get rid of thr{1,2,3}, which are obsolete at the moment. Keep the
[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 $
41a01a4d 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.26 2004/03/01 06:33:17 dillon Exp $
984263bc
MD
41 */
42
43/*
44 * External virtual filesystem routines
45 */
46#include "opt_ddb.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/buf.h>
51#include <sys/conf.h>
52#include <sys/dirent.h>
53#include <sys/domain.h>
54#include <sys/eventhandler.h>
55#include <sys/fcntl.h>
56#include <sys/kernel.h>
57#include <sys/kthread.h>
58#include <sys/malloc.h>
59#include <sys/mbuf.h>
60#include <sys/mount.h>
984263bc 61#include <sys/proc.h>
dadab5e9 62#include <sys/namei.h>
984263bc
MD
63#include <sys/reboot.h>
64#include <sys/socket.h>
65#include <sys/stat.h>
66#include <sys/sysctl.h>
67#include <sys/syslog.h>
68#include <sys/vmmeter.h>
69#include <sys/vnode.h>
70
71#include <machine/limits.h>
72
73#include <vm/vm.h>
74#include <vm/vm_object.h>
75#include <vm/vm_extern.h>
6ef943a3 76#include <vm/vm_kern.h>
984263bc
MD
77#include <vm/pmap.h>
78#include <vm/vm_map.h>
79#include <vm/vm_page.h>
80#include <vm/vm_pager.h>
81#include <vm/vnode_pager.h>
82#include <vm/vm_zone.h>
83
3020e3be 84#include <sys/buf2.h>
f5865223 85#include <sys/thread2.h>
3020e3be 86
984263bc
MD
87static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
88
402ed7e1 89static void insmntque (struct vnode *vp, struct mount *mp);
41a01a4d 90static void vclean (struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td);
984263bc 91static unsigned long numvnodes;
984263bc
MD
92SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93
94enum vtype iftovt_tab[16] = {
95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
97};
98int vttoif_tab[9] = {
99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
100 S_IFSOCK, S_IFIFO, S_IFMT,
101};
102
103static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
104
105static u_long wantfreevnodes = 25;
106SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
107static u_long freevnodes = 0;
108SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
109
110static int reassignbufcalls;
111SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
112static int reassignbufloops;
113SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
114static int reassignbufsortgood;
115SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
116static int reassignbufsortbad;
117SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
118static int reassignbufmethod = 1;
119SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
984263bc
MD
120
121#ifdef ENABLE_VFS_IOOPT
122int vfs_ioopt = 0;
123SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
124#endif
125
126struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
8a8d5d85
MD
127struct lwkt_token mountlist_token;
128struct lwkt_token mntvnode_token;
984263bc 129int nfs_mount_type = -1;
8a8d5d85
MD
130static struct lwkt_token mntid_token;
131static struct lwkt_token vnode_free_list_token;
132static struct lwkt_token spechash_token;
984263bc
MD
133struct nfs_public nfs_pub; /* publicly exported FS */
134static vm_zone_t vnode_zone;
135
136/*
137 * The workitem queue.
138 */
139#define SYNCER_MAXDELAY 32
140static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
141time_t syncdelay = 30; /* max time to delay syncing data */
99138c2f
HP
142SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
143 "VFS data synchronization delay");
984263bc 144time_t filedelay = 30; /* time to delay syncing files */
99138c2f
HP
145SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
146 "File synchronization delay");
984263bc 147time_t dirdelay = 29; /* time to delay syncing directories */
99138c2f
HP
148SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
149 "Directory synchronization delay");
984263bc 150time_t metadelay = 28; /* time to delay syncing metadata */
99138c2f
HP
151SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
152 "VFS metadata synchronization delay");
984263bc
MD
153static int rushjob; /* number of slots to run ASAP */
154static int stat_rush_requests; /* number of times I/O speeded up */
155SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
156
157static int syncer_delayno = 0;
158static long syncer_mask;
159LIST_HEAD(synclist, vnode);
160static struct synclist *syncer_workitem_pending;
161
162int desiredvnodes;
163SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
164 &desiredvnodes, 0, "Maximum number of vnodes");
165static int minvnodes;
166SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
167 &minvnodes, 0, "Minimum number of vnodes");
168static int vnlru_nowhere = 0;
169SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
170 "Number of times the vnlru process ran without success");
171
402ed7e1
RG
172static void vfs_free_addrlist (struct netexport *nep);
173static int vfs_free_netcred (struct radix_node *rn, void *w);
174static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
175 struct export_args *argp);
984263bc 176
41a01a4d
MD
177#define VSHOULDFREE(vp) \
178 (!((vp)->v_flag & (VFREE|VDOOMED)) && \
179 !(vp)->v_holdcnt && !(vp)->v_usecount && \
180 (!(vp)->v_object || \
181 !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
182
183#define VMIGHTFREE(vp) \
184 (((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) == 0 && \
185 cache_leaf_test(vp) == 0 && (vp)->v_usecount == 0)
186
187#define VSHOULDBUSY(vp) \
188 (((vp)->v_flag & VFREE) && \
189 ((vp)->v_holdcnt || (vp)->v_usecount))
190
191static void vbusy(struct vnode *vp);
192static void vfree(struct vnode *vp);
193static void vmaybefree(struct vnode *vp);
194
195/*
196 * NOTE: the vnode interlock must be held on call.
197 */
198static __inline void
199vmaybefree(struct vnode *vp)
200{
201 if (VSHOULDFREE(vp))
202 vfree(vp);
203}
204
984263bc
MD
205/*
206 * Initialize the vnode management data structures.
207 */
208void
209vntblinit()
210{
211
6ef943a3
MD
212 /*
213 * Desired vnodes is a result of the physical page count
214 * and the size of kernel's heap. It scales in proportion
215 * to the amount of available physical memory. This can
216 * cause trouble on 64-bit and large memory platforms.
217 */
218 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
219 desiredvnodes =
220 min(maxproc + vmstats.v_page_count /4,
221 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
222 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
223
984263bc 224 minvnodes = desiredvnodes / 4;
41a01a4d
MD
225 lwkt_token_init(&mountlist_token);
226 lwkt_token_init(&mntvnode_token);
227 lwkt_token_init(&mntid_token);
228 lwkt_token_init(&spechash_token);
984263bc 229 TAILQ_INIT(&vnode_free_list);
41a01a4d 230 lwkt_token_init(&vnode_free_list_token);
984263bc
MD
231 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
232 /*
233 * Initialize the filesystem syncer.
234 */
235 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
236 &syncer_mask);
237 syncer_maxdelay = syncer_mask + 1;
238}
239
240/*
241 * Mark a mount point as busy. Used to synchronize access and to delay
242 * unmounting. Interlock is not released on failure.
243 */
244int
41a01a4d 245vfs_busy(struct mount *mp, int flags, lwkt_tokref_t interlkp, struct thread *td)
984263bc
MD
246{
247 int lkflags;
248
249 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
250 if (flags & LK_NOWAIT)
251 return (ENOENT);
252 mp->mnt_kern_flag |= MNTK_MWAIT;
984263bc
MD
253 /*
254 * Since all busy locks are shared except the exclusive
255 * lock granted when unmounting, the only place that a
256 * wakeup needs to be done is at the release of the
257 * exclusive lock at the end of dounmount.
41a01a4d
MD
258 *
259 * note: interlkp is a serializer and thus can be safely
260 * held through any sleep
984263bc 261 */
377d4740 262 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
984263bc
MD
263 return (ENOENT);
264 }
265 lkflags = LK_SHARED | LK_NOPAUSE;
266 if (interlkp)
267 lkflags |= LK_INTERLOCK;
dadab5e9 268 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
984263bc
MD
269 panic("vfs_busy: unexpected lock failure");
270 return (0);
271}
272
273/*
274 * Free a busy filesystem.
275 */
276void
dadab5e9 277vfs_unbusy(struct mount *mp, struct thread *td)
984263bc 278{
dadab5e9 279 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
984263bc
MD
280}
281
282/*
283 * Lookup a filesystem type, and if found allocate and initialize
284 * a mount structure for it.
285 *
286 * Devname is usually updated by mount(8) after booting.
287 */
288int
dadab5e9 289vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
984263bc 290{
dadab5e9 291 struct thread *td = curthread; /* XXX */
984263bc
MD
292 struct vfsconf *vfsp;
293 struct mount *mp;
294
295 if (fstypename == NULL)
296 return (ENODEV);
297 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
298 if (!strcmp(vfsp->vfc_name, fstypename))
299 break;
300 if (vfsp == NULL)
301 return (ENODEV);
302 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
303 bzero((char *)mp, (u_long)sizeof(struct mount));
377d4740 304 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
41a01a4d 305 vfs_busy(mp, LK_NOWAIT, NULL, td);
984263bc
MD
306 TAILQ_INIT(&mp->mnt_nvnodelist);
307 TAILQ_INIT(&mp->mnt_reservedvnlist);
308 mp->mnt_nvnodelistsize = 0;
309 mp->mnt_vfc = vfsp;
310 mp->mnt_op = vfsp->vfc_vfsops;
311 mp->mnt_flag = MNT_RDONLY;
312 mp->mnt_vnodecovered = NULLVP;
313 vfsp->vfc_refcount++;
314 mp->mnt_iosize_max = DFLTPHYS;
315 mp->mnt_stat.f_type = vfsp->vfc_typenum;
316 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
317 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
318 mp->mnt_stat.f_mntonname[0] = '/';
319 mp->mnt_stat.f_mntonname[1] = 0;
320 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
321 *mpp = mp;
322 return (0);
323}
324
325/*
326 * Find an appropriate filesystem to use for the root. If a filesystem
327 * has not been preselected, walk through the list of known filesystems
328 * trying those that have mountroot routines, and try them until one
329 * works or we have tried them all.
330 */
331#ifdef notdef /* XXX JH */
332int
333lite2_vfs_mountroot()
334{
335 struct vfsconf *vfsp;
402ed7e1 336 extern int (*lite2_mountroot) (void);
984263bc
MD
337 int error;
338
339 if (lite2_mountroot != NULL)
340 return ((*lite2_mountroot)());
341 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
342 if (vfsp->vfc_mountroot == NULL)
343 continue;
344 if ((error = (*vfsp->vfc_mountroot)()) == 0)
345 return (0);
346 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
347 }
348 return (ENODEV);
349}
350#endif
351
352/*
353 * Lookup a mount point by filesystem identifier.
354 */
355struct mount *
356vfs_getvfs(fsid)
357 fsid_t *fsid;
358{
1fd87d54 359 struct mount *mp;
41a01a4d 360 lwkt_tokref ilock;
984263bc 361
41a01a4d 362 lwkt_gettoken(&ilock, &mountlist_token);
984263bc
MD
363 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
364 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
365 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
41a01a4d 366 break;
984263bc
MD
367 }
368 }
41a01a4d
MD
369 lwkt_reltoken(&ilock);
370 return (mp);
984263bc
MD
371}
372
373/*
374 * Get a new unique fsid. Try to make its val[0] unique, since this value
375 * will be used to create fake device numbers for stat(). Also try (but
376 * not so hard) make its val[0] unique mod 2^16, since some emulators only
377 * support 16-bit device numbers. We end up with unique val[0]'s for the
378 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
379 *
380 * Keep in mind that several mounts may be running in parallel. Starting
381 * the search one past where the previous search terminated is both a
382 * micro-optimization and a defense against returning the same fsid to
383 * different mounts.
384 */
385void
386vfs_getnewfsid(mp)
387 struct mount *mp;
388{
389 static u_int16_t mntid_base;
41a01a4d 390 lwkt_tokref ilock;
984263bc
MD
391 fsid_t tfsid;
392 int mtype;
393
41a01a4d 394 lwkt_gettoken(&ilock, &mntid_token);
984263bc
MD
395 mtype = mp->mnt_vfc->vfc_typenum;
396 tfsid.val[1] = mtype;
397 mtype = (mtype & 0xFF) << 24;
398 for (;;) {
399 tfsid.val[0] = makeudev(255,
400 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
401 mntid_base++;
402 if (vfs_getvfs(&tfsid) == NULL)
403 break;
404 }
405 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
406 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
41a01a4d 407 lwkt_reltoken(&ilock);
984263bc
MD
408}
409
410/*
411 * Knob to control the precision of file timestamps:
412 *
413 * 0 = seconds only; nanoseconds zeroed.
414 * 1 = seconds and nanoseconds, accurate within 1/HZ.
415 * 2 = seconds and nanoseconds, truncated to microseconds.
416 * >=3 = seconds and nanoseconds, maximum precision.
417 */
418enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
419
420static int timestamp_precision = TSP_SEC;
421SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
422 &timestamp_precision, 0, "");
423
424/*
425 * Get a current timestamp.
426 */
427void
428vfs_timestamp(tsp)
429 struct timespec *tsp;
430{
431 struct timeval tv;
432
433 switch (timestamp_precision) {
434 case TSP_SEC:
435 tsp->tv_sec = time_second;
436 tsp->tv_nsec = 0;
437 break;
438 case TSP_HZ:
439 getnanotime(tsp);
440 break;
441 case TSP_USEC:
442 microtime(&tv);
443 TIMEVAL_TO_TIMESPEC(&tv, tsp);
444 break;
445 case TSP_NSEC:
446 default:
447 nanotime(tsp);
448 break;
449 }
450}
451
452/*
453 * Set vnode attributes to VNOVAL
454 */
455void
456vattr_null(vap)
1fd87d54 457 struct vattr *vap;
984263bc
MD
458{
459
460 vap->va_type = VNON;
461 vap->va_size = VNOVAL;
462 vap->va_bytes = VNOVAL;
463 vap->va_mode = VNOVAL;
464 vap->va_nlink = VNOVAL;
465 vap->va_uid = VNOVAL;
466 vap->va_gid = VNOVAL;
467 vap->va_fsid = VNOVAL;
468 vap->va_fileid = VNOVAL;
469 vap->va_blocksize = VNOVAL;
470 vap->va_rdev = VNOVAL;
471 vap->va_atime.tv_sec = VNOVAL;
472 vap->va_atime.tv_nsec = VNOVAL;
473 vap->va_mtime.tv_sec = VNOVAL;
474 vap->va_mtime.tv_nsec = VNOVAL;
475 vap->va_ctime.tv_sec = VNOVAL;
476 vap->va_ctime.tv_nsec = VNOVAL;
477 vap->va_flags = VNOVAL;
478 vap->va_gen = VNOVAL;
479 vap->va_vaflags = 0;
480}
481
482/*
483 * This routine is called when we have too many vnodes. It attempts
484 * to free <count> vnodes and will potentially free vnodes that still
485 * have VM backing store (VM backing store is typically the cause
486 * of a vnode blowout so we want to do this). Therefore, this operation
487 * is not considered cheap.
488 *
489 * A number of conditions may prevent a vnode from being reclaimed.
490 * the buffer cache may have references on the vnode, a directory
491 * vnode may still have references due to the namei cache representing
492 * underlying files, or the vnode may be in active use. It is not
493 * desireable to reuse such vnodes. These conditions may cause the
494 * number of vnodes to reach some minimum value regardless of what
495 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
496 */
497static int
498vlrureclaim(struct mount *mp)
499{
500 struct vnode *vp;
41a01a4d
MD
501 lwkt_tokref ilock;
502 lwkt_tokref vlock;
984263bc
MD
503 int done;
504 int trigger;
505 int usevnodes;
506 int count;
507
508 /*
509 * Calculate the trigger point, don't allow user
510 * screwups to blow us up. This prevents us from
511 * recycling vnodes with lots of resident pages. We
512 * aren't trying to free memory, we are trying to
513 * free vnodes.
514 */
515 usevnodes = desiredvnodes;
516 if (usevnodes <= 0)
517 usevnodes = 1;
12e4aaff 518 trigger = vmstats.v_page_count * 2 / usevnodes;
984263bc
MD
519
520 done = 0;
41a01a4d 521 lwkt_gettoken(&ilock, &mntvnode_token);
984263bc
MD
522 count = mp->mnt_nvnodelistsize / 10 + 1;
523 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
41a01a4d
MD
524 /*
525 * __VNODESCAN__
526 *
527 * The VP will stick around while we hold mntvnode_token,
528 * at least until we block, so we can safely do an initial
529 * check. But we have to check again after obtaining
530 * the vnode interlock. vp->v_interlock points to stable
531 * storage so it's ok if the vp gets ripped out from
532 * under us while we are blocked.
533 */
534 if (vp->v_type == VNON ||
535 vp->v_type == VBAD ||
536 !VMIGHTFREE(vp) || /* critical path opt */
537 (vp->v_object &&
538 vp->v_object->resident_page_count >= trigger)
539 ) {
540 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
541 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
542 --count;
543 continue;
544 }
545
546 /*
547 * Get the interlock, delay moving the node to the tail so
548 * we don't race against new additions to the mountlist.
549 */
550 lwkt_gettoken(&vlock, vp->v_interlock);
551 if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
552 lwkt_reltoken(&vlock);
553 continue;
554 }
984263bc 555 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
41a01a4d 556 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
984263bc 557
41a01a4d
MD
558 /*
559 * Must check again
560 */
561 if (vp->v_type == VNON ||
562 vp->v_type == VBAD ||
563 !VMIGHTFREE(vp) || /* critical path opt */
564 (vp->v_object &&
565 vp->v_object->resident_page_count >= trigger)
984263bc 566 ) {
41a01a4d
MD
567 lwkt_reltoken(&vlock);
568 --count;
569 continue;
984263bc 570 }
41a01a4d
MD
571 vgonel(vp, &vlock, curthread);
572 ++done;
984263bc
MD
573 --count;
574 }
41a01a4d 575 lwkt_reltoken(&ilock);
984263bc
MD
576 return done;
577}
578
579/*
580 * Attempt to recycle vnodes in a context that is always safe to block.
581 * Calling vlrurecycle() from the bowels of file system code has some
582 * interesting deadlock problems.
583 */
bc6dffab 584static struct thread *vnlruthread;
984263bc
MD
585static int vnlruproc_sig;
586
587static void
588vnlru_proc(void)
589{
590 struct mount *mp, *nmp;
41a01a4d 591 lwkt_tokref ilock;
984263bc
MD
592 int s;
593 int done;
0cfcada1 594 struct thread *td = curthread;
984263bc 595
bc6dffab 596 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
597 SHUTDOWN_PRI_FIRST);
598
599 s = splbio();
600 for (;;) {
0cfcada1 601 kproc_suspend_loop();
984263bc
MD
602 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
603 vnlruproc_sig = 0;
604 wakeup(&vnlruproc_sig);
377d4740 605 tsleep(td, 0, "vlruwt", hz);
984263bc
MD
606 continue;
607 }
608 done = 0;
41a01a4d 609 lwkt_gettoken(&ilock, &mountlist_token);
984263bc 610 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
41a01a4d 611 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
984263bc
MD
612 nmp = TAILQ_NEXT(mp, mnt_list);
613 continue;
614 }
615 done += vlrureclaim(mp);
41a01a4d 616 lwkt_gettokref(&ilock);
984263bc 617 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 618 vfs_unbusy(mp, td);
984263bc 619 }
41a01a4d 620 lwkt_reltoken(&ilock);
984263bc
MD
621 if (done == 0) {
622 vnlru_nowhere++;
377d4740 623 tsleep(td, 0, "vlrup", hz * 3);
984263bc
MD
624 }
625 }
626 splx(s);
627}
628
629static struct kproc_desc vnlru_kp = {
630 "vnlru",
631 vnlru_proc,
bc6dffab 632 &vnlruthread
984263bc
MD
633};
634SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
635
636/*
637 * Routines having to do with the management of the vnode table.
638 */
639extern vop_t **dead_vnodeop_p;
640
641/*
642 * Return the next vnode from the free list.
643 */
644int
645getnewvnode(tag, mp, vops, vpp)
646 enum vtagtype tag;
647 struct mount *mp;
648 vop_t **vops;
649 struct vnode **vpp;
650{
651 int s;
dadab5e9 652 struct thread *td = curthread; /* XXX */
984263bc 653 struct vnode *vp = NULL;
41a01a4d 654 struct vnode *xvp;
984263bc 655 vm_object_t object;
41a01a4d
MD
656 lwkt_tokref ilock;
657 lwkt_tokref vlock;
984263bc
MD
658
659 s = splbio();
660
661 /*
662 * Try to reuse vnodes if we hit the max. This situation only
663 * occurs in certain large-memory (2G+) situations. We cannot
664 * attempt to directly reclaim vnodes due to nasty recursion
665 * problems.
666 */
667 while (numvnodes - freevnodes > desiredvnodes) {
668 if (vnlruproc_sig == 0) {
669 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
bc6dffab 670 wakeup(vnlruthread);
984263bc 671 }
377d4740 672 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
984263bc
MD
673 }
674
675
676 /*
677 * Attempt to reuse a vnode already on the free list, allocating
678 * a new vnode if we can't find one or if we have not reached a
679 * good minimum for good LRU performance.
680 */
41a01a4d 681 lwkt_gettoken(&ilock, &vnode_free_list_token);
984263bc
MD
682 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
683 int count;
684
685 for (count = 0; count < freevnodes; count++) {
41a01a4d
MD
686 /*
687 * __VNODESCAN__
688 *
689 * Pull the next vnode off the free list and do some
690 * sanity checks. Note that regardless of how we
691 * block, if freevnodes is non-zero there had better
692 * be something on the list.
693 */
984263bc 694 vp = TAILQ_FIRST(&vnode_free_list);
41a01a4d 695 if (vp == NULL)
984263bc
MD
696 panic("getnewvnode: free vnode isn't");
697
8a8d5d85 698 /*
41a01a4d
MD
699 * Move the vnode to the end of the list so other
700 * processes do not double-block trying to recycle
701 * the same vnode (as an optimization), then get
702 * the interlock.
8a8d5d85 703 */
41a01a4d
MD
704 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
705 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
706
707 /*
708 * Skip vnodes that are in the process of being
709 * held or referenced. Since the act of adding or
710 * removing a vnode on the freelist requires a token
711 * and may block, the ref count may be adjusted
712 * prior to its addition or removal.
713 */
714 if (VSHOULDBUSY(vp)) {
8a8d5d85
MD
715 vp = NULL;
716 continue;
717 }
718
41a01a4d 719
8a8d5d85 720 /*
41a01a4d
MD
721 * Obtain the vnode interlock and check that the
722 * vnode is still on the free list.
723 *
724 * This normally devolves into a degenerate case so
725 * it is optimal. Loop up if it isn't. Note that
726 * the vnode could be in the middle of being moved
727 * off the free list (the VSHOULDBUSY() check) and
728 * must be skipped if so.
8a8d5d85 729 */
41a01a4d
MD
730 lwkt_gettoken(&vlock, vp->v_interlock);
731 TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
732 freelst, v_freelist) {
733 if (vp == xvp)
734 break;
735 }
736 if (vp != xvp || VSHOULDBUSY(vp)) {
8a8d5d85
MD
737 vp = NULL;
738 continue;
739 }
740
741 /*
41a01a4d
MD
742 * We now safely own the vnode. If the vnode has
743 * an object do not recycle it if its VM object
744 * has resident pages or references.
8a8d5d85 745 */
41a01a4d
MD
746 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
747 (object->resident_page_count || object->ref_count))
748 ) {
749 lwkt_reltoken(&vlock);
984263bc
MD
750 vp = NULL;
751 continue;
752 }
8a8d5d85 753
7ea21ed1 754 /*
41a01a4d
MD
755 * We can almost reuse this vnode. But we don't want
756 * to recycle it if the vnode has children in the
757 * namecache because that breaks the namecache's
758 * path element chain. (YYY use nc_refs for the
759 * check?)
7ea21ed1 760 */
41a01a4d
MD
761 KKASSERT(vp->v_flag & VFREE);
762 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
763
764 if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
765 cache_leaf_test(vp) >= 0) {
766 /* ok, we can reuse this vnode */
767 break;
984263bc 768 }
41a01a4d
MD
769 lwkt_reltoken(&vlock);
770 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
771 vp = NULL;
984263bc
MD
772 }
773 }
774
41a01a4d
MD
775 /*
776 * If vp is non-NULL we hold it's interlock.
777 */
984263bc
MD
778 if (vp) {
779 vp->v_flag |= VDOOMED;
780 vp->v_flag &= ~VFREE;
781 freevnodes--;
41a01a4d 782 lwkt_reltoken(&ilock);
8a8d5d85 783 cache_purge(vp); /* YYY may block */
984263bc
MD
784 vp->v_lease = NULL;
785 if (vp->v_type != VBAD) {
41a01a4d 786 vgonel(vp, &vlock, td);
984263bc 787 } else {
41a01a4d 788 lwkt_reltoken(&vlock);
984263bc
MD
789 }
790
791#ifdef INVARIANTS
792 {
793 int s;
794
795 if (vp->v_data)
796 panic("cleaned vnode isn't");
797 s = splbio();
798 if (vp->v_numoutput)
799 panic("Clean vnode has pending I/O's");
800 splx(s);
801 }
802#endif
803 vp->v_flag = 0;
804 vp->v_lastw = 0;
805 vp->v_lasta = 0;
806 vp->v_cstart = 0;
807 vp->v_clen = 0;
808 vp->v_socket = 0;
809 vp->v_writecount = 0; /* XXX */
810 } else {
41a01a4d
MD
811 lwkt_reltoken(&ilock);
812 vp = zalloc(vnode_zone);
813 bzero(vp, sizeof(*vp));
814 vp->v_interlock = lwkt_token_pool_get(vp);
815 lwkt_token_init(&vp->v_pollinfo.vpi_token);
984263bc
MD
816 vp->v_dd = vp;
817 cache_purge(vp);
7ea21ed1 818 TAILQ_INIT(&vp->v_namecache);
984263bc
MD
819 numvnodes++;
820 }
821
822 TAILQ_INIT(&vp->v_cleanblkhd);
823 TAILQ_INIT(&vp->v_dirtyblkhd);
824 vp->v_type = VNON;
825 vp->v_tag = tag;
826 vp->v_op = vops;
827 insmntque(vp, mp);
828 *vpp = vp;
829 vp->v_usecount = 1;
830 vp->v_data = 0;
831 splx(s);
832
3b568787 833 vfs_object_create(vp, td);
984263bc
MD
834 return (0);
835}
836
837/*
838 * Move a vnode from one mount queue to another.
839 */
840static void
841insmntque(vp, mp)
1fd87d54
RG
842 struct vnode *vp;
843 struct mount *mp;
984263bc 844{
41a01a4d 845 lwkt_tokref ilock;
984263bc 846
41a01a4d 847 lwkt_gettoken(&ilock, &mntvnode_token);
984263bc
MD
848 /*
849 * Delete from old mount point vnode list, if on one.
850 */
851 if (vp->v_mount != NULL) {
852 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
853 ("bad mount point vnode list size"));
854 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
855 vp->v_mount->mnt_nvnodelistsize--;
856 }
857 /*
858 * Insert into list of vnodes for the new mount point, if available.
859 */
860 if ((vp->v_mount = mp) == NULL) {
41a01a4d 861 lwkt_reltoken(&ilock);
984263bc
MD
862 return;
863 }
864 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
865 mp->mnt_nvnodelistsize++;
41a01a4d 866 lwkt_reltoken(&ilock);
984263bc
MD
867}
868
869/*
870 * Update outstanding I/O count and do wakeup if requested.
871 */
872void
873vwakeup(bp)
1fd87d54 874 struct buf *bp;
984263bc 875{
1fd87d54 876 struct vnode *vp;
984263bc
MD
877
878 bp->b_flags &= ~B_WRITEINPROG;
879 if ((vp = bp->b_vp)) {
880 vp->v_numoutput--;
881 if (vp->v_numoutput < 0)
882 panic("vwakeup: neg numoutput");
883 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
884 vp->v_flag &= ~VBWAIT;
885 wakeup((caddr_t) &vp->v_numoutput);
886 }
887 }
888}
889
890/*
891 * Flush out and invalidate all buffers associated with a vnode.
892 * Called with the underlying object locked.
893 */
894int
3b568787
MD
895vinvalbuf(struct vnode *vp, int flags, struct thread *td,
896 int slpflag, int slptimeo)
984263bc 897{
1fd87d54 898 struct buf *bp;
984263bc
MD
899 struct buf *nbp, *blist;
900 int s, error;
901 vm_object_t object;
41a01a4d 902 lwkt_tokref vlock;
984263bc
MD
903
904 if (flags & V_SAVE) {
905 s = splbio();
906 while (vp->v_numoutput) {
907 vp->v_flag |= VBWAIT;
908 error = tsleep((caddr_t)&vp->v_numoutput,
377d4740 909 slpflag, "vinvlbuf", slptimeo);
984263bc
MD
910 if (error) {
911 splx(s);
912 return (error);
913 }
914 }
915 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
916 splx(s);
3b568787 917 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
984263bc
MD
918 return (error);
919 s = splbio();
920 if (vp->v_numoutput > 0 ||
921 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
922 panic("vinvalbuf: dirty bufs");
923 }
924 splx(s);
925 }
926 s = splbio();
927 for (;;) {
928 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
929 if (!blist)
930 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
931 if (!blist)
932 break;
933
934 for (bp = blist; bp; bp = nbp) {
935 nbp = TAILQ_NEXT(bp, b_vnbufs);
936 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
937 error = BUF_TIMELOCK(bp,
938 LK_EXCLUSIVE | LK_SLEEPFAIL,
939 "vinvalbuf", slpflag, slptimeo);
940 if (error == ENOLCK)
941 break;
942 splx(s);
943 return (error);
944 }
945 /*
946 * XXX Since there are no node locks for NFS, I
947 * believe there is a slight chance that a delayed
948 * write will occur while sleeping just above, so
949 * check for it. Note that vfs_bio_awrite expects
950 * buffers to reside on a queue, while VOP_BWRITE and
951 * brelse do not.
952 */
953 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
954 (flags & V_SAVE)) {
955
956 if (bp->b_vp == vp) {
957 if (bp->b_flags & B_CLUSTEROK) {
958 BUF_UNLOCK(bp);
959 vfs_bio_awrite(bp);
960 } else {
961 bremfree(bp);
962 bp->b_flags |= B_ASYNC;
963 VOP_BWRITE(bp->b_vp, bp);
964 }
965 } else {
966 bremfree(bp);
967 (void) VOP_BWRITE(bp->b_vp, bp);
968 }
969 break;
970 }
971 bremfree(bp);
972 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
973 bp->b_flags &= ~B_ASYNC;
974 brelse(bp);
975 }
976 }
977
978 /*
979 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
980 * have write I/O in-progress but if there is a VM object then the
981 * VM object can also have read-I/O in-progress.
982 */
983 do {
984 while (vp->v_numoutput > 0) {
985 vp->v_flag |= VBWAIT;
377d4740 986 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
984263bc
MD
987 }
988 if (VOP_GETVOBJECT(vp, &object) == 0) {
989 while (object->paging_in_progress)
990 vm_object_pip_sleep(object, "vnvlbx");
991 }
992 } while (vp->v_numoutput > 0);
993
994 splx(s);
995
996 /*
997 * Destroy the copy in the VM cache, too.
998 */
41a01a4d 999 lwkt_gettoken(&vlock, vp->v_interlock);
984263bc
MD
1000 if (VOP_GETVOBJECT(vp, &object) == 0) {
1001 vm_object_page_remove(object, 0, 0,
1002 (flags & V_SAVE) ? TRUE : FALSE);
1003 }
41a01a4d 1004 lwkt_reltoken(&vlock);
984263bc
MD
1005
1006 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1007 panic("vinvalbuf: flush failed");
1008 return (0);
1009}
1010
1011/*
1012 * Truncate a file's buffer and pages to a specified length. This
1013 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1014 * sync activity.
1015 */
1016int
3b568787 1017vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
984263bc 1018{
dadab5e9 1019 struct buf *bp;
984263bc
MD
1020 struct buf *nbp;
1021 int s, anyfreed;
1022 int trunclbn;
1023
1024 /*
1025 * Round up to the *next* lbn.
1026 */
1027 trunclbn = (length + blksize - 1) / blksize;
1028
1029 s = splbio();
1030restart:
1031 anyfreed = 1;
1032 for (;anyfreed;) {
1033 anyfreed = 0;
1034 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1035 nbp = TAILQ_NEXT(bp, b_vnbufs);
1036 if (bp->b_lblkno >= trunclbn) {
1037 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1038 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1039 goto restart;
1040 } else {
1041 bremfree(bp);
1042 bp->b_flags |= (B_INVAL | B_RELBUF);
1043 bp->b_flags &= ~B_ASYNC;
1044 brelse(bp);
1045 anyfreed = 1;
1046 }
1047 if (nbp &&
1048 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1049 (nbp->b_vp != vp) ||
1050 (nbp->b_flags & B_DELWRI))) {
1051 goto restart;
1052 }
1053 }
1054 }
1055
1056 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1057 nbp = TAILQ_NEXT(bp, b_vnbufs);
1058 if (bp->b_lblkno >= trunclbn) {
1059 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1060 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1061 goto restart;
1062 } else {
1063 bremfree(bp);
1064 bp->b_flags |= (B_INVAL | B_RELBUF);
1065 bp->b_flags &= ~B_ASYNC;
1066 brelse(bp);
1067 anyfreed = 1;
1068 }
1069 if (nbp &&
1070 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1071 (nbp->b_vp != vp) ||
1072 (nbp->b_flags & B_DELWRI) == 0)) {
1073 goto restart;
1074 }
1075 }
1076 }
1077 }
1078
1079 if (length > 0) {
1080restartsync:
1081 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1082 nbp = TAILQ_NEXT(bp, b_vnbufs);
1083 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1084 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1085 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1086 goto restart;
1087 } else {
1088 bremfree(bp);
1089 if (bp->b_vp == vp) {
1090 bp->b_flags |= B_ASYNC;
1091 } else {
1092 bp->b_flags &= ~B_ASYNC;
1093 }
1094 VOP_BWRITE(bp->b_vp, bp);
1095 }
1096 goto restartsync;
1097 }
1098
1099 }
1100 }
1101
1102 while (vp->v_numoutput > 0) {
1103 vp->v_flag |= VBWAIT;
377d4740 1104 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
984263bc
MD
1105 }
1106
1107 splx(s);
1108
1109 vnode_pager_setsize(vp, length);
1110
1111 return (0);
1112}
1113
1114/*
1115 * Associate a buffer with a vnode.
1116 */
1117void
1118bgetvp(vp, bp)
1fd87d54
RG
1119 struct vnode *vp;
1120 struct buf *bp;
984263bc
MD
1121{
1122 int s;
1123
1124 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1125
1126 vhold(vp);
1127 bp->b_vp = vp;
1128 bp->b_dev = vn_todev(vp);
1129 /*
1130 * Insert onto list for new vnode.
1131 */
1132 s = splbio();
1133 bp->b_xflags |= BX_VNCLEAN;
1134 bp->b_xflags &= ~BX_VNDIRTY;
1135 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1136 splx(s);
1137}
1138
1139/*
1140 * Disassociate a buffer from a vnode.
1141 */
1142void
1143brelvp(bp)
1fd87d54 1144 struct buf *bp;
984263bc
MD
1145{
1146 struct vnode *vp;
1147 struct buflists *listheadp;
1148 int s;
1149
1150 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1151
1152 /*
1153 * Delete from old vnode list, if on one.
1154 */
1155 vp = bp->b_vp;
1156 s = splbio();
1157 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1158 if (bp->b_xflags & BX_VNDIRTY)
1159 listheadp = &vp->v_dirtyblkhd;
1160 else
1161 listheadp = &vp->v_cleanblkhd;
1162 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1163 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1164 }
1165 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1166 vp->v_flag &= ~VONWORKLST;
1167 LIST_REMOVE(vp, v_synclist);
1168 }
1169 splx(s);
1170 bp->b_vp = (struct vnode *) 0;
1171 vdrop(vp);
1172}
1173
1174/*
1175 * The workitem queue.
1176 *
1177 * It is useful to delay writes of file data and filesystem metadata
1178 * for tens of seconds so that quickly created and deleted files need
1179 * not waste disk bandwidth being created and removed. To realize this,
1180 * we append vnodes to a "workitem" queue. When running with a soft
1181 * updates implementation, most pending metadata dependencies should
1182 * not wait for more than a few seconds. Thus, mounted on block devices
1183 * are delayed only about a half the time that file data is delayed.
1184 * Similarly, directory updates are more critical, so are only delayed
1185 * about a third the time that file data is delayed. Thus, there are
1186 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1187 * one each second (driven off the filesystem syncer process). The
1188 * syncer_delayno variable indicates the next queue that is to be processed.
1189 * Items that need to be processed soon are placed in this queue:
1190 *
1191 * syncer_workitem_pending[syncer_delayno]
1192 *
1193 * A delay of fifteen seconds is done by placing the request fifteen
1194 * entries later in the queue:
1195 *
1196 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1197 *
1198 */
1199
1200/*
1201 * Add an item to the syncer work queue.
1202 */
1203static void
1204vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1205{
1206 int s, slot;
1207
1208 s = splbio();
1209
1210 if (vp->v_flag & VONWORKLST) {
1211 LIST_REMOVE(vp, v_synclist);
1212 }
1213
1214 if (delay > syncer_maxdelay - 2)
1215 delay = syncer_maxdelay - 2;
1216 slot = (syncer_delayno + delay) & syncer_mask;
1217
1218 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1219 vp->v_flag |= VONWORKLST;
1220 splx(s);
1221}
1222
bc6dffab 1223struct thread *updatethread;
402ed7e1 1224static void sched_sync (void);
984263bc
MD
1225static struct kproc_desc up_kp = {
1226 "syncer",
1227 sched_sync,
bc6dffab 1228 &updatethread
984263bc
MD
1229};
1230SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1231
1232/*
1233 * System filesystem synchronizer daemon.
1234 */
1235void
1236sched_sync(void)
1237{
1238 struct synclist *slp;
1239 struct vnode *vp;
1240 long starttime;
1241 int s;
0cfcada1 1242 struct thread *td = curthread;
984263bc 1243
bc6dffab 1244 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
984263bc
MD
1245 SHUTDOWN_PRI_LAST);
1246
1247 for (;;) {
0cfcada1 1248 kproc_suspend_loop();
984263bc
MD
1249
1250 starttime = time_second;
1251
1252 /*
1253 * Push files whose dirty time has expired. Be careful
1254 * of interrupt race on slp queue.
1255 */
1256 s = splbio();
1257 slp = &syncer_workitem_pending[syncer_delayno];
1258 syncer_delayno += 1;
1259 if (syncer_delayno == syncer_maxdelay)
1260 syncer_delayno = 0;
1261 splx(s);
1262
1263 while ((vp = LIST_FIRST(slp)) != NULL) {
1264 if (VOP_ISLOCKED(vp, NULL) == 0) {
41a01a4d 1265 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
3b568787 1266 (void) VOP_FSYNC(vp, MNT_LAZY, td);
41a01a4d 1267 VOP_UNLOCK(vp, NULL, 0, td);
984263bc
MD
1268 }
1269 s = splbio();
1270 if (LIST_FIRST(slp) == vp) {
1271 /*
1272 * Note: v_tag VT_VFS vps can remain on the
1273 * worklist too with no dirty blocks, but
1274 * since sync_fsync() moves it to a different
1275 * slot we are safe.
1276 */
1277 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1278 !vn_isdisk(vp, NULL))
1279 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1280 /*
1281 * Put us back on the worklist. The worklist
1282 * routine will remove us from our current
1283 * position and then add us back in at a later
1284 * position.
1285 */
1286 vn_syncer_add_to_worklist(vp, syncdelay);
1287 }
1288 splx(s);
1289 }
1290
1291 /*
1292 * Do soft update processing.
1293 */
1294 if (bioops.io_sync)
1295 (*bioops.io_sync)(NULL);
1296
1297 /*
1298 * The variable rushjob allows the kernel to speed up the
1299 * processing of the filesystem syncer process. A rushjob
1300 * value of N tells the filesystem syncer to process the next
1301 * N seconds worth of work on its queue ASAP. Currently rushjob
1302 * is used by the soft update code to speed up the filesystem
1303 * syncer process when the incore state is getting so far
1304 * ahead of the disk that the kernel memory pool is being
1305 * threatened with exhaustion.
1306 */
1307 if (rushjob > 0) {
1308 rushjob -= 1;
1309 continue;
1310 }
1311 /*
1312 * If it has taken us less than a second to process the
1313 * current work, then wait. Otherwise start right over
1314 * again. We can still lose time if any single round
1315 * takes more than two seconds, but it does not really
1316 * matter as we are just trying to generally pace the
1317 * filesystem activity.
1318 */
1319 if (time_second == starttime)
377d4740 1320 tsleep(&lbolt, 0, "syncer", 0);
984263bc
MD
1321 }
1322}
1323
1324/*
1325 * Request the syncer daemon to speed up its work.
1326 * We never push it to speed up more than half of its
1327 * normal turn time, otherwise it could take over the cpu.
f5865223
MD
1328 *
1329 * YYY wchan field protected by the BGL.
984263bc
MD
1330 */
1331int
1332speedup_syncer()
1333{
f5865223
MD
1334 crit_enter();
1335 if (updatethread->td_wchan == &lbolt) { /* YYY */
1336 unsleep(updatethread);
1337 lwkt_schedule(updatethread);
1338 }
1339 crit_exit();
984263bc
MD
1340 if (rushjob < syncdelay / 2) {
1341 rushjob += 1;
1342 stat_rush_requests += 1;
1343 return (1);
1344 }
1345 return(0);
1346}
1347
1348/*
1349 * Associate a p-buffer with a vnode.
1350 *
1351 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1352 * with the buffer. i.e. the bp has not been linked into the vnode or
1353 * ref-counted.
1354 */
1355void
1356pbgetvp(vp, bp)
1fd87d54
RG
1357 struct vnode *vp;
1358 struct buf *bp;
984263bc
MD
1359{
1360
1361 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1362
1363 bp->b_vp = vp;
1364 bp->b_flags |= B_PAGING;
1365 bp->b_dev = vn_todev(vp);
1366}
1367
1368/*
1369 * Disassociate a p-buffer from a vnode.
1370 */
1371void
1372pbrelvp(bp)
1fd87d54 1373 struct buf *bp;
984263bc
MD
1374{
1375
1376 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1377
1378 /* XXX REMOVE ME */
1379 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1380 panic(
1381 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1382 bp,
1383 (int)bp->b_flags
1384 );
1385 }
1386 bp->b_vp = (struct vnode *) 0;
1387 bp->b_flags &= ~B_PAGING;
1388}
1389
1390void
1391pbreassignbuf(bp, newvp)
1392 struct buf *bp;
1393 struct vnode *newvp;
1394{
1395 if ((bp->b_flags & B_PAGING) == 0) {
1396 panic(
1397 "pbreassignbuf() on non phys bp %p",
1398 bp
1399 );
1400 }
1401 bp->b_vp = newvp;
1402}
1403
1404/*
1405 * Reassign a buffer from one vnode to another.
1406 * Used to assign file specific control information
1407 * (indirect blocks) to the vnode to which they belong.
1408 */
1409void
1410reassignbuf(bp, newvp)
1fd87d54
RG
1411 struct buf *bp;
1412 struct vnode *newvp;
984263bc
MD
1413{
1414 struct buflists *listheadp;
1415 int delay;
1416 int s;
1417
1418 if (newvp == NULL) {
1419 printf("reassignbuf: NULL");
1420 return;
1421 }
1422 ++reassignbufcalls;
1423
1424 /*
1425 * B_PAGING flagged buffers cannot be reassigned because their vp
1426 * is not fully linked in.
1427 */
1428 if (bp->b_flags & B_PAGING)
1429 panic("cannot reassign paging buffer");
1430
1431 s = splbio();
1432 /*
1433 * Delete from old vnode list, if on one.
1434 */
1435 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1436 if (bp->b_xflags & BX_VNDIRTY)
1437 listheadp = &bp->b_vp->v_dirtyblkhd;
1438 else
1439 listheadp = &bp->b_vp->v_cleanblkhd;
1440 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1441 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1442 if (bp->b_vp != newvp) {
1443 vdrop(bp->b_vp);
1444 bp->b_vp = NULL; /* for clarification */
1445 }
1446 }
1447 /*
1448 * If dirty, put on list of dirty buffers; otherwise insert onto list
1449 * of clean buffers.
1450 */
1451 if (bp->b_flags & B_DELWRI) {
1452 struct buf *tbp;
1453
1454 listheadp = &newvp->v_dirtyblkhd;
1455 if ((newvp->v_flag & VONWORKLST) == 0) {
1456 switch (newvp->v_type) {
1457 case VDIR:
1458 delay = dirdelay;
1459 break;
1460 case VCHR:
1461 case VBLK:
1462 if (newvp->v_specmountpoint != NULL) {
1463 delay = metadelay;
1464 break;
1465 }
1466 /* fall through */
1467 default:
1468 delay = filedelay;
1469 }
1470 vn_syncer_add_to_worklist(newvp, delay);
1471 }
1472 bp->b_xflags |= BX_VNDIRTY;
1473 tbp = TAILQ_FIRST(listheadp);
1474 if (tbp == NULL ||
1475 bp->b_lblkno == 0 ||
1476 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1477 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1478 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1479 ++reassignbufsortgood;
1480 } else if (bp->b_lblkno < 0) {
1481 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1482 ++reassignbufsortgood;
1483 } else if (reassignbufmethod == 1) {
1484 /*
1485 * New sorting algorithm, only handle sequential case,
1486 * otherwise append to end (but before metadata)
1487 */
1488 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1489 (tbp->b_xflags & BX_VNDIRTY)) {
1490 /*
1491 * Found the best place to insert the buffer
1492 */
1493 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1494 ++reassignbufsortgood;
1495 } else {
1496 /*
1497 * Missed, append to end, but before meta-data.
1498 * We know that the head buffer in the list is
1499 * not meta-data due to prior conditionals.
1500 *
1501 * Indirect effects: NFS second stage write
1502 * tends to wind up here, giving maximum
1503 * distance between the unstable write and the
1504 * commit rpc.
1505 */
1506 tbp = TAILQ_LAST(listheadp, buflists);
1507 while (tbp && tbp->b_lblkno < 0)
1508 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1509 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1510 ++reassignbufsortbad;
1511 }
1512 } else {
1513 /*
1514 * Old sorting algorithm, scan queue and insert
1515 */
1516 struct buf *ttbp;
1517 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1518 (ttbp->b_lblkno < bp->b_lblkno)) {
1519 ++reassignbufloops;
1520 tbp = ttbp;
1521 }
1522 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1523 }
1524 } else {
1525 bp->b_xflags |= BX_VNCLEAN;
1526 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1527 if ((newvp->v_flag & VONWORKLST) &&
1528 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1529 newvp->v_flag &= ~VONWORKLST;
1530 LIST_REMOVE(newvp, v_synclist);
1531 }
1532 }
1533 if (bp->b_vp != newvp) {
1534 bp->b_vp = newvp;
1535 vhold(bp->b_vp);
1536 }
1537 splx(s);
1538}
1539
1540/*
1541 * Create a vnode for a block device.
1542 * Used for mounting the root file system.
1543 */
1544int
1545bdevvp(dev, vpp)
1546 dev_t dev;
1547 struct vnode **vpp;
1548{
1fd87d54 1549 struct vnode *vp;
984263bc
MD
1550 struct vnode *nvp;
1551 int error;
1552
1553 if (dev == NODEV) {
1554 *vpp = NULLVP;
1555 return (ENXIO);
1556 }
1557 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1558 if (error) {
1559 *vpp = NULLVP;
1560 return (error);
1561 }
1562 vp = nvp;
1563 vp->v_type = VBLK;
1564 addalias(vp, dev);
1565 *vpp = vp;
1566 return (0);
1567}
1568
1569/*
7732c8cd 1570 * Add a vnode to the alias list hung off the dev_t.
984263bc
MD
1571 *
1572 * The reason for this gunk is that multiple vnodes can reference
1573 * the same physical device, so checking vp->v_usecount to see
1574 * how many users there are is inadequate; the v_usecount for
1575 * the vnodes need to be accumulated. vcount() does that.
1576 */
1577void
7732c8cd 1578addaliasu(struct vnode *nvp, udev_t nvp_rdev)
984263bc 1579{
7732c8cd 1580 dev_t dev;
984263bc
MD
1581
1582 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1583 panic("addaliasu on non-special vnode");
7732c8cd 1584 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
4908c260
JH
1585 if (dev != NODEV) {
1586 nvp->v_rdev = dev;
7732c8cd 1587 addalias(nvp, dev);
4908c260
JH
1588 } else
1589 nvp->v_rdev = NULL;
984263bc
MD
1590}
1591
1592void
7732c8cd 1593addalias(struct vnode *nvp, dev_t dev)
984263bc 1594{
41a01a4d 1595 lwkt_tokref ilock;
984263bc
MD
1596
1597 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1598 panic("addalias on non-special vnode");
1599
1600 nvp->v_rdev = dev;
41a01a4d 1601 lwkt_gettoken(&ilock, &spechash_token);
984263bc 1602 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
41a01a4d 1603 lwkt_reltoken(&ilock);
984263bc
MD
1604}
1605
1606/*
1607 * Grab a particular vnode from the free list, increment its
1608 * reference count and lock it. The vnode lock bit is set if the
1609 * vnode is being eliminated in vgone. The process is awakened
1610 * when the transition is completed, and an error returned to
1611 * indicate that the vnode is no longer usable (possibly having
1612 * been changed to a new file system type).
41a01a4d
MD
1613 *
1614 * This code is very sensitive. We are depending on the vnode interlock
1615 * to be maintained through to the vn_lock() call, which means that we
1616 * cannot block which means that we cannot call vbusy() until after vn_lock().
1617 * If the interlock is not maintained, the VXLOCK check will not properly
1618 * interlock against a vclean()'s LK_DRAIN operation on the lock.
984263bc
MD
1619 */
1620int
41a01a4d 1621vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
984263bc
MD
1622{
1623 int error;
41a01a4d
MD
1624 lwkt_tokref vvlock;
1625
1626 /*
1627 * We need the interlock to safely modify the v_ fields. ZZZ it is
1628 * only legal to pass (1) the vnode's interlock and (2) only pass
1629 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1630 * held.
1631 */
1632 if ((flags & LK_INTERLOCK) == 0) {
1633 lwkt_gettoken(&vvlock, vp->v_interlock);
1634 vlock = &vvlock;
1635 }
984263bc
MD
1636
1637 /*
1638 * If the vnode is in the process of being cleaned out for
1639 * another use, we wait for the cleaning to finish and then
1640 * return failure. Cleaning is determined by checking that
41a01a4d
MD
1641 * the VXLOCK flag is set. It is possible for the vnode to be
1642 * self-referenced during the cleaning operation.
984263bc 1643 */
984263bc 1644 if (vp->v_flag & VXLOCK) {
41a01a4d 1645 if (vp->v_vxthread == curthread) {
984263bc
MD
1646#if 0
1647 /* this can now occur in normal operation */
1648 log(LOG_INFO, "VXLOCK interlock avoided\n");
1649#endif
1650 } else {
1651 vp->v_flag |= VXWANT;
41a01a4d 1652 lwkt_reltoken(vlock);
377d4740 1653 tsleep((caddr_t)vp, 0, "vget", 0);
984263bc
MD
1654 return (ENOENT);
1655 }
1656 }
1657
2811dfbd 1658 /*
41a01a4d
MD
1659 * Bump v_usecount to prevent the vnode from being recycled. The
1660 * usecount needs to be bumped before we successfully get our lock.
2811dfbd 1661 */
41a01a4d 1662 vp->v_usecount++;
984263bc 1663 if (flags & LK_TYPE_MASK) {
41a01a4d 1664 if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
984263bc
MD
1665 /*
1666 * must expand vrele here because we do not want
1667 * to call VOP_INACTIVE if the reference count
1668 * drops back to zero since it was never really
1669 * active. We must remove it from the free list
1670 * before sleeping so that multiple processes do
1671 * not try to recycle it.
1672 */
41a01a4d 1673 lwkt_gettokref(vlock);
984263bc 1674 vp->v_usecount--;
41a01a4d
MD
1675 vmaybefree(vp);
1676 lwkt_reltoken(vlock);
984263bc
MD
1677 }
1678 return (error);
1679 }
41a01a4d
MD
1680 if (VSHOULDBUSY(vp))
1681 vbusy(vp); /* interlock must be held on call */
1682 lwkt_reltoken(vlock);
984263bc
MD
1683 return (0);
1684}
1685
1686void
1687vref(struct vnode *vp)
1688{
2811dfbd 1689 vp->v_usecount++; /* XXX MP */
984263bc
MD
1690}
1691
1692/*
1693 * Vnode put/release.
1694 * If count drops to zero, call inactive routine and return to freelist.
1695 */
1696void
dadab5e9 1697vrele(struct vnode *vp)
984263bc 1698{
dadab5e9 1699 struct thread *td = curthread; /* XXX */
41a01a4d 1700 lwkt_tokref vlock;
984263bc
MD
1701
1702 KASSERT(vp != NULL, ("vrele: null vp"));
1703
41a01a4d 1704 lwkt_gettoken(&vlock, vp->v_interlock);
984263bc
MD
1705
1706 if (vp->v_usecount > 1) {
984263bc 1707 vp->v_usecount--;
41a01a4d 1708 lwkt_reltoken(&vlock);
984263bc
MD
1709 return;
1710 }
1711
1712 if (vp->v_usecount == 1) {
1713 vp->v_usecount--;
1714 /*
41a01a4d
MD
1715 * We must call VOP_INACTIVE with the node locked and the
1716 * usecount 0. If we are doing a vpu, the node is already
1717 * locked, but, in the case of vrele, we must explicitly lock
1718 * the vnode before calling VOP_INACTIVE.
984263bc
MD
1719 */
1720
41a01a4d 1721 if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
dadab5e9 1722 VOP_INACTIVE(vp, td);
41a01a4d
MD
1723 vmaybefree(vp);
1724 lwkt_reltoken(&vlock);
984263bc
MD
1725 } else {
1726#ifdef DIAGNOSTIC
1727 vprint("vrele: negative ref count", vp);
984263bc 1728#endif
41a01a4d 1729 lwkt_reltoken(&vlock);
984263bc
MD
1730 panic("vrele: negative ref cnt");
1731 }
1732}
1733
1734void
dadab5e9 1735vput(struct vnode *vp)
984263bc 1736{
dadab5e9 1737 struct thread *td = curthread; /* XXX */
41a01a4d 1738 lwkt_tokref vlock;
984263bc
MD
1739
1740 KASSERT(vp != NULL, ("vput: null vp"));
1741
41a01a4d 1742 lwkt_gettoken(&vlock, vp->v_interlock);
984263bc
MD
1743
1744 if (vp->v_usecount > 1) {
1745 vp->v_usecount--;
41a01a4d 1746 VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
984263bc
MD
1747 return;
1748 }
1749
1750 if (vp->v_usecount == 1) {
1751 vp->v_usecount--;
1752 /*
1753 * We must call VOP_INACTIVE with the node locked.
1754 * If we are doing a vpu, the node is already locked,
1755 * so we just need to release the vnode mutex.
1756 */
dadab5e9 1757 VOP_INACTIVE(vp, td);
41a01a4d
MD
1758 vmaybefree(vp);
1759 lwkt_reltoken(&vlock);
984263bc
MD
1760 } else {
1761#ifdef DIAGNOSTIC
1762 vprint("vput: negative ref count", vp);
1763#endif
41a01a4d 1764 lwkt_reltoken(&vlock);
984263bc
MD
1765 panic("vput: negative ref cnt");
1766 }
1767}
1768
1769/*
41a01a4d
MD
1770 * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1771 * be held but isn't.
984263bc
MD
1772 */
1773void
1774vhold(vp)
1fd87d54 1775 struct vnode *vp;
984263bc
MD
1776{
1777 int s;
1778
1779 s = splbio();
1780 vp->v_holdcnt++;
1781 if (VSHOULDBUSY(vp))
41a01a4d 1782 vbusy(vp); /* interlock must be held on call */
984263bc
MD
1783 splx(s);
1784}
1785
1786/*
1787 * One less who cares about this vnode.
1788 */
1789void
1790vdrop(vp)
1fd87d54 1791 struct vnode *vp;
984263bc 1792{
41a01a4d 1793 lwkt_tokref vlock;
984263bc 1794
41a01a4d 1795 lwkt_gettoken(&vlock, vp->v_interlock);
984263bc
MD
1796 if (vp->v_holdcnt <= 0)
1797 panic("vdrop: holdcnt");
1798 vp->v_holdcnt--;
41a01a4d
MD
1799 vmaybefree(vp);
1800 lwkt_reltoken(&vlock);
1801}
1802
1803int
1804vmntvnodescan(
1805 struct mount *mp,
1806 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1807 int (*slowfunc)(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data),
1808 void *data
1809) {
1810 lwkt_tokref ilock;
1811 lwkt_tokref vlock;
1812 struct vnode *pvp;
1813 struct vnode *vp;
1814 int r = 0;
1815
1816 /*
1817 * Scan the vnodes on the mount's vnode list. Use a placemarker
1818 */
1819 pvp = zalloc(vnode_zone);
1820 pvp->v_flag |= VPLACEMARKER;
1821
1822 lwkt_gettoken(&ilock, &mntvnode_token);
1823 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1824
1825 while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1826 /*
1827 * Move the placemarker and skip other placemarkers we
1828 * encounter. The nothing can get in our way so the
1829 * mount point on the vp must be valid.
1830 */
1831 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1832 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1833 if (vp->v_flag & VPLACEMARKER)
1834 continue;
1835 KKASSERT(vp->v_mount == mp);
1836
1837 /*
1838 * Quick test
1839 */
1840 if (fastfunc) {
1841 if ((r = fastfunc(mp, vp, data)) < 0)
1842 continue;
1843 if (r)
1844 break;
1845 }
1846
1847 /*
1848 * Get the vnodes interlock and make sure it is still on the
1849 * mount list. Skip it if it has moved (we may encounter it
1850 * later). Then do the with-interlock test. The callback
1851 * is responsible for releasing the vnode interlock.
1852 *
1853 * The interlock is type-stable.
1854 */
1855 if (slowfunc) {
1856 lwkt_gettoken(&vlock, vp->v_interlock);
1857 if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1858 printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1859 lwkt_reltoken(&vlock);
1860 continue;
1861 }
1862 if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1863 KKASSERT(lwkt_havetokref(&vlock) == 0);
1864 break;
1865 }
1866 KKASSERT(lwkt_havetokref(&vlock) == 0);
1867 }
1868 }
1869 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1870 zfree(vnode_zone, pvp);
1871 lwkt_reltoken(&ilock);
1872 return(r);
984263bc
MD
1873}
1874
1875/*
1876 * Remove any vnodes in the vnode table belonging to mount point mp.
1877 *
1878 * If FORCECLOSE is not specified, there should not be any active ones,
1879 * return error if any are found (nb: this is a user error, not a
1880 * system error). If FORCECLOSE is specified, detach any active vnodes
1881 * that are found.
1882 *
1883 * If WRITECLOSE is set, only flush out regular file vnodes open for
1884 * writing.
1885 *
1886 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1887 *
1888 * `rootrefs' specifies the base reference count for the root vnode
1889 * of this filesystem. The root vnode is considered busy if its
1890 * v_usecount exceeds this value. On a successful return, vflush()
1891 * will call vrele() on the root vnode exactly rootrefs times.
1892 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1893 * be zero.
1894 */
1895#ifdef DIAGNOSTIC
1896static int busyprt = 0; /* print out busy vnodes */
1897SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1898#endif
1899
41a01a4d
MD
1900static int vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data);
1901
1902struct vflush_info {
1903 int flags;
1904 int busy;
1905 thread_t td;
1906};
1907
984263bc
MD
1908int
1909vflush(mp, rootrefs, flags)
1910 struct mount *mp;
1911 int rootrefs;
1912 int flags;
1913{
dadab5e9 1914 struct thread *td = curthread; /* XXX */
41a01a4d
MD
1915 struct vnode *rootvp = NULL;
1916 int error;
1917 lwkt_tokref vlock;
1918 struct vflush_info vflush_info;
984263bc
MD
1919
1920 if (rootrefs > 0) {
1921 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1922 ("vflush: bad args"));
1923 /*
1924 * Get the filesystem root vnode. We can vput() it
1925 * immediately, since with rootrefs > 0, it won't go away.
1926 */
1927 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1928 return (error);
1929 vput(rootvp);
1930 }
984263bc 1931
41a01a4d
MD
1932 vflush_info.busy = 0;
1933 vflush_info.flags = flags;
1934 vflush_info.td = td;
1935 vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
984263bc 1936
984263bc
MD
1937 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1938 /*
1939 * If just the root vnode is busy, and if its refcount
1940 * is equal to `rootrefs', then go ahead and kill it.
1941 */
41a01a4d
MD
1942 lwkt_gettoken(&vlock, rootvp->v_interlock);
1943 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
984263bc 1944 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
41a01a4d
MD
1945 if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1946 vgonel(rootvp, &vlock, td);
1947 vflush_info.busy = 0;
1948 } else {
1949 lwkt_reltoken(&vlock);
1950 }
984263bc 1951 }
41a01a4d 1952 if (vflush_info.busy)
984263bc
MD
1953 return (EBUSY);
1954 for (; rootrefs > 0; rootrefs--)
1955 vrele(rootvp);
1956 return (0);
1957}
1958
1959/*
41a01a4d 1960 * The scan callback is made with an interlocked vnode.
984263bc 1961 */
41a01a4d
MD
1962static int
1963vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
984263bc 1964{
41a01a4d
MD
1965 struct vflush_info *info = data;
1966 struct vattr vattr;
984263bc 1967
41a01a4d
MD
1968 /*
1969 * Skip over a vnodes marked VSYSTEM.
1970 */
1971 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1972 lwkt_reltoken(vlock);
1973 return(0);
1974 }
1975
1976 /*
1977 * If WRITECLOSE is set, flush out unlinked but still open
1978 * files (even if open only for reading) and regular file
1979 * vnodes open for writing.
1980 */
1981 if ((info->flags & WRITECLOSE) &&
1982 (vp->v_type == VNON ||
1983 (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
1984 vattr.va_nlink > 0)) &&
1985 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1986 lwkt_reltoken(vlock);
1987 return(0);
1988 }
1989
1990 /*
1991 * With v_usecount == 0, all we need to do is clear out the
1992 * vnode data structures and we are done.
1993 */
1994 if (vp->v_usecount == 0) {
1995 vgonel(vp, vlock, info->td);
1996 return(0);
1997 }
1998
1999 /*
2000 * If FORCECLOSE is set, forcibly close the vnode. For block
2001 * or character devices, revert to an anonymous device. For
2002 * all other files, just kill them.
2003 */
2004 if (info->flags & FORCECLOSE) {
2005 if (vp->v_type != VBLK && vp->v_type != VCHR) {
2006 vgonel(vp, vlock, info->td);
2007 } else {
2008 vclean(vp, vlock, 0, info->td);
2009 vp->v_op = spec_vnodeop_p;
2010 insmntque(vp, (struct mount *) 0);
2011 }
2012 return(0);
984263bc 2013 }
41a01a4d
MD
2014#ifdef DIAGNOSTIC
2015 if (busyprt)
2016 vprint("vflush: busy vnode", vp);
984263bc 2017#endif
41a01a4d
MD
2018 lwkt_reltoken(vlock);
2019 ++info->busy;
2020 return(0);
984263bc
MD
2021}
2022
2023/*
2024 * Disassociate the underlying file system from a vnode.
2025 */
2026static void
41a01a4d 2027vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
984263bc
MD
2028{
2029 int active;
2030
2031 /*
2032 * Check to see if the vnode is in use. If so we have to reference it
2033 * before we clean it out so that its count cannot fall to zero and
2034 * generate a race against ourselves to recycle it.
2035 */
2036 if ((active = vp->v_usecount))
2037 vp->v_usecount++;
2038
2039 /*
2040 * Prevent the vnode from being recycled or brought into use while we
2041 * clean it out.
2042 */
2043 if (vp->v_flag & VXLOCK)
2044 panic("vclean: deadlock");
2045 vp->v_flag |= VXLOCK;
41a01a4d
MD
2046 vp->v_vxthread = curthread;
2047
984263bc
MD
2048 /*
2049 * Even if the count is zero, the VOP_INACTIVE routine may still
2050 * have the object locked while it cleans it out. The VOP_LOCK
2051 * ensures that the VOP_INACTIVE routine is done with its work.
2052 * For active vnodes, it ensures that no other activity can
2053 * occur while the underlying object is being cleaned out.
41a01a4d
MD
2054 *
2055 * NOTE: we continue to hold the vnode interlock through to the
2056 * end of vclean().
984263bc 2057 */
41a01a4d 2058 VOP_LOCK(vp, NULL, LK_DRAIN, td);
984263bc
MD
2059
2060 /*
2061 * Clean out any buffers associated with the vnode.
2062 */
3b568787 2063 vinvalbuf(vp, V_SAVE, td, 0, 0);
984263bc
MD
2064 VOP_DESTROYVOBJECT(vp);
2065
2066 /*
2067 * If purging an active vnode, it must be closed and
2068 * deactivated before being reclaimed. Note that the
2069 * VOP_INACTIVE will unlock the vnode.
2070 */
2071 if (active) {
2072 if (flags & DOCLOSE)
3b568787 2073 VOP_CLOSE(vp, FNONBLOCK, td);
dadab5e9 2074 VOP_INACTIVE(vp, td);
984263bc
MD
2075 } else {
2076 /*
2077 * Any other processes trying to obtain this lock must first
2078 * wait for VXLOCK to clear, then call the new lock operation.
2079 */
41a01a4d 2080 VOP_UNLOCK(vp, NULL, 0, td);
984263bc
MD
2081 }
2082 /*
2083 * Reclaim the vnode.
2084 */
dadab5e9 2085 if (VOP_RECLAIM(vp, td))
984263bc
MD
2086 panic("vclean: cannot reclaim");
2087
2088 if (active) {
2089 /*
2090 * Inline copy of vrele() since VOP_INACTIVE
2091 * has already been called.
2092 */
984263bc
MD
2093 if (--vp->v_usecount <= 0) {
2094#ifdef DIAGNOSTIC
2095 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2096 vprint("vclean: bad ref count", vp);
2097 panic("vclean: ref cnt");
2098 }
2099#endif
2100 vfree(vp);
2101 }
984263bc
MD
2102 }
2103
2104 cache_purge(vp);
2105 vp->v_vnlock = NULL;
41a01a4d 2106 vmaybefree(vp);
984263bc
MD
2107
2108 /*
2109 * Done with purge, notify sleepers of the grim news.
2110 */
2111 vp->v_op = dead_vnodeop_p;
2112 vn_pollgone(vp);
2113 vp->v_tag = VT_NON;
2114 vp->v_flag &= ~VXLOCK;
41a01a4d 2115 vp->v_vxthread = NULL;
984263bc
MD
2116 if (vp->v_flag & VXWANT) {
2117 vp->v_flag &= ~VXWANT;
2118 wakeup((caddr_t) vp);
2119 }
41a01a4d 2120 lwkt_reltoken(vlock);
984263bc
MD
2121}
2122
2123/*
2124 * Eliminate all activity associated with the requested vnode
2125 * and with all vnodes aliased to the requested vnode.
2126 */
2127int
2128vop_revoke(ap)
2129 struct vop_revoke_args /* {
2130 struct vnode *a_vp;
2131 int a_flags;
2132 } */ *ap;
2133{
2134 struct vnode *vp, *vq;
41a01a4d 2135 lwkt_tokref ilock;
984263bc
MD
2136 dev_t dev;
2137
2138 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2139
2140 vp = ap->a_vp;
2141 /*
2142 * If a vgone (or vclean) is already in progress,
2143 * wait until it is done and return.
2144 */
2145 if (vp->v_flag & VXLOCK) {
2146 vp->v_flag |= VXWANT;
41a01a4d 2147 /*lwkt_reltoken(vlock); ZZZ */
377d4740 2148 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
984263bc
MD
2149 return (0);
2150 }
2151 dev = vp->v_rdev;
2152 for (;;) {
41a01a4d 2153 lwkt_gettoken(&ilock, &spechash_token);
984263bc 2154 vq = SLIST_FIRST(&dev->si_hlist);
41a01a4d 2155 lwkt_reltoken(&ilock);
984263bc
MD
2156 if (!vq)
2157 break;
2158 vgone(vq);
2159 }
2160 return (0);
2161}
2162
2163/*
2164 * Recycle an unused vnode to the front of the free list.
2165 * Release the passed interlock if the vnode will be recycled.
2166 */
2167int
41a01a4d 2168vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
984263bc 2169{
41a01a4d
MD
2170 lwkt_tokref vlock;
2171
2172 lwkt_gettoken(&vlock, vp->v_interlock);
984263bc 2173 if (vp->v_usecount == 0) {
41a01a4d 2174 if (inter_lkp)
8a8d5d85 2175 lwkt_reltoken(inter_lkp);
41a01a4d 2176 vgonel(vp, &vlock, td);
984263bc
MD
2177 return (1);
2178 }
41a01a4d 2179 lwkt_reltoken(&vlock);
984263bc
MD
2180 return (0);
2181}
2182
2183/*
2184 * Eliminate all activity associated with a vnode
2185 * in preparation for reuse.
2186 */
2187void
dadab5e9 2188vgone(struct vnode *vp)
984263bc 2189{
dadab5e9 2190 struct thread *td = curthread; /* XXX */
41a01a4d 2191 lwkt_tokref vlock;
984263bc 2192
41a01a4d
MD
2193 lwkt_gettoken(&vlock, vp->v_interlock);
2194 vgonel(vp, &vlock, td);
984263bc
MD
2195}
2196
2197/*
2198 * vgone, with the vp interlock held.
2199 */
2200void
41a01a4d 2201vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
984263bc 2202{
41a01a4d 2203 lwkt_tokref ilock;
984263bc
MD
2204 int s;
2205
2206 /*
2207 * If a vgone (or vclean) is already in progress,
2208 * wait until it is done and return.
2209 */
2210 if (vp->v_flag & VXLOCK) {
2211 vp->v_flag |= VXWANT;
41a01a4d 2212 lwkt_reltoken(vlock);
377d4740 2213 tsleep((caddr_t)vp, 0, "vgone", 0);
984263bc
MD
2214 return;
2215 }
2216
2217 /*
2218 * Clean out the filesystem specific data.
2219 */
41a01a4d
MD
2220 vclean(vp, vlock, DOCLOSE, td);
2221 lwkt_gettokref(vlock);
984263bc
MD
2222
2223 /*
2224 * Delete from old mount point vnode list, if on one.
2225 */
2226 if (vp->v_mount != NULL)
2227 insmntque(vp, (struct mount *)0);
2228 /*
2229 * If special device, remove it from special device alias list
2230 * if it is on one.
2231 */
2232 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
41a01a4d 2233 lwkt_gettoken(&ilock, &spechash_token);
984263bc
MD
2234 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2235 freedev(vp->v_rdev);
41a01a4d 2236 lwkt_reltoken(&ilock);
984263bc
MD
2237 vp->v_rdev = NULL;
2238 }
2239
2240 /*
2241 * If it is on the freelist and not already at the head,
2242 * move it to the head of the list. The test of the
2243 * VDOOMED flag and the reference count of zero is because
2244 * it will be removed from the free list by getnewvnode,
2245 * but will not have its reference count incremented until
2246 * after calling vgone. If the reference count were
2247 * incremented first, vgone would (incorrectly) try to
2248 * close the previous instance of the underlying object.
2249 */
2250 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2251 s = splbio();
41a01a4d 2252 lwkt_gettoken(&ilock, &vnode_free_list_token);
984263bc
MD
2253 if (vp->v_flag & VFREE)
2254 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2255 else
2256 freevnodes++;
2257 vp->v_flag |= VFREE;
2258 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
41a01a4d 2259 lwkt_reltoken(&ilock);
984263bc
MD
2260 splx(s);
2261 }
984263bc 2262 vp->v_type = VBAD;
41a01a4d 2263 lwkt_reltoken(vlock);
984263bc
MD
2264}
2265
2266/*
2267 * Lookup a vnode by device number.
2268 */
2269int
2270vfinddev(dev, type, vpp)
2271 dev_t dev;
2272 enum vtype type;
2273 struct vnode **vpp;
2274{
41a01a4d 2275 lwkt_tokref ilock;
984263bc
MD
2276 struct vnode *vp;
2277
41a01a4d 2278 lwkt_gettoken(&ilock, &spechash_token);
984263bc
MD
2279 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2280 if (type == vp->v_type) {
2281 *vpp = vp;
41a01a4d 2282 lwkt_reltoken(&ilock);
984263bc
MD
2283 return (1);
2284 }
2285 }
41a01a4d 2286 lwkt_reltoken(&ilock);
984263bc
MD
2287 return (0);
2288}
2289
2290/*
2291 * Calculate the total number of references to a special device.
2292 */
2293int
2294vcount(vp)
2295 struct vnode *vp;
2296{
41a01a4d 2297 lwkt_tokref ilock;
984263bc
MD
2298 struct vnode *vq;
2299 int count;
2300
2301 count = 0;
41a01a4d 2302 lwkt_gettoken(&ilock, &spechash_token);
984263bc
MD
2303 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2304 count += vq->v_usecount;
41a01a4d 2305 lwkt_reltoken(&ilock);
984263bc
MD
2306 return (count);
2307}
2308
2309/*
2310 * Same as above, but using the dev_t as argument
2311 */
2312
2313int
2314count_dev(dev)
2315 dev_t dev;
2316{
2317 struct vnode *vp;
2318
2319 vp = SLIST_FIRST(&dev->si_hlist);
2320 if (vp == NULL)
2321 return (0);
2322 return(vcount(vp));
2323}
2324
2325/*
2326 * Print out a description of a vnode.
2327 */
2328static char *typename[] =
2329{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2330
2331void
2332vprint(label, vp)
2333 char *label;
2334 struct vnode *vp;
2335{
2336 char buf[96];
2337
2338 if (label != NULL)
2339 printf("%s: %p: ", label, (void *)vp);
2340 else
2341 printf("%p: ", (void *)vp);
2342 printf("type %s, usecount %d, writecount %d, refcount %d,",
2343 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2344 vp->v_holdcnt);
2345 buf[0] = '\0';
2346 if (vp->v_flag & VROOT)
2347 strcat(buf, "|VROOT");
2348 if (vp->v_flag & VTEXT)
2349 strcat(buf, "|VTEXT");
2350 if (vp->v_flag & VSYSTEM)
2351 strcat(buf, "|VSYSTEM");
2352 if (vp->v_flag & VXLOCK)
2353 strcat(buf, "|VXLOCK");
2354 if (vp->v_flag & VXWANT)
2355 strcat(buf, "|VXWANT");
2356 if (vp->v_flag & VBWAIT)
2357 strcat(buf, "|VBWAIT");
2358 if (vp->v_flag & VDOOMED)
2359 strcat(buf, "|VDOOMED");
2360 if (vp->v_flag & VFREE)
2361 strcat(buf, "|VFREE");
2362 if (vp->v_flag & VOBJBUF)
2363 strcat(buf, "|VOBJBUF");
2364 if (buf[0] != '\0')
2365 printf(" flags (%s)", &buf[1]);
2366 if (vp->v_data == NULL) {
2367 printf("\n");
2368 } else {
2369 printf("\n\t");
2370 VOP_PRINT(vp);
2371 }
2372}
2373
2374#ifdef DDB
2375#include <ddb/ddb.h>
2376/*
2377 * List all of the locked vnodes in the system.
2378 * Called when debugging the kernel.
2379 */
2380DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2381{
dadab5e9 2382 struct thread *td = curthread; /* XXX */
41a01a4d 2383 lwkt_tokref ilock;
984263bc
MD
2384 struct mount *mp, *nmp;
2385 struct vnode *vp;
2386
2387 printf("Locked vnodes\n");
41a01a4d 2388 lwkt_gettoken(&ilock, &mountlist_token);
984263bc 2389 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
41a01a4d 2390 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
984263bc
MD
2391 nmp = TAILQ_NEXT(mp, mnt_list);
2392 continue;
2393 }
2394 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2395 if (VOP_ISLOCKED(vp, NULL))
2396 vprint((char *)0, vp);
2397 }
41a01a4d 2398 lwkt_gettokref(&ilock);
984263bc 2399 nmp = TAILQ_NEXT(mp, mnt_list);
dadab5e9 2400 vfs_unbusy(mp, td);
984263bc 2401 }
41a01a4d 2402 lwkt_reltoken(&ilock);
984263bc
MD
2403}
2404#endif
2405
2406/*
2407 * Top level filesystem related information gathering.
2408 */
402ed7e1 2409static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
984263bc
MD
2410
2411static int
2412vfs_sysctl(SYSCTL_HANDLER_ARGS)
2413{
2414 int *name = (int *)arg1 - 1; /* XXX */
2415 u_int namelen = arg2 + 1; /* XXX */
2416 struct vfsconf *vfsp;
2417
2418#if 1 || defined(COMPAT_PRELITE2)
2419 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2420 if (namelen == 1)
2421 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2422#endif
2423
2424#ifdef notyet
2425 /* all sysctl names at this level are at least name and field */
2426 if (namelen < 2)
2427 return (ENOTDIR); /* overloaded */
2428 if (name[0] != VFS_GENERIC) {
2429 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2430 if (vfsp->vfc_typenum == name[0])
2431 break;
2432 if (vfsp == NULL)
2433 return (EOPNOTSUPP);
2434 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2435 oldp, oldlenp, newp, newlen, p));
2436 }
2437#endif
2438 switch (name[1]) {
2439 case VFS_MAXTYPENUM:
2440 if (namelen != 2)
2441 return (ENOTDIR);
2442 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2443 case VFS_CONF:
2444 if (namelen != 3)
2445 return (ENOTDIR); /* overloaded */
2446 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2447 if (vfsp->vfc_typenum == name[2])
2448 break;
2449 if (vfsp == NULL)
2450 return (EOPNOTSUPP);
2451 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2452 }
2453 return (EOPNOTSUPP);
2454}
2455
2456SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2457 "Generic filesystem");
2458
2459#if 1 || defined(COMPAT_PRELITE2)
2460
2461static int
2462sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2463{
2464 int error;
2465 struct vfsconf *vfsp;
2466 struct ovfsconf ovfs;
2467
2468 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2469 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2470 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2471 ovfs.vfc_index = vfsp->vfc_typenum;
2472 ovfs.vfc_refcount = vfsp->vfc_refcount;
2473 ovfs.vfc_flags = vfsp->vfc_flags;
2474 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2475 if (error)
2476 return error;
2477 }
2478 return 0;
2479}
2480
2481#endif /* 1 || COMPAT_PRELITE2 */
2482
2483#if 0
2484#define KINFO_VNODESLOP 10
2485/*
2486 * Dump vnode list (via sysctl).
2487 * Copyout address of vnode followed by vnode.
2488 */
2489/* ARGSUSED */
2490static int
2491sysctl_vnode(SYSCTL_HANDLER_ARGS)
2492{
2493 struct proc *p = curproc; /* XXX */
2494 struct mount *mp, *nmp;
2495 struct vnode *nvp, *vp;
41a01a4d
MD
2496 lwkt_tokref ilock;
2497 lwkt_tokref jlock;
984263bc
MD
2498 int error;
2499
2500#define VPTRSZ sizeof (struct vnode *)
2501#define VNODESZ sizeof (struct vnode)
2502
2503 req->lock = 0;
2504 if (!req->oldptr) /* Make an estimate */
2505 return (SYSCTL_OUT(req, 0,
2506 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2507
41a01a4d 2508 lwkt_gettoken(&ilock, &mountlist_token);
984263bc 2509 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
41a01a4d 2510 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
984263bc
MD
2511 nmp = TAILQ_NEXT(mp, mnt_list);
2512 continue;
2513 }
41a01a4d 2514 lwkt_gettoken(&jlock, &mntvnode_token);
984263bc 2515again:
984263bc
MD
2516 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2517 vp != NULL;
2518 vp = nvp) {
2519 /*
2520 * Check that the vp is still associated with
2521 * this filesystem. RACE: could have been
2522 * recycled onto the same filesystem.
2523 */
41a01a4d 2524 if (vp->v_mount != mp)
984263bc 2525 goto again;
984263bc 2526 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
984263bc 2527 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
41a01a4d
MD
2528 (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2529 lwkt_reltoken(&jlock);
984263bc 2530 return (error);
41a01a4d 2531 }
984263bc 2532 }
41a01a4d
MD
2533 lwkt_reltoken(&jlock);
2534 lwkt_gettokref(&ilock);
2535 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */
984263bc
MD
2536 vfs_unbusy(mp, p);
2537 }
41a01a4d 2538 lwkt_reltoken(&ilock);
984263bc
MD
2539
2540 return (0);
2541}
2542#endif
2543
2544/*
2545 * XXX
2546 * Exporting the vnode list on large systems causes them to crash.
2547 * Exporting the vnode list on medium systems causes sysctl to coredump.
2548 */
2549#if 0
2550SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2551 0, 0, sysctl_vnode, "S,vnode", "");
2552#endif
2553
2554/*
2555 * Check to see if a filesystem is mounted on a block device.
2556 */
2557int
2558vfs_mountedon(vp)
2559 struct vnode *vp;
2560{
2561
2562 if (vp->v_specmountpoint != NULL)
2563 return (EBUSY);
2564 return (0);
2565}
2566
2567/*
2568 * Unmount all filesystems. The list is traversed in reverse order
2569 * of mounting to avoid dependencies.
2570 */
2571void
2572vfs_unmountall()
2573{
2574 struct mount *mp;
dadab5e9 2575 struct thread *td = curthread;
984263bc
MD
2576 int error;
2577
dadab5e9
MD
2578 if (td->td_proc == NULL)
2579 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2580
984263bc
MD
2581 /*
2582 * Since this only runs when rebooting, it is not interlocked.
2583 */
2584 while(!TAILQ_EMPTY(&mountlist)) {
2585 mp = TAILQ_LAST(&mountlist, mntlist);
dadab5e9 2586 error = dounmount(mp, MNT_FORCE, td);
984263bc
MD
2587 if (error) {
2588 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2589 printf("unmount of %s failed (",
2590 mp->mnt_stat.f_mntonname);
2591 if (error == EBUSY)
2592 printf("BUSY)\n");
2593 else
2594 printf("%d)\n", error);
2595 } else {
2596 /* The unmount has removed mp from the mountlist */
2597 }
2598 }
2599}
2600
2601/*
2602 * Build hash lists of net addresses and hang them off the mount point.
2603 * Called by ufs_mount() to set up the lists of export addresses.
2604 */
2605static int
2606vfs_hang_addrlist(mp, nep, argp)
2607 struct mount *mp;
2608 struct netexport *nep;
2609 struct export_args *argp;
2610{
1fd87d54
RG
2611 struct netcred *np;
2612 struct radix_node_head *rnh;
2613 int i;
984263bc
MD
2614 struct radix_node *rn;
2615 struct sockaddr *saddr, *smask = 0;
2616 struct domain *dom;
2617 int error;
2618
2619 if (argp->ex_addrlen == 0) {
2620 if (mp->mnt_flag & MNT_DEFEXPORTED)
2621 return (EPERM);
2622 np = &nep->ne_defexported;
2623 np->netc_exflags = argp->ex_flags;
2624 np->netc_anon = argp->ex_anon;
2625 np->netc_anon.cr_ref = 1;
2626 mp->mnt_flag |= MNT_DEFEXPORTED;
2627 return (0);
2628 }
2629
0260ddf9
MD
2630 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2631 return (EINVAL);
2632 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
984263bc
MD
2633 return (EINVAL);
2634
2635 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2636 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2637 bzero((caddr_t) np, i);
2638 saddr = (struct sockaddr *) (np + 1);
2639 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2640 goto out;
2641 if (saddr->sa_len > argp->ex_addrlen)
2642 saddr->sa_len = argp->ex_addrlen;
2643 if (argp->ex_masklen) {
2644 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2645 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2646 if (error)
2647 goto out;
2648 if (smask->sa_len > argp->ex_masklen)
2649 smask->sa_len = argp->ex_masklen;
2650 }
2651 i = saddr->sa_family;
2652 if ((rnh = nep->ne_rtable[i]) == 0) {
2653 /*
2654 * Seems silly to initialize every AF when most are not used,
2655 * do so on demand here
2656 */
2657 for (dom = domains; dom; dom = dom->dom_next)
2658 if (dom->dom_family == i && dom->dom_rtattach) {
2659 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2660 dom->dom_rtoffset);
2661 break;
2662 }
2663 if ((rnh = nep->ne_rtable[i]) == 0) {
2664 error = ENOBUFS;
2665 goto out;
2666 }
2667 }
2668 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2669 np->netc_rnodes);
2670 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2671 error = EPERM;
2672 goto out;
2673 }
2674 np->netc_exflags = argp->ex_flags;
2675 np->netc_anon = argp->ex_anon;
2676 np->netc_anon.cr_ref = 1;
2677 return (0);
2678out:
2679 free(np, M_NETADDR);
2680 return (error);
2681}
2682
2683/* ARGSUSED */
2684static int
2685vfs_free_netcred(rn, w)
2686 struct radix_node *rn;
2687 void *w;
2688{
1fd87d54 2689 struct radix_node_head *rnh = (struct radix_node_head *) w;
984263bc
MD
2690
2691 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2692 free((caddr_t) rn, M_NETADDR);
2693 return (0);
2694}
2695
2696/*
2697 * Free the net address hash lists that are hanging off the mount points.
2698 */
2699static void
2700vfs_free_addrlist(nep)
2701 struct netexport *nep;
2702{
1fd87d54
RG
2703 int i;
2704 struct radix_node_head *rnh;
984263bc
MD
2705
2706 for (i = 0; i <= AF_MAX; i++)
2707 if ((rnh = nep->ne_rtable[i])) {
2708 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2709 (caddr_t) rnh);
2710 free((caddr_t) rnh, M_RTABLE);
2711 nep->ne_rtable[i] = 0;
2712 }
2713}
2714
2715int
2716vfs_export(mp, nep, argp)
2717 struct mount *mp;
2718 struct netexport *nep;
2719 struct export_args *argp;
2720{
2721 int error;
2722
2723 if (argp->ex_flags & MNT_DELEXPORT) {
2724 if (mp->mnt_flag & MNT_EXPUBLIC) {
2725 vfs_setpublicfs(NULL, NULL, NULL);
2726 mp->mnt_flag &= ~MNT_EXPUBLIC;
2727 }
2728 vfs_free_addrlist(nep);
2729 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2730 }
2731 if (argp->ex_flags & MNT_EXPORTED) {
2732 if (argp->ex_flags & MNT_EXPUBLIC) {
2733 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2734 return (error);
2735 mp->mnt_flag |= MNT_EXPUBLIC;
2736 }
2737 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2738 return (error);
2739 mp->mnt_flag |= MNT_EXPORTED;
2740 }
2741 return (0);
2742}
2743
2744
2745/*
2746 * Set the publicly exported filesystem (WebNFS). Currently, only
2747 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2748 */
2749int
2750vfs_setpublicfs(mp, nep, argp)
2751 struct mount *mp;
2752 struct netexport *nep;
2753 struct export_args *argp;
2754{
2755 int error;
2756 struct vnode *rvp;
2757 char *cp;
2758
2759 /*
2760 * mp == NULL -> invalidate the current info, the FS is
2761 * no longer exported. May be called from either vfs_export
2762 * or unmount, so check if it hasn't already been done.
2763 */
2764 if (mp == NULL) {
2765 if (nfs_pub.np_valid) {
2766 nfs_pub.np_valid = 0;
2767 if (nfs_pub.np_index != NULL) {
2768 FREE(nfs_pub.np_index, M_TEMP);
2769 nfs_pub.np_index = NULL;
2770 }
2771 }
2772 return (0);
2773 }
2774
2775 /*
2776 * Only one allowed at a time.
2777 */
2778 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2779 return (EBUSY);
2780
2781 /*
2782 * Get real filehandle for root of exported FS.
2783 */
2784 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2785 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2786
2787 if ((error = VFS_ROOT(mp, &rvp)))
2788 return (error);
2789
2790 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2791 return (error);
2792
2793 vput(rvp);
2794
2795 /*
2796 * If an indexfile was specified, pull it in.
2797 */
2798 if (argp->ex_indexfile != NULL) {
2799 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2800 M_WAITOK);
2801 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2802 MAXNAMLEN, (size_t *)0);
2803 if (!error) {
2804 /*
2805 * Check for illegal filenames.
2806 */
2807 for (cp = nfs_pub.np_index; *cp; cp++) {
2808 if (*cp == '/') {
2809 error = EINVAL;
2810 break;
2811 }
2812 }
2813 }
2814 if (error) {
2815 FREE(nfs_pub.np_index, M_TEMP);
2816 return (error);
2817 }
2818 }
2819
2820 nfs_pub.np_mount = mp;
2821 nfs_pub.np_valid = 1;
2822 return (0);
2823}
2824
2825struct netcred *
2826vfs_export_lookup(mp, nep, nam)
1fd87d54 2827 struct mount *mp;
984263bc
MD
2828 struct netexport *nep;
2829 struct sockaddr *nam;
2830{
1fd87d54
RG
2831 struct netcred *np;
2832 struct radix_node_head *rnh;
984263bc
MD
2833 struct sockaddr *saddr;
2834
2835 np = NULL;
2836 if (mp->mnt_flag & MNT_EXPORTED) {
2837 /*
2838 * Lookup in the export list first.
2839 */
2840 if (nam != NULL) {
2841 saddr = nam;
2842 rnh = nep->ne_rtable[saddr->sa_family];
2843 if (rnh != NULL) {
2844 np = (struct netcred *)
2845 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2846 rnh);
2847 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2848 np = NULL;
2849 }
2850 }
2851 /*
2852 * If no address match, use the default if it exists.
2853 */
2854 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2855 np = &nep->ne_defexported;
2856 }
2857 return (np);
2858}
2859
2860/*
41a01a4d
MD
2861 * perform msync on all vnodes under a mount point. The mount point must
2862 * be locked. This code is also responsible for lazy-freeing unreferenced
2863 * vnodes whos VM objects no longer contain pages.
2864 *
2865 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
984263bc 2866 */
41a01a4d
MD
2867static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2868static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2869 lwkt_tokref_t vlock, void *data);
2870
984263bc
MD
2871void
2872vfs_msync(struct mount *mp, int flags)
2873{
41a01a4d
MD
2874 vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2875}
984263bc 2876
41a01a4d
MD
2877/*
2878 * scan1 is a fast pre-check. There could be hundreds of thousands of
2879 * vnodes, we cannot afford to do anything heavy weight until we have a
2880 * fairly good indication that there is work to do.
2881 */
2882static
2883int
2884vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2885{
2886 int flags = (int)data;
984263bc 2887
41a01a4d
MD
2888 if ((vp->v_flag & VXLOCK) == 0) {
2889 if (VSHOULDFREE(vp))
2890 return(0);
2891 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2892 (vp->v_flag & VOBJDIRTY) &&
984263bc 2893 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
41a01a4d
MD
2894 return(0);
2895 }
2896 }
2897 return(-1);
2898}
2899
2900static
2901int
2902vfs_msync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
2903{
2904 vm_object_t obj;
2905 int error;
2906 int flags = (int)data;
2907
2908 if (vp->v_flag & VXLOCK)
2909 return(0);
2910
2911 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2912 (vp->v_flag & VOBJDIRTY) &&
2913 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2914 error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2915 if (error == 0) {
2916 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2917 vm_object_page_clean(obj, 0, 0,
2918 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
984263bc 2919 }
41a01a4d 2920 vput(vp);
984263bc 2921 }
41a01a4d 2922 return(0);
984263bc 2923 }
41a01a4d
MD
2924 vmaybefree(vp);
2925 lwkt_reltoken(vlock);
2926 return(0);
984263bc
MD
2927}
2928
2929/*
2930 * Create the VM object needed for VMIO and mmap support. This
2931 * is done for all VREG files in the system. Some filesystems might
2932 * afford the additional metadata buffering capability of the
2933 * VMIO code by making the device node be VMIO mode also.
2934 *
2935 * vp must be locked when vfs_object_create is called.
2936 */
2937int
3b568787 2938vfs_object_create(struct vnode *vp, struct thread *td)
984263bc 2939{
3b568787 2940 return (VOP_CREATEVOBJECT(vp, td));
984263bc
MD
2941}
2942
41a01a4d
MD
2943/*
2944 * NOTE: the vnode interlock must be held during the call. We have to recheck
2945 * the VFREE flag since the vnode may have been removed from the free list
2946 * while we were blocked on vnode_free_list_token. The use or hold count
2947 * must have already been bumped by the caller.
2948 */
2949static void
2950vbusy(struct vnode *vp)
984263bc 2951{
41a01a4d 2952 lwkt_tokref ilock;
984263bc 2953
41a01a4d
MD
2954 lwkt_gettoken(&ilock, &vnode_free_list_token);
2955 if ((vp->v_flag & VFREE) != 0) {
2956 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2957 freevnodes--;
2958 vp->v_flag &= ~(VFREE|VAGE);
984263bc 2959 }
41a01a4d 2960 lwkt_reltoken(&ilock);
984263bc
MD
2961}
2962
41a01a4d
MD
2963/*
2964 * NOTE: the vnode interlock must be held during the call. The use or hold
2965 * count must have already been bumped by the caller. We use a VINFREE to
2966 * interlock against other calls to vfree() which might occur while we
2967 * are blocked. The vnode cannot be reused until it has actually been
2968 * placed on the free list, so there are no other races even though the
2969 * use and hold counts are 0.
2970 */
2971static void
2972vfree(struct vnode *vp)
984263bc 2973{
41a01a4d 2974 lwkt_tokref ilock;
984263bc 2975
41a01a4d
MD
2976 if ((vp->v_flag & VINFREE) == 0) {
2977 vp->v_flag |= VINFREE;
2978 lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
2979 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2980 if (vp->v_flag & VAGE) {
2981 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2982 } else {
2983 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2984 }
2985 freevnodes++;
2986 vp->v_flag &= ~(VAGE|VINFREE);
2987 vp->v_flag |= VFREE;
2988 lwkt_reltoken(&ilock); /* can block */
2989 }
984263bc
MD
2990}
2991
41a01a4d 2992
984263bc
MD
2993/*
2994 * Record a process's interest in events which might happen to
2995 * a vnode. Because poll uses the historic select-style interface
2996 * internally, this routine serves as both the ``check for any
2997 * pending events'' and the ``record my interest in future events''
2998 * functions. (These are done together, while the lock is held,
2999 * to avoid race conditions.)
3000 */
3001int
dadab5e9 3002vn_pollrecord(struct vnode *vp, struct thread *td, int events)
984263bc 3003{
41a01a4d
MD
3004 lwkt_tokref ilock;
3005
3006 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
3007 if (vp->v_pollinfo.vpi_revents & events) {
3008 /*
3009 * This leaves events we are not interested
3010 * in available for the other process which
3011 * which presumably had requested them
3012 * (otherwise they would never have been
3013 * recorded).
3014 */
3015 events &= vp->v_pollinfo.vpi_revents;
3016 vp->v_pollinfo.vpi_revents &= ~events;
3017
41a01a4d 3018 lwkt_reltoken(&ilock);
984263bc
MD
3019 return events;
3020 }
3021 vp->v_pollinfo.vpi_events |= events;
dadab5e9 3022 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
41a01a4d 3023 lwkt_reltoken(&ilock);
984263bc
MD
3024 return 0;
3025}
3026
3027/*
3028 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3029 * it is possible for us to miss an event due to race conditions, but
3030 * that condition is expected to be rare, so for the moment it is the
3031 * preferred interface.
3032 */
3033void
3034vn_pollevent(vp, events)
3035 struct vnode *vp;
3036 short events;
3037{
41a01a4d
MD
3038 lwkt_tokref ilock;
3039
3040 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
3041 if (vp->v_pollinfo.vpi_events & events) {
3042 /*
3043 * We clear vpi_events so that we don't
3044 * call selwakeup() twice if two events are
3045 * posted before the polling process(es) is
3046 * awakened. This also ensures that we take at
3047 * most one selwakeup() if the polling process
3048 * is no longer interested. However, it does
3049 * mean that only one event can be noticed at
3050 * a time. (Perhaps we should only clear those
3051 * event bits which we note?) XXX
3052 */
3053 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
3054 vp->v_pollinfo.vpi_revents |= events;
3055 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3056 }
41a01a4d 3057 lwkt_reltoken(&ilock);
984263bc
MD
3058}
3059
3060/*
3061 * Wake up anyone polling on vp because it is being revoked.
3062 * This depends on dead_poll() returning POLLHUP for correct
3063 * behavior.
3064 */
3065void
3066vn_pollgone(vp)
3067 struct vnode *vp;
3068{
41a01a4d
MD
3069 lwkt_tokref ilock;
3070
3071 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
984263bc
MD
3072 if (vp->v_pollinfo.vpi_events) {
3073 vp->v_pollinfo.vpi_events = 0;
3074 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3075 }
41a01a4d 3076 lwkt_reltoken(&ilock);
984263bc
MD
3077}
3078
3079
3080
3081/*
3082 * Routine to create and manage a filesystem syncer vnode.
3083 */
402ed7e1
RG
3084#define sync_close ((int (*) (struct vop_close_args *))nullop)
3085static int sync_fsync (struct vop_fsync_args *);
3086static int sync_inactive (struct vop_inactive_args *);
3087static int sync_reclaim (struct vop_reclaim_args *);
3088#define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
3089#define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
3090static int sync_print (struct vop_print_args *);
3091#define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
984263bc
MD
3092
3093static vop_t **sync_vnodeop_p;
3094static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3095 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3096 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3097 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3098 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3099 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3100 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
3101 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
3102 { &vop_print_desc, (vop_t *) sync_print }, /* print */
3103 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
3104 { NULL, NULL }
3105};
3106static struct vnodeopv_desc sync_vnodeop_opv_desc =
3107 { &sync_vnodeop_p, sync_vnodeop_entries };
3108
3109VNODEOP_SET(sync_vnodeop_opv_desc);
3110
3111/*
3112 * Create a new filesystem syncer vnode for the specified mount point.
41a01a4d
MD
3113 * This vnode is placed on the worklist and is responsible for sync'ing
3114 * the filesystem.
3115 *
3116 * NOTE: read-only mounts are also placed on the worklist. The filesystem
3117 * sync code is also responsible for cleaning up vnodes.
984263bc
MD
3118 */
3119int
41a01a4d 3120vfs_allocate_syncvnode(struct mount *mp)
984263bc
MD
3121{
3122 struct vnode *vp;
3123 static long start, incr, next;
3124 int error;
3125
3126 /* Allocate a new vnode */
3127 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3128 mp->mnt_syncer = NULL;
3129 return (error);
3130 }
3131 vp->v_type = VNON;
3132 /*
3133 * Place the vnode onto the syncer worklist. We attempt to
3134 * scatter them about on the list so that they will go off
3135 * at evenly distributed times even if all the filesystems
3136 * are mounted at once.
3137 */
3138 next += incr;
3139 if (next == 0 || next > syncer_maxdelay) {
3140 start /= 2;
3141 incr /= 2;
3142 if (start == 0) {
3143 start = syncer_maxdelay / 2;
3144 incr = syncer_maxdelay;
3145 }
3146 next = start;
3147 }
3148 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3149 mp->mnt_syncer = vp;
3150 return (0);
3151}
3152
3153/*
3154 * Do a lazy sync of the filesystem.
3155 */
3156static int
3157sync_fsync(ap)
3158 struct vop_fsync_args /* {
3159 struct vnode *a_vp;
3160 struct ucred *a_cred;
3161 int a_waitfor;
dadab5e9 3162 struct thread *a_td;
984263bc
MD
3163 } */ *ap;
3164{
3165 struct vnode *syncvp = ap->a_vp;
3166 struct mount *mp = syncvp->v_mount;
dadab5e9 3167 struct thread *td = ap->a_td;
41a01a4d 3168 lwkt_tokref ilock;
984263bc
MD
3169 int asyncflag;
3170
3171 /*
3172 * We only need to do something if this is a lazy evaluation.
3173 */
3174 if (ap->a_waitfor != MNT_LAZY)
3175 return (0);
3176
3177 /*
3178 * Move ourselves to the back of the sync list.
3179 */
3180 vn_syncer_add_to_worklist(syncvp, syncdelay);
3181
3182 /*
3183 * Walk the list of vnodes pushing all that are dirty and
41a01a4d
MD
3184 * not already on the sync list, and freeing vnodes which have
3185 * no refs and whos VM objects are empty. vfs_msync() handles
3186 * the VM issues and must be called whether the mount is readonly
3187 * or not.
984263bc 3188 */
41a01a4d
MD
3189 lwkt_gettoken(&ilock, &mountlist_token);
3190 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3191 lwkt_reltoken(&ilock);
984263bc
MD
3192 return (0);
3193 }
41a01a4d
MD
3194 if (mp->mnt_flag & MNT_RDONLY) {
3195 vfs_msync(mp, MNT_NOWAIT);
3196 } else {
3197 asyncflag = mp->mnt_flag & MNT_ASYNC;
3198 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */
3199 vfs_msync(mp, MNT_NOWAIT);
3200 VFS_SYNC(mp, MNT_LAZY, td);
3201 if (asyncflag)
3202 mp->mnt_flag |= MNT_ASYNC;
3203 }
dadab5e9 3204 vfs_unbusy(mp, td);
984263bc
MD
3205 return (0);
3206}
3207
3208/*
3209 * The syncer vnode is no referenced.
3210 */
3211static int
3212sync_inactive(ap)
3213 struct vop_inactive_args /* {
3214 struct vnode *a_vp;
3215 struct proc *a_p;
3216 } */ *ap;
3217{
3218
3219 vgone(ap->a_vp);
3220 return (0);
3221}
3222
3223/*
3224 * The syncer vnode is no longer needed and is being decommissioned.
3225 *
3226 * Modifications to the worklist must be protected at splbio().
3227 */
3228static int
3229sync_reclaim(ap)
3230 struct vop_reclaim_args /* {
3231 struct vnode *a_vp;
3232 } */ *ap;
3233{
3234 struct vnode *vp = ap->a_vp;
3235 int s;
3236
3237 s = splbio();
3238 vp->v_mount->mnt_syncer = NULL;
3239 if (vp->v_flag & VONWORKLST) {
3240 LIST_REMOVE(vp, v_synclist);
3241 vp->v_flag &= ~VONWORKLST;
3242 }
3243 splx(s);
3244
3245 return (0);
3246}
3247
3248/*
3249 * Print out a syncer vnode.
3250 */
3251static int
3252sync_print(ap)
3253 struct vop_print_args /* {
3254 struct vnode *a_vp;
3255 } */ *ap;
3256{
3257 struct vnode *vp = ap->a_vp;
3258
3259 printf("syncer vnode");
3260 if (vp->v_vnlock != NULL)
3261 lockmgr_printinfo(vp->v_vnlock);
3262 printf("\n");
3263 return (0);
3264}
3265
3266/*
3267 * extract the dev_t from a VBLK or VCHR
3268 */
3269dev_t
3270vn_todev(vp)
3271 struct vnode *vp;
3272{
3273 if (vp->v_type != VBLK && vp->v_type != VCHR)
3274 return (NODEV);
3275 return (vp->v_rdev);
3276}
3277
3278/*
3279 * Check if vnode represents a disk device
3280 */
3281int
3282vn_isdisk(vp, errp)
3283 struct vnode *vp;
3284 int *errp;
3285{
3286 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3287 if (errp != NULL)
3288 *errp = ENOTBLK;
3289 return (0);
3290 }
3291 if (vp->v_rdev == NULL) {
3292 if (errp != NULL)
3293 *errp = ENXIO;
3294 return (0);
3295 }
335dda38 3296 if (!dev_dport(vp->v_rdev)) {
984263bc
MD
3297 if (errp != NULL)
3298 *errp = ENXIO;
3299 return (0);
3300 }
335dda38 3301 if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
984263bc
MD
3302 if (errp != NULL)
3303 *errp = ENOTBLK;
3304 return (0);
3305 }
3306 if (errp != NULL)
3307 *errp = 0;
3308 return (1);
3309}
3310
3311void
3312NDFREE(ndp, flags)
3313 struct nameidata *ndp;
3314 const uint flags;
3315{
3316 if (!(flags & NDF_NO_FREE_PNBUF) &&
2b69e610 3317 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
984263bc 3318 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
2b69e610 3319 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
984263bc 3320 }
bc0c094e
MD
3321 if (!(flags & NDF_NO_DNCP_RELE) &&
3322 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3323 ndp->ni_dncp) {
3324 cache_drop(ndp->ni_dncp);
3325 ndp->ni_dncp = NULL;
3326 }
3327 if (!(flags & NDF_NO_NCP_RELE) &&
3328 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3329 ndp->ni_ncp) {
3330 cache_drop(ndp->ni_ncp);
3331 ndp->ni_ncp = NULL;
3332 }
984263bc 3333 if (!(flags & NDF_NO_DVP_UNLOCK) &&
2b69e610 3334 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
bc0c094e 3335 ndp->ni_dvp != ndp->ni_vp) {
41a01a4d 3336 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
bc0c094e 3337 }
984263bc 3338 if (!(flags & NDF_NO_DVP_RELE) &&
2b69e610 3339 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
984263bc
MD
3340 vrele(ndp->ni_dvp);
3341 ndp->ni_dvp = NULL;
3342 }
3343 if (!(flags & NDF_NO_VP_UNLOCK) &&
bc0c094e 3344 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
41a01a4d 3345 VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
bc0c094e 3346 }
984263bc
MD
3347 if (!(flags & NDF_NO_VP_RELE) &&
3348 ndp->ni_vp) {
3349 vrele(ndp->ni_vp);
3350 ndp->ni_vp = NULL;
3351 }
3352 if (!(flags & NDF_NO_STARTDIR_RELE) &&
2b69e610 3353 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
984263bc
MD
3354 vrele(ndp->ni_startdir);
3355 ndp->ni_startdir = NULL;
3356 }
3357}
bc0c094e 3358