AMD64 - Fix many compile-time warnings. int/ptr type mismatches, %llx, etc.
[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 $
67863d04 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.118 2008/09/17 21:44:18 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>
b8477cda 56#include <sys/file.h>
984263bc
MD
57#include <sys/kernel.h>
58#include <sys/kthread.h>
59#include <sys/malloc.h>
60#include <sys/mbuf.h>
61#include <sys/mount.h>
984263bc
MD
62#include <sys/proc.h>
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>
5d72d6ed 68#include <sys/unistd.h>
984263bc
MD
69#include <sys/vmmeter.h>
70#include <sys/vnode.h>
71
72#include <machine/limits.h>
73
74#include <vm/vm.h>
75#include <vm/vm_object.h>
76#include <vm/vm_extern.h>
6ef943a3 77#include <vm/vm_kern.h>
984263bc
MD
78#include <vm/pmap.h>
79#include <vm/vm_map.h>
80#include <vm/vm_page.h>
81#include <vm/vm_pager.h>
82#include <vm/vnode_pager.h>
83#include <vm/vm_zone.h>
84
3020e3be 85#include <sys/buf2.h>
f5865223 86#include <sys/thread2.h>
3c37c940 87#include <sys/sysref2.h>
3020e3be 88
984263bc
MD
89static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
90
5fd012e0 91int numvnodes;
984263bc 92SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
36dbf6bd 93int vfs_fastdev = 1;
fad57d0e 94SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
984263bc
MD
95
96enum vtype iftovt_tab[16] = {
97 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99};
100int vttoif_tab[9] = {
101 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 S_IFSOCK, S_IFIFO, S_IFMT,
103};
104
984263bc 105static int reassignbufcalls;
dd98570a
MD
106SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
107 &reassignbufcalls, 0, "");
984263bc 108static int reassignbufloops;
dd98570a
MD
109SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
110 &reassignbufloops, 0, "");
984263bc 111static int reassignbufsortgood;
dd98570a
MD
112SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
113 &reassignbufsortgood, 0, "");
984263bc 114static int reassignbufsortbad;
dd98570a
MD
115SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
116 &reassignbufsortbad, 0, "");
984263bc 117static int reassignbufmethod = 1;
dd98570a
MD
118SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
119 &reassignbufmethod, 0, "");
984263bc 120
984263bc 121int nfs_mount_type = -1;
8a8d5d85 122static struct lwkt_token spechash_token;
984263bc 123struct nfs_public nfs_pub; /* publicly exported FS */
984263bc
MD
124
125int desiredvnodes;
126SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
dd98570a 127 &desiredvnodes, 0, "Maximum number of vnodes");
984263bc 128
402ed7e1
RG
129static void vfs_free_addrlist (struct netexport *nep);
130static int vfs_free_netcred (struct radix_node *rn, void *w);
131static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
1aa89f17 132 const struct export_args *argp);
984263bc 133
e4c9c0c8
MD
134extern int dev_ref_debug;
135
41a01a4d 136/*
6bae6177
MD
137 * Red black tree functions
138 */
139static int rb_buf_compare(struct buf *b1, struct buf *b2);
54078292
MD
140RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
141RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
6bae6177
MD
142
143static int
144rb_buf_compare(struct buf *b1, struct buf *b2)
145{
54078292 146 if (b1->b_loffset < b2->b_loffset)
6bae6177 147 return(-1);
54078292 148 if (b1->b_loffset > b2->b_loffset)
6bae6177
MD
149 return(1);
150 return(0);
151}
152
153/*
44b1cf3d 154 * Returns non-zero if the vnode is a candidate for lazy msyncing.
41a01a4d 155 */
5fd012e0 156static __inline int
3c37c940 157vshouldmsync(struct vnode *vp)
41a01a4d 158{
3c37c940 159 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
44b1cf3d 160 return (0); /* other holders */
5fd012e0
MD
161 if (vp->v_object &&
162 (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
163 return (0);
164 }
165 return (1);
41a01a4d 166}
5fd012e0 167
984263bc 168/*
5fd012e0
MD
169 * Initialize the vnode management data structures.
170 *
171 * Called from vfsinit()
984263bc
MD
172 */
173void
5fd012e0 174vfs_subr_init(void)
984263bc 175{
6ef943a3 176 /*
55d3a838
MD
177 * Desiredvnodes is kern.maxvnodes. We want to scale it
178 * according to available system memory but we may also have
179 * to limit it based on available KVM, which is capped on 32 bit
180 * systems.
6ef943a3 181 */
55d3a838
MD
182 desiredvnodes = min(maxproc + vmstats.v_page_count / 4,
183 KvaSize / (20 *
184 (sizeof(struct vm_object) + sizeof(struct vnode))));
6ef943a3 185
41a01a4d 186 lwkt_token_init(&spechash_token);
984263bc
MD
187}
188
189/*
190 * Knob to control the precision of file timestamps:
191 *
192 * 0 = seconds only; nanoseconds zeroed.
193 * 1 = seconds and nanoseconds, accurate within 1/HZ.
194 * 2 = seconds and nanoseconds, truncated to microseconds.
195 * >=3 = seconds and nanoseconds, maximum precision.
196 */
197enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
198
199static int timestamp_precision = TSP_SEC;
200SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
dd98570a 201 &timestamp_precision, 0, "");
984263bc
MD
202
203/*
204 * Get a current timestamp.
205 */
206void
dd98570a 207vfs_timestamp(struct timespec *tsp)
984263bc
MD
208{
209 struct timeval tv;
210
211 switch (timestamp_precision) {
212 case TSP_SEC:
213 tsp->tv_sec = time_second;
214 tsp->tv_nsec = 0;
215 break;
216 case TSP_HZ:
217 getnanotime(tsp);
218 break;
219 case TSP_USEC:
220 microtime(&tv);
221 TIMEVAL_TO_TIMESPEC(&tv, tsp);
222 break;
223 case TSP_NSEC:
224 default:
225 nanotime(tsp);
226 break;
227 }
228}
229
230/*
231 * Set vnode attributes to VNOVAL
232 */
233void
dd98570a 234vattr_null(struct vattr *vap)
984263bc 235{
984263bc
MD
236 vap->va_type = VNON;
237 vap->va_size = VNOVAL;
238 vap->va_bytes = VNOVAL;
239 vap->va_mode = VNOVAL;
240 vap->va_nlink = VNOVAL;
241 vap->va_uid = VNOVAL;
242 vap->va_gid = VNOVAL;
243 vap->va_fsid = VNOVAL;
244 vap->va_fileid = VNOVAL;
245 vap->va_blocksize = VNOVAL;
0e9b9130
MD
246 vap->va_rmajor = VNOVAL;
247 vap->va_rminor = VNOVAL;
984263bc
MD
248 vap->va_atime.tv_sec = VNOVAL;
249 vap->va_atime.tv_nsec = VNOVAL;
250 vap->va_mtime.tv_sec = VNOVAL;
251 vap->va_mtime.tv_nsec = VNOVAL;
252 vap->va_ctime.tv_sec = VNOVAL;
253 vap->va_ctime.tv_nsec = VNOVAL;
254 vap->va_flags = VNOVAL;
255 vap->va_gen = VNOVAL;
256 vap->va_vaflags = 0;
dc1be39c 257 vap->va_fsmid = VNOVAL;
50626622 258 /* va_*_uuid fields are only valid if related flags are set */
984263bc
MD
259}
260
261/*
984263bc 262 * Flush out and invalidate all buffers associated with a vnode.
5fd012e0
MD
263 *
264 * vp must be locked.
984263bc 265 */
6bae6177
MD
266static int vinvalbuf_bp(struct buf *bp, void *data);
267
268struct vinvalbuf_bp_info {
269 struct vnode *vp;
270 int slptimeo;
f2770c70 271 int lkflags;
6bae6177
MD
272 int flags;
273};
274
6b008938
MD
275void
276vupdatefsmid(struct vnode *vp)
277{
278 atomic_set_int(&vp->v_flag, VFSMID);
279}
280
984263bc 281int
87de5057 282vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
984263bc 283{
6bae6177 284 struct vinvalbuf_bp_info info;
e43a034f 285 int error;
984263bc
MD
286 vm_object_t object;
287
6bae6177
MD
288 /*
289 * If we are being asked to save, call fsync to ensure that the inode
290 * is updated.
291 */
984263bc 292 if (flags & V_SAVE) {
e43a034f 293 crit_enter();
81b5c339
MD
294 while (vp->v_track_write.bk_active) {
295 vp->v_track_write.bk_waitflag = 1;
296 error = tsleep(&vp->v_track_write, slpflag,
297 "vinvlbuf", slptimeo);
984263bc 298 if (error) {
e43a034f 299 crit_exit();
984263bc
MD
300 return (error);
301 }
302 }
6bae6177 303 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
e43a034f 304 crit_exit();
87de5057 305 if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0)
984263bc 306 return (error);
e43a034f 307 crit_enter();
1bb61199
MD
308
309 /*
310 * Dirty bufs may be left or generated via races
311 * in circumstances where vinvalbuf() is called on
312 * a vnode not undergoing reclamation. Only
313 * panic if we are trying to reclaim the vnode.
314 */
315 if ((vp->v_flag & VRECLAIMED) &&
316 (vp->v_track_write.bk_active > 0 ||
317 !RB_EMPTY(&vp->v_rbdirty_tree))) {
984263bc 318 panic("vinvalbuf: dirty bufs");
1bb61199 319 }
984263bc 320 }
e43a034f 321 crit_exit();
984263bc 322 }
e43a034f 323 crit_enter();
6bae6177 324 info.slptimeo = slptimeo;
f2770c70
MD
325 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
326 if (slpflag & PCATCH)
327 info.lkflags |= LK_PCATCH;
6bae6177
MD
328 info.flags = flags;
329 info.vp = vp;
330
331 /*
332 * Flush the buffer cache until nothing is left.
333 */
334 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
335 !RB_EMPTY(&vp->v_rbdirty_tree)) {
336 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
65c6c519 337 vinvalbuf_bp, &info);
6bae6177
MD
338 if (error == 0) {
339 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
340 vinvalbuf_bp, &info);
984263bc
MD
341 }
342 }
343
344 /*
345 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
346 * have write I/O in-progress but if there is a VM object then the
347 * VM object can also have read-I/O in-progress.
348 */
349 do {
81b5c339
MD
350 while (vp->v_track_write.bk_active > 0) {
351 vp->v_track_write.bk_waitflag = 1;
352 tsleep(&vp->v_track_write, 0, "vnvlbv", 0);
984263bc 353 }
7540ab49 354 if ((object = vp->v_object) != NULL) {
984263bc
MD
355 while (object->paging_in_progress)
356 vm_object_pip_sleep(object, "vnvlbx");
357 }
81b5c339 358 } while (vp->v_track_write.bk_active > 0);
984263bc 359
e43a034f 360 crit_exit();
984263bc
MD
361
362 /*
363 * Destroy the copy in the VM cache, too.
364 */
7540ab49 365 if ((object = vp->v_object) != NULL) {
984263bc
MD
366 vm_object_page_remove(object, 0, 0,
367 (flags & V_SAVE) ? TRUE : FALSE);
368 }
984263bc 369
6bae6177 370 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
984263bc 371 panic("vinvalbuf: flush failed");
1f1ea522
MD
372 if (!RB_EMPTY(&vp->v_rbhash_tree))
373 panic("vinvalbuf: flush failed, buffers still present");
984263bc
MD
374 return (0);
375}
376
6bae6177
MD
377static int
378vinvalbuf_bp(struct buf *bp, void *data)
379{
380 struct vinvalbuf_bp_info *info = data;
381 int error;
382
383 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
f2770c70
MD
384 error = BUF_TIMELOCK(bp, info->lkflags,
385 "vinvalbuf", info->slptimeo);
6bae6177
MD
386 if (error == 0) {
387 BUF_UNLOCK(bp);
388 error = ENOLCK;
389 }
390 if (error == ENOLCK)
391 return(0);
392 return (-error);
393 }
65c6c519
MD
394
395 KKASSERT(bp->b_vp == info->vp);
396
6bae6177
MD
397 /*
398 * XXX Since there are no node locks for NFS, I
399 * believe there is a slight chance that a delayed
400 * write will occur while sleeping just above, so
401 * check for it. Note that vfs_bio_awrite expects
62cfda27
MD
402 * buffers to reside on a queue, while bwrite() and
403 * brelse() do not.
27bc0cb1
MD
404 *
405 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
406 * check. This code will write out the buffer, period.
6bae6177
MD
407 */
408 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
409 (info->flags & V_SAVE)) {
410 if (bp->b_vp == info->vp) {
411 if (bp->b_flags & B_CLUSTEROK) {
6bae6177
MD
412 vfs_bio_awrite(bp);
413 } else {
414 bremfree(bp);
415 bp->b_flags |= B_ASYNC;
62cfda27 416 bwrite(bp);
6bae6177
MD
417 }
418 } else {
419 bremfree(bp);
62cfda27 420 bwrite(bp);
6bae6177 421 }
135bd6a8
MD
422 } else if (info->flags & V_SAVE) {
423 /*
424 * Cannot set B_NOCACHE on a clean buffer as this will
425 * destroy the VM backing store which might actually
426 * be dirty (and unsynchronized).
427 */
428 bremfree(bp);
429 bp->b_flags |= (B_INVAL | B_RELBUF);
430 bp->b_flags &= ~B_ASYNC;
431 brelse(bp);
6bae6177
MD
432 } else {
433 bremfree(bp);
434 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
435 bp->b_flags &= ~B_ASYNC;
436 brelse(bp);
437 }
438 return(0);
439}
440
984263bc
MD
441/*
442 * Truncate a file's buffer and pages to a specified length. This
443 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
444 * sync activity.
5fd012e0
MD
445 *
446 * The vnode must be locked.
984263bc 447 */
6bae6177
MD
448static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
449static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
450static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
451static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
452
984263bc 453int
87de5057 454vtruncbuf(struct vnode *vp, off_t length, int blksize)
984263bc 455{
54078292 456 off_t truncloffset;
6bae6177 457 int count;
c4b46cb4 458 const char *filename;
984263bc
MD
459
460 /*
54078292 461 * Round up to the *next* block, then destroy the buffers in question.
6bae6177
MD
462 * Since we are only removing some of the buffers we must rely on the
463 * scan count to determine whether a loop is necessary.
984263bc 464 */
54078292
MD
465 if ((count = (int)(length % blksize)) != 0)
466 truncloffset = length + (blksize - count);
467 else
468 truncloffset = length;
984263bc 469
e43a034f 470 crit_enter();
6bae6177
MD
471 do {
472 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
473 vtruncbuf_bp_trunc_cmp,
54078292 474 vtruncbuf_bp_trunc, &truncloffset);
6bae6177
MD
475 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
476 vtruncbuf_bp_trunc_cmp,
54078292 477 vtruncbuf_bp_trunc, &truncloffset);
6bae6177 478 } while(count);
984263bc 479
6bae6177
MD
480 /*
481 * For safety, fsync any remaining metadata if the file is not being
482 * truncated to 0. Since the metadata does not represent the entire
483 * dirty list we have to rely on the hit count to ensure that we get
484 * all of it.
485 */
984263bc 486 if (length > 0) {
6bae6177
MD
487 do {
488 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
489 vtruncbuf_bp_metasync_cmp,
490 vtruncbuf_bp_metasync, vp);
491 } while (count);
984263bc
MD
492 }
493
6bae6177 494 /*
c4b46cb4
MD
495 * Clean out any left over VM backing store.
496 */
497 crit_exit();
b1f72a5c 498
c4b46cb4 499 vnode_pager_setsize(vp, length);
b1f72a5c 500
c4b46cb4
MD
501 crit_enter();
502
503 /*
135bd6a8
MD
504 * It is possible to have in-progress I/O from buffers that were
505 * not part of the truncation. This should not happen if we
506 * are truncating to 0-length.
6bae6177 507 */
c4b46cb4
MD
508 filename = TAILQ_FIRST(&vp->v_namecache) ?
509 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
510
511 while ((count = vp->v_track_write.bk_active) > 0) {
81b5c339
MD
512 vp->v_track_write.bk_waitflag = 1;
513 tsleep(&vp->v_track_write, 0, "vbtrunc", 0);
135bd6a8 514 if (length == 0) {
6ea70f76 515 kprintf("Warning: vtruncbuf(): Had to wait for "
135bd6a8
MD
516 "%d buffer I/Os to finish in %s\n",
517 count, filename);
518 }
984263bc
MD
519 }
520
c4b46cb4
MD
521 /*
522 * Make sure no buffers were instantiated while we were trying
523 * to clean out the remaining VM pages. This could occur due
524 * to busy dirty VM pages being flushed out to disk.
525 */
526 do {
527 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
528 vtruncbuf_bp_trunc_cmp,
529 vtruncbuf_bp_trunc, &truncloffset);
530 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
531 vtruncbuf_bp_trunc_cmp,
532 vtruncbuf_bp_trunc, &truncloffset);
533 if (count) {
6ea70f76 534 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
c4b46cb4
MD
535 "left over buffers in %s\n", count, filename);
536 }
537 } while(count);
984263bc 538
b1f72a5c
MD
539 crit_exit();
540
984263bc
MD
541 return (0);
542}
543
544/*
6bae6177
MD
545 * The callback buffer is beyond the new file EOF and must be destroyed.
546 * Note that the compare function must conform to the RB_SCAN's requirements.
547 */
548static
549int
550vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
551{
54078292 552 if (bp->b_loffset >= *(off_t *)data)
6bae6177
MD
553 return(0);
554 return(-1);
555}
556
557static
558int
559vtruncbuf_bp_trunc(struct buf *bp, void *data)
560{
561 /*
562 * Do not try to use a buffer we cannot immediately lock, but sleep
563 * anyway to prevent a livelock. The code will loop until all buffers
564 * can be acted upon.
565 */
566 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
567 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
568 BUF_UNLOCK(bp);
569 } else {
570 bremfree(bp);
135bd6a8 571 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
6bae6177
MD
572 bp->b_flags &= ~B_ASYNC;
573 brelse(bp);
574 }
575 return(1);
576}
577
578/*
579 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
54078292 580 * blocks (with a negative loffset) are scanned.
6bae6177
MD
581 * Note that the compare function must conform to the RB_SCAN's requirements.
582 */
583static int
584vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
585{
54078292 586 if (bp->b_loffset < 0)
6bae6177
MD
587 return(0);
588 return(1);
589}
590
591static int
592vtruncbuf_bp_metasync(struct buf *bp, void *data)
593{
594 struct vnode *vp = data;
595
596 if (bp->b_flags & B_DELWRI) {
597 /*
598 * Do not try to use a buffer we cannot immediately lock,
599 * but sleep anyway to prevent a livelock. The code will
600 * loop until all buffers can be acted upon.
601 */
602 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
603 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
604 BUF_UNLOCK(bp);
605 } else {
606 bremfree(bp);
607 if (bp->b_vp == vp) {
608 bp->b_flags |= B_ASYNC;
609 } else {
610 bp->b_flags &= ~B_ASYNC;
611 }
62cfda27 612 bwrite(bp);
6bae6177
MD
613 }
614 return(1);
615 } else {
616 return(0);
617 }
618}
619
620/*
621 * vfsync - implements a multipass fsync on a file which understands
622 * dependancies and meta-data. The passed vnode must be locked. The
623 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
624 *
625 * When fsyncing data asynchronously just do one consolidated pass starting
626 * with the most negative block number. This may not get all the data due
627 * to dependancies.
628 *
629 * When fsyncing data synchronously do a data pass, then a metadata pass,
630 * then do additional data+metadata passes to try to get all the data out.
631 */
632static int vfsync_wait_output(struct vnode *vp,
633 int (*waitoutput)(struct vnode *, struct thread *));
634static int vfsync_data_only_cmp(struct buf *bp, void *data);
635static int vfsync_meta_only_cmp(struct buf *bp, void *data);
636static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
637static int vfsync_bp(struct buf *bp, void *data);
638
639struct vfsync_info {
640 struct vnode *vp;
641 int synchronous;
642 int syncdeps;
643 int lazycount;
644 int lazylimit;
54078292 645 int skippedbufs;
6bae6177
MD
646 int (*checkdef)(struct buf *);
647};
648
649int
4e0ecc94 650vfsync(struct vnode *vp, int waitfor, int passes,
6bae6177
MD
651 int (*checkdef)(struct buf *),
652 int (*waitoutput)(struct vnode *, struct thread *))
653{
654 struct vfsync_info info;
655 int error;
656
657 bzero(&info, sizeof(info));
658 info.vp = vp;
6bae6177
MD
659 if ((info.checkdef = checkdef) == NULL)
660 info.syncdeps = 1;
661
a975cf3a 662 crit_enter_id("vfsync");
e43a034f 663
6bae6177
MD
664 switch(waitfor) {
665 case MNT_LAZY:
666 /*
667 * Lazy (filesystem syncer typ) Asynchronous plus limit the
668 * number of data (not meta) pages we try to flush to 1MB.
669 * A non-zero return means that lazy limit was reached.
670 */
671 info.lazylimit = 1024 * 1024;
672 info.syncdeps = 1;
673 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
674 vfsync_lazy_range_cmp, vfsync_bp, &info);
675 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
676 vfsync_meta_only_cmp, vfsync_bp, &info);
677 if (error == 0)
678 vp->v_lazyw = 0;
679 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
680 vn_syncer_add_to_worklist(vp, 1);
681 error = 0;
682 break;
683 case MNT_NOWAIT:
684 /*
685 * Asynchronous. Do a data-only pass and a meta-only pass.
686 */
687 info.syncdeps = 1;
688 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
689 vfsync_bp, &info);
690 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
691 vfsync_bp, &info);
692 error = 0;
693 break;
694 default:
695 /*
696 * Synchronous. Do a data-only pass, then a meta-data+data
697 * pass, then additional integrated passes to try to get
698 * all the dependancies flushed.
699 */
700 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
701 vfsync_bp, &info);
702 error = vfsync_wait_output(vp, waitoutput);
703 if (error == 0) {
54078292 704 info.skippedbufs = 0;
6bae6177
MD
705 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
706 vfsync_bp, &info);
707 error = vfsync_wait_output(vp, waitoutput);
54078292 708 if (info.skippedbufs)
6ea70f76 709 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
6bae6177
MD
710 }
711 while (error == 0 && passes > 0 &&
712 !RB_EMPTY(&vp->v_rbdirty_tree)) {
713 if (--passes == 0) {
714 info.synchronous = 1;
715 info.syncdeps = 1;
716 }
717 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
718 vfsync_bp, &info);
719 if (error < 0)
720 error = -error;
721 info.syncdeps = 1;
722 if (error == 0)
723 error = vfsync_wait_output(vp, waitoutput);
724 }
725 break;
726 }
a975cf3a 727 crit_exit_id("vfsync");
6bae6177
MD
728 return(error);
729}
730
731static int
732vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
733{
734 int error = 0;
735
81b5c339
MD
736 while (vp->v_track_write.bk_active) {
737 vp->v_track_write.bk_waitflag = 1;
738 tsleep(&vp->v_track_write, 0, "fsfsn", 0);
6bae6177
MD
739 }
740 if (waitoutput)
741 error = waitoutput(vp, curthread);
742 return(error);
743}
744
745static int
746vfsync_data_only_cmp(struct buf *bp, void *data)
747{
54078292 748 if (bp->b_loffset < 0)
6bae6177
MD
749 return(-1);
750 return(0);
751}
752
753static int
754vfsync_meta_only_cmp(struct buf *bp, void *data)
755{
54078292 756 if (bp->b_loffset < 0)
6bae6177
MD
757 return(0);
758 return(1);
759}
760
761static int
762vfsync_lazy_range_cmp(struct buf *bp, void *data)
763{
764 struct vfsync_info *info = data;
54078292 765 if (bp->b_loffset < info->vp->v_lazyw)
6bae6177
MD
766 return(-1);
767 return(0);
768}
769
770static int
771vfsync_bp(struct buf *bp, void *data)
772{
773 struct vfsync_info *info = data;
774 struct vnode *vp = info->vp;
775 int error;
776
777 /*
778 * if syncdeps is not set we do not try to write buffers which have
779 * dependancies.
780 */
781 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
782 return(0);
783
784 /*
785 * Ignore buffers that we cannot immediately lock. XXX
786 */
54078292 787 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
6ea70f76 788 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
54078292 789 ++info->skippedbufs;
6bae6177 790 return(0);
54078292 791 }
6bae6177
MD
792 if ((bp->b_flags & B_DELWRI) == 0)
793 panic("vfsync_bp: buffer not dirty");
794 if (vp != bp->b_vp)
795 panic("vfsync_bp: buffer vp mismatch");
796
797 /*
798 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
799 * has been written but an additional handshake with the device
800 * is required before we can dispose of the buffer. We have no idea
801 * how to do this so we have to skip these buffers.
802 */
803 if (bp->b_flags & B_NEEDCOMMIT) {
804 BUF_UNLOCK(bp);
805 return(0);
806 }
807
17a8ba12
MD
808 /*
809 * Ask bioops if it is ok to sync
810 */
811 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
812 bremfree(bp);
813 brelse(bp);
814 return(0);
815 }
816
6bae6177
MD
817 if (info->synchronous) {
818 /*
819 * Synchronous flushing. An error may be returned.
820 */
821 bremfree(bp);
b1f72a5c 822 crit_exit_id("vfsync");
6bae6177 823 error = bwrite(bp);
b1f72a5c 824 crit_enter_id("vfsync");
6bae6177
MD
825 } else {
826 /*
827 * Asynchronous flushing. A negative return value simply
828 * stops the scan and is not considered an error. We use
829 * this to support limited MNT_LAZY flushes.
830 */
54078292 831 vp->v_lazyw = bp->b_loffset;
6bae6177 832 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
6bae6177
MD
833 info->lazycount += vfs_bio_awrite(bp);
834 } else {
835 info->lazycount += bp->b_bufsize;
836 bremfree(bp);
b1f72a5c 837 crit_exit_id("vfsync");
6bae6177 838 bawrite(bp);
b1f72a5c 839 crit_enter_id("vfsync");
6bae6177
MD
840 }
841 if (info->lazylimit && info->lazycount >= info->lazylimit)
842 error = 1;
843 else
844 error = 0;
845 }
846 return(-error);
847}
848
849/*
984263bc
MD
850 * Associate a buffer with a vnode.
851 */
852void
dd98570a 853bgetvp(struct vnode *vp, struct buf *bp)
984263bc 854{
984263bc 855 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
9e45bec7 856 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
984263bc
MD
857
858 vhold(vp);
984263bc
MD
859 /*
860 * Insert onto list for new vnode.
861 */
5fd012e0 862 crit_enter();
1f1ea522
MD
863 bp->b_vp = vp;
864 bp->b_flags |= B_HASHED;
865 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp))
866 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
867
9e45bec7 868 bp->b_flags |= B_VNCLEAN;
6bae6177 869 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
1f1ea522 870 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
5fd012e0 871 crit_exit();
984263bc
MD
872}
873
874/*
875 * Disassociate a buffer from a vnode.
876 */
877void
dd98570a 878brelvp(struct buf *bp)
984263bc
MD
879{
880 struct vnode *vp;
984263bc
MD
881
882 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
883
884 /*
885 * Delete from old vnode list, if on one.
886 */
887 vp = bp->b_vp;
5fd012e0 888 crit_enter();
9e45bec7
MD
889 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
890 if (bp->b_flags & B_VNDIRTY)
6bae6177
MD
891 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
892 else
893 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
9e45bec7 894 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
984263bc 895 }
1f1ea522
MD
896 if (bp->b_flags & B_HASHED) {
897 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
898 bp->b_flags &= ~B_HASHED;
899 }
6bae6177 900 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
984263bc
MD
901 vp->v_flag &= ~VONWORKLST;
902 LIST_REMOVE(vp, v_synclist);
903 }
f5865223 904 crit_exit();
5fd012e0
MD
905 bp->b_vp = NULL;
906 vdrop(vp);
984263bc
MD
907}
908
909/*
1f1ea522
MD
910 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
911 * This routine is called when the state of the B_DELWRI bit is changed.
984263bc
MD
912 */
913void
1f1ea522 914reassignbuf(struct buf *bp)
984263bc 915{
1f1ea522 916 struct vnode *vp = bp->b_vp;
984263bc 917 int delay;
984263bc 918
1f1ea522 919 KKASSERT(vp != NULL);
984263bc
MD
920 ++reassignbufcalls;
921
922 /*
923 * B_PAGING flagged buffers cannot be reassigned because their vp
924 * is not fully linked in.
925 */
926 if (bp->b_flags & B_PAGING)
927 panic("cannot reassign paging buffer");
928
5fd012e0 929 crit_enter();
984263bc 930 if (bp->b_flags & B_DELWRI) {
1f1ea522
MD
931 /*
932 * Move to the dirty list, add the vnode to the worklist
933 */
9e45bec7 934 if (bp->b_flags & B_VNCLEAN) {
1f1ea522 935 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
9e45bec7 936 bp->b_flags &= ~B_VNCLEAN;
1f1ea522 937 }
9e45bec7 938 if ((bp->b_flags & B_VNDIRTY) == 0) {
1f1ea522
MD
939 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
940 panic("reassignbuf: dup lblk vp %p bp %p",
941 vp, bp);
942 }
9e45bec7 943 bp->b_flags |= B_VNDIRTY;
1f1ea522
MD
944 }
945 if ((vp->v_flag & VONWORKLST) == 0) {
946 switch (vp->v_type) {
984263bc
MD
947 case VDIR:
948 delay = dirdelay;
949 break;
950 case VCHR:
951 case VBLK:
1f1ea522
MD
952 if (vp->v_rdev &&
953 vp->v_rdev->si_mountpoint != NULL) {
984263bc
MD
954 delay = metadelay;
955 break;
956 }
957 /* fall through */
958 default:
959 delay = filedelay;
960 }
1f1ea522 961 vn_syncer_add_to_worklist(vp, delay);
984263bc 962 }
984263bc 963 } else {
1f1ea522
MD
964 /*
965 * Move to the clean list, remove the vnode from the worklist
966 * if no dirty blocks remain.
967 */
9e45bec7 968 if (bp->b_flags & B_VNDIRTY) {
1f1ea522 969 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
9e45bec7 970 bp->b_flags &= ~B_VNDIRTY;
1f1ea522 971 }
9e45bec7 972 if ((bp->b_flags & B_VNCLEAN) == 0) {
1f1ea522
MD
973 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
974 panic("reassignbuf: dup lblk vp %p bp %p",
975 vp, bp);
976 }
9e45bec7 977 bp->b_flags |= B_VNCLEAN;
1f1ea522
MD
978 }
979 if ((vp->v_flag & VONWORKLST) &&
980 RB_EMPTY(&vp->v_rbdirty_tree)) {
981 vp->v_flag &= ~VONWORKLST;
982 LIST_REMOVE(vp, v_synclist);
984263bc 983 }
984263bc 984 }
5fd012e0 985 crit_exit();
984263bc
MD
986}
987
988/*
989 * Create a vnode for a block device.
990 * Used for mounting the root file system.
991 */
992int
b13267a5 993bdevvp(cdev_t dev, struct vnode **vpp)
984263bc 994{
1fd87d54 995 struct vnode *vp;
984263bc
MD
996 struct vnode *nvp;
997 int error;
998
028066b1 999 if (dev == NULL) {
984263bc
MD
1000 *vpp = NULLVP;
1001 return (ENXIO);
1002 }
66a1ddf5 1003 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0);
984263bc
MD
1004 if (error) {
1005 *vpp = NULLVP;
1006 return (error);
1007 }
1008 vp = nvp;
e4c9c0c8 1009 vp->v_type = VCHR;
0e9b9130
MD
1010 vp->v_umajor = dev->si_umajor;
1011 vp->v_uminor = dev->si_uminor;
5fd012e0 1012 vx_unlock(vp);
984263bc
MD
1013 *vpp = vp;
1014 return (0);
5fd012e0 1015}
41a01a4d 1016
984263bc 1017int
b13267a5 1018v_associate_rdev(struct vnode *vp, cdev_t dev)
984263bc 1019{
5fd012e0 1020 lwkt_tokref ilock;
984263bc 1021
0e9b9130 1022 if (dev == NULL)
5fd012e0
MD
1023 return(ENXIO);
1024 if (dev_is_good(dev) == 0)
1025 return(ENXIO);
1026 KKASSERT(vp->v_rdev == NULL);
1027 if (dev_ref_debug)
6ea70f76 1028 kprintf("Z1");
5fd012e0
MD
1029 vp->v_rdev = reference_dev(dev);
1030 lwkt_gettoken(&ilock, &spechash_token);
0de08e6d 1031 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
5fd012e0
MD
1032 lwkt_reltoken(&ilock);
1033 return(0);
1034}
984263bc 1035
5fd012e0
MD
1036void
1037v_release_rdev(struct vnode *vp)
1038{
1039 lwkt_tokref ilock;
b13267a5 1040 cdev_t dev;
984263bc 1041
5fd012e0
MD
1042 if ((dev = vp->v_rdev) != NULL) {
1043 lwkt_gettoken(&ilock, &spechash_token);
0de08e6d 1044 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
5fd012e0 1045 vp->v_rdev = NULL;
5fd012e0
MD
1046 release_dev(dev);
1047 lwkt_reltoken(&ilock);
984263bc 1048 }
984263bc
MD
1049}
1050
1051/*
b13267a5 1052 * Add a vnode to the alias list hung off the cdev_t. We only associate
5fd012e0
MD
1053 * the device number with the vnode. The actual device is not associated
1054 * until the vnode is opened (usually in spec_open()), and will be
1055 * disassociated on last close.
984263bc 1056 */
5fd012e0 1057void
0e9b9130 1058addaliasu(struct vnode *nvp, int x, int y)
984263bc 1059{
5fd012e0
MD
1060 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1061 panic("addaliasu on non-special vnode");
0e9b9130
MD
1062 nvp->v_umajor = x;
1063 nvp->v_uminor = y;
984263bc
MD
1064}
1065
1066/*
cf683bae
MD
1067 * Simple call that a filesystem can make to try to get rid of a
1068 * vnode. It will fail if anyone is referencing the vnode (including
1069 * the caller).
1070 *
1071 * The filesystem can check whether its in-memory inode structure still
1072 * references the vp on return.
1073 */
1074void
1075vclean_unlocked(struct vnode *vp)
1076{
1077 vx_get(vp);
1078 if (sysref_isactive(&vp->v_sysref) == 0)
2b4ed70b 1079 vgone_vxlocked(vp);
cf683bae
MD
1080 vx_put(vp);
1081}
1082
1083/*
5fd012e0
MD
1084 * Disassociate a vnode from its underlying filesystem.
1085 *
3c37c940
MD
1086 * The vnode must be VX locked and referenced. In all normal situations
1087 * there are no active references. If vclean_vxlocked() is called while
1088 * there are active references, the vnode is being ripped out and we have
1089 * to call VOP_CLOSE() as appropriate before we can reclaim it.
984263bc 1090 */
5fd012e0 1091void
3c37c940 1092vclean_vxlocked(struct vnode *vp, int flags)
984263bc
MD
1093{
1094 int active;
8ddc6004 1095 int n;
7540ab49 1096 vm_object_t object;
984263bc
MD
1097
1098 /*
5fd012e0 1099 * If the vnode has already been reclaimed we have nothing to do.
984263bc 1100 */
3c37c940 1101 if (vp->v_flag & VRECLAIMED)
5fd012e0
MD
1102 return;
1103 vp->v_flag |= VRECLAIMED;
984263bc
MD
1104
1105 /*
5fd012e0 1106 * Scrap the vfs cache
984263bc 1107 */
6b008938 1108 while (cache_inval_vp(vp, 0) != 0) {
6ea70f76 1109 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
25cb3304
MD
1110 tsleep(vp, 0, "vclninv", 2);
1111 }
41a01a4d 1112
984263bc 1113 /*
5fd012e0
MD
1114 * Check to see if the vnode is in use. If so we have to reference it
1115 * before we clean it out so that its count cannot fall to zero and
1116 * generate a race against ourselves to recycle it.
984263bc 1117 */
3c37c940 1118 active = sysref_isactive(&vp->v_sysref);
984263bc
MD
1119
1120 /*
5fd012e0 1121 * Clean out any buffers associated with the vnode and destroy its
7540ab49 1122 * object, if it has one.
984263bc 1123 */
87de5057 1124 vinvalbuf(vp, V_SAVE, 0, 0);
7540ab49 1125
984263bc 1126 /*
8ddc6004
MD
1127 * If purging an active vnode (typically during a forced unmount
1128 * or reboot), it must be closed and deactivated before being
1129 * reclaimed. This isn't really all that safe, but what can
1130 * we do? XXX.
5fd012e0
MD
1131 *
1132 * Note that neither of these routines unlocks the vnode.
984263bc 1133 */
8ddc6004
MD
1134 if (active && (flags & DOCLOSE)) {
1135 while ((n = vp->v_opencount) != 0) {
1136 if (vp->v_writecount)
87de5057 1137 VOP_CLOSE(vp, FWRITE|FNONBLOCK);
8ddc6004 1138 else
87de5057 1139 VOP_CLOSE(vp, FNONBLOCK);
8ddc6004 1140 if (vp->v_opencount == n) {
6ea70f76 1141 kprintf("Warning: unable to force-close"
8ddc6004
MD
1142 " vnode %p\n", vp);
1143 break;
1144 }
1145 }
5fd012e0
MD
1146 }
1147
1148 /*
64e0b2d3 1149 * If the vnode has not been deactivated, deactivated it. Deactivation
e3bc9a94
MD
1150 * can create new buffers and VM pages so we have to call vinvalbuf()
1151 * again to make sure they all get flushed.
1152 *
1153 * This can occur if a file with a link count of 0 needs to be
1154 * truncated.
5fd012e0
MD
1155 */
1156 if ((vp->v_flag & VINACTIVE) == 0) {
1157 vp->v_flag |= VINACTIVE;
87de5057 1158 VOP_INACTIVE(vp);
e3bc9a94
MD
1159 vinvalbuf(vp, V_SAVE, 0, 0);
1160 }
1161
1162 /*
1163 * If the vnode has an object, destroy it.
1164 */
1165 if ((object = vp->v_object) != NULL) {
1166 if (object->ref_count == 0) {
1167 if ((object->flags & OBJ_DEAD) == 0)
1168 vm_object_terminate(object);
1169 } else {
1170 vm_pager_deallocate(object);
1171 }
1172 vp->v_flag &= ~VOBJBUF;
984263bc 1173 }
e3bc9a94
MD
1174 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1175
984263bc
MD
1176 /*
1177 * Reclaim the vnode.
1178 */
87de5057 1179 if (VOP_RECLAIM(vp))
984263bc
MD
1180 panic("vclean: cannot reclaim");
1181
984263bc
MD
1182 /*
1183 * Done with purge, notify sleepers of the grim news.
1184 */
66a1ddf5 1185 vp->v_ops = &dead_vnode_vops_p;
984263bc
MD
1186 vn_pollgone(vp);
1187 vp->v_tag = VT_NON;
64e0b2d3
MD
1188
1189 /*
1190 * If we are destroying an active vnode, reactivate it now that
1191 * we have reassociated it with deadfs. This prevents the system
1192 * from crashing on the vnode due to it being unexpectedly marked
1193 * as inactive or reclaimed.
1194 */
1195 if (active && (flags & DOCLOSE)) {
1196 vp->v_flag &= ~(VINACTIVE|VRECLAIMED);
1197 }
984263bc
MD
1198}
1199
1200/*
1201 * Eliminate all activity associated with the requested vnode
1202 * and with all vnodes aliased to the requested vnode.
dd98570a 1203 *
b8477cda 1204 * The vnode must be referenced but should not be locked.
984263bc
MD
1205 */
1206int
b8477cda 1207vrevoke(struct vnode *vp, struct ucred *cred)
984263bc 1208{
b8477cda 1209 struct vnode *vq;
41a01a4d 1210 lwkt_tokref ilock;
b13267a5 1211 cdev_t dev;
b8477cda 1212 int error;
e4c9c0c8
MD
1213
1214 /*
1215 * If the vnode has a device association, scrap all vnodes associated
1216 * with the device. Don't let the device disappear on us while we
1217 * are scrapping the vnodes.
5fd012e0
MD
1218 *
1219 * The passed vp will probably show up in the list, do not VX lock
1220 * it twice!
e4c9c0c8 1221 */
b8477cda
MD
1222 if (vp->v_type != VCHR) {
1223 error = fdrevoke(vp, DTYPE_VNODE, cred);
1224 return (error);
1225 }
e4c9c0c8 1226 if ((dev = vp->v_rdev) == NULL) {
0e9b9130 1227 if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL)
e4c9c0c8
MD
1228 return(0);
1229 }
1230 reference_dev(dev);
5fd012e0
MD
1231 lwkt_gettoken(&ilock, &spechash_token);
1232 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
b8477cda
MD
1233 vref(vq);
1234 fdrevoke(vq, DTYPE_VNODE, cred);
1235 v_release_rdev(vq);
1236 vrele(vq);
984263bc 1237 }
5fd012e0 1238 lwkt_reltoken(&ilock);
e4c9c0c8 1239 release_dev(dev);
984263bc
MD
1240 return (0);
1241}
1242
1243/*
3c37c940
MD
1244 * This is called when the object underlying a vnode is being destroyed,
1245 * such as in a remove(). Try to recycle the vnode immediately if the
1246 * only active reference is our reference.
c0c70b27
MD
1247 *
1248 * Directory vnodes in the namecache with children cannot be immediately
1249 * recycled because numerous VOP_N*() ops require them to be stable.
984263bc
MD
1250 */
1251int
87de5057 1252vrecycle(struct vnode *vp)
984263bc 1253{
5954ac07 1254 if (vp->v_sysref.refcnt <= 1) {
c0c70b27
MD
1255 if (cache_inval_vp_nonblock(vp))
1256 return(0);
3c37c940 1257 vgone_vxlocked(vp);
984263bc
MD
1258 return (1);
1259 }
984263bc
MD
1260 return (0);
1261}
1262
1263/*
2ec4b00d
MD
1264 * Return the maximum I/O size allowed for strategy calls on VP.
1265 *
1266 * If vp is VCHR or VBLK we dive the device, otherwise we use
1267 * the vp's mount info.
1268 */
1269int
1270vmaxiosize(struct vnode *vp)
1271{
1272 if (vp->v_type == VBLK || vp->v_type == VCHR) {
1273 return(vp->v_rdev->si_iosize_max);
1274 } else {
1275 return(vp->v_mount->mnt_iosize_max);
1276 }
1277}
1278
1279/*
5fd012e0
MD
1280 * Eliminate all activity associated with a vnode in preparation for reuse.
1281 *
57ac0c99
MD
1282 * The vnode must be VX locked and refd and will remain VX locked and refd
1283 * on return. This routine may be called with the vnode in any state, as
1284 * long as it is VX locked. The vnode will be cleaned out and marked
1285 * VRECLAIMED but will not actually be reused until all existing refs and
1286 * holds go away.
5fd012e0
MD
1287 *
1288 * NOTE: This routine may be called on a vnode which has not yet been
1289 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1290 * already been reclaimed.
1291 *
1292 * This routine is not responsible for placing us back on the freelist.
1293 * Instead, it happens automatically when the caller releases the VX lock
1294 * (assuming there aren't any other references).
984263bc 1295 */
e3332475
MD
1296
1297void
3c37c940 1298vgone_vxlocked(struct vnode *vp)
e3332475 1299{
984263bc 1300 /*
5fd012e0 1301 * assert that the VX lock is held. This is an absolute requirement
3c37c940 1302 * now for vgone_vxlocked() to be called.
984263bc 1303 */
5fd012e0 1304 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
984263bc
MD
1305
1306 /*
5fd012e0 1307 * Clean out the filesystem specific data and set the VRECLAIMED
e3332475 1308 * bit. Also deactivate the vnode if necessary.
984263bc 1309 */
3c37c940 1310 vclean_vxlocked(vp, DOCLOSE);
984263bc
MD
1311
1312 /*
1313 * Delete from old mount point vnode list, if on one.
1314 */
1315 if (vp->v_mount != NULL)
5fd012e0 1316 insmntque(vp, NULL);
dd98570a 1317
984263bc
MD
1318 /*
1319 * If special device, remove it from special device alias list
1fbb5fc0
MD
1320 * if it is on one. This should normally only occur if a vnode is
1321 * being revoked as the device should otherwise have been released
1322 * naturally.
984263bc
MD
1323 */
1324 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
e4c9c0c8 1325 v_release_rdev(vp);
984263bc
MD
1326 }
1327
1328 /*
5fd012e0 1329 * Set us to VBAD
984263bc 1330 */
984263bc 1331 vp->v_type = VBAD;
984263bc
MD
1332}
1333
1334/*
1335 * Lookup a vnode by device number.
3875f5b0
MD
1336 *
1337 * Returns non-zero and *vpp set to a vref'd vnode on success.
1338 * Returns zero on failure.
984263bc
MD
1339 */
1340int
b13267a5 1341vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
984263bc 1342{
41a01a4d 1343 lwkt_tokref ilock;
984263bc
MD
1344 struct vnode *vp;
1345
41a01a4d 1346 lwkt_gettoken(&ilock, &spechash_token);
0de08e6d 1347 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
984263bc
MD
1348 if (type == vp->v_type) {
1349 *vpp = vp;
3875f5b0 1350 vref(vp);
41a01a4d 1351 lwkt_reltoken(&ilock);
984263bc
MD
1352 return (1);
1353 }
1354 }
41a01a4d 1355 lwkt_reltoken(&ilock);
984263bc
MD
1356 return (0);
1357}
1358
1359/*
e4c9c0c8
MD
1360 * Calculate the total number of references to a special device. This
1361 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
028066b1 1362 * an overloaded field. Since udev2dev can now return NULL, we have
e4c9c0c8 1363 * to check for a NULL v_rdev.
984263bc
MD
1364 */
1365int
b13267a5 1366count_dev(cdev_t dev)
984263bc 1367{
41a01a4d 1368 lwkt_tokref ilock;
e4c9c0c8
MD
1369 struct vnode *vp;
1370 int count = 0;
984263bc 1371
e4c9c0c8
MD
1372 if (SLIST_FIRST(&dev->si_hlist)) {
1373 lwkt_gettoken(&ilock, &spechash_token);
0de08e6d 1374 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
3c37c940
MD
1375 if (vp->v_sysref.refcnt > 0)
1376 count += vp->v_sysref.refcnt;
e4c9c0c8
MD
1377 }
1378 lwkt_reltoken(&ilock);
1379 }
1380 return(count);
984263bc
MD
1381}
1382
984263bc 1383int
0e9b9130 1384count_udev(int x, int y)
984263bc 1385{
b13267a5 1386 cdev_t dev;
984263bc 1387
0e9b9130 1388 if ((dev = get_dev(x, y)) == NULL)
e4c9c0c8
MD
1389 return(0);
1390 return(count_dev(dev));
1391}
1392
1393int
1394vcount(struct vnode *vp)
1395{
1396 if (vp->v_rdev == NULL)
1397 return(0);
1398 return(count_dev(vp->v_rdev));
984263bc
MD
1399}
1400
1401/*
1c843a13
MD
1402 * Initialize VMIO for a vnode. This routine MUST be called before a
1403 * VFS can issue buffer cache ops on a vnode. It is typically called
1404 * when a vnode is initialized from its inode.
7540ab49
MD
1405 */
1406int
1c843a13 1407vinitvmio(struct vnode *vp, off_t filesize)
7540ab49 1408{
7540ab49
MD
1409 vm_object_t object;
1410 int error = 0;
1411
1412retry:
1413 if ((object = vp->v_object) == NULL) {
1c843a13 1414 object = vnode_pager_alloc(vp, filesize, 0, 0);
7540ab49
MD
1415 /*
1416 * Dereference the reference we just created. This assumes
1417 * that the object is associated with the vp.
1418 */
1419 object->ref_count--;
3c37c940 1420 vrele(vp);
7540ab49
MD
1421 } else {
1422 if (object->flags & OBJ_DEAD) {
a11aaa81 1423 vn_unlock(vp);
9e12ff11 1424 vm_object_dead_sleep(object, "vodead");
ca466bae 1425 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
7540ab49
MD
1426 goto retry;
1427 }
1428 }
1429 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1430 vp->v_flag |= VOBJBUF;
7540ab49
MD
1431 return (error);
1432}
1433
1434
1435/*
984263bc
MD
1436 * Print out a description of a vnode.
1437 */
1438static char *typename[] =
1439{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1440
1441void
dd98570a 1442vprint(char *label, struct vnode *vp)
984263bc
MD
1443{
1444 char buf[96];
1445
1446 if (label != NULL)
6ea70f76 1447 kprintf("%s: %p: ", label, (void *)vp);
984263bc 1448 else
6ea70f76 1449 kprintf("%p: ", (void *)vp);
3c37c940
MD
1450 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1451 typename[vp->v_type],
1452 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
984263bc
MD
1453 buf[0] = '\0';
1454 if (vp->v_flag & VROOT)
1455 strcat(buf, "|VROOT");
67863d04
MD
1456 if (vp->v_flag & VPFSROOT)
1457 strcat(buf, "|VPFSROOT");
984263bc
MD
1458 if (vp->v_flag & VTEXT)
1459 strcat(buf, "|VTEXT");
1460 if (vp->v_flag & VSYSTEM)
1461 strcat(buf, "|VSYSTEM");
984263bc
MD
1462 if (vp->v_flag & VFREE)
1463 strcat(buf, "|VFREE");
1464 if (vp->v_flag & VOBJBUF)
1465 strcat(buf, "|VOBJBUF");
1466 if (buf[0] != '\0')
6ea70f76 1467 kprintf(" flags (%s)", &buf[1]);
984263bc 1468 if (vp->v_data == NULL) {
6ea70f76 1469 kprintf("\n");
984263bc 1470 } else {
6ea70f76 1471 kprintf("\n\t");
984263bc
MD
1472 VOP_PRINT(vp);
1473 }
1474}
1475
1476#ifdef DDB
1477#include <ddb/ddb.h>
861905fb
MD
1478
1479static int db_show_locked_vnodes(struct mount *mp, void *data);
1480
984263bc
MD
1481/*
1482 * List all of the locked vnodes in the system.
1483 * Called when debugging the kernel.
1484 */
1485DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1486{
6ea70f76 1487 kprintf("Locked vnodes\n");
861905fb
MD
1488 mountlist_scan(db_show_locked_vnodes, NULL,
1489 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1490}
1491
1492static int
1493db_show_locked_vnodes(struct mount *mp, void *data __unused)
1494{
984263bc
MD
1495 struct vnode *vp;
1496
861905fb 1497 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
a11aaa81 1498 if (vn_islocked(vp))
60233e58 1499 vprint(NULL, vp);
984263bc 1500 }
861905fb 1501 return(0);
984263bc
MD
1502}
1503#endif
1504
1505/*
1506 * Top level filesystem related information gathering.
1507 */
402ed7e1 1508static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
984263bc
MD
1509
1510static int
1511vfs_sysctl(SYSCTL_HANDLER_ARGS)
1512{
1513 int *name = (int *)arg1 - 1; /* XXX */
1514 u_int namelen = arg2 + 1; /* XXX */
1515 struct vfsconf *vfsp;
2613053d 1516 int maxtypenum;
984263bc
MD
1517
1518#if 1 || defined(COMPAT_PRELITE2)
1519 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1520 if (namelen == 1)
1521 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1522#endif
1523
1524#ifdef notyet
1525 /* all sysctl names at this level are at least name and field */
1526 if (namelen < 2)
1527 return (ENOTDIR); /* overloaded */
1528 if (name[0] != VFS_GENERIC) {
2613053d 1529 vfsp = vfsconf_find_by_typenum(name[0]);
984263bc
MD
1530 if (vfsp == NULL)
1531 return (EOPNOTSUPP);
1532 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1533 oldp, oldlenp, newp, newlen, p));
1534 }
1535#endif
1536 switch (name[1]) {
1537 case VFS_MAXTYPENUM:
1538 if (namelen != 2)
1539 return (ENOTDIR);
2613053d
MN
1540 maxtypenum = vfsconf_get_maxtypenum();
1541 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
984263bc
MD
1542 case VFS_CONF:
1543 if (namelen != 3)
1544 return (ENOTDIR); /* overloaded */
2613053d 1545 vfsp = vfsconf_find_by_typenum(name[2]);
984263bc
MD
1546 if (vfsp == NULL)
1547 return (EOPNOTSUPP);
1548 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1549 }
1550 return (EOPNOTSUPP);
1551}
1552
1553SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1554 "Generic filesystem");
1555
1556#if 1 || defined(COMPAT_PRELITE2)
1557
1558static int
2613053d 1559sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
984263bc
MD
1560{
1561 int error;
984263bc 1562 struct ovfsconf ovfs;
2613053d
MN
1563 struct sysctl_req *req = (struct sysctl_req*) data;
1564
1565 bzero(&ovfs, sizeof(ovfs));
1566 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1567 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1568 ovfs.vfc_index = vfsp->vfc_typenum;
1569 ovfs.vfc_refcount = vfsp->vfc_refcount;
1570 ovfs.vfc_flags = vfsp->vfc_flags;
1571 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1572 if (error)
1573 return error; /* abort iteration with error code */
1574 else
1575 return 0; /* continue iterating with next element */
1576}
984263bc 1577
2613053d
MN
1578static int
1579sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1580{
1581 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
984263bc
MD
1582}
1583
1584#endif /* 1 || COMPAT_PRELITE2 */
1585
984263bc
MD
1586/*
1587 * Check to see if a filesystem is mounted on a block device.
1588 */
1589int
e4c9c0c8 1590vfs_mountedon(struct vnode *vp)
984263bc 1591{
b13267a5 1592 cdev_t dev;
984263bc 1593
0e9b9130
MD
1594 if ((dev = vp->v_rdev) == NULL) {
1595 if (vp->v_type != VBLK)
1596 dev = get_dev(vp->v_uminor, vp->v_umajor);
1597 }
028066b1 1598 if (dev != NULL && dev->si_mountpoint)
984263bc
MD
1599 return (EBUSY);
1600 return (0);
1601}
1602
1603/*
1604 * Unmount all filesystems. The list is traversed in reverse order
1605 * of mounting to avoid dependencies.
1606 */
861905fb
MD
1607
1608static int vfs_umountall_callback(struct mount *mp, void *data);
1609
984263bc 1610void
dd98570a 1611vfs_unmountall(void)
984263bc 1612{
861905fb 1613 int count;
984263bc 1614
861905fb
MD
1615 do {
1616 count = mountlist_scan(vfs_umountall_callback,
acde96db 1617 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
861905fb
MD
1618 } while (count);
1619}
1620
1621static
1622int
1623vfs_umountall_callback(struct mount *mp, void *data)
1624{
861905fb
MD
1625 int error;
1626
acde96db 1627 error = dounmount(mp, MNT_FORCE);
861905fb
MD
1628 if (error) {
1629 mountlist_remove(mp);
6ea70f76 1630 kprintf("unmount of filesystem mounted from %s failed (",
861905fb
MD
1631 mp->mnt_stat.f_mntfromname);
1632 if (error == EBUSY)
6ea70f76 1633 kprintf("BUSY)\n");
861905fb 1634 else
6ea70f76 1635 kprintf("%d)\n", error);
984263bc 1636 }
861905fb 1637 return(1);
984263bc
MD
1638}
1639
1640/*
1641 * Build hash lists of net addresses and hang them off the mount point.
1642 * Called by ufs_mount() to set up the lists of export addresses.
1643 */
1644static int
dd98570a 1645vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1aa89f17 1646 const struct export_args *argp)
984263bc 1647{
1fd87d54
RG
1648 struct netcred *np;
1649 struct radix_node_head *rnh;
1650 int i;
984263bc
MD
1651 struct radix_node *rn;
1652 struct sockaddr *saddr, *smask = 0;
1653 struct domain *dom;
1654 int error;
1655
1656 if (argp->ex_addrlen == 0) {
1657 if (mp->mnt_flag & MNT_DEFEXPORTED)
1658 return (EPERM);
1659 np = &nep->ne_defexported;
1660 np->netc_exflags = argp->ex_flags;
1661 np->netc_anon = argp->ex_anon;
1662 np->netc_anon.cr_ref = 1;
1663 mp->mnt_flag |= MNT_DEFEXPORTED;
1664 return (0);
1665 }
1666
0260ddf9
MD
1667 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1668 return (EINVAL);
1669 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
984263bc
MD
1670 return (EINVAL);
1671
1672 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
e7b4468c 1673 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
984263bc
MD
1674 saddr = (struct sockaddr *) (np + 1);
1675 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1676 goto out;
1677 if (saddr->sa_len > argp->ex_addrlen)
1678 saddr->sa_len = argp->ex_addrlen;
1679 if (argp->ex_masklen) {
dd98570a
MD
1680 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1681 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
984263bc
MD
1682 if (error)
1683 goto out;
1684 if (smask->sa_len > argp->ex_masklen)
1685 smask->sa_len = argp->ex_masklen;
1686 }
1687 i = saddr->sa_family;
1688 if ((rnh = nep->ne_rtable[i]) == 0) {
1689 /*
1690 * Seems silly to initialize every AF when most are not used,
1691 * do so on demand here
1692 */
9c70fe43 1693 SLIST_FOREACH(dom, &domains, dom_next)
984263bc
MD
1694 if (dom->dom_family == i && dom->dom_rtattach) {
1695 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1696 dom->dom_rtoffset);
1697 break;
1698 }
1699 if ((rnh = nep->ne_rtable[i]) == 0) {
1700 error = ENOBUFS;
1701 goto out;
1702 }
1703 }
2e9572df 1704 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
984263bc
MD
1705 np->netc_rnodes);
1706 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1707 error = EPERM;
1708 goto out;
1709 }
1710 np->netc_exflags = argp->ex_flags;
1711 np->netc_anon = argp->ex_anon;
1712 np->netc_anon.cr_ref = 1;
1713 return (0);
1714out:
efda3bd0 1715 kfree(np, M_NETADDR);
984263bc
MD
1716 return (error);
1717}
1718
1719/* ARGSUSED */
1720static int
dd98570a 1721vfs_free_netcred(struct radix_node *rn, void *w)
984263bc 1722{
1fd87d54 1723 struct radix_node_head *rnh = (struct radix_node_head *) w;
984263bc
MD
1724
1725 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
efda3bd0 1726 kfree((caddr_t) rn, M_NETADDR);
984263bc
MD
1727 return (0);
1728}
1729
1730/*
1731 * Free the net address hash lists that are hanging off the mount points.
1732 */
1733static void
dd98570a 1734vfs_free_addrlist(struct netexport *nep)
984263bc 1735{
1fd87d54
RG
1736 int i;
1737 struct radix_node_head *rnh;
984263bc
MD
1738
1739 for (i = 0; i <= AF_MAX; i++)
1740 if ((rnh = nep->ne_rtable[i])) {
1741 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1742 (caddr_t) rnh);
efda3bd0 1743 kfree((caddr_t) rnh, M_RTABLE);
984263bc
MD
1744 nep->ne_rtable[i] = 0;
1745 }
1746}
1747
1748int
1aa89f17
MD
1749vfs_export(struct mount *mp, struct netexport *nep,
1750 const struct export_args *argp)
984263bc
MD
1751{
1752 int error;
1753
1754 if (argp->ex_flags & MNT_DELEXPORT) {
1755 if (mp->mnt_flag & MNT_EXPUBLIC) {
1756 vfs_setpublicfs(NULL, NULL, NULL);
1757 mp->mnt_flag &= ~MNT_EXPUBLIC;
1758 }
1759 vfs_free_addrlist(nep);
1760 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1761 }
1762 if (argp->ex_flags & MNT_EXPORTED) {
1763 if (argp->ex_flags & MNT_EXPUBLIC) {
1764 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1765 return (error);
1766 mp->mnt_flag |= MNT_EXPUBLIC;
1767 }
1768 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1769 return (error);
1770 mp->mnt_flag |= MNT_EXPORTED;
1771 }
1772 return (0);
1773}
1774
1775
1776/*
1777 * Set the publicly exported filesystem (WebNFS). Currently, only
1778 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1779 */
1780int
dd98570a 1781vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1aa89f17 1782 const struct export_args *argp)
984263bc
MD
1783{
1784 int error;
1785 struct vnode *rvp;
1786 char *cp;
1787
1788 /*
1789 * mp == NULL -> invalidate the current info, the FS is
1790 * no longer exported. May be called from either vfs_export
1791 * or unmount, so check if it hasn't already been done.
1792 */
1793 if (mp == NULL) {
1794 if (nfs_pub.np_valid) {
1795 nfs_pub.np_valid = 0;
1796 if (nfs_pub.np_index != NULL) {
1797 FREE(nfs_pub.np_index, M_TEMP);
1798 nfs_pub.np_index = NULL;
1799 }
1800 }
1801 return (0);
1802 }
1803
1804 /*
1805 * Only one allowed at a time.
1806 */
1807 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1808 return (EBUSY);
1809
1810 /*
1811 * Get real filehandle for root of exported FS.
1812 */
1813 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1814 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1815
1816 if ((error = VFS_ROOT(mp, &rvp)))
1817 return (error);
1818
1819 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1820 return (error);
1821
1822 vput(rvp);
1823
1824 /*
1825 * If an indexfile was specified, pull it in.
1826 */
1827 if (argp->ex_indexfile != NULL) {
b80c9733
JS
1828 int namelen;
1829
1830 error = vn_get_namelen(rvp, &namelen);
1831 if (error)
1832 return (error);
1833 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
984263bc
MD
1834 M_WAITOK);
1835 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
60233e58 1836 namelen, NULL);
984263bc
MD
1837 if (!error) {
1838 /*
1839 * Check for illegal filenames.
1840 */
1841 for (cp = nfs_pub.np_index; *cp; cp++) {
1842 if (*cp == '/') {
1843 error = EINVAL;
1844 break;
1845 }
1846 }
1847 }
1848 if (error) {
1849 FREE(nfs_pub.np_index, M_TEMP);
1850 return (error);
1851 }
1852 }
1853
1854 nfs_pub.np_mount = mp;
1855 nfs_pub.np_valid = 1;
1856 return (0);
1857}
1858
1859struct netcred *
dd98570a
MD
1860vfs_export_lookup(struct mount *mp, struct netexport *nep,
1861 struct sockaddr *nam)
984263bc 1862{
1fd87d54
RG
1863 struct netcred *np;
1864 struct radix_node_head *rnh;
984263bc
MD
1865 struct sockaddr *saddr;
1866
1867 np = NULL;
1868 if (mp->mnt_flag & MNT_EXPORTED) {
1869 /*
1870 * Lookup in the export list first.
1871 */
1872 if (nam != NULL) {
1873 saddr = nam;
1874 rnh = nep->ne_rtable[saddr->sa_family];
1875 if (rnh != NULL) {
1876 np = (struct netcred *)
2e9572df 1877 (*rnh->rnh_matchaddr)((char *)saddr,
984263bc
MD
1878 rnh);
1879 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1880 np = NULL;
1881 }
1882 }
1883 /*
1884 * If no address match, use the default if it exists.
1885 */
1886 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1887 np = &nep->ne_defexported;
1888 }
1889 return (np);
1890}
1891
1892/*
41a01a4d
MD
1893 * perform msync on all vnodes under a mount point. The mount point must
1894 * be locked. This code is also responsible for lazy-freeing unreferenced
1895 * vnodes whos VM objects no longer contain pages.
1896 *
1897 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
03a964e9
MD
1898 *
1899 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1900 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1901 * way up in this high level function.
984263bc 1902 */
41a01a4d 1903static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
5fd012e0 1904static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
41a01a4d 1905
984263bc
MD
1906void
1907vfs_msync(struct mount *mp, int flags)
1908{
03a964e9
MD
1909 int vmsc_flags;
1910
1911 vmsc_flags = VMSC_GETVP;
1912 if (flags != MNT_WAIT)
1913 vmsc_flags |= VMSC_NOWAIT;
1914 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
973c11b9 1915 (void *)(intptr_t)flags);
41a01a4d 1916}
984263bc 1917
41a01a4d
MD
1918/*
1919 * scan1 is a fast pre-check. There could be hundreds of thousands of
1920 * vnodes, we cannot afford to do anything heavy weight until we have a
1921 * fairly good indication that there is work to do.
1922 */
1923static
1924int
1925vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1926{
973c11b9 1927 int flags = (int)(intptr_t)data;
984263bc 1928
5fd012e0 1929 if ((vp->v_flag & VRECLAIMED) == 0) {
3c37c940 1930 if (vshouldmsync(vp))
5fd012e0 1931 return(0); /* call scan2 */
41a01a4d
MD
1932 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1933 (vp->v_flag & VOBJDIRTY) &&
a11aaa81 1934 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
5fd012e0 1935 return(0); /* call scan2 */
41a01a4d
MD
1936 }
1937 }
5fd012e0
MD
1938
1939 /*
1940 * do not call scan2, continue the loop
1941 */
41a01a4d
MD
1942 return(-1);
1943}
1944
03a964e9
MD
1945/*
1946 * This callback is handed a locked vnode.
1947 */
41a01a4d
MD
1948static
1949int
5fd012e0 1950vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
41a01a4d
MD
1951{
1952 vm_object_t obj;
973c11b9 1953 int flags = (int)(intptr_t)data;
41a01a4d 1954
5fd012e0 1955 if (vp->v_flag & VRECLAIMED)
41a01a4d
MD
1956 return(0);
1957
7540ab49
MD
1958 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
1959 if ((obj = vp->v_object) != NULL) {
5fd012e0
MD
1960 vm_object_page_clean(obj, 0, 0,
1961 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
984263bc
MD
1962 }
1963 }
41a01a4d 1964 return(0);
984263bc
MD
1965}
1966
1967/*
984263bc
MD
1968 * Record a process's interest in events which might happen to
1969 * a vnode. Because poll uses the historic select-style interface
1970 * internally, this routine serves as both the ``check for any
1971 * pending events'' and the ``record my interest in future events''
1972 * functions. (These are done together, while the lock is held,
1973 * to avoid race conditions.)
1974 */
1975int
87de5057 1976vn_pollrecord(struct vnode *vp, int events)
984263bc 1977{
41a01a4d
MD
1978 lwkt_tokref ilock;
1979
87de5057
MD
1980 KKASSERT(curthread->td_proc != NULL);
1981
41a01a4d 1982 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
1983 if (vp->v_pollinfo.vpi_revents & events) {
1984 /*
1985 * This leaves events we are not interested
1986 * in available for the other process which
1987 * which presumably had requested them
1988 * (otherwise they would never have been
1989 * recorded).
1990 */
1991 events &= vp->v_pollinfo.vpi_revents;
1992 vp->v_pollinfo.vpi_revents &= ~events;
1993
41a01a4d 1994 lwkt_reltoken(&ilock);
984263bc
MD
1995 return events;
1996 }
1997 vp->v_pollinfo.vpi_events |= events;
87de5057 1998 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
41a01a4d 1999 lwkt_reltoken(&ilock);
984263bc
MD
2000 return 0;
2001}
2002
2003/*
2004 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2005 * it is possible for us to miss an event due to race conditions, but
2006 * that condition is expected to be rare, so for the moment it is the
2007 * preferred interface.
2008 */
2009void
dd98570a 2010vn_pollevent(struct vnode *vp, int events)
984263bc 2011{
41a01a4d
MD
2012 lwkt_tokref ilock;
2013
2014 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
2015 if (vp->v_pollinfo.vpi_events & events) {
2016 /*
2017 * We clear vpi_events so that we don't
2018 * call selwakeup() twice if two events are
2019 * posted before the polling process(es) is
2020 * awakened. This also ensures that we take at
2021 * most one selwakeup() if the polling process
2022 * is no longer interested. However, it does
2023 * mean that only one event can be noticed at
2024 * a time. (Perhaps we should only clear those
2025 * event bits which we note?) XXX
2026 */
2027 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
2028 vp->v_pollinfo.vpi_revents |= events;
2029 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2030 }
41a01a4d 2031 lwkt_reltoken(&ilock);
984263bc
MD
2032}
2033
2034/*
2035 * Wake up anyone polling on vp because it is being revoked.
2036 * This depends on dead_poll() returning POLLHUP for correct
2037 * behavior.
2038 */
2039void
dd98570a 2040vn_pollgone(struct vnode *vp)
984263bc 2041{
41a01a4d
MD
2042 lwkt_tokref ilock;
2043
2044 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
2045 if (vp->v_pollinfo.vpi_events) {
2046 vp->v_pollinfo.vpi_events = 0;
2047 selwakeup(&vp->v_pollinfo.vpi_selinfo);
2048 }
41a01a4d 2049 lwkt_reltoken(&ilock);
984263bc
MD
2050}
2051
984263bc 2052/*
b13267a5 2053 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
e4c9c0c8 2054 * (or v_rdev might be NULL).
984263bc 2055 */
b13267a5 2056cdev_t
e4c9c0c8 2057vn_todev(struct vnode *vp)
984263bc
MD
2058{
2059 if (vp->v_type != VBLK && vp->v_type != VCHR)
028066b1 2060 return (NULL);
e4c9c0c8 2061 KKASSERT(vp->v_rdev != NULL);
984263bc
MD
2062 return (vp->v_rdev);
2063}
2064
2065/*
e4c9c0c8
MD
2066 * Check if vnode represents a disk device. The vnode does not need to be
2067 * opened.
984263bc
MD
2068 */
2069int
e4c9c0c8 2070vn_isdisk(struct vnode *vp, int *errp)
984263bc 2071{
b13267a5 2072 cdev_t dev;
e4c9c0c8 2073
0e9b9130 2074 if (vp->v_type != VCHR) {
984263bc
MD
2075 if (errp != NULL)
2076 *errp = ENOTBLK;
2077 return (0);
2078 }
e4c9c0c8
MD
2079
2080 if ((dev = vp->v_rdev) == NULL)
0e9b9130
MD
2081 dev = get_dev(vp->v_umajor, vp->v_uminor);
2082
028066b1 2083 if (dev == NULL) {
984263bc
MD
2084 if (errp != NULL)
2085 *errp = ENXIO;
2086 return (0);
2087 }
e4c9c0c8 2088 if (dev_is_good(dev) == 0) {
984263bc
MD
2089 if (errp != NULL)
2090 *errp = ENXIO;
2091 return (0);
2092 }
e4c9c0c8 2093 if ((dev_dflags(dev) & D_DISK) == 0) {
984263bc
MD
2094 if (errp != NULL)
2095 *errp = ENOTBLK;
2096 return (0);
2097 }
2098 if (errp != NULL)
2099 *errp = 0;
2100 return (1);
2101}
2102
5d72d6ed
JS
2103int
2104vn_get_namelen(struct vnode *vp, int *namelen)
2105{
973c11b9
MD
2106 int error;
2107 register_t retval[2];
5d72d6ed
JS
2108
2109 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2110 if (error)
2111 return (error);
973c11b9 2112 *namelen = (int)retval[0];
5d72d6ed
JS
2113 return (0);
2114}
fc46f680
JS
2115
2116int
b45c5139
MD
2117vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2118 uint16_t d_namlen, const char *d_name)
fc46f680 2119{
01f31ab3
JS
2120 struct dirent *dp;
2121 size_t len;
fc46f680 2122
01f31ab3
JS
2123 len = _DIRENT_RECLEN(d_namlen);
2124 if (len > uio->uio_resid)
fc46f680
JS
2125 return(1);
2126
efda3bd0 2127 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
01f31ab3
JS
2128
2129 dp->d_ino = d_ino;
2130 dp->d_namlen = d_namlen;
fc46f680
JS
2131 dp->d_type = d_type;
2132 bcopy(d_name, dp->d_name, d_namlen);
fc46f680 2133
01f31ab3
JS
2134 *error = uiomove((caddr_t)dp, len, uio);
2135
efda3bd0 2136 kfree(dp, M_TEMP);
fc46f680
JS
2137
2138 return(0);
2139}
7540ab49 2140
349433c9
MD
2141void
2142vn_mark_atime(struct vnode *vp, struct thread *td)
2143{
2144 struct proc *p = td->td_proc;
2145 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2146
2147 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2148 VOP_MARKATIME(vp, cred);
2149 }
2150}