anon;
1356 np->netc_anon.cr_ref = 1;
1357 mp->mnt_flag |= MNT_DEFEXPORTED;
1358 return (0);
1359 }
1360
0260ddf9
MD
1361 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1362 return (EINVAL);
1363 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
984263bc
MD
1364 return (EINVAL);
1365
1366 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1367 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1368 bzero((caddr_t) np, i);
1369 saddr = (struct sockaddr *) (np + 1);
1370 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1371 goto out;
1372 if (saddr->sa_len > argp->ex_addrlen)
1373 saddr->sa_len = argp->ex_addrlen;
1374 if (argp->ex_masklen) {
dd98570a
MD
1375 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1376 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
984263bc
MD
1377 if (error)
1378 goto out;
1379 if (smask->sa_len > argp->ex_masklen)
1380 smask->sa_len = argp->ex_masklen;
1381 }
1382 i = saddr->sa_family;
1383 if ((rnh = nep->ne_rtable[i]) == 0) {
1384 /*
1385 * Seems silly to initialize every AF when most are not used,
1386 * do so on demand here
1387 */
1388 for (dom = domains; dom; dom = dom->dom_next)
1389 if (dom->dom_family == i && dom->dom_rtattach) {
1390 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1391 dom->dom_rtoffset);
1392 break;
1393 }
1394 if ((rnh = nep->ne_rtable[i]) == 0) {
1395 error = ENOBUFS;
1396 goto out;
1397 }
1398 }
2e9572df 1399 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
984263bc
MD
1400 np->netc_rnodes);
1401 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1402 error = EPERM;
1403 goto out;
1404 }
1405 np->netc_exflags = argp->ex_flags;
1406 np->netc_anon = argp->ex_anon;
1407 np->netc_anon.cr_ref = 1;
1408 return (0);
1409out:
1410 free(np, M_NETADDR);
1411 return (error);
1412}
1413
1414/* ARGSUSED */
1415static int
dd98570a 1416vfs_free_netcred(struct radix_node *rn, void *w)
984263bc 1417{
1fd87d54 1418 struct radix_node_head *rnh = (struct radix_node_head *) w;
984263bc
MD
1419
1420 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1421 free((caddr_t) rn, M_NETADDR);
1422 return (0);
1423}
1424
1425/*
1426 * Free the net address hash lists that are hanging off the mount points.
1427 */
1428static void
dd98570a 1429vfs_free_addrlist(struct netexport *nep)
984263bc 1430{
1fd87d54
RG
1431 int i;
1432 struct radix_node_head *rnh;
984263bc
MD
1433
1434 for (i = 0; i <= AF_MAX; i++)
1435 if ((rnh = nep->ne_rtable[i])) {
1436 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1437 (caddr_t) rnh);
1438 free((caddr_t) rnh, M_RTABLE);
1439 nep->ne_rtable[i] = 0;
1440 }
1441}
1442
1443int
dd98570a 1444vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
984263bc
MD
1445{
1446 int error;
1447
1448 if (argp->ex_flags & MNT_DELEXPORT) {
1449 if (mp->mnt_flag & MNT_EXPUBLIC) {
1450 vfs_setpublicfs(NULL, NULL, NULL);
1451 mp->mnt_flag &= ~MNT_EXPUBLIC;
1452 }
1453 vfs_free_addrlist(nep);
1454 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1455 }
1456 if (argp->ex_flags & MNT_EXPORTED) {
1457 if (argp->ex_flags & MNT_EXPUBLIC) {
1458 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1459 return (error);
1460 mp->mnt_flag |= MNT_EXPUBLIC;
1461 }
1462 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1463 return (error);
1464 mp->mnt_flag |= MNT_EXPORTED;
1465 }
1466 return (0);
1467}
1468
1469
1470/*
1471 * Set the publicly exported filesystem (WebNFS). Currently, only
1472 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1473 */
1474int
dd98570a
MD
1475vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1476 struct export_args *argp)
984263bc
MD
1477{
1478 int error;
1479 struct vnode *rvp;
1480 char *cp;
1481
1482 /*
1483 * mp == NULL -> invalidate the current info, the FS is
1484 * no longer exported. May be called from either vfs_export
1485 * or unmount, so check if it hasn't already been done.
1486 */
1487 if (mp == NULL) {
1488 if (nfs_pub.np_valid) {
1489 nfs_pub.np_valid = 0;
1490 if (nfs_pub.np_index != NULL) {
1491 FREE(nfs_pub.np_index, M_TEMP);
1492 nfs_pub.np_index = NULL;
1493 }
1494 }
1495 return (0);
1496 }
1497
1498 /*
1499 * Only one allowed at a time.
1500 */
1501 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1502 return (EBUSY);
1503
1504 /*
1505 * Get real filehandle for root of exported FS.
1506 */
1507 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1508 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1509
1510 if ((error = VFS_ROOT(mp, &rvp)))
1511 return (error);
1512
1513 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1514 return (error);
1515
1516 vput(rvp);
1517
1518 /*
1519 * If an indexfile was specified, pull it in.
1520 */
1521 if (argp->ex_indexfile != NULL) {
1522 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
1523 M_WAITOK);
1524 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1525 MAXNAMLEN, (size_t *)0);
1526 if (!error) {
1527 /*
1528 * Check for illegal filenames.
1529 */
1530 for (cp = nfs_pub.np_index; *cp; cp++) {
1531 if (*cp == '/') {
1532 error = EINVAL;
1533 break;
1534 }
1535 }
1536 }
1537 if (error) {
1538 FREE(nfs_pub.np_index, M_TEMP);
1539 return (error);
1540 }
1541 }
1542
1543 nfs_pub.np_mount = mp;
1544 nfs_pub.np_valid = 1;
1545 return (0);
1546}
1547
1548struct netcred *
dd98570a
MD
1549vfs_export_lookup(struct mount *mp, struct netexport *nep,
1550 struct sockaddr *nam)
984263bc 1551{
1fd87d54
RG
1552 struct netcred *np;
1553 struct radix_node_head *rnh;
984263bc
MD
1554 struct sockaddr *saddr;
1555
1556 np = NULL;
1557 if (mp->mnt_flag & MNT_EXPORTED) {
1558 /*
1559 * Lookup in the export list first.
1560 */
1561 if (nam != NULL) {
1562 saddr = nam;
1563 rnh = nep->ne_rtable[saddr->sa_family];
1564 if (rnh != NULL) {
1565 np = (struct netcred *)
2e9572df 1566 (*rnh->rnh_matchaddr)((char *)saddr,
984263bc
MD
1567 rnh);
1568 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1569 np = NULL;
1570 }
1571 }
1572 /*
1573 * If no address match, use the default if it exists.
1574 */
1575 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1576 np = &nep->ne_defexported;
1577 }
1578 return (np);
1579}
1580
1581/*
41a01a4d
MD
1582 * perform msync on all vnodes under a mount point. The mount point must
1583 * be locked. This code is also responsible for lazy-freeing unreferenced
1584 * vnodes whos VM objects no longer contain pages.
1585 *
1586 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
984263bc 1587 */
41a01a4d 1588static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
5fd012e0 1589static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
41a01a4d 1590
984263bc
MD
1591void
1592vfs_msync(struct mount *mp, int flags)
1593{
5fd012e0
MD
1594 vmntvnodescan(mp, VMSC_REFVP, vfs_msync_scan1, vfs_msync_scan2,
1595 (void *)flags);
41a01a4d 1596}
984263bc 1597
41a01a4d
MD
1598/*
1599 * scan1 is a fast pre-check. There could be hundreds of thousands of
1600 * vnodes, we cannot afford to do anything heavy weight until we have a
1601 * fairly good indication that there is work to do.
1602 */
1603static
1604int
1605vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1606{
1607 int flags = (int)data;
984263bc 1608
5fd012e0
MD
1609 if ((vp->v_flag & VRECLAIMED) == 0) {
1610 if (vshouldfree(vp, 0))
1611 return(0); /* call scan2 */
41a01a4d
MD
1612 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1613 (vp->v_flag & VOBJDIRTY) &&
984263bc 1614 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
5fd012e0 1615 return(0); /* call scan2 */
41a01a4d
MD
1616 }
1617 }
5fd012e0
MD
1618
1619 /*
1620 * do not call scan2, continue the loop
1621 */
41a01a4d
MD
1622 return(-1);
1623}
1624
1625static
1626int
5fd012e0 1627vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
41a01a4d
MD
1628{
1629 vm_object_t obj;
41a01a4d
MD
1630 int flags = (int)data;
1631
5fd012e0 1632 if (vp->v_flag & VRECLAIMED)
41a01a4d
MD
1633 return(0);
1634
1635 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1636 (vp->v_flag & VOBJDIRTY) &&
1637 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
5fd012e0
MD
1638 if (VOP_GETVOBJECT(vp, &obj) == 0) {
1639 vm_object_page_clean(obj, 0, 0,
1640 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
984263bc
MD
1641 }
1642 }
41a01a4d 1643 return(0);
984263bc
MD
1644}
1645
1646/*
1647 * Create the VM object needed for VMIO and mmap support. This
1648 * is done for all VREG files in the system. Some filesystems might
1649 * afford the additional metadata buffering capability of the
1650 * VMIO code by making the device node be VMIO mode also.
1651 *
1652 * vp must be locked when vfs_object_create is called.
1653 */
1654int
3b568787 1655vfs_object_create(struct vnode *vp, struct thread *td)
984263bc 1656{
3b568787 1657 return (VOP_CREATEVOBJECT(vp, td));
984263bc
MD
1658}
1659
984263bc
MD
1660/*
1661 * Record a process's interest in events which might happen to
1662 * a vnode. Because poll uses the historic select-style interface
1663 * internally, this routine serves as both the ``check for any
1664 * pending events'' and the ``record my interest in future events''
1665 * functions. (These are done together, while the lock is held,
1666 * to avoid race conditions.)
1667 */
1668int
dadab5e9 1669vn_pollrecord(struct vnode *vp, struct thread *td, int events)
984263bc 1670{
41a01a4d
MD
1671 lwkt_tokref ilock;
1672
1673 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
1674 if (vp->v_pollinfo.vpi_revents & events) {
1675 /*
1676 * This leaves events we are not interested
1677 * in available for the other process which
1678 * which presumably had requested them
1679 * (otherwise they would never have been
1680 * recorded).
1681 */
1682 events &= vp->v_pollinfo.vpi_revents;
1683 vp->v_pollinfo.vpi_revents &= ~events;
1684
41a01a4d 1685 lwkt_reltoken(&ilock);
984263bc
MD
1686 return events;
1687 }
1688 vp->v_pollinfo.vpi_events |= events;
dadab5e9 1689 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
41a01a4d 1690 lwkt_reltoken(&ilock);
984263bc
MD
1691 return 0;
1692}
1693
1694/*
1695 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1696 * it is possible for us to miss an event due to race conditions, but
1697 * that condition is expected to be rare, so for the moment it is the
1698 * preferred interface.
1699 */
1700void
dd98570a 1701vn_pollevent(struct vnode *vp, int events)
984263bc 1702{
41a01a4d
MD
1703 lwkt_tokref ilock;
1704
1705 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
1706 if (vp->v_pollinfo.vpi_events & events) {
1707 /*
1708 * We clear vpi_events so that we don't
1709 * call selwakeup() twice if two events are
1710 * posted before the polling process(es) is
1711 * awakened. This also ensures that we take at
1712 * most one selwakeup() if the polling process
1713 * is no longer interested. However, it does
1714 * mean that only one event can be noticed at
1715 * a time. (Perhaps we should only clear those
1716 * event bits which we note?) XXX
1717 */
1718 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
1719 vp->v_pollinfo.vpi_revents |= events;
1720 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1721 }
41a01a4d 1722 lwkt_reltoken(&ilock);
984263bc
MD
1723}
1724
1725/*
1726 * Wake up anyone polling on vp because it is being revoked.
1727 * This depends on dead_poll() returning POLLHUP for correct
1728 * behavior.
1729 */
1730void
dd98570a 1731vn_pollgone(struct vnode *vp)
984263bc 1732{
41a01a4d
MD
1733 lwkt_tokref ilock;
1734
1735 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
1736 if (vp->v_pollinfo.vpi_events) {
1737 vp->v_pollinfo.vpi_events = 0;
1738 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1739 }
41a01a4d 1740 lwkt_reltoken(&ilock);
984263bc
MD
1741}
1742
984263bc 1743/*
e4c9c0c8
MD
1744 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1745 * (or v_rdev might be NULL).
984263bc
MD
1746 */
1747dev_t
e4c9c0c8 1748vn_todev(struct vnode *vp)
984263bc
MD
1749{
1750 if (vp->v_type != VBLK && vp->v_type != VCHR)
1751 return (NODEV);
e4c9c0c8 1752 KKASSERT(vp->v_rdev != NULL);
984263bc
MD
1753 return (vp->v_rdev);
1754}
1755
1756/*
e4c9c0c8
MD
1757 * Check if vnode represents a disk device. The vnode does not need to be
1758 * opened.
984263bc
MD
1759 */
1760int
e4c9c0c8 1761vn_isdisk(struct vnode *vp, int *errp)
984263bc 1762{
e4c9c0c8
MD
1763 dev_t dev;
1764
984263bc
MD
1765 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1766 if (errp != NULL)
1767 *errp = ENOTBLK;
1768 return (0);
1769 }
e4c9c0c8
MD
1770
1771 if ((dev = vp->v_rdev) == NULL)
1772 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1773 if (dev == NULL || dev == NODEV) {
984263bc
MD
1774 if (errp != NULL)
1775 *errp = ENXIO;
1776 return (0);
1777 }
e4c9c0c8 1778 if (dev_is_good(dev) == 0) {
984263bc
MD
1779 if (errp != NULL)
1780 *errp = ENXIO;
1781 return (0);
1782 }
e4c9c0c8 1783 if ((dev_dflags(dev) & D_DISK) == 0) {
984263bc
MD
1784 if (errp != NULL)
1785 *errp = ENOTBLK;
1786 return (0);
1787 }
1788 if (errp != NULL)
1789 *errp = 0;
1790 return (1);
1791}
1792
4f322a84
EN
1793#ifdef DEBUG_VFS_LOCKS
1794
1795void
1796assert_vop_locked(struct vnode *vp, const char *str)
1797{
4f322a84
EN
1798 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
1799 panic("%s: %p is not locked shared but should be", str, vp);
1800 }
1801}
1802
1803void
1804assert_vop_unlocked(struct vnode *vp, const char *str)
1805{
4f322a84
EN
1806 if (vp && IS_LOCKING_VFS(vp)) {
1807 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
1808 panic("%s: %p is locked but should not be", str, vp);
1809 }
1810 }
1811}
1812
1813#endif