kernel - namecache MPSAFE work
[dragonfly.git] / sys / kern / vfs_mount.c
CommitLineData
5fd012e0
MD
1/*
2 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * Copyright (c) 1989, 1993
35 * The Regents of the University of California. All rights reserved.
36 * (c) UNIX System Laboratories, Inc.
37 * All or some portions of this file are derived from material licensed
38 * to the University of California by American Telephone and Telegraph
39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40 * the permission of UNIX System Laboratories, Inc.
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. All advertising materials mentioning features or use of this software
51 * must display the following acknowledgement:
52 * This product includes software developed by the University of
53 * California, Berkeley and its contributors.
54 * 4. Neither the name of the University nor the names of its contributors
55 * may be used to endorse or promote products derived from this software
56 * without specific prior written permission.
57 *
58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68 * SUCH DAMAGE.
69 *
67863d04 70 * $DragonFly: src/sys/kern/vfs_mount.c,v 1.37 2008/09/17 21:44:18 dillon Exp $
5fd012e0
MD
71 */
72
73/*
74 * External virtual filesystem routines
75 */
76#include "opt_ddb.h"
77
78#include <sys/param.h>
79#include <sys/systm.h>
80#include <sys/kernel.h>
81#include <sys/malloc.h>
82#include <sys/mount.h>
83#include <sys/proc.h>
84#include <sys/vnode.h>
85#include <sys/buf.h>
86#include <sys/eventhandler.h>
87#include <sys/kthread.h>
88#include <sys/sysctl.h>
89
90#include <machine/limits.h>
91
92#include <sys/buf2.h>
93#include <sys/thread2.h>
3c37c940 94#include <sys/sysref2.h>
5fd012e0
MD
95
96#include <vm/vm.h>
97#include <vm/vm_object.h>
98
861905fb
MD
99struct mountscan_info {
100 TAILQ_ENTRY(mountscan_info) msi_entry;
101 int msi_how;
102 struct mount *msi_node;
103};
104
be6c08cb
MD
105struct vmntvnodescan_info {
106 TAILQ_ENTRY(vmntvnodescan_info) entry;
107 struct vnode *vp;
108};
109
0e8bd897
MD
110struct vnlru_info {
111 int pass;
112};
113
5fd012e0 114static int vnlru_nowhere = 0;
58552887 115SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD,
5fd012e0
MD
116 &vnlru_nowhere, 0,
117 "Number of times the vnlru process ran without success");
118
119
120static struct lwkt_token mntid_token;
121
460426e6
MD
122/* note: mountlist exported to pstat */
123struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
861905fb
MD
124static TAILQ_HEAD(,mountscan_info) mountscan_list;
125static struct lwkt_token mountlist_token;
be6c08cb 126static TAILQ_HEAD(,vmntvnodescan_info) mntvnodescan_list;
5fd012e0
MD
127struct lwkt_token mntvnode_token;
128
408357d8
MD
129static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list);
130
5fd012e0
MD
131/*
132 * Called from vfsinit()
133 */
134void
135vfs_mount_init(void)
136{
137 lwkt_token_init(&mountlist_token);
138 lwkt_token_init(&mntvnode_token);
139 lwkt_token_init(&mntid_token);
861905fb 140 TAILQ_INIT(&mountscan_list);
be6c08cb
MD
141 TAILQ_INIT(&mntvnodescan_list);
142}
143
144/*
145 * Support function called with mntvnode_token held to remove a vnode
146 * from the mountlist. We must update any list scans which are in progress.
147 */
148static void
149vremovevnodemnt(struct vnode *vp)
150{
151 struct vmntvnodescan_info *info;
152
153 TAILQ_FOREACH(info, &mntvnodescan_list, entry) {
154 if (info->vp == vp)
155 info->vp = TAILQ_NEXT(vp, v_nmntvnodes);
156 }
157 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
158}
159
160/*
5fd012e0
MD
161 * Allocate a new vnode and associate it with a tag, mount point, and
162 * operations vector.
163 *
164 * A VX locked and refd vnode is returned. The caller should setup the
165 * remaining fields and vx_put() or, if he wishes to leave a vref,
166 * vx_unlock() the vnode.
167 */
168int
6ddb7618
MD
169getnewvnode(enum vtagtype tag, struct mount *mp,
170 struct vnode **vpp, int lktimeout, int lkflags)
171{
172 struct vnode *vp;
173
174 KKASSERT(mp != NULL);
175
176 vp = allocvnode(lktimeout, lkflags);
177 vp->v_tag = tag;
178 vp->v_data = NULL;
179
180 /*
181 * By default the vnode is assigned the mount point's normal
182 * operations vector.
183 */
184 vp->v_ops = &mp->mnt_vn_use_ops;
185
186 /*
187 * Placing the vnode on the mount point's queue makes it visible.
188 * VNON prevents it from being messed with, however.
189 */
190 insmntque(vp, mp);
6ddb7618
MD
191
192 /*
193 * A VX locked & refd vnode is returned.
194 */
195 *vpp = vp;
196 return (0);
197}
198
199/*
200 * This function creates vnodes with special operations vectors. The
201 * mount point is optional.
202 *
203 * This routine is being phased out.
204 */
205int
206getspecialvnode(enum vtagtype tag, struct mount *mp,
66a1ddf5 207 struct vop_ops **ops,
5fd012e0
MD
208 struct vnode **vpp, int lktimeout, int lkflags)
209{
210 struct vnode *vp;
211
212 vp = allocvnode(lktimeout, lkflags);
213 vp->v_tag = tag;
5fd012e0 214 vp->v_data = NULL;
66a1ddf5 215 vp->v_ops = ops;
5fd012e0
MD
216
217 /*
218 * Placing the vnode on the mount point's queue makes it visible.
219 * VNON prevents it from being messed with, however.
220 */
221 insmntque(vp, mp);
5fd012e0
MD
222
223 /*
224 * A VX locked & refd vnode is returned.
225 */
226 *vpp = vp;
227 return (0);
228}
229
230/*
861905fb
MD
231 * Interlock against an unmount, return 0 on success, non-zero on failure.
232 *
233 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount
234 * is in-progress.
235 *
236 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits
237 * are used. A shared locked will be obtained and the filesystem will not
238 * be unmountable until the lock is released.
5fd012e0
MD
239 */
240int
f9642f56 241vfs_busy(struct mount *mp, int flags)
5fd012e0
MD
242{
243 int lkflags;
244
245 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
246 if (flags & LK_NOWAIT)
247 return (ENOENT);
861905fb 248 /* XXX not MP safe */
5fd012e0
MD
249 mp->mnt_kern_flag |= MNTK_MWAIT;
250 /*
251 * Since all busy locks are shared except the exclusive
252 * lock granted when unmounting, the only place that a
253 * wakeup needs to be done is at the release of the
254 * exclusive lock at the end of dounmount.
5fd012e0
MD
255 */
256 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
257 return (ENOENT);
258 }
ab6f251b 259 lkflags = LK_SHARED;
df4f70a6 260 if (lockmgr(&mp->mnt_lock, lkflags))
5fd012e0
MD
261 panic("vfs_busy: unexpected lock failure");
262 return (0);
263}
264
265/*
266 * Free a busy filesystem.
267 */
268void
f9642f56 269vfs_unbusy(struct mount *mp)
5fd012e0 270{
df4f70a6 271 lockmgr(&mp->mnt_lock, LK_RELEASE);
5fd012e0
MD
272}
273
274/*
275 * Lookup a filesystem type, and if found allocate and initialize
276 * a mount structure for it.
277 *
278 * Devname is usually updated by mount(8) after booting.
279 */
280int
281vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
282{
5fd012e0
MD
283 struct vfsconf *vfsp;
284 struct mount *mp;
285
286 if (fstypename == NULL)
287 return (ENODEV);
2613053d
MN
288
289 vfsp = vfsconf_find_by_name(fstypename);
5fd012e0
MD
290 if (vfsp == NULL)
291 return (ENODEV);
e7b4468c 292 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
ab6f251b 293 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0);
f9642f56 294 vfs_busy(mp, LK_NOWAIT);
5fd012e0
MD
295 TAILQ_INIT(&mp->mnt_nvnodelist);
296 TAILQ_INIT(&mp->mnt_reservedvnlist);
2281065e 297 TAILQ_INIT(&mp->mnt_jlist);
5fd012e0
MD
298 mp->mnt_nvnodelistsize = 0;
299 mp->mnt_vfc = vfsp;
300 mp->mnt_op = vfsp->vfc_vfsops;
301 mp->mnt_flag = MNT_RDONLY;
5fd012e0
MD
302 vfsp->vfc_refcount++;
303 mp->mnt_iosize_max = DFLTPHYS;
304 mp->mnt_stat.f_type = vfsp->vfc_typenum;
305 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
306 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
069b825e 307 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
5fd012e0
MD
308 *mpp = mp;
309 return (0);
310}
311
312/*
313 * Lookup a mount point by filesystem identifier.
314 */
315struct mount *
316vfs_getvfs(fsid_t *fsid)
317{
318 struct mount *mp;
319 lwkt_tokref ilock;
320
321 lwkt_gettoken(&ilock, &mountlist_token);
322 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
323 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
324 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
325 break;
02dede15 326 }
5fd012e0
MD
327 }
328 lwkt_reltoken(&ilock);
329 return (mp);
330}
331
332/*
333 * Get a new unique fsid. Try to make its val[0] unique, since this value
334 * will be used to create fake device numbers for stat(). Also try (but
335 * not so hard) make its val[0] unique mod 2^16, since some emulators only
336 * support 16-bit device numbers. We end up with unique val[0]'s for the
337 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
338 *
339 * Keep in mind that several mounts may be running in parallel. Starting
340 * the search one past where the previous search terminated is both a
341 * micro-optimization and a defense against returning the same fsid to
342 * different mounts.
343 */
344void
345vfs_getnewfsid(struct mount *mp)
346{
347 static u_int16_t mntid_base;
348 lwkt_tokref ilock;
349 fsid_t tfsid;
350 int mtype;
351
352 lwkt_gettoken(&ilock, &mntid_token);
353 mtype = mp->mnt_vfc->vfc_typenum;
354 tfsid.val[1] = mtype;
355 mtype = (mtype & 0xFF) << 24;
356 for (;;) {
357 tfsid.val[0] = makeudev(255,
358 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
359 mntid_base++;
360 if (vfs_getvfs(&tfsid) == NULL)
361 break;
362 }
363 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
364 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
365 lwkt_reltoken(&ilock);
366}
367
368/*
67863d04
MD
369 * Set the FSID for a new mount point to the template. Adjust
370 * the FSID to avoid collisions.
371 */
372int
373vfs_setfsid(struct mount *mp, fsid_t *template)
374{
375 int didmunge = 0;
376
377 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid));
378 for (;;) {
379 if (vfs_getvfs(template) == NULL)
380 break;
381 didmunge = 1;
382 ++template->val[1];
383 }
384 mp->mnt_stat.f_fsid = *template;
385 return(didmunge);
386}
387
388/*
5fd012e0
MD
389 * This routine is called when we have too many vnodes. It attempts
390 * to free <count> vnodes and will potentially free vnodes that still
391 * have VM backing store (VM backing store is typically the cause
392 * of a vnode blowout so we want to do this). Therefore, this operation
393 * is not considered cheap.
394 *
395 * A number of conditions may prevent a vnode from being reclaimed.
396 * the buffer cache may have references on the vnode, a directory
397 * vnode may still have references due to the namei cache representing
398 * underlying files, or the vnode may be in active use. It is not
399 * desireable to reuse such vnodes. These conditions may cause the
400 * number of vnodes to reach some minimum value regardless of what
401 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
402 */
403
404/*
58552887
MD
405 * This is a quick non-blocking check to determine if the vnode is a good
406 * candidate for being (eventually) vgone()'d. Returns 0 if the vnode is
407 * not a good candidate, 1 if it is.
5fd012e0
MD
408 */
409static __inline int
0e8bd897 410vmightfree(struct vnode *vp, int page_count, int pass)
5fd012e0 411{
58552887
MD
412 if (vp->v_flag & VRECLAIMED)
413 return (0);
44b1cf3d 414#if 0
58552887 415 if ((vp->v_flag & VFREE) && TAILQ_EMPTY(&vp->v_namecache))
5fd012e0 416 return (0);
44b1cf3d 417#endif
3c37c940 418 if (sysref_isactive(&vp->v_sysref))
5fd012e0
MD
419 return (0);
420 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
421 return (0);
0e8bd897
MD
422
423 /*
424 * XXX horrible hack. Up to four passes will be taken. Each pass
425 * makes a larger set of vnodes eligible. For now what this really
426 * means is that we try to recycle files opened only once before
427 * recycling files opened multiple times.
428 */
429 switch(vp->v_flag & (VAGE0 | VAGE1)) {
430 case 0:
431 if (pass < 3)
432 return(0);
433 break;
434 case VAGE0:
435 if (pass < 2)
436 return(0);
437 break;
438 case VAGE1:
439 if (pass < 1)
440 return(0);
441 break;
442 case VAGE0 | VAGE1:
443 break;
444 }
5fd012e0
MD
445 return (1);
446}
447
250d127c 448/*
58552887 449 * The vnode was found to be possibly vgone()able and the caller has locked it
250d127c 450 * (thus the usecount should be 1 now). Determine if the vnode is actually
58552887
MD
451 * vgone()able, doing some cleanups in the process. Returns 1 if the vnode
452 * can be vgone()'d, 0 otherwise.
250d127c 453 *
3c37c940 454 * Note that v_auxrefs may be non-zero because (A) this vnode is not a leaf
250d127c
MD
455 * in the namecache topology and (B) this vnode has buffer cache bufs.
456 * We cannot remove vnodes with non-leaf namecache associations. We do a
457 * tentitive leaf check prior to attempting to flush out any buffers but the
3c37c940 458 * 'real' test when all is said in done is that v_auxrefs must become 0 for
250d127c
MD
459 * the vnode to be freeable.
460 *
3c37c940 461 * We could theoretically just unconditionally flush when v_auxrefs != 0,
250d127c
MD
462 * but flushing data associated with non-leaf nodes (which are always
463 * directories), just throws it away for no benefit. It is the buffer
464 * cache's responsibility to choose buffers to recycle from the cached
465 * data point of view.
466 */
467static int
468visleaf(struct vnode *vp)
469{
470 struct namecache *ncp;
471
f63911bf 472 spin_lock_wr(&vp->v_spinlock);
250d127c 473 TAILQ_FOREACH(ncp, &vp->v_namecache, nc_vnode) {
f63911bf
MD
474 if (!TAILQ_EMPTY(&ncp->nc_list)) {
475 spin_unlock_wr(&vp->v_spinlock);
250d127c 476 return(0);
f63911bf 477 }
250d127c 478 }
f63911bf 479 spin_unlock_wr(&vp->v_spinlock);
250d127c
MD
480 return(1);
481}
482
58552887
MD
483/*
484 * Try to clean up the vnode to the point where it can be vgone()'d, returning
485 * 0 if it cannot be vgone()'d (or already has been), 1 if it can. Unlike
486 * vmightfree() this routine may flush the vnode and block. Vnodes marked
487 * VFREE are still candidates for vgone()ing because they may hold namecache
488 * resources and could be blocking the namecache directory hierarchy (and
489 * related vnodes) from being freed.
490 */
250d127c 491static int
58552887 492vtrytomakegoneable(struct vnode *vp, int page_count)
250d127c 493{
58552887 494 if (vp->v_flag & VRECLAIMED)
250d127c 495 return (0);
3c37c940 496 if (vp->v_sysref.refcnt > 1)
250d127c
MD
497 return (0);
498 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
499 return (0);
3c37c940 500 if (vp->v_auxrefs && visleaf(vp)) {
87de5057 501 vinvalbuf(vp, V_SAVE, 0, 0);
250d127c 502#if 0 /* DEBUG */
3c37c940 503 kprintf((vp->v_auxrefs ? "vrecycle: vp %p failed: %s\n" :
250d127c
MD
504 "vrecycle: vp %p succeeded: %s\n"), vp,
505 (TAILQ_FIRST(&vp->v_namecache) ?
506 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"));
507#endif
508 }
5c6c3cac
MD
509
510 /*
511 * This sequence may seem a little strange, but we need to optimize
512 * the critical path a bit. We can't recycle vnodes with other
513 * references and because we are trying to recycle an otherwise
514 * perfectly fine vnode we have to invalidate the namecache in a
515 * way that avoids possible deadlocks (since the vnode lock is being
516 * held here). Finally, we have to check for other references one
517 * last time in case something snuck in during the inval.
518 */
519 if (vp->v_sysref.refcnt > 1 || vp->v_auxrefs != 0)
520 return (0);
521 if (cache_inval_vp_nonblock(vp))
522 return (0);
523 return (vp->v_sysref.refcnt <= 1 && vp->v_auxrefs == 0);
250d127c 524}
5fd012e0 525
58552887
MD
526/*
527 * Reclaim up to 1/10 of the vnodes associated with a mount point. Try
528 * to avoid vnodes which have lots of resident pages (we are trying to free
861905fb
MD
529 * vnodes, not memory).
530 *
531 * This routine is a callback from the mountlist scan. The mount point
532 * in question will be busied.
0e8bd897
MD
533 *
534 * NOTE: The 1/10 reclamation also ensures that the inactive data set
535 * (the vnodes being recycled by the one-time use) does not degenerate
536 * into too-small a set. This is important because once a vnode is
537 * marked as not being one-time-use (VAGE0/VAGE1 both 0) that vnode
538 * will not be destroyed EXCEPT by this mechanism. VM pages can still
539 * be cleaned/freed by the pageout daemon.
58552887 540 */
5fd012e0 541static int
861905fb 542vlrureclaim(struct mount *mp, void *data)
5fd012e0 543{
0e8bd897 544 struct vnlru_info *info = data;
5fd012e0
MD
545 struct vnode *vp;
546 lwkt_tokref ilock;
547 int done;
548 int trigger;
549 int usevnodes;
550 int count;
861905fb 551 int trigger_mult = vnlru_nowhere;
5fd012e0
MD
552
553 /*
58552887
MD
554 * Calculate the trigger point for the resident pages check. The
555 * minimum trigger value is approximately the number of pages in
556 * the system divded by the number of vnodes. However, due to
557 * various other system memory overheads unrelated to data caching
558 * it is a good idea to double the trigger (at least).
559 *
560 * trigger_mult starts at 0. If the recycler is having problems
561 * finding enough freeable vnodes it will increase trigger_mult.
562 * This should not happen in normal operation, even on machines with
563 * low amounts of memory, but extraordinary memory use by the system
564 * verses the amount of cached data can trigger it.
5fd012e0
MD
565 */
566 usevnodes = desiredvnodes;
567 if (usevnodes <= 0)
568 usevnodes = 1;
58552887 569 trigger = vmstats.v_page_count * (trigger_mult + 2) / usevnodes;
5fd012e0
MD
570
571 done = 0;
572 lwkt_gettoken(&ilock, &mntvnode_token);
573 count = mp->mnt_nvnodelistsize / 10 + 1;
0e8bd897 574
2ec4b00d
MD
575 while (count && mp->mnt_syncer) {
576 /*
577 * Next vnode. Use the special syncer vnode to placemark
578 * the LRU. This way the LRU code does not interfere with
579 * vmntvnodescan().
580 */
581 vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
582 TAILQ_REMOVE(&mp->mnt_nvnodelist, mp->mnt_syncer, v_nmntvnodes);
583 if (vp) {
584 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp,
585 mp->mnt_syncer, v_nmntvnodes);
586 } else {
587 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, mp->mnt_syncer,
588 v_nmntvnodes);
589 vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
590 if (vp == NULL)
591 break;
592 }
593
5fd012e0
MD
594 /*
595 * __VNODESCAN__
596 *
597 * The VP will stick around while we hold mntvnode_token,
598 * at least until we block, so we can safely do an initial
599 * check, and then must check again after we lock the vnode.
600 */
986e7cda 601 if (vp->v_type == VNON || /* syncer or indeterminant */
0e8bd897 602 !vmightfree(vp, trigger, info->pass) /* critical path opt */
5fd012e0 603 ) {
5fd012e0
MD
604 --count;
605 continue;
606 }
607
608 /*
609 * VX get the candidate vnode. If the VX get fails the
610 * vnode might still be on the mountlist. Our loop depends
611 * on us at least cycling the vnode to the end of the
612 * mountlist.
613 */
614 if (vx_get_nonblock(vp) != 0) {
5fd012e0
MD
615 --count;
616 continue;
617 }
618
619 /*
620 * Since we blocked locking the vp, make sure it is still
621 * a candidate for reclamation. That is, it has not already
622 * been reclaimed and only has our VX reference associated
623 * with it.
624 */
986e7cda 625 if (vp->v_type == VNON || /* syncer or indeterminant */
5fd012e0
MD
626 (vp->v_flag & VRECLAIMED) ||
627 vp->v_mount != mp ||
58552887 628 !vtrytomakegoneable(vp, trigger) /* critical path opt */
5fd012e0 629 ) {
5fd012e0
MD
630 --count;
631 vx_put(vp);
632 continue;
633 }
634
635 /*
636 * All right, we are good, move the vp to the end of the
637 * mountlist and clean it out. The vget will have returned
638 * an error if the vnode was destroyed (VRECLAIMED set), so we
639 * do not have to check again. The vput() will move the
640 * vnode to the free list if the vgone() was successful.
641 */
642 KKASSERT(vp->v_mount == mp);
3c37c940 643 vgone_vxlocked(vp);
5fd012e0
MD
644 vx_put(vp);
645 ++done;
646 --count;
647 }
648 lwkt_reltoken(&ilock);
649 return (done);
650}
651
652/*
653 * Attempt to recycle vnodes in a context that is always safe to block.
654 * Calling vlrurecycle() from the bowels of file system code has some
655 * interesting deadlock problems.
656 */
657static struct thread *vnlruthread;
658static int vnlruproc_sig;
659
660void
661vnlru_proc_wait(void)
662{
663 if (vnlruproc_sig == 0) {
664 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
665 wakeup(vnlruthread);
666 }
667 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
668}
669
670static void
671vnlru_proc(void)
672{
5fd012e0 673 struct thread *td = curthread;
0e8bd897 674 struct vnlru_info info;
861905fb 675 int done;
5fd012e0
MD
676
677 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
678 SHUTDOWN_PRI_FIRST);
679
e43a034f 680 crit_enter();
5fd012e0
MD
681 for (;;) {
682 kproc_suspend_loop();
3c37c940
MD
683
684 /*
685 * Try to free some vnodes if we have too many
686 */
687 if (numvnodes > desiredvnodes &&
688 freevnodes > desiredvnodes * 2 / 10) {
689 int count = numvnodes - desiredvnodes;
690
691 if (count > freevnodes / 100)
692 count = freevnodes / 100;
693 if (count < 5)
694 count = 5;
695 freesomevnodes(count);
696 }
697
698 /*
699 * Nothing to do if most of our vnodes are already on
700 * the free list.
701 */
5fd012e0
MD
702 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
703 vnlruproc_sig = 0;
704 wakeup(&vnlruproc_sig);
705 tsleep(td, 0, "vlruwt", hz);
706 continue;
707 }
62d0f1f0 708 cache_cleanneg(0);
0e8bd897
MD
709
710 /*
711 * The pass iterates through the four combinations of
712 * VAGE0/VAGE1. We want to get rid of aged small files
713 * first.
714 */
715 info.pass = 0;
716 done = 0;
717 while (done == 0 && info.pass < 4) {
718 done = mountlist_scan(vlrureclaim, &info,
719 MNTSCAN_FORWARD);
720 ++info.pass;
721 }
58552887
MD
722
723 /*
724 * The vlrureclaim() call only processes 1/10 of the vnodes
725 * on each mount. If we couldn't find any repeat the loop
726 * at least enough times to cover all available vnodes before
727 * we start sleeping. Complain if the failure extends past
728 * 30 second, every 30 seconds.
729 */
5fd012e0
MD
730 if (done == 0) {
731 ++vnlru_nowhere;
5fd012e0 732 if (vnlru_nowhere % 10 == 0)
58552887
MD
733 tsleep(td, 0, "vlrup", hz * 3);
734 if (vnlru_nowhere % 100 == 0)
6ea70f76 735 kprintf("vnlru_proc: vnode recycler stopped working!\n");
58552887
MD
736 if (vnlru_nowhere == 1000)
737 vnlru_nowhere = 900;
5fd012e0
MD
738 } else {
739 vnlru_nowhere = 0;
740 }
741 }
e43a034f 742 crit_exit();
5fd012e0
MD
743}
744
861905fb
MD
745/*
746 * MOUNTLIST FUNCTIONS
747 */
748
749/*
750 * mountlist_insert (MP SAFE)
751 *
752 * Add a new mount point to the mount list.
753 */
754void
755mountlist_insert(struct mount *mp, int how)
756{
757 lwkt_tokref ilock;
758
759 lwkt_gettoken(&ilock, &mountlist_token);
760 if (how == MNTINS_FIRST)
761 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
762 else
763 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
764 lwkt_reltoken(&ilock);
765}
766
767/*
768 * mountlist_interlock (MP SAFE)
769 *
770 * Execute the specified interlock function with the mountlist token
771 * held. The function will be called in a serialized fashion verses
772 * other functions called through this mechanism.
773 */
774int
775mountlist_interlock(int (*callback)(struct mount *), struct mount *mp)
776{
777 lwkt_tokref ilock;
778 int error;
779
780 lwkt_gettoken(&ilock, &mountlist_token);
781 error = callback(mp);
782 lwkt_reltoken(&ilock);
783 return (error);
784}
785
786/*
787 * mountlist_boot_getfirst (DURING BOOT ONLY)
788 *
789 * This function returns the first mount on the mountlist, which is
790 * expected to be the root mount. Since no interlocks are obtained
791 * this function is only safe to use during booting.
792 */
793
794struct mount *
795mountlist_boot_getfirst(void)
796{
797 return(TAILQ_FIRST(&mountlist));
798}
799
800/*
801 * mountlist_remove (MP SAFE)
802 *
803 * Remove a node from the mountlist. If this node is the next scan node
804 * for any active mountlist scans, the active mountlist scan will be
805 * adjusted to skip the node, thus allowing removals during mountlist
806 * scans.
807 */
808void
809mountlist_remove(struct mount *mp)
810{
811 struct mountscan_info *msi;
812 lwkt_tokref ilock;
813
814 lwkt_gettoken(&ilock, &mountlist_token);
815 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) {
816 if (msi->msi_node == mp) {
817 if (msi->msi_how & MNTSCAN_FORWARD)
818 msi->msi_node = TAILQ_NEXT(mp, mnt_list);
819 else
820 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
821 }
822 }
823 TAILQ_REMOVE(&mountlist, mp, mnt_list);
824 lwkt_reltoken(&ilock);
825}
826
827/*
828 * mountlist_scan (MP SAFE)
829 *
830 * Safely scan the mount points on the mount list. Unless otherwise
831 * specified each mount point will be busied prior to the callback and
832 * unbusied afterwords. The callback may safely remove any mount point
833 * without interfering with the scan. If the current callback
834 * mount is removed the scanner will not attempt to unbusy it.
835 *
836 * If a mount node cannot be busied it is silently skipped.
837 *
838 * The callback return value is aggregated and a total is returned. A return
839 * value of < 0 is not aggregated and will terminate the scan.
840 *
841 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction
842 * MNTSCAN_REVERSE - the mountlist is scanned in reverse
843 * MNTSCAN_NOBUSY - the scanner will make the callback without busying
844 * the mount node.
845 */
846int
847mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how)
848{
849 struct mountscan_info info;
850 lwkt_tokref ilock;
851 struct mount *mp;
852 thread_t td;
853 int count;
854 int res;
855
856 lwkt_gettoken(&ilock, &mountlist_token);
857
858 info.msi_how = how;
859 info.msi_node = NULL; /* paranoia */
860 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry);
861
862 res = 0;
863 td = curthread;
864
865 if (how & MNTSCAN_FORWARD) {
866 info.msi_node = TAILQ_FIRST(&mountlist);
867 while ((mp = info.msi_node) != NULL) {
868 if (how & MNTSCAN_NOBUSY) {
869 count = callback(mp, data);
f9642f56 870 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
861905fb
MD
871 count = callback(mp, data);
872 if (mp == info.msi_node)
f9642f56 873 vfs_unbusy(mp);
861905fb
MD
874 } else {
875 count = 0;
876 }
877 if (count < 0)
878 break;
879 res += count;
880 if (mp == info.msi_node)
881 info.msi_node = TAILQ_NEXT(mp, mnt_list);
882 }
883 } else if (how & MNTSCAN_REVERSE) {
884 info.msi_node = TAILQ_LAST(&mountlist, mntlist);
885 while ((mp = info.msi_node) != NULL) {
886 if (how & MNTSCAN_NOBUSY) {
887 count = callback(mp, data);
f9642f56 888 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
861905fb
MD
889 count = callback(mp, data);
890 if (mp == info.msi_node)
f9642f56 891 vfs_unbusy(mp);
861905fb
MD
892 } else {
893 count = 0;
894 }
895 if (count < 0)
896 break;
897 res += count;
898 if (mp == info.msi_node)
899 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
900 }
901 }
902 TAILQ_REMOVE(&mountscan_list, &info, msi_entry);
903 lwkt_reltoken(&ilock);
904 return(res);
905}
906
907/*
908 * MOUNT RELATED VNODE FUNCTIONS
909 */
910
5fd012e0
MD
911static struct kproc_desc vnlru_kp = {
912 "vnlru",
913 vnlru_proc,
914 &vnlruthread
915};
916SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
917
918/*
919 * Move a vnode from one mount queue to another.
920 */
921void
922insmntque(struct vnode *vp, struct mount *mp)
923{
924 lwkt_tokref ilock;
925
926 lwkt_gettoken(&ilock, &mntvnode_token);
927 /*
928 * Delete from old mount point vnode list, if on one.
929 */
930 if (vp->v_mount != NULL) {
931 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
932 ("bad mount point vnode list size"));
be6c08cb 933 vremovevnodemnt(vp);
5fd012e0
MD
934 vp->v_mount->mnt_nvnodelistsize--;
935 }
936 /*
937 * Insert into list of vnodes for the new mount point, if available.
2ec4b00d 938 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer.
5fd012e0
MD
939 */
940 if ((vp->v_mount = mp) == NULL) {
941 lwkt_reltoken(&ilock);
942 return;
943 }
2ec4b00d
MD
944 if (mp->mnt_syncer) {
945 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes);
946 } else {
947 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
948 }
5fd012e0
MD
949 mp->mnt_nvnodelistsize++;
950 lwkt_reltoken(&ilock);
951}
952
953
954/*
be6c08cb
MD
955 * Scan the vnodes under a mount point and issue appropriate callbacks.
956 *
957 * The fastfunc() callback is called with just the mountlist token held
958 * (no vnode lock). It may not block and the vnode may be undergoing
959 * modifications while the caller is processing it. The vnode will
960 * not be entirely destroyed, however, due to the fact that the mountlist
961 * token is held. A return value < 0 skips to the next vnode without calling
962 * the slowfunc(), a return value > 0 terminates the loop.
963 *
964 * The slowfunc() callback is called after the vnode has been successfully
965 * locked based on passed flags. The vnode is skipped if it gets rearranged
966 * or destroyed while blocking on the lock. A non-zero return value from
967 * the slow function terminates the loop. The slow function is allowed to
968 * arbitrarily block. The scanning code guarentees consistency of operation
969 * even if the slow function deletes or moves the node, or blocks and some
970 * other thread deletes or moves the node.
5fd012e0
MD
971 */
972int
973vmntvnodescan(
974 struct mount *mp,
975 int flags,
976 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
977 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
978 void *data
979) {
be6c08cb 980 struct vmntvnodescan_info info;
5fd012e0 981 lwkt_tokref ilock;
5fd012e0
MD
982 struct vnode *vp;
983 int r = 0;
be6c08cb 984 int maxcount = 1000000;
19b97e01 985 int stopcount = 0;
9fe8385f 986 int count = 0;
5fd012e0
MD
987
988 lwkt_gettoken(&ilock, &mntvnode_token);
5fd012e0 989
19b97e01
MD
990 /*
991 * If asked to do one pass stop after iterating available vnodes.
992 * Under heavy loads new vnodes can be added while we are scanning,
993 * so this isn't perfect. Create a slop factor of 2x.
994 */
995 if (flags & VMSC_ONEPASS)
996 stopcount = mp->mnt_nvnodelistsize * 2;
997
be6c08cb
MD
998 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
999 TAILQ_INSERT_TAIL(&mntvnodescan_list, &info, entry);
1000 while ((vp = info.vp) != NULL) {
1001 if (--maxcount == 0)
1002 panic("maxcount reached during vmntvnodescan");
1003
2ec4b00d
MD
1004 /*
1005 * Skip if visible but not ready, or special (e.g.
1006 * mp->mnt_syncer)
1007 */
1008 if (vp->v_type == VNON)
be6c08cb 1009 goto next;
5fd012e0
MD
1010 KKASSERT(vp->v_mount == mp);
1011
1012 /*
1013 * Quick test. A negative return continues the loop without
1014 * calling the slow test. 0 continues onto the slow test.
1015 * A positive number aborts the loop.
1016 */
1017 if (fastfunc) {
c66c09cf
MD
1018 if ((r = fastfunc(mp, vp, data)) < 0) {
1019 r = 0;
be6c08cb 1020 goto next;
c66c09cf 1021 }
5fd012e0
MD
1022 if (r)
1023 break;
1024 }
1025
1026 /*
1027 * Get a vxlock on the vnode, retry if it has moved or isn't
1028 * in the mountlist where we expect it.
1029 */
1030 if (slowfunc) {
1031 int error;
1032
19b97e01 1033 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
5fd012e0 1034 case VMSC_GETVP:
87de5057 1035 error = vget(vp, LK_EXCLUSIVE);
5fd012e0
MD
1036 break;
1037 case VMSC_GETVP|VMSC_NOWAIT:
87de5057 1038 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT);
5fd012e0
MD
1039 break;
1040 case VMSC_GETVX:
e3332475
MD
1041 vx_get(vp);
1042 error = 0;
5fd012e0 1043 break;
5fd012e0
MD
1044 default:
1045 error = 0;
1046 break;
1047 }
1048 if (error)
be6c08cb
MD
1049 goto next;
1050 /*
1051 * Do not call the slow function if the vnode is
1052 * invalid or if it was ripped out from under us
1053 * while we (potentially) blocked.
1054 */
1055 if (info.vp == vp && vp->v_type != VNON)
1056 r = slowfunc(mp, vp, data);
1057
1058 /*
1059 * Cleanup
1060 */
19b97e01 1061 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
5fd012e0
MD
1062 case VMSC_GETVP:
1063 case VMSC_GETVP|VMSC_NOWAIT:
1064 vput(vp);
1065 break;
1066 case VMSC_GETVX:
1067 vx_put(vp);
1068 break;
5fd012e0
MD
1069 default:
1070 break;
1071 }
1072 if (r != 0)
1073 break;
1074 }
be6c08cb 1075
9fe8385f
SS
1076next:
1077 /*
1078 * Yield after some processing. Depending on the number
1079 * of vnodes, we might wind up running for a long time.
1080 * Because threads are not preemptable, time critical
1081 * userland processes might starve. Give them a chance
1082 * now and then.
1083 */
1084 if (++count == 10000) {
5a96e837
SS
1085 /* We really want to yield a bit, so we simply sleep a tick */
1086 tsleep(mp, 0, "vnodescn", 1);
9fe8385f
SS
1087 count = 0;
1088 }
1089
be6c08cb 1090 /*
19b97e01
MD
1091 * If doing one pass this decrements to zero. If it starts
1092 * at zero it is effectively unlimited for the purposes of
1093 * this loop.
1094 */
1095 if (--stopcount == 0)
1096 break;
1097
1098 /*
be6c08cb
MD
1099 * Iterate. If the vnode was ripped out from under us
1100 * info.vp will already point to the next vnode, otherwise
1101 * we have to obtain the next valid vnode ourselves.
1102 */
be6c08cb
MD
1103 if (info.vp == vp)
1104 info.vp = TAILQ_NEXT(vp, v_nmntvnodes);
5fd012e0 1105 }
be6c08cb 1106 TAILQ_REMOVE(&mntvnodescan_list, &info, entry);
5fd012e0
MD
1107 lwkt_reltoken(&ilock);
1108 return(r);
1109}
1110
1111/*
1112 * Remove any vnodes in the vnode table belonging to mount point mp.
1113 *
1114 * If FORCECLOSE is not specified, there should not be any active ones,
1115 * return error if any are found (nb: this is a user error, not a
1116 * system error). If FORCECLOSE is specified, detach any active vnodes
1117 * that are found.
1118 *
1119 * If WRITECLOSE is set, only flush out regular file vnodes open for
1120 * writing.
1121 *
1122 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1123 *
1124 * `rootrefs' specifies the base reference count for the root vnode
1125 * of this filesystem. The root vnode is considered busy if its
3c37c940 1126 * v_sysref.refcnt exceeds this value. On a successful return, vflush()
5fd012e0
MD
1127 * will call vrele() on the root vnode exactly rootrefs times.
1128 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1129 * be zero.
1130 */
1131#ifdef DIAGNOSTIC
1132static int busyprt = 0; /* print out busy vnodes */
1133SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1134#endif
1135
1136static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
1137
1138struct vflush_info {
1139 int flags;
1140 int busy;
1141 thread_t td;
1142};
1143
1144int
1145vflush(struct mount *mp, int rootrefs, int flags)
1146{
1147 struct thread *td = curthread; /* XXX */
1148 struct vnode *rootvp = NULL;
1149 int error;
1150 struct vflush_info vflush_info;
1151
1152 if (rootrefs > 0) {
1153 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1154 ("vflush: bad args"));
1155 /*
1156 * Get the filesystem root vnode. We can vput() it
1157 * immediately, since with rootrefs > 0, it won't go away.
1158 */
d9adbeaf
MD
1159 if ((error = VFS_ROOT(mp, &rootvp)) != 0) {
1160 if ((flags & FORCECLOSE) == 0)
1161 return (error);
1162 rootrefs = 0;
1163 /* continue anyway */
1164 }
1165 if (rootrefs)
1166 vput(rootvp);
5fd012e0
MD
1167 }
1168
1169 vflush_info.busy = 0;
1170 vflush_info.flags = flags;
1171 vflush_info.td = td;
1172 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
1173
1174 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1175 /*
1176 * If just the root vnode is busy, and if its refcount
1177 * is equal to `rootrefs', then go ahead and kill it.
1178 */
1179 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
3c37c940
MD
1180 KASSERT(rootvp->v_sysref.refcnt >= rootrefs, ("vflush: rootrefs"));
1181 if (vflush_info.busy == 1 && rootvp->v_sysref.refcnt == rootrefs) {
e3332475 1182 vx_lock(rootvp);
3c37c940 1183 vgone_vxlocked(rootvp);
e3332475
MD
1184 vx_unlock(rootvp);
1185 vflush_info.busy = 0;
5fd012e0
MD
1186 }
1187 }
1188 if (vflush_info.busy)
1189 return (EBUSY);
1190 for (; rootrefs > 0; rootrefs--)
1191 vrele(rootvp);
1192 return (0);
1193}
1194
1195/*
1196 * The scan callback is made with an VX locked vnode.
1197 */
1198static int
1199vflush_scan(struct mount *mp, struct vnode *vp, void *data)
1200{
1201 struct vflush_info *info = data;
1202 struct vattr vattr;
1203
1204 /*
1205 * Skip over a vnodes marked VSYSTEM.
1206 */
1207 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1208 return(0);
1209 }
1210
1211 /*
1212 * If WRITECLOSE is set, flush out unlinked but still open
1213 * files (even if open only for reading) and regular file
1214 * vnodes open for writing.
1215 */
1216 if ((info->flags & WRITECLOSE) &&
1217 (vp->v_type == VNON ||
87de5057 1218 (VOP_GETATTR(vp, &vattr) == 0 &&
5fd012e0
MD
1219 vattr.va_nlink > 0)) &&
1220 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1221 return(0);
1222 }
1223
1224 /*
3c37c940
MD
1225 * If we are the only holder (refcnt of 1) or the vnode is in
1226 * termination (refcnt < 0), we can vgone the vnode.
5fd012e0 1227 */
3c37c940
MD
1228 if (vp->v_sysref.refcnt <= 1) {
1229 vgone_vxlocked(vp);
5fd012e0
MD
1230 return(0);
1231 }
1232
1233 /*
1234 * If FORCECLOSE is set, forcibly close the vnode. For block
aec8eea4
MD
1235 * or character devices we just clean and leave the vp
1236 * associated with devfs. For all other files, just kill them.
1237 *
1238 * XXX we need to do something about devfs here, I'd rather not
1239 * blow away device associations.
5fd012e0
MD
1240 */
1241 if (info->flags & FORCECLOSE) {
aec8eea4
MD
1242 vgone_vxlocked(vp);
1243#if 0
5fd012e0 1244 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3c37c940 1245 vgone_vxlocked(vp);
5fd012e0 1246 } else {
3c37c940 1247 vclean_vxlocked(vp, 0);
aec8eea4 1248 /*vp->v_ops = &devfs_vnode_dev_vops_p;*/
5fd012e0
MD
1249 insmntque(vp, NULL);
1250 }
aec8eea4 1251#endif
5fd012e0
MD
1252 return(0);
1253 }
1254#ifdef DIAGNOSTIC
1255 if (busyprt)
1256 vprint("vflush: busy vnode", vp);
1257#endif
1258 ++info->busy;
1259 return(0);
1260}
1261
408357d8
MD
1262void
1263add_bio_ops(struct bio_ops *ops)
1264{
1265 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry);
1266}
1267
1268void
1269rem_bio_ops(struct bio_ops *ops)
1270{
1271 TAILQ_REMOVE(&bio_ops_list, ops, entry);
1272}
1273
1274/*
1275 * This calls the bio_ops io_sync function either for a mount point
1276 * or generally.
1277 *
1278 * WARNING: softdeps is weirdly coded and just isn't happy unless
1279 * io_sync is called with a NULL mount from the general syncing code.
1280 */
1281void
1282bio_ops_sync(struct mount *mp)
1283{
1284 struct bio_ops *ops;
1285
1286 if (mp) {
1287 if ((ops = mp->mnt_bioops) != NULL)
1288 ops->io_sync(mp);
1289 } else {
1290 TAILQ_FOREACH(ops, &bio_ops_list, entry) {
1291 ops->io_sync(NULL);
1292 }
1293 }
1294}
1295