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 */
40
41/*
42 * External virtual filesystem routines
43 */
44
45#include <sys/cdefs.h>
46__FBSDID("$FreeBSD: head/sys/kern/vfs_subr.c 117879 2003-07-22 10:36:36Z phk $");
47
48#include "opt_ddb.h"
49#include "opt_mac.h"
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/bio.h>
54#include <sys/buf.h>
55#include <sys/conf.h>
56#include <sys/eventhandler.h>
57#include <sys/extattr.h>
58#include <sys/fcntl.h>
59#include <sys/kernel.h>
60#include <sys/kthread.h>
61#include <sys/mac.h>
62#include <sys/malloc.h>
63#include <sys/mount.h>
64#include <sys/namei.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 <vm/vm.h>
72#include <vm/vm_object.h>
73#include <vm/vm_extern.h>
74#include <vm/pmap.h>
75#include <vm/vm_map.h>
76#include <vm/vm_page.h>
77#include <vm/vm_kern.h>
78#include <vm/uma.h>
79
80static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
81
82static void addalias(struct vnode *vp, dev_t nvp_rdev);
83static void insmntque(struct vnode *vp, struct mount *mp);
84static void vclean(struct vnode *vp, int flags, struct thread *td);
85static void vlruvp(struct vnode *vp);
86static int flushbuflist(struct buf *blist, int flags, struct vnode *vp,
87 int slpflag, int slptimeo, int *errorp);
88static int vcanrecycle(struct vnode *vp, struct mount **vnmpp);
89
90
91/*
92 * Number of vnodes in existence. Increased whenever getnewvnode()
93 * allocates a new vnode, never decreased.
94 */
95static unsigned long numvnodes;
96
97SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
98
99/*
100 * Conversion tables for conversion from vnode types to inode formats
101 * and back.
102 */
103enum vtype iftovt_tab[16] = {
104 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
105 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
106};
107int vttoif_tab[9] = {
108 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
109 S_IFSOCK, S_IFIFO, S_IFMT,
110};
111
112/*
113 * List of vnodes that are ready for recycling.
114 */
115static TAILQ_HEAD(freelst, vnode) vnode_free_list;
116
117/*
118 * Minimum number of free vnodes. If there are fewer than this free vnodes,
119 * getnewvnode() will return a newly allocated vnode.
120 */
121static u_long wantfreevnodes = 25;
122SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
123/* Number of vnodes in the free list. */
124static u_long freevnodes;
125SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
126
127/*
128 * Various variables used for debugging the new implementation of
129 * reassignbuf().
130 * XXX these are probably of (very) limited utility now.
131 */
132static int reassignbufcalls;
133SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
134static int nameileafonly;
135SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
136
137/*
138 * Cache for the mount type id assigned to NFS. This is used for
139 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
140 */
141int nfs_mount_type = -1;
142
143/* To keep more than one thread at a time from running vfs_getnewfsid */
144static struct mtx mntid_mtx;
145
146/*
147 * Lock for any access to the following:
148 * vnode_free_list
149 * numvnodes
150 * freevnodes
151 */
152static struct mtx vnode_free_list_mtx;
153
154/*
155 * For any iteration/modification of dev->si_hlist (linked through
156 * v_specnext)
157 */
158static struct mtx spechash_mtx;
159
160/* Publicly exported FS */
161struct nfs_public nfs_pub;
162
163/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
164static uma_zone_t vnode_zone;
165static uma_zone_t vnodepoll_zone;
166
167/* Set to 1 to print out reclaim of active vnodes */
168int prtactive;
169
170/*
171 * The workitem queue.
172 *
173 * It is useful to delay writes of file data and filesystem metadata
174 * for tens of seconds so that quickly created and deleted files need
175 * not waste disk bandwidth being created and removed. To realize this,
176 * we append vnodes to a "workitem" queue. When running with a soft
177 * updates implementation, most pending metadata dependencies should
178 * not wait for more than a few seconds. Thus, mounted on block devices
179 * are delayed only about a half the time that file data is delayed.
180 * Similarly, directory updates are more critical, so are only delayed
181 * about a third the time that file data is delayed. Thus, there are
182 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
183 * one each second (driven off the filesystem syncer process). The
184 * syncer_delayno variable indicates the next queue that is to be processed.
185 * Items that need to be processed soon are placed in this queue:
186 *
187 * syncer_workitem_pending[syncer_delayno]
188 *
189 * A delay of fifteen seconds is done by placing the request fifteen
190 * entries later in the queue:
191 *
192 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
193 *
194 */
195static int syncer_delayno;
196static long syncer_mask;
197LIST_HEAD(synclist, vnode);
198static struct synclist *syncer_workitem_pending;
199/*
200 * The sync_mtx protects:
201 * vp->v_synclist
202 * syncer_delayno
203 * syncer_workitem_pending
204 * rushjob
205 */
206static struct mtx sync_mtx;
207
208#define SYNCER_MAXDELAY 32
209static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
210static int syncdelay = 30; /* max time to delay syncing data */
211static int filedelay = 30; /* time to delay syncing files */
212SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
213static int dirdelay = 29; /* time to delay syncing directories */
214SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
215static int metadelay = 28; /* time to delay syncing metadata */
216SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
217static int rushjob; /* number of slots to run ASAP */
218static int stat_rush_requests; /* number of times I/O speeded up */
219SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
220
221/*
222 * Number of vnodes we want to exist at any one time. This is mostly used
223 * to size hash tables in vnode-related code. It is normally not used in
224 * getnewvnode(), as wantfreevnodes is normally nonzero.)
225 *
226 * XXX desiredvnodes is historical cruft and should not exist.
227 */
228int desiredvnodes;
229SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
230 &desiredvnodes, 0, "Maximum number of vnodes");
231static int minvnodes;
232SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
233 &minvnodes, 0, "Minimum number of vnodes");
234static int vnlru_nowhere;
235SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
236 "Number of times the vnlru process ran without success");
237
238/* Hook for calling soft updates */
239int (*softdep_process_worklist_hook)(struct mount *);
240
241/*
242 * This only exists to supress warnings from unlocked specfs accesses. It is
243 * no longer ok to have an unlocked VFS.
244 */
245#define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
246
247/* Print lock violations */
248int vfs_badlock_print = 1;
249
250/* Panic on violation */
251int vfs_badlock_panic = 1;
252
253/* Check for interlock across VOPs */
254int vfs_badlock_mutex = 1;
255
256static void
257vfs_badlock(char *msg, char *str, struct vnode *vp)
258{
259 if (vfs_badlock_print)
260 printf("%s: %p %s\n", str, vp, msg);
261 if (vfs_badlock_panic)
262 Debugger("Lock violation.\n");
263}
264
265void
266assert_vi_unlocked(struct vnode *vp, char *str)
267{
268 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
269 vfs_badlock("interlock is locked but should not be", str, vp);
270}
271
272void
273assert_vi_locked(struct vnode *vp, char *str)
274{
275 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
276 vfs_badlock("interlock is not locked but should be", str, vp);
277}
278
279void
280assert_vop_locked(struct vnode *vp, char *str)
281{
282 if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
283 vfs_badlock("is not locked but should be", str, vp);
284}
285
286void
287assert_vop_unlocked(struct vnode *vp, char *str)
288{
289 if (vp && !IGNORE_LOCK(vp) &&
290 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
291 vfs_badlock("is locked but should not be", str, vp);
292}
293
294void
295assert_vop_elocked(struct vnode *vp, char *str)
296{
297 if (vp && !IGNORE_LOCK(vp) &&
298 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
299 vfs_badlock("is not exclusive locked but should be", str, vp);
300}
301
302void
303assert_vop_elocked_other(struct vnode *vp, char *str)
304{
305 if (vp && !IGNORE_LOCK(vp) &&
306 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
307 vfs_badlock("is not exclusive locked by another thread",
308 str, vp);
309}
310
311void
312assert_vop_slocked(struct vnode *vp, char *str)
313{
314 if (vp && !IGNORE_LOCK(vp) &&
315 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
316 vfs_badlock("is not locked shared but should be", str, vp);
317}
318
319void
320vop_rename_pre(void *ap)
321{
322 struct vop_rename_args *a = ap;
323
324 if (a->a_tvp)
325 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
326 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
327 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
328 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
329
330 /* Check the source (from) */
331 if (a->a_tdvp != a->a_fdvp)
332 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
333 if (a->a_tvp != a->a_fvp)
334 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
335
336 /* Check the target */
337 if (a->a_tvp)
338 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
339
340 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
341}
342
343void
344vop_strategy_pre(void *ap)
345{
346 struct vop_strategy_args *a = ap;
347 struct buf *bp;
348
349 bp = a->a_bp;
350
351 /*
352 * Cluster ops lock their component buffers but not the IO container.
353 */
354 if ((bp->b_flags & B_CLUSTER) != 0)
355 return;
356
357 if (BUF_REFCNT(bp) < 1) {
358 if (vfs_badlock_print)
359 printf("VOP_STRATEGY: bp is not locked but should be.\n");
360 if (vfs_badlock_panic)
361 Debugger("Lock violation.\n");
362 }
363}
364
365void
366vop_lookup_pre(void *ap)
367{
368 struct vop_lookup_args *a = ap;
369 struct vnode *dvp;
370
371 dvp = a->a_dvp;
372
373 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
374 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
375}
376
377void
378vop_lookup_post(void *ap, int rc)
379{
380 struct vop_lookup_args *a = ap;
381 struct componentname *cnp;
382 struct vnode *dvp;
383 struct vnode *vp;
384 int flags;
385
386 dvp = a->a_dvp;
387 cnp = a->a_cnp;
388 vp = *(a->a_vpp);
389 flags = cnp->cn_flags;
390
391
392 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
393 /*
394 * If this is the last path component for this lookup and LOCPARENT
395 * is set, OR if there is an error the directory has to be locked.
396 */
397 if ((flags & LOCKPARENT) && (flags & ISLASTCN))
398 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
399 else if (rc != 0)
400 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
401 else if (dvp != vp)
402 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
403
404 if (flags & PDIRUNLOCK)
405 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
406}
407
408void
409vop_unlock_pre(void *ap)
410{
411 struct vop_unlock_args *a = ap;
412
413 if (a->a_flags & LK_INTERLOCK)
414 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
415
416 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
417}
418
419void
420vop_unlock_post(void *ap, int rc)
421{
422 struct vop_unlock_args *a = ap;
423
424 if (a->a_flags & LK_INTERLOCK)
425 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
426}
427
428void
429vop_lock_pre(void *ap)
430{
431 struct vop_lock_args *a = ap;
432
433 if ((a->a_flags & LK_INTERLOCK) == 0)
434 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
435 else
436 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
437}
438
439void
440vop_lock_post(void *ap, int rc)
441{
442 struct vop_lock_args *a;
443
444 a = ap;
445
446 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
447 if (rc == 0)
448 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
449}
450
451void
452v_addpollinfo(struct vnode *vp)
453{
454 vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
455 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
456}
457
458/*
459 * Initialize the vnode management data structures.
460 */
461static void
462vntblinit(void *dummy __unused)
463{
464
465 /*
466 * Desiredvnodes is a function of the physical memory size and
467 * the kernel's heap size. Specifically, desiredvnodes scales
468 * in proportion to the physical memory size until two fifths
469 * of the kernel's heap size is consumed by vnodes and vm
470 * objects.
471 */
472 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
473 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
474 minvnodes = desiredvnodes / 4;
475 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
476 mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
477 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
478 mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
479 TAILQ_INIT(&vnode_free_list);
480 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
481 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
482 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
483 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
484 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
485 /*
486 * Initialize the filesystem syncer.
487 */
488 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
489 &syncer_mask);
490 syncer_maxdelay = syncer_mask + 1;
491 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
492}
493SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
494
495
496/*
497 * Mark a mount point as busy. Used to synchronize access and to delay
498 * unmounting. Interlock is not released on failure.
499 */
500int
501vfs_busy(mp, flags, interlkp, td)
502 struct mount *mp;
503 int flags;
504 struct mtx *interlkp;
505 struct thread *td;
506{
507 int lkflags;
508
509 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
510 if (flags & LK_NOWAIT)
511 return (ENOENT);
512 mp->mnt_kern_flag |= MNTK_MWAIT;
513 /*
514 * Since all busy locks are shared except the exclusive
515 * lock granted when unmounting, the only place that a
516 * wakeup needs to be done is at the release of the
517 * exclusive lock at the end of dounmount.
518 */
519 msleep(mp, interlkp, PVFS, "vfs_busy", 0);
520 return (ENOENT);
521 }
522 lkflags = LK_SHARED | LK_NOPAUSE;
523 if (interlkp)
524 lkflags |= LK_INTERLOCK;
525 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
526 panic("vfs_busy: unexpected lock failure");
527 return (0);
528}
529
530/*
531 * Free a busy filesystem.
532 */
533void
534vfs_unbusy(mp, td)
535 struct mount *mp;
536 struct thread *td;
537{
538
539 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
540}
541
542/*
543 * Lookup a mount point by filesystem identifier.
544 */
545struct mount *
546vfs_getvfs(fsid)
547 fsid_t *fsid;
548{
549 register struct mount *mp;
550
551 mtx_lock(&mountlist_mtx);
552 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
553 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
554 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
555 mtx_unlock(&mountlist_mtx);
556 return (mp);
557 }
558 }
559 mtx_unlock(&mountlist_mtx);
560 return ((struct mount *) 0);
561}
562
563/*
564 * Get a new unique fsid. Try to make its val[0] unique, since this value
565 * will be used to create fake device numbers for stat(). Also try (but
566 * not so hard) make its val[0] unique mod 2^16, since some emulators only
567 * support 16-bit device numbers. We end up with unique val[0]'s for the
568 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
569 *
570 * Keep in mind that several mounts may be running in parallel. Starting
571 * the search one past where the previous search terminated is both a
572 * micro-optimization and a defense against returning the same fsid to
573 * different mounts.
574 */
575void
576vfs_getnewfsid(mp)
577 struct mount *mp;
578{
579 static u_int16_t mntid_base;
580 fsid_t tfsid;
581 int mtype;
582
583 mtx_lock(&mntid_mtx);
584 mtype = mp->mnt_vfc->vfc_typenum;
585 tfsid.val[1] = mtype;
586 mtype = (mtype & 0xFF) << 24;
587 for (;;) {
588 tfsid.val[0] = makeudev(255,
589 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
590 mntid_base++;
591 if (vfs_getvfs(&tfsid) == NULL)
592 break;
593 }
594 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
595 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
596 mtx_unlock(&mntid_mtx);
597}
598
599/*
600 * Knob to control the precision of file timestamps:
601 *
602 * 0 = seconds only; nanoseconds zeroed.
603 * 1 = seconds and nanoseconds, accurate within 1/HZ.
604 * 2 = seconds and nanoseconds, truncated to microseconds.
605 * >=3 = seconds and nanoseconds, maximum precision.
606 */
607enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
608
609static int timestamp_precision = TSP_SEC;
610SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
611 ×tamp_precision, 0, "");
612
613/*
614 * Get a current timestamp.
615 */
616void
617vfs_timestamp(tsp)
618 struct timespec *tsp;
619{
620 struct timeval tv;
621
622 switch (timestamp_precision) {
623 case TSP_SEC:
624 tsp->tv_sec = time_second;
625 tsp->tv_nsec = 0;
626 break;
627 case TSP_HZ:
628 getnanotime(tsp);
629 break;
630 case TSP_USEC:
631 microtime(&tv);
632 TIMEVAL_TO_TIMESPEC(&tv, tsp);
633 break;
634 case TSP_NSEC:
635 default:
636 nanotime(tsp);
637 break;
638 }
639}
640
641/*
642 * Set vnode attributes to VNOVAL
643 */
644void
645vattr_null(vap)
646 register struct vattr *vap;
647{
648
649 vap->va_type = VNON;
650 vap->va_size = VNOVAL;
651 vap->va_bytes = VNOVAL;
652 vap->va_mode = VNOVAL;
653 vap->va_nlink = VNOVAL;
654 vap->va_uid = VNOVAL;
655 vap->va_gid = VNOVAL;
656 vap->va_fsid = VNOVAL;
657 vap->va_fileid = VNOVAL;
658 vap->va_blocksize = VNOVAL;
659 vap->va_rdev = VNOVAL;
660 vap->va_atime.tv_sec = VNOVAL;
661 vap->va_atime.tv_nsec = VNOVAL;
662 vap->va_mtime.tv_sec = VNOVAL;
663 vap->va_mtime.tv_nsec = VNOVAL;
664 vap->va_ctime.tv_sec = VNOVAL;
665 vap->va_ctime.tv_nsec = VNOVAL;
666 vap->va_birthtime.tv_sec = VNOVAL;
667 vap->va_birthtime.tv_nsec = VNOVAL;
668 vap->va_flags = VNOVAL;
669 vap->va_gen = VNOVAL;
670 vap->va_vaflags = 0;
671}
672
673/*
674 * This routine is called when we have too many vnodes. It attempts
675 * to free <count> vnodes and will potentially free vnodes that still
676 * have VM backing store (VM backing store is typically the cause
677 * of a vnode blowout so we want to do this). Therefore, this operation
678 * is not considered cheap.
679 *
680 * A number of conditions may prevent a vnode from being reclaimed.
681 * the buffer cache may have references on the vnode, a directory
682 * vnode may still have references due to the namei cache representing
683 * underlying files, or the vnode may be in active use. It is not
684 * desireable to reuse such vnodes. These conditions may cause the
685 * number of vnodes to reach some minimum value regardless of what
686 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
687 */
688static int
689vlrureclaim(struct mount *mp)
690{
691 struct vnode *vp;
692 int done;
693 int trigger;
694 int usevnodes;
695 int count;
696
697 /*
698 * Calculate the trigger point, don't allow user
699 * screwups to blow us up. This prevents us from
700 * recycling vnodes with lots of resident pages. We
701 * aren't trying to free memory, we are trying to
702 * free vnodes.
703 */
704 usevnodes = desiredvnodes;
705 if (usevnodes <= 0)
706 usevnodes = 1;
707 trigger = cnt.v_page_count * 2 / usevnodes;
708
709 done = 0;
710 mtx_lock(&mntvnode_mtx);
711 count = mp->mnt_nvnodelistsize / 10 + 1;
712 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
713 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
714 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
715
716 if (vp->v_type != VNON &&
717 vp->v_type != VBAD &&
718 VI_TRYLOCK(vp)) {
719 if (VMIGHTFREE(vp) && /* critical path opt */
720 (vp->v_object == NULL ||
721 vp->v_object->resident_page_count < trigger)) {
722 mtx_unlock(&mntvnode_mtx);
723 vgonel(vp, curthread);
724 done++;
725 mtx_lock(&mntvnode_mtx);
726 } else
727 VI_UNLOCK(vp);
728 }
729 --count;
730 }
731 mtx_unlock(&mntvnode_mtx);
732 return done;
733}
734
735/*
736 * Attempt to recycle vnodes in a context that is always safe to block.
737 * Calling vlrurecycle() from the bowels of filesystem code has some
738 * interesting deadlock problems.
739 */
740static struct proc *vnlruproc;
741static int vnlruproc_sig;
742
743static void
744vnlru_proc(void)
745{
746 struct mount *mp, *nmp;
747 int s;
748 int done;
749 struct proc *p = vnlruproc;
750 struct thread *td = FIRST_THREAD_IN_PROC(p); /* XXXKSE */
751
752 mtx_lock(&Giant);
753
754 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
755 SHUTDOWN_PRI_FIRST);
756
757 s = splbio();
758 for (;;) {
759 kthread_suspend_check(p);
760 mtx_lock(&vnode_free_list_mtx);
761 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
762 mtx_unlock(&vnode_free_list_mtx);
763 vnlruproc_sig = 0;
764 wakeup(&vnlruproc_sig);
765 tsleep(vnlruproc, PVFS, "vlruwt", hz);
766 continue;
767 }
768 mtx_unlock(&vnode_free_list_mtx);
769 done = 0;
770 mtx_lock(&mountlist_mtx);
771 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
772 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
773 nmp = TAILQ_NEXT(mp, mnt_list);
774 continue;
775 }
776 done += vlrureclaim(mp);
777 mtx_lock(&mountlist_mtx);
778 nmp = TAILQ_NEXT(mp, mnt_list);
779 vfs_unbusy(mp, td);
780 }
781 mtx_unlock(&mountlist_mtx);
782 if (done == 0) {
783#if 0
784 /* These messages are temporary debugging aids */
785 if (vnlru_nowhere < 5)
786 printf("vnlru process getting nowhere..\n");
787 else if (vnlru_nowhere == 5)
788 printf("vnlru process messages stopped.\n");
789#endif
790 vnlru_nowhere++;
791 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
792 }
793 }
794 splx(s);
795}
796
797static struct kproc_desc vnlru_kp = {
798 "vnlru",
799 vnlru_proc,
800 &vnlruproc
801};
802SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
803
804
805/*
806 * Routines having to do with the management of the vnode table.
807 */
808
809/*
810 * Check to see if a free vnode can be recycled. If it can,
811 * return it locked with the vn lock, but not interlock. Also
812 * get the vn_start_write lock. Otherwise indicate the error.
813 */
814static int
815vcanrecycle(struct vnode *vp, struct mount **vnmpp)
816{
817 struct thread *td = curthread;
818 vm_object_t object;
819 int error;
820
821 /* Don't recycle if we can't get the interlock */
822 if (!VI_TRYLOCK(vp))
823 return (EWOULDBLOCK);
824
825 /* We should be able to immediately acquire this */
826 /* XXX This looks like it should panic if it fails */
827 if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
828 if (VOP_ISLOCKED(vp, td))
829 panic("vcanrecycle: locked vnode");
830 return (EWOULDBLOCK);
831 }
832
833 /*
834 * Don't recycle if its filesystem is being suspended.
835 */
836 if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
837 error = EBUSY;
838 goto done;
839 }
840
841 /*
842 * Don't recycle if we still have cached pages.
843 */
844 if (VOP_GETVOBJECT(vp, &object) == 0) {
845 VM_OBJECT_LOCK(object);
846 if (object->resident_page_count ||
847 object->ref_count) {
848 VM_OBJECT_UNLOCK(object);
849 error = EBUSY;
850 goto done;
851 }
852 VM_OBJECT_UNLOCK(object);
853 }
854 if (LIST_FIRST(&vp->v_cache_src)) {
855 /*
856 * note: nameileafonly sysctl is temporary,
857 * for debugging only, and will eventually be
858 * removed.
859 */
860 if (nameileafonly > 0) {
861 /*
862 * Do not reuse namei-cached directory
863 * vnodes that have cached
864 * subdirectories.
865 */
866 if (cache_leaf_test(vp) < 0) {
867 error = EISDIR;
868 goto done;
869 }
870 } else if (nameileafonly < 0 ||
871 vmiodirenable == 0) {
872 /*
873 * Do not reuse namei-cached directory
874 * vnodes if nameileafonly is -1 or
875 * if VMIO backing for directories is
876 * turned off (otherwise we reuse them
877 * too quickly).
878 */
879 error = EBUSY;
880 goto done;
881 }
882 }
883 return (0);
884done:
885 VOP_UNLOCK(vp, 0, td);
886 return (error);
887}
888
889/*
890 * Return the next vnode from the free list.
891 */
892int
893getnewvnode(tag, mp, vops, vpp)
894 const char *tag;
895 struct mount *mp;
896 vop_t **vops;
897 struct vnode **vpp;
898{
899 struct thread *td = curthread; /* XXX */
900 struct vnode *vp = NULL;
901 struct vpollinfo *pollinfo = NULL;
902 struct mount *vnmp;
903
904 mtx_lock(&vnode_free_list_mtx);
905
906 /*
907 * Try to reuse vnodes if we hit the max. This situation only
908 * occurs in certain large-memory (2G+) situations. We cannot
909 * attempt to directly reclaim vnodes due to nasty recursion
910 * problems.
911 */
912 while (numvnodes - freevnodes > desiredvnodes) {
913 if (vnlruproc_sig == 0) {
914 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
915 wakeup(vnlruproc);
916 }
917 mtx_unlock(&vnode_free_list_mtx);
918 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
919 mtx_lock(&vnode_free_list_mtx);
920 }
921
922 /*
923 * Attempt to reuse a vnode already on the free list, allocating
924 * a new vnode if we can't find one or if we have not reached a
925 * good minimum for good LRU performance.
926 */
927
928 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
929 int error;
930 int count;
931
932 for (count = 0; count < freevnodes; count++) {
933 vp = TAILQ_FIRST(&vnode_free_list);
934
935 KASSERT(vp->v_usecount == 0 &&
936 (vp->v_iflag & VI_DOINGINACT) == 0,
937 ("getnewvnode: free vnode isn't"));
938
939 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
940 /*
941 * We have to drop the free list mtx to avoid lock
942 * order reversals with interlock.
943 */
944 mtx_unlock(&vnode_free_list_mtx);
945 error = vcanrecycle(vp, &vnmp);
946 mtx_lock(&vnode_free_list_mtx);
947 if (error == 0)
948 break;
949 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
950 vp = NULL;
951 }
952 }
953 if (vp) {
954 freevnodes--;
955 mtx_unlock(&vnode_free_list_mtx);
956
957 cache_purge(vp);
958 VI_LOCK(vp);
959 vp->v_iflag |= VI_DOOMED;
960 vp->v_iflag &= ~VI_FREE;
961 if (vp->v_type != VBAD) {
962 VOP_UNLOCK(vp, 0, td);
963 vgonel(vp, td);
964 VI_LOCK(vp);
965 } else {
966 VOP_UNLOCK(vp, 0, td);
967 }
968 vn_finished_write(vnmp);
969
970#ifdef INVARIANTS
971 {
972 if (vp->v_data)
973 panic("cleaned vnode isn't");
974 if (vp->v_numoutput)
975 panic("Clean vnode has pending I/O's");
976 if (vp->v_writecount != 0)
977 panic("Non-zero write count");
978 }
979#endif
980 if ((pollinfo = vp->v_pollinfo) != NULL) {
981 /*
982 * To avoid lock order reversals, the call to
983 * uma_zfree() must be delayed until the vnode
984 * interlock is released.
985 */
986 vp->v_pollinfo = NULL;
987 }
988#ifdef MAC
989 mac_destroy_vnode(vp);
990#endif
991 vp->v_iflag = 0;
992 vp->v_vflag = 0;
993 vp->v_lastw = 0;
994 vp->v_lasta = 0;
995 vp->v_cstart = 0;
996 vp->v_clen = 0;
997 vp->v_socket = 0;
998 lockdestroy(vp->v_vnlock);
999 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1000 KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
1001 KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
1002 KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
1003 KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
1004 } else {
1005 numvnodes++;
1006 mtx_unlock(&vnode_free_list_mtx);
1007
1008 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1009 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1010 VI_LOCK(vp);
1011 vp->v_dd = vp;
1012 vp->v_vnlock = &vp->v_lock;
1013 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1014 cache_purge(vp);
1015 LIST_INIT(&vp->v_cache_src);
1016 TAILQ_INIT(&vp->v_cache_dst);
1017 }
1018
1019 TAILQ_INIT(&vp->v_cleanblkhd);
1020 TAILQ_INIT(&vp->v_dirtyblkhd);
1021 vp->v_type = VNON;
1022 vp->v_tag = tag;
1023 vp->v_op = vops;
1024 *vpp = vp;
1025 vp->v_usecount = 1;
1026 vp->v_data = 0;
1027 vp->v_cachedid = -1;
1028 VI_UNLOCK(vp);
1029 if (pollinfo != NULL) {
1030 mtx_destroy(&pollinfo->vpi_lock);
1031 uma_zfree(vnodepoll_zone, pollinfo);
1032 }
1033#ifdef MAC
1034 mac_init_vnode(vp);
1035 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1036 mac_associate_vnode_singlelabel(mp, vp);
1037#endif
1038 insmntque(vp, mp);
1039
1040 return (0);
1041}
1042
1043/*
1044 * Move a vnode from one mount queue to another.
1045 */
1046static void
1047insmntque(vp, mp)
1048 register struct vnode *vp;
1049 register struct mount *mp;
1050{
1051
1052 mtx_lock(&mntvnode_mtx);
1053 /*
1054 * Delete from old mount point vnode list, if on one.
1055 */
1056 if (vp->v_mount != NULL) {
1057 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
1058 ("bad mount point vnode list size"));
1059 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1060 vp->v_mount->mnt_nvnodelistsize--;
1061 }
1062 /*
1063 * Insert into list of vnodes for the new mount point, if available.
1064 */
1065 if ((vp->v_mount = mp) == NULL) {
1066 mtx_unlock(&mntvnode_mtx);
1067 return;
1068 }
1069 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1070 mp->mnt_nvnodelistsize++;
1071 mtx_unlock(&mntvnode_mtx);
1072}
1073
1074/*
1075 * Update outstanding I/O count and do wakeup if requested.
1076 */
1077void
1078vwakeup(bp)
1079 register struct buf *bp;
1080{
1081 register struct vnode *vp;
1082
1083 bp->b_flags &= ~B_WRITEINPROG;
1084 if ((vp = bp->b_vp)) {
1085 VI_LOCK(vp);
1086 vp->v_numoutput--;
1087 if (vp->v_numoutput < 0)
1088 panic("vwakeup: neg numoutput");
1089 if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1090 vp->v_iflag &= ~VI_BWAIT;
1091 wakeup(&vp->v_numoutput);
1092 }
1093 VI_UNLOCK(vp);
1094 }
1095}
1096
1097/*
1098 * Flush out and invalidate all buffers associated with a vnode.
1099 * Called with the underlying object locked.
1100 */
1101int
1102vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1103 struct vnode *vp;
1104 int flags;
1105 struct ucred *cred;
1106 struct thread *td;
1107 int slpflag, slptimeo;
1108{
1109 struct buf *blist;
1110 int s, error;
1111 vm_object_t object;
1112
1113 GIANT_REQUIRED;
1114
1115 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1116
1117 VI_LOCK(vp);
1118 if (flags & V_SAVE) {
1119 s = splbio();
1120 while (vp->v_numoutput) {
1121 vp->v_iflag |= VI_BWAIT;
1122 error = msleep(&vp->v_numoutput, VI_MTX(vp),
1123 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1124 if (error) {
1125 VI_UNLOCK(vp);
1126 splx(s);
1127 return (error);
1128 }
1129 }
1130 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1131 splx(s);
1132 VI_UNLOCK(vp);
1133 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1134 return (error);
1135 /*
1136 * XXX We could save a lock/unlock if this was only
1137 * enabled under INVARIANTS
1138 */
1139 VI_LOCK(vp);
1140 s = splbio();
1141 if (vp->v_numoutput > 0 ||
1142 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1143 panic("vinvalbuf: dirty bufs");
1144 }
1145 splx(s);
1146 }
1147 s = splbio();
1148 /*
1149 * If you alter this loop please notice that interlock is dropped and
1150 * reacquired in flushbuflist. Special care is needed to ensure that
1151 * no race conditions occur from this.
1152 */
1153 for (error = 0;;) {
1154 if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1155 flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1156 if (error)
1157 break;
1158 continue;
1159 }
1160 if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1161 flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1162 if (error)
1163 break;
1164 continue;
1165 }
1166 break;
1167 }
1168 if (error) {
1169 splx(s);
1170 VI_UNLOCK(vp);
1171 return (error);
1172 }
1173
1174 /*
1175 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1176 * have write I/O in-progress but if there is a VM object then the
1177 * VM object can also have read-I/O in-progress.
1178 */
1179 do {
1180 while (vp->v_numoutput > 0) {
1181 vp->v_iflag |= VI_BWAIT;
1182 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1183 }
1184 VI_UNLOCK(vp);
1185 if (VOP_GETVOBJECT(vp, &object) == 0) {
1186 VM_OBJECT_LOCK(object);
1187 vm_object_pip_wait(object, "vnvlbx");
1188 VM_OBJECT_UNLOCK(object);
1189 }
1190 VI_LOCK(vp);
1191 } while (vp->v_numoutput > 0);
1192 VI_UNLOCK(vp);
1193
1194 splx(s);
1195
1196 /*
1197 * Destroy the copy in the VM cache, too.
1198 */
1199 if (VOP_GETVOBJECT(vp, &object) == 0) {
1200 VM_OBJECT_LOCK(object);
1201 vm_object_page_remove(object, 0, 0,
1202 (flags & V_SAVE) ? TRUE : FALSE);
1203 VM_OBJECT_UNLOCK(object);
1204 }
1205
1206#ifdef INVARIANTS
1207 VI_LOCK(vp);
1208 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1209 (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1210 !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1211 panic("vinvalbuf: flush failed");
1212 VI_UNLOCK(vp);
1213#endif
1214 return (0);
1215}
1216
1217/*
1218 * Flush out buffers on the specified list.
1219 *
1220 */
1221static int
1222flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1223 struct buf *blist;
1224 int flags;
1225 struct vnode *vp;
1226 int slpflag, slptimeo;
1227 int *errorp;
1228{
1229 struct buf *bp, *nbp;
1230 int found, error;
1231
1232 ASSERT_VI_LOCKED(vp, "flushbuflist");
1233
1234 for (found = 0, bp = blist; bp; bp = nbp) {
1235 nbp = TAILQ_NEXT(bp, b_vnbufs);
1236 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1237 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1238 continue;
1239 }
1240 found += 1;
1241 error = BUF_TIMELOCK(bp,
1242 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1243 "flushbuf", slpflag, slptimeo);
1244 if (error) {
1245 if (error != ENOLCK)
1246 *errorp = error;
1247 goto done;
1248 }
1249 /*
1250 * XXX Since there are no node locks for NFS, I
1251 * believe there is a slight chance that a delayed
1252 * write will occur while sleeping just above, so
1253 * check for it. Note that vfs_bio_awrite expects
1254 * buffers to reside on a queue, while BUF_WRITE and
1255 * brelse do not.
1256 */
1257 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1258 (flags & V_SAVE)) {
1259
1260 if (bp->b_vp == vp) {
1261 if (bp->b_flags & B_CLUSTEROK) {
1262 vfs_bio_awrite(bp);
1263 } else {
1264 bremfree(bp);
1265 bp->b_flags |= B_ASYNC;
1266 BUF_WRITE(bp);
1267 }
1268 } else {
1269 bremfree(bp);
1270 (void) BUF_WRITE(bp);
1271 }
1272 goto done;
1273 }
1274 bremfree(bp);
1275 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1276 bp->b_flags &= ~B_ASYNC;
1277 brelse(bp);
1278 VI_LOCK(vp);
1279 }
1280 return (found);
1281done:
1282 VI_LOCK(vp);
1283 return (found);
1284}
1285
1286/*
1287 * Truncate a file's buffer and pages to a specified length. This
1288 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1289 * sync activity.
1290 */
1291int
1292vtruncbuf(vp, cred, td, length, blksize)
1293 register struct vnode *vp;
1294 struct ucred *cred;
1295 struct thread *td;
1296 off_t length;
1297 int blksize;
1298{
1299 register struct buf *bp;
1300 struct buf *nbp;
1301 int s, anyfreed;
1302 int trunclbn;
1303
1304 /*
1305 * Round up to the *next* lbn.
1306 */
1307 trunclbn = (length + blksize - 1) / blksize;
1308
1309 s = splbio();
1310 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1311restart:
1312 VI_LOCK(vp);
1313 anyfreed = 1;
1314 for (;anyfreed;) {
1315 anyfreed = 0;
1316 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1317 nbp = TAILQ_NEXT(bp, b_vnbufs);
1318 if (bp->b_lblkno >= trunclbn) {
1319 if (BUF_LOCK(bp,
1320 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1321 VI_MTX(vp)) == ENOLCK)
1322 goto restart;
1323
1324 bremfree(bp);
1325 bp->b_flags |= (B_INVAL | B_RELBUF);
1326 bp->b_flags &= ~B_ASYNC;
1327 brelse(bp);
1328 anyfreed = 1;
1329
1330 if (nbp &&
1331 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1332 (nbp->b_vp != vp) ||
1333 (nbp->b_flags & B_DELWRI))) {
1334 goto restart;
1335 }
1336 VI_LOCK(vp);
1337 }
1338 }
1339
1340 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1341 nbp = TAILQ_NEXT(bp, b_vnbufs);
1342 if (bp->b_lblkno >= trunclbn) {
1343 if (BUF_LOCK(bp,
1344 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1345 VI_MTX(vp)) == ENOLCK)
1346 goto restart;
1347 bremfree(bp);
1348 bp->b_flags |= (B_INVAL | B_RELBUF);
1349 bp->b_flags &= ~B_ASYNC;
1350 brelse(bp);
1351 anyfreed = 1;
1352 if (nbp &&
1353 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1354 (nbp->b_vp != vp) ||
1355 (nbp->b_flags & B_DELWRI) == 0)) {
1356 goto restart;
1357 }
1358 VI_LOCK(vp);
1359 }
1360 }
1361 }
1362
1363 if (length > 0) {
1364restartsync:
1365 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1366 nbp = TAILQ_NEXT(bp, b_vnbufs);
1367 if (bp->b_lblkno > 0)
1368 continue;
1369 /*
1370 * Since we hold the vnode lock this should only
1371 * fail if we're racing with the buf daemon.
1372 */
1373 if (BUF_LOCK(bp,
1374 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1375 VI_MTX(vp)) == ENOLCK) {
1376 goto restart;
1377 }
1378 KASSERT((bp->b_flags & B_DELWRI),
1379 ("buf(%p) on dirty queue without DELWRI.", bp));
1380
1381 bremfree(bp);
1382 bawrite(bp);
1383 VI_LOCK(vp);
1384 goto restartsync;
1385 }
1386 }
1387
1388 while (vp->v_numoutput > 0) {
1389 vp->v_iflag |= VI_BWAIT;
1390 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1391 }
1392 VI_UNLOCK(vp);
1393 splx(s);
1394
1395 vnode_pager_setsize(vp, length);
1396
1397 return (0);
1398}
1399
1400/*
1401 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1402 * a vnode.
1403 *
1404 * NOTE: We have to deal with the special case of a background bitmap
1405 * buffer, a situation where two buffers will have the same logical
1406 * block offset. We want (1) only the foreground buffer to be accessed
1407 * in a lookup and (2) must differentiate between the foreground and
1408 * background buffer in the splay tree algorithm because the splay
1409 * tree cannot normally handle multiple entities with the same 'index'.
1410 * We accomplish this by adding differentiating flags to the splay tree's
1411 * numerical domain.
1412 */
1413static
1414struct buf *
1415buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1416{
1417 struct buf dummy;
1418 struct buf *lefttreemax, *righttreemin, *y;
1419
1420 if (root == NULL)
1421 return (NULL);
1422 lefttreemax = righttreemin = &dummy;
1423 for (;;) {
1424 if (lblkno < root->b_lblkno ||
1425 (lblkno == root->b_lblkno &&
1426 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1427 if ((y = root->b_left) == NULL)
1428 break;
1429 if (lblkno < y->b_lblkno) {
1430 /* Rotate right. */
1431 root->b_left = y->b_right;
1432 y->b_right = root;
1433 root = y;
1434 if ((y = root->b_left) == NULL)
1435 break;
1436 }
1437 /* Link into the new root's right tree. */
1438 righttreemin->b_left = root;
1439 righttreemin = root;
1440 } else if (lblkno > root->b_lblkno ||
1441 (lblkno == root->b_lblkno &&
1442 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1443 if ((y = root->b_right) == NULL)
1444 break;
1445 if (lblkno > y->b_lblkno) {
1446 /* Rotate left. */
1447 root->b_right = y->b_left;
1448 y->b_left = root;
1449 root = y;
1450 if ((y = root->b_right) == NULL)
1451 break;
1452 }
1453 /* Link into the new root's left tree. */
1454 lefttreemax->b_right = root;
1455 lefttreemax = root;
1456 } else {
1457 break;
1458 }
1459 root = y;
1460 }
1461 /* Assemble the new root. */
1462 lefttreemax->b_right = root->b_left;
1463 righttreemin->b_left = root->b_right;
1464 root->b_left = dummy.b_right;
1465 root->b_right = dummy.b_left;
1466 return (root);
1467}
1468
1469static
1470void
1471buf_vlist_remove(struct buf *bp)
1472{
1473 struct vnode *vp = bp->b_vp;
1474 struct buf *root;
1475
1476 ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1477 if (bp->b_xflags & BX_VNDIRTY) {
1478 if (bp != vp->v_dirtyblkroot) {
1479 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1480 KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1481 }
1482 if (bp->b_left == NULL) {
1483 root = bp->b_right;
1484 } else {
1485 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1486 root->b_right = bp->b_right;
1487 }
1488 vp->v_dirtyblkroot = root;
1489 TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1490 vp->v_dirtybufcnt--;
1491 } else {
1492 /* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1493 if (bp != vp->v_cleanblkroot) {
1494 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1495 KASSERT(root == bp, ("splay lookup failed during clean remove"));
1496 }
1497 if (bp->b_left == NULL) {
1498 root = bp->b_right;
1499 } else {
1500 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1501 root->b_right = bp->b_right;
1502 }
1503 vp->v_cleanblkroot = root;
1504 TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1505 vp->v_cleanbufcnt--;
1506 }
1507 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1508}
1509
1510/*
1511 * Add the buffer to the sorted clean or dirty block list using a
1512 * splay tree algorithm.
1513 *
1514 * NOTE: xflags is passed as a constant, optimizing this inline function!
1515 */
1516static
1517void
1518buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1519{
1520 struct buf *root;
1521
1522 ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1523 bp->b_xflags |= xflags;
1524 if (xflags & BX_VNDIRTY) {
1525 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1526 if (root == NULL) {
1527 bp->b_left = NULL;
1528 bp->b_right = NULL;
1529 TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1530 } else if (bp->b_lblkno < root->b_lblkno ||
1531 (bp->b_lblkno == root->b_lblkno &&
1532 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1533 bp->b_left = root->b_left;
1534 bp->b_right = root;
1535 root->b_left = NULL;
1536 TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1537 } else {
1538 bp->b_right = root->b_right;
1539 bp->b_left = root;
1540 root->b_right = NULL;
1541 TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1542 root, bp, b_vnbufs);
1543 }
1544 vp->v_dirtybufcnt++;
1545 vp->v_dirtyblkroot = bp;
1546 } else {
1547 /* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1548 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1549 if (root == NULL) {
1550 bp->b_left = NULL;
1551 bp->b_right = NULL;
1552 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1553 } else if (bp->b_lblkno < root->b_lblkno ||
1554 (bp->b_lblkno == root->b_lblkno &&
1555 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1556 bp->b_left = root->b_left;
1557 bp->b_right = root;
1558 root->b_left = NULL;
1559 TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1560 } else {
1561 bp->b_right = root->b_right;
1562 bp->b_left = root;
1563 root->b_right = NULL;
1564 TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1565 root, bp, b_vnbufs);
1566 }
1567 vp->v_cleanbufcnt++;
1568 vp->v_cleanblkroot = bp;
1569 }
1570}
1571
1572/*
1573 * Lookup a buffer using the splay tree. Note that we specifically avoid
1574 * shadow buffers used in background bitmap writes.
1575 *
1576 * This code isn't quite efficient as it could be because we are maintaining
1577 * two sorted lists and do not know which list the block resides in.
1578 *
1579 * During a "make buildworld" the desired buffer is found at one of
1580 * the roots more than 60% of the time. Thus, checking both roots
1581 * before performing either splay eliminates unnecessary splays on the
1582 * first tree splayed.
1583 */
1584struct buf *
1585gbincore(struct vnode *vp, daddr_t lblkno)
1586{
1587 struct buf *bp;
1588
1589 GIANT_REQUIRED;
1590
1591 ASSERT_VI_LOCKED(vp, "gbincore");
1592 if ((bp = vp->v_cleanblkroot) != NULL &&
1593 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1594 return (bp);
1595 if ((bp = vp->v_dirtyblkroot) != NULL &&
1596 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1597 return (bp);
1598 if ((bp = vp->v_cleanblkroot) != NULL) {
1599 vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
1600 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1601 return (bp);
1602 }
1603 if ((bp = vp->v_dirtyblkroot) != NULL) {
1604 vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
1605 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1606 return (bp);
1607 }
1608 return (NULL);
1609}
1610
1611/*
1612 * Associate a buffer with a vnode.
1613 */
1614void
1615bgetvp(vp, bp)
1616 register struct vnode *vp;
1617 register struct buf *bp;
1618{
1619 int s;
1620
1621 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1622
1623 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1624 ("bgetvp: bp already attached! %p", bp));
1625
1626 ASSERT_VI_LOCKED(vp, "bgetvp");
1627 vholdl(vp);
1628 bp->b_vp = vp;
1629 bp->b_dev = vn_todev(vp);
1630 /*
1631 * Insert onto list for new vnode.
1632 */
1633 s = splbio();
1634 buf_vlist_add(bp, vp, BX_VNCLEAN);
1635 splx(s);
1636}
1637
1638/*
1639 * Disassociate a buffer from a vnode.
1640 */
1641void
1642brelvp(bp)
1643 register struct buf *bp;
1644{
1645 struct vnode *vp;
1646 int s;
1647
1648 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1649
1650 /*
1651 * Delete from old vnode list, if on one.
1652 */
1653 vp = bp->b_vp;
1654 s = splbio();
1655 VI_LOCK(vp);
1656 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1657 buf_vlist_remove(bp);
1658 if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1659 vp->v_iflag &= ~VI_ONWORKLST;
1660 mtx_lock(&sync_mtx);
1661 LIST_REMOVE(vp, v_synclist);
1662 mtx_unlock(&sync_mtx);
1663 }
1664 vdropl(vp);
1665 VI_UNLOCK(vp);
1666 bp->b_vp = (struct vnode *) 0;
1667 if (bp->b_object)
1668 bp->b_object = NULL;
1669 splx(s);
1670}
1671
1672/*
1673 * Add an item to the syncer work queue.
1674 */
1675static void
1676vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1677{
1678 int s, slot;
1679
1680 s = splbio();
1681 ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1682
1683 mtx_lock(&sync_mtx);
1684 if (vp->v_iflag & VI_ONWORKLST)
1685 LIST_REMOVE(vp, v_synclist);
1686 else
1687 vp->v_iflag |= VI_ONWORKLST;
1688
1689 if (delay > syncer_maxdelay - 2)
1690 delay = syncer_maxdelay - 2;
1691 slot = (syncer_delayno + delay) & syncer_mask;
1692
1693 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1694 mtx_unlock(&sync_mtx);
1695
1696 splx(s);
1697}
1698
1699struct proc *updateproc;
1700static void sched_sync(void);
1701static struct kproc_desc up_kp = {
1702 "syncer",
1703 sched_sync,
1704 &updateproc
1705};
1706SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1707
1708/*
1709 * System filesystem synchronizer daemon.
1710 */
1711static void
1712sched_sync(void)
1713{
1714 struct synclist *slp;
1715 struct vnode *vp;
1716 struct mount *mp;
1717 long starttime;
1718 int s;
1719 struct thread *td = FIRST_THREAD_IN_PROC(updateproc); /* XXXKSE */
1720
1721 mtx_lock(&Giant);
1722
1723 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1724 SHUTDOWN_PRI_LAST);
1725
1726 for (;;) {
1727 kthread_suspend_check(td->td_proc);
1728
1729 starttime = time_second;
1730
1731 /*
1732 * Push files whose dirty time has expired. Be careful
1733 * of interrupt race on slp queue.
1734 */
1735 s = splbio();
1736 mtx_lock(&sync_mtx);
1737 slp = &syncer_workitem_pending[syncer_delayno];
1738 syncer_delayno += 1;
1739 if (syncer_delayno == syncer_maxdelay)
1740 syncer_delayno = 0;
1741 splx(s);
1742
1743 while ((vp = LIST_FIRST(slp)) != NULL) {
1744 mtx_unlock(&sync_mtx);
1745 if (VOP_ISLOCKED(vp, NULL) == 0 &&
1746 vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1747 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1748 (void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1749 VOP_UNLOCK(vp, 0, td);
1750 vn_finished_write(mp);
1751 }
1752 s = splbio();
1753 mtx_lock(&sync_mtx);
1754 if (LIST_FIRST(slp) == vp) {
1755 mtx_unlock(&sync_mtx);
1756 /*
1757 * Note: VFS vnodes can remain on the
1758 * worklist too with no dirty blocks, but
1759 * since sync_fsync() moves it to a different
1760 * slot we are safe.
1761 */
1762 VI_LOCK(vp);
1763 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1764 !vn_isdisk(vp, NULL)) {
1765 panic("sched_sync: fsync failed "
1766 "vp %p tag %s", vp, vp->v_tag);
1767 }
1768 /*
1769 * Put us back on the worklist. The worklist
1770 * routine will remove us from our current
1771 * position and then add us back in at a later
1772 * position.
1773 */
1774 vn_syncer_add_to_worklist(vp, syncdelay);
1775 VI_UNLOCK(vp);
1776 mtx_lock(&sync_mtx);
1777 }
1778 splx(s);
1779 }
1780 mtx_unlock(&sync_mtx);
1781
1782 /*
1783 * Do soft update processing.
1784 */
1785 if (softdep_process_worklist_hook != NULL)
1786 (*softdep_process_worklist_hook)(NULL);
1787
1788 /*
1789 * The variable rushjob allows the kernel to speed up the
1790 * processing of the filesystem syncer process. A rushjob
1791 * value of N tells the filesystem syncer to process the next
1792 * N seconds worth of work on its queue ASAP. Currently rushjob
1793 * is used by the soft update code to speed up the filesystem
1794 * syncer process when the incore state is getting so far
1795 * ahead of the disk that the kernel memory pool is being
1796 * threatened with exhaustion.
1797 */
1798 mtx_lock(&sync_mtx);
1799 if (rushjob > 0) {
1800 rushjob -= 1;
1801 mtx_unlock(&sync_mtx);
1802 continue;
1803 }
1804 mtx_unlock(&sync_mtx);
1805 /*
1806 * If it has taken us less than a second to process the
1807 * current work, then wait. Otherwise start right over
1808 * again. We can still lose time if any single round
1809 * takes more than two seconds, but it does not really
1810 * matter as we are just trying to generally pace the
1811 * filesystem activity.
1812 */
1813 if (time_second == starttime)
1814 tsleep(&lbolt, PPAUSE, "syncer", 0);
1815 }
1816}
1817
1818/*
1819 * Request the syncer daemon to speed up its work.
1820 * We never push it to speed up more than half of its
1821 * normal turn time, otherwise it could take over the cpu.
1822 * XXXKSE only one update?
1823 */
1824int
1825speedup_syncer()
1826{
1827 struct thread *td;
1828 int ret = 0;
1829
1830 td = FIRST_THREAD_IN_PROC(updateproc);
1831 mtx_lock_spin(&sched_lock);
1832 if (td->td_wchan == &lbolt) {
1833 unsleep(td);
1834 TD_CLR_SLEEPING(td);
1835 setrunnable(td);
1836 }
1837 mtx_unlock_spin(&sched_lock);
1838 mtx_lock(&sync_mtx);
1839 if (rushjob < syncdelay / 2) {
1840 rushjob += 1;
1841 stat_rush_requests += 1;
1842 ret = 1;
1843 }
1844 mtx_unlock(&sync_mtx);
1845 return (ret);
1846}
1847
1848/*
1849 * Associate a p-buffer with a vnode.
1850 *
1851 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1852 * with the buffer. i.e. the bp has not been linked into the vnode or
1853 * ref-counted.
1854 */
1855void
1856pbgetvp(vp, bp)
1857 register struct vnode *vp;
1858 register struct buf *bp;
1859{
1860
1861 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1862
1863 bp->b_vp = vp;
1864 bp->b_flags |= B_PAGING;
1865 bp->b_dev = vn_todev(vp);
1866}
1867
1868/*
1869 * Disassociate a p-buffer from a vnode.
1870 */
1871void
1872pbrelvp(bp)
1873 register struct buf *bp;
1874{
1875
1876 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1877
1878 /* XXX REMOVE ME */
1879 VI_LOCK(bp->b_vp);
1880 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1881 panic(
1882 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1883 bp,
1884 (int)bp->b_flags
1885 );
1886 }
1887 VI_UNLOCK(bp->b_vp);
1888 bp->b_vp = (struct vnode *) 0;
1889 bp->b_flags &= ~B_PAGING;
1890}
1891
1892/*
1893 * Reassign a buffer from one vnode to another.
1894 * Used to assign file specific control information
1895 * (indirect blocks) to the vnode to which they belong.
1896 */
1897void
1898reassignbuf(bp, newvp)
1899 register struct buf *bp;
1900 register struct vnode *newvp;
1901{
1902 int delay;
1903 int s;
1904
1905 if (newvp == NULL) {
1906 printf("reassignbuf: NULL");
1907 return;
1908 }
1909 ++reassignbufcalls;
1910
1911 /*
1912 * B_PAGING flagged buffers cannot be reassigned because their vp
1913 * is not fully linked in.
1914 */
1915 if (bp->b_flags & B_PAGING)
1916 panic("cannot reassign paging buffer");
1917
1918 s = splbio();
1919 /*
1920 * Delete from old vnode list, if on one.
1921 */
1922 VI_LOCK(bp->b_vp);
1923 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1924 buf_vlist_remove(bp);
1925 if (bp->b_vp != newvp) {
1926 vdropl(bp->b_vp);
1927 bp->b_vp = NULL; /* for clarification */
1928 }
1929 }
1930 VI_UNLOCK(bp->b_vp);
1931 /*
1932 * If dirty, put on list of dirty buffers; otherwise insert onto list
1933 * of clean buffers.
1934 */
1935 VI_LOCK(newvp);
1936 if (bp->b_flags & B_DELWRI) {
1937 if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1938 switch (newvp->v_type) {
1939 case VDIR:
1940 delay = dirdelay;
1941 break;
1942 case VCHR:
1943 if (newvp->v_rdev->si_mountpoint != NULL) {
1944 delay = metadelay;
1945 break;
1946 }
1947 /* FALLTHROUGH */
1948 default:
1949 delay = filedelay;
1950 }
1951 vn_syncer_add_to_worklist(newvp, delay);
1952 }
1953 buf_vlist_add(bp, newvp, BX_VNDIRTY);
1954 } else {
1955 buf_vlist_add(bp, newvp, BX_VNCLEAN);
1956
1957 if ((newvp->v_iflag & VI_ONWORKLST) &&
1958 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1959 mtx_lock(&sync_mtx);
1960 LIST_REMOVE(newvp, v_synclist);
1961 mtx_unlock(&sync_mtx);
1962 newvp->v_iflag &= ~VI_ONWORKLST;
1963 }
1964 }
1965 if (bp->b_vp != newvp) {
1966 bp->b_vp = newvp;
1967 vholdl(bp->b_vp);
1968 }
1969 VI_UNLOCK(newvp);
1970 splx(s);
1971}
1972
1973/*
1974 * Create a vnode for a device.
1975 * Used for mounting the root filesystem.
1976 */
1977int
1978bdevvp(dev, vpp)
1979 dev_t dev;
1980 struct vnode **vpp;
1981{
1982 register struct vnode *vp;
1983 struct vnode *nvp;
1984 int error;
1985
1986 if (dev == NODEV) {
1987 *vpp = NULLVP;
1988 return (ENXIO);
1989 }
1990 if (vfinddev(dev, VCHR, vpp))
1991 return (0);
1992 error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1993 if (error) {
1994 *vpp = NULLVP;
1995 return (error);
1996 }
1997 vp = nvp;
1998 vp->v_type = VCHR;
1999 addalias(vp, dev);
2000 *vpp = vp;
2001 return (0);
2002}
2003
2004static void
2005v_incr_usecount(struct vnode *vp, int delta)
2006{
2007 vp->v_usecount += delta;
2008 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2009 mtx_lock(&spechash_mtx);
2010 vp->v_rdev->si_usecount += delta;
2011 mtx_unlock(&spechash_mtx);
2012 }
2013}
2014
2015/*
2016 * Add vnode to the alias list hung off the dev_t.
2017 *
2018 * The reason for this gunk is that multiple vnodes can reference
2019 * the same physical device, so checking vp->v_usecount to see
2020 * how many users there are is inadequate; the v_usecount for
2021 * the vnodes need to be accumulated. vcount() does that.
2022 */
2023struct vnode *
2024addaliasu(nvp, nvp_rdev)
2025 struct vnode *nvp;
2026 udev_t nvp_rdev;
2027{
2028 struct vnode *ovp;
2029 vop_t **ops;
2030 dev_t dev;
2031
2032 if (nvp->v_type == VBLK)
2033 return (nvp);
2034 if (nvp->v_type != VCHR)
2035 panic("addaliasu on non-special vnode");
2036 dev = udev2dev(nvp_rdev, 0);
2037 /*
2038 * Check to see if we have a bdevvp vnode with no associated
2039 * filesystem. If so, we want to associate the filesystem of
2040 * the new newly instigated vnode with the bdevvp vnode and
2041 * discard the newly created vnode rather than leaving the
2042 * bdevvp vnode lying around with no associated filesystem.
2043 */
2044 if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
2045 addalias(nvp, dev);
2046 return (nvp);
2047 }
2048 /*
2049 * Discard unneeded vnode, but save its node specific data.
2050 * Note that if there is a lock, it is carried over in the
2051 * node specific data to the replacement vnode.
2052 */
2053 vref(ovp);
2054 ovp->v_data = nvp->v_data;
2055 ovp->v_tag = nvp->v_tag;
2056 nvp->v_data = NULL;
2057 lockdestroy(ovp->v_vnlock);
2058 lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2059 nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2060 ops = ovp->v_op;
2061 ovp->v_op = nvp->v_op;
2062 if (VOP_ISLOCKED(nvp, curthread)) {
2063 VOP_UNLOCK(nvp, 0, curthread);
2064 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2065 }
2066 nvp->v_op = ops;
2067 insmntque(ovp, nvp->v_mount);
2068 vrele(nvp);
2069 vgone(nvp);
2070 return (ovp);
2071}
2072
2073/* This is a local helper function that do the same as addaliasu, but for a
2074 * dev_t instead of an udev_t. */
2075static void
2076addalias(nvp, dev)
2077 struct vnode *nvp;
2078 dev_t dev;
2079{
2080
2081 KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2082 nvp->v_rdev = dev;
2083 VI_LOCK(nvp);
2084 mtx_lock(&spechash_mtx);
2085 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2086 dev->si_usecount += nvp->v_usecount;
2087 mtx_unlock(&spechash_mtx);
2088 VI_UNLOCK(nvp);
2089}
2090
2091/*
2092 * Grab a particular vnode from the free list, increment its
2093 * reference count and lock it. The vnode lock bit is set if the
2094 * vnode is being eliminated in vgone. The process is awakened
2095 * when the transition is completed, and an error returned to
2096 * indicate that the vnode is no longer usable (possibly having
2097 * been changed to a new filesystem type).
2098 */
2099int
2100vget(vp, flags, td)
2101 register struct vnode *vp;
2102 int flags;
2103 struct thread *td;
2104{
2105 int error;
2106
2107 /*
2108 * If the vnode is in the process of being cleaned out for
2109 * another use, we wait for the cleaning to finish and then
2110 * return failure. Cleaning is determined by checking that
2111 * the VI_XLOCK flag is set.
2112 */
2113 if ((flags & LK_INTERLOCK) == 0)
2114 VI_LOCK(vp);
2115 if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2116 vp->v_iflag |= VI_XWANT;
2117 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2118 return (ENOENT);
2119 }
2120
2121 v_incr_usecount(vp, 1);
2122
2123 if (VSHOULDBUSY(vp))
2124 vbusy(vp);
2125 if (flags & LK_TYPE_MASK) {
2126 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2127 /*
2128 * must expand vrele here because we do not want
2129 * to call VOP_INACTIVE if the reference count
2130 * drops back to zero since it was never really
2131 * active. We must remove it from the free list
2132 * before sleeping so that multiple processes do
2133 * not try to recycle it.
2134 */
2135 VI_LOCK(vp);
2136 v_incr_usecount(vp, -1);
2137 if (VSHOULDFREE(vp))
2138 vfree(vp);
2139 else
2140 vlruvp(vp);
2141 VI_UNLOCK(vp);
2142 }
2143 return (error);
2144 }
2145 VI_UNLOCK(vp);
2146 return (0);
2147}
2148
2149/*
2150 * Increase the reference count of a vnode.
2151 */
2152void
2153vref(struct vnode *vp)
2154{
2155 VI_LOCK(vp);
2156 v_incr_usecount(vp, 1);
2157 VI_UNLOCK(vp);
2158}
2159
2160/*
2161 * Return reference count of a vnode.
2162 *
2163 * The results of this call are only guaranteed when some mechanism other
2164 * than the VI lock is used to stop other processes from gaining references
2165 * to the vnode. This may be the case if the caller holds the only reference.
2166 * This is also useful when stale data is acceptable as race conditions may
2167 * be accounted for by some other means.
2168 */
2169int
2170vrefcnt(struct vnode *vp)
2171{
2172 int usecnt;
2173
2174 VI_LOCK(vp);
2175 usecnt = vp->v_usecount;
2176 VI_UNLOCK(vp);
2177
2178 return (usecnt);
2179}
2180
2181
2182/*
2183 * Vnode put/release.
2184 * If count drops to zero, call inactive routine and return to freelist.
2185 */
2186void
2187vrele(vp)
2188 struct vnode *vp;
2189{
2190 struct thread *td = curthread; /* XXX */
2191
2192 KASSERT(vp != NULL, ("vrele: null vp"));
2193
2194 VI_LOCK(vp);
2195
2196 /* Skip this v_writecount check if we're going to panic below. */
2197 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2198 ("vrele: missed vn_close"));
2199
2200 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2201 vp->v_usecount == 1)) {
2202 v_incr_usecount(vp, -1);
2203 VI_UNLOCK(vp);
2204
2205 return;
2206 }
2207
2208 if (vp->v_usecount == 1) {
2209 v_incr_usecount(vp, -1);
2210 /*
2211 * We must call VOP_INACTIVE with the node locked. Mark
2212 * as VI_DOINGINACT to avoid recursion.
2213 */
2214 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2215 VI_LOCK(vp);
2216 vp->v_iflag |= VI_DOINGINACT;
2217 VI_UNLOCK(vp);
2218 VOP_INACTIVE(vp, td);
2219 VI_LOCK(vp);
2220 KASSERT(vp->v_iflag & VI_DOINGINACT,
2221 ("vrele: lost VI_DOINGINACT"));
2222 vp->v_iflag &= ~VI_DOINGINACT;
2223 VI_UNLOCK(vp);
2224 }
2225 VI_LOCK(vp);
2226 if (VSHOULDFREE(vp))
2227 vfree(vp);
2228 else
2229 vlruvp(vp);
2230 VI_UNLOCK(vp);
2231
2232 } else {
2233#ifdef DIAGNOSTIC
2234 vprint("vrele: negative ref count", vp);
2235#endif
2236 VI_UNLOCK(vp);
2237 panic("vrele: negative ref cnt");
2238 }
2239}
2240
2241/*
2242 * Release an already locked vnode. This give the same effects as
2243 * unlock+vrele(), but takes less time and avoids releasing and
2244 * re-aquiring the lock (as vrele() aquires the lock internally.)
2245 */
2246void
2247vput(vp)
2248 struct vnode *vp;
2249{
2250 struct thread *td = curthread; /* XXX */
2251
2252 GIANT_REQUIRED;
2253
2254 KASSERT(vp != NULL, ("vput: null vp"));
2255 VI_LOCK(vp);
2256 /* Skip this v_writecount check if we're going to panic below. */
2257 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2258 ("vput: missed vn_close"));
2259
2260 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2261 vp->v_usecount == 1)) {
2262 v_incr_usecount(vp, -1);
2263 VOP_UNLOCK(vp, LK_INTERLOCK, td);
2264 return;
2265 }
2266
2267 if (vp->v_usecount == 1) {
2268 v_incr_usecount(vp, -1);
2269 /*
2270 * We must call VOP_INACTIVE with the node locked, so
2271 * we just need to release the vnode mutex. Mark as
2272 * as VI_DOINGINACT to avoid recursion.
2273 */
2274 vp->v_iflag |= VI_DOINGINACT;
2275 VI_UNLOCK(vp);
2276 VOP_INACTIVE(vp, td);
2277 VI_LOCK(vp);
2278 KASSERT(vp->v_iflag & VI_DOINGINACT,
2279 ("vput: lost VI_DOINGINACT"));
2280 vp->v_iflag &= ~VI_DOINGINACT;
2281 if (VSHOULDFREE(vp))
2282 vfree(vp);
2283 else
2284 vlruvp(vp);
2285 VI_UNLOCK(vp);
2286
2287 } else {
2288#ifdef DIAGNOSTIC
2289 vprint("vput: negative ref count", vp);
2290#endif
2291 panic("vput: negative ref cnt");
2292 }
2293}
2294
2295/*
2296 * Somebody doesn't want the vnode recycled.
2297 */
2298void
2299vhold(struct vnode *vp)
2300{
2301 VI_LOCK(vp);
2302 vholdl(vp);
2303 VI_UNLOCK(vp);
2304}
2305
2306void
2307vholdl(vp)
2308 register struct vnode *vp;
2309{
2310 int s;
2311
2312 s = splbio();
2313 vp->v_holdcnt++;
2314 if (VSHOULDBUSY(vp))
2315 vbusy(vp);
2316 splx(s);
2317}
2318
2319/*
2320 * Note that there is one less who cares about this vnode. vdrop() is the
2321 * opposite of vhold().
2322 */
2323void
2324vdrop(struct vnode *vp)
2325{
2326 VI_LOCK(vp);
2327 vdropl(vp);
2328 VI_UNLOCK(vp);
2329}
2330
2331void
2332vdropl(vp)
2333 register struct vnode *vp;
2334{
2335 int s;
2336
2337 s = splbio();
2338 if (vp->v_holdcnt <= 0)
2339 panic("vdrop: holdcnt");
2340 vp->v_holdcnt--;
2341 if (VSHOULDFREE(vp))
2342 vfree(vp);
2343 else
2344 vlruvp(vp);
2345 splx(s);
2346}
2347
2348/*
2349 * Remove any vnodes in the vnode table belonging to mount point mp.
2350 *
2351 * If FORCECLOSE is not specified, there should not be any active ones,
2352 * return error if any are found (nb: this is a user error, not a
2353 * system error). If FORCECLOSE is specified, detach any active vnodes
2354 * that are found.
2355 *
2356 * If WRITECLOSE is set, only flush out regular file vnodes open for
2357 * writing.
2358 *
2359 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2360 *
2361 * `rootrefs' specifies the base reference count for the root vnode
2362 * of this filesystem. The root vnode is considered busy if its
2363 * v_usecount exceeds this value. On a successful return, vflush()
2364 * will call vrele() on the root vnode exactly rootrefs times.
2365 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2366 * be zero.
2367 */
2368#ifdef DIAGNOSTIC
2369static int busyprt = 0; /* print out busy vnodes */
2370SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2371#endif
2372
2373int
2374vflush(mp, rootrefs, flags)
2375 struct mount *mp;
2376 int rootrefs;
2377 int flags;
2378{
2379 struct thread *td = curthread; /* XXX */
2380 struct vnode *vp, *nvp, *rootvp = NULL;
2381 struct vattr vattr;
2382 int busy = 0, error;
2383
2384 if (rootrefs > 0) {
2385 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2386 ("vflush: bad args"));
2387 /*
2388 * Get the filesystem root vnode. We can vput() it
2389 * immediately, since with rootrefs > 0, it won't go away.
2390 */
2391 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2392 return (error);
2393 vput(rootvp);
2394
2395 }
2396 mtx_lock(&mntvnode_mtx);
2397loop:
2398 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2399 /*
2400 * Make sure this vnode wasn't reclaimed in getnewvnode().
2401 * Start over if it has (it won't be on the list anymore).
2402 */
2403 if (vp->v_mount != mp)
2404 goto loop;
2405 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2406
2407 VI_LOCK(vp);
2408 mtx_unlock(&mntvnode_mtx);
2409 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
2410 /*
2411 * This vnode could have been reclaimed while we were
2412 * waiting for the lock since we are not holding a
2413 * reference.
2414 * Start over if the vnode was reclaimed.
2415 */
2416 if (vp->v_mount != mp) {
2417 VOP_UNLOCK(vp, 0, td);
2418 mtx_lock(&mntvnode_mtx);
2419 goto loop;
2420 }
2421 /*
2422 * Skip over a vnodes marked VV_SYSTEM.
2423 */
2424 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2425 VOP_UNLOCK(vp, 0, td);
2426 mtx_lock(&mntvnode_mtx);
2427 continue;
2428 }
2429 /*
2430 * If WRITECLOSE is set, flush out unlinked but still open
2431 * files (even if open only for reading) and regular file
2432 * vnodes open for writing.
2433 */
2434 if (flags & WRITECLOSE) {
2435 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2436 VI_LOCK(vp);
2437
2438 if ((vp->v_type == VNON ||
2439 (error == 0 && vattr.va_nlink > 0)) &&
2440 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2441 VOP_UNLOCK(vp, LK_INTERLOCK, td);
2442 mtx_lock(&mntvnode_mtx);
2443 continue;
2444 }
2445 } else
2446 VI_LOCK(vp);
2447
2448 VOP_UNLOCK(vp, 0, td);
2449
2450 /*
2451 * With v_usecount == 0, all we need to do is clear out the
2452 * vnode data structures and we are done.
2453 */
2454 if (vp->v_usecount == 0) {
2455 vgonel(vp, td);
2456 mtx_lock(&mntvnode_mtx);
2457 continue;
2458 }
2459
2460 /*
2461 * If FORCECLOSE is set, forcibly close the vnode. For block
2462 * or character devices, revert to an anonymous device. For
2463 * all other files, just kill them.
2464 */
2465 if (flags & FORCECLOSE) {
2466 if (vp->v_type != VCHR) {
2467 vgonel(vp, td);
2468 } else {
2469 vclean(vp, 0, td);
2470 VI_UNLOCK(vp);
2471 vp->v_op = spec_vnodeop_p;
2472 insmntque(vp, (struct mount *) 0);
2473 }
2474 mtx_lock(&mntvnode_mtx);
2475 continue;
2476 }
2477#ifdef DIAGNOSTIC
2478 if (busyprt)
2479 vprint("vflush: busy vnode", vp);
2480#endif
2481 VI_UNLOCK(vp);
2482 mtx_lock(&mntvnode_mtx);
2483 busy++;
2484 }
2485 mtx_unlock(&mntvnode_mtx);
2486 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2487 /*
2488 * If just the root vnode is busy, and if its refcount
2489 * is equal to `rootrefs', then go ahead and kill it.
2490 */
2491 VI_LOCK(rootvp);
2492 KASSERT(busy > 0, ("vflush: not busy"));
2493 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2494 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2495 vgonel(rootvp, td);
2496 busy = 0;
2497 } else
2498 VI_UNLOCK(rootvp);
2499 }
2500 if (busy)
2501 return (EBUSY);
2502 for (; rootrefs > 0; rootrefs--)
2503 vrele(rootvp);
2504 return (0);
2505}
2506
2507/*
2508 * This moves a now (likely recyclable) vnode to the end of the
2509 * mountlist. XXX However, it is temporarily disabled until we
2510 * can clean up ffs_sync() and friends, which have loop restart
2511 * conditions which this code causes to operate O(N^2).
2512 */
2513static void
2514vlruvp(struct vnode *vp)
2515{
2516#if 0
2517 struct mount *mp;
2518
2519 if ((mp = vp->v_mount) != NULL) {
2520 mtx_lock(&mntvnode_mtx);
2521 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2522 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2523 mtx_unlock(&mntvnode_mtx);
2524 }
2525#endif
2526}
2527
2528/*
2529 * Disassociate the underlying filesystem from a vnode.
2530 */
2531static void
2532vclean(vp, flags, td)
2533 struct vnode *vp;
2534 int flags;
2535 struct thread *td;
2536{
2537 int active;
2538
2539 ASSERT_VI_LOCKED(vp, "vclean");
2540 /*
2541 * Check to see if the vnode is in use. If so we have to reference it
2542 * before we clean it out so that its count cannot fall to zero and
2543 * generate a race against ourselves to recycle it.
2544 */
2545 if ((active = vp->v_usecount))
2546 v_incr_usecount(vp, 1);
2547
2548 /*
2549 * Prevent the vnode from being recycled or brought into use while we
2550 * clean it out.
2551 */
2552 if (vp->v_iflag & VI_XLOCK)
2553 panic("vclean: deadlock");
2554 vp->v_iflag |= VI_XLOCK;
2555 vp->v_vxproc = curthread;
2556 /*
2557 * Even if the count is zero, the VOP_INACTIVE routine may still
2558 * have the object locked while it cleans it out. The VOP_LOCK
2559 * ensures that the VOP_INACTIVE routine is done with its work.
2560 * For active vnodes, it ensures that no other activity can
2561 * occur while the underlying object is being cleaned out.
2562 */
2563 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2564
2565 /*
2566 * Clean out any buffers associated with the vnode.
2567 * If the flush fails, just toss the buffers.
2568 */
2569 if (flags & DOCLOSE) {
2570 struct buf *bp;
2571 VI_LOCK(vp);
2572 bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2573 VI_UNLOCK(vp);
2574 if (bp != NULL)
2575 (void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2576 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2577 vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2578 }
2579
2580 VOP_DESTROYVOBJECT(vp);
2581
2582 /*
2583 * Any other processes trying to obtain this lock must first
2584 * wait for VXLOCK to clear, then call the new lock operation.
2585 */
2586 VOP_UNLOCK(vp, 0, td);
2587
2588 /*
2589 * If purging an active vnode, it must be closed and
2590 * deactivated before being reclaimed. Note that the
2591 * VOP_INACTIVE will unlock the vnode.
2592 */
2593 if (active) {
2594 if (flags & DOCLOSE)
2595 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2596 VI_LOCK(vp);
2597 if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2598 vp->v_iflag |= VI_DOINGINACT;
2599 VI_UNLOCK(vp);
2600 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2601 panic("vclean: cannot relock.");
2602 VOP_INACTIVE(vp, td);
2603 VI_LOCK(vp);
2604 KASSERT(vp->v_iflag & VI_DOINGINACT,
2605 ("vclean: lost VI_DOINGINACT"));
2606 vp->v_iflag &= ~VI_DOINGINACT;
2607 }
2608 VI_UNLOCK(vp);
2609 }
2610
2611 /*
2612 * Reclaim the vnode.
2613 */
2614 if (VOP_RECLAIM(vp, td))
2615 panic("vclean: cannot reclaim");
2616
2617 if (active) {
2618 /*
2619 * Inline copy of vrele() since VOP_INACTIVE
2620 * has already been called.
2621 */
2622 VI_LOCK(vp);
2623 v_incr_usecount(vp, -1);
2624 if (vp->v_usecount <= 0) {
2625#ifdef DIAGNOSTIC
2626 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2627 vprint("vclean: bad ref count", vp);
2628 panic("vclean: ref cnt");
2629 }
2630#endif
2631 vfree(vp);
2632 }
2633 VI_UNLOCK(vp);
2634 }
2635
2636 cache_purge(vp);
2637 VI_LOCK(vp);
2638 if (VSHOULDFREE(vp))
2639 vfree(vp);
2640
2641 /*
2642 * Done with purge, reset to the standard lock and
2643 * notify sleepers of the grim news.
2644 */
2645 vp->v_vnlock = &vp->v_lock;
2646 vp->v_op = dead_vnodeop_p;
2647 if (vp->v_pollinfo != NULL)
2648 vn_pollgone(vp);
2649 vp->v_tag = "none";
2650 vp->v_iflag &= ~VI_XLOCK;
2651 vp->v_vxproc = NULL;
2652 if (vp->v_iflag & VI_XWANT) {
2653 vp->v_iflag &= ~VI_XWANT;
2654 wakeup(vp);
2655 }
2656}
2657
2658/*
2659 * Eliminate all activity associated with the requested vnode
2660 * and with all vnodes aliased to the requested vnode.
2661 */
2662int
2663vop_revoke(ap)
2664 struct vop_revoke_args /* {
2665 struct vnode *a_vp;
2666 int a_flags;
2667 } */ *ap;
2668{
2669 struct vnode *vp, *vq;
2670 dev_t dev;
2671
2672 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2673 vp = ap->a_vp;
2674 KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2675
2676 VI_LOCK(vp);
2677 /*
2678 * If a vgone (or vclean) is already in progress,
2679 * wait until it is done and return.
2680 */
2681 if (vp->v_iflag & VI_XLOCK) {
2682 vp->v_iflag |= VI_XWANT;
2683 msleep(vp, VI_MTX(vp), PINOD | PDROP,
2684 "vop_revokeall", 0);
2685 return (0);
2686 }
2687 VI_UNLOCK(vp);
2688 dev = vp->v_rdev;
2689 for (;;) {
2690 mtx_lock(&spechash_mtx);
2691 vq = SLIST_FIRST(&dev->si_hlist);
2692 mtx_unlock(&spechash_mtx);
2693 if (!vq)
2694 break;
2695 vgone(vq);
2696 }
2697 return (0);
2698}
2699
2700/*
2701 * Recycle an unused vnode to the front of the free list.
2702 * Release the passed interlock if the vnode will be recycled.
2703 */
2704int
2705vrecycle(vp, inter_lkp, td)
2706 struct vnode *vp;
2707 struct mtx *inter_lkp;
2708 struct thread *td;
2709{
2710
2711 VI_LOCK(vp);
2712 if (vp->v_usecount == 0) {
2713 if (inter_lkp) {
2714 mtx_unlock(inter_lkp);
2715 }
2716 vgonel(vp, td);
2717 return (1);
2718 }
2719 VI_UNLOCK(vp);
2720 return (0);
2721}
2722
2723/*
2724 * Eliminate all activity associated with a vnode
2725 * in preparation for reuse.
2726 */
2727void
2728vgone(vp)
2729 register struct vnode *vp;
2730{
2731 struct thread *td = curthread; /* XXX */
2732
2733 VI_LOCK(vp);
2734 vgonel(vp, td);
2735}
2736
2737/*
2738 * vgone, with the vp interlock held.
2739 */
2740void
2741vgonel(vp, td)
2742 struct vnode *vp;
2743 struct thread *td;
2744{
2745 int s;
2746
2747 /*
2748 * If a vgone (or vclean) is already in progress,
2749 * wait until it is done and return.
2750 */
2751 ASSERT_VI_LOCKED(vp, "vgonel");
2752 if (vp->v_iflag & VI_XLOCK) {
2753 vp->v_iflag |= VI_XWANT;
2754 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2755 return;
2756 }
2757
2758 /*
2759 * Clean out the filesystem specific data.
2760 */
2761 vclean(vp, DOCLOSE, td);
2762 VI_UNLOCK(vp);
2763
2764 /*
2765 * Delete from old mount point vnode list, if on one.
2766 */
2767 if (vp->v_mount != NULL)
2768 insmntque(vp, (struct mount *)0);
2769 /*
2770 * If special device, remove it from special device alias list
2771 * if it is on one.
2772 */
2773 if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2774 VI_LOCK(vp);
2775 mtx_lock(&spechash_mtx);
2776 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2777 vp->v_rdev->si_usecount -= vp->v_usecount;
2778 mtx_unlock(&spechash_mtx);
2779 VI_UNLOCK(vp);
2780 vp->v_rdev = NULL;
2781 }
2782
2783 /*
2784 * If it is on the freelist and not already at the head,
2785 * move it to the head of the list. The test of the
2786 * VDOOMED flag and the reference count of zero is because
2787 * it will be removed from the free list by getnewvnode,
2788 * but will not have its reference count incremented until
2789 * after calling vgone. If the reference count were
2790 * incremented first, vgone would (incorrectly) try to
2791 * close the previous instance of the underlying object.
2792 */
2793 VI_LOCK(vp);
2794 if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2795 s = splbio();
2796 mtx_lock(&vnode_free_list_mtx);
2797 if (vp->v_iflag & VI_FREE) {
2798 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2799 } else {
2800 vp->v_iflag |= VI_FREE;
2801 freevnodes++;
2802 }
2803 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2804 mtx_unlock(&vnode_free_list_mtx);
2805 splx(s);
2806 }
2807
2808 vp->v_type = VBAD;
2809 VI_UNLOCK(vp);
2810}
2811
2812/*
2813 * Lookup a vnode by device number.
2814 */
2815int
2816vfinddev(dev, type, vpp)
2817 dev_t dev;
2818 enum vtype type;
2819 struct vnode **vpp;
2820{
2821 struct vnode *vp;
2822
2823 mtx_lock(&spechash_mtx);
2824 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2825 if (type == vp->v_type) {
2826 *vpp = vp;
2827 mtx_unlock(&spechash_mtx);
2828 return (1);
2829 }
2830 }
2831 mtx_unlock(&spechash_mtx);
2832 return (0);
2833}
2834
2835/*
2836 * Calculate the total number of references to a special device.
2837 */
2838int
2839vcount(vp)
2840 struct vnode *vp;
2841{
2842 int count;
2843
2844 mtx_lock(&spechash_mtx);
2845 count = vp->v_rdev->si_usecount;
2846 mtx_unlock(&spechash_mtx);
2847 return (count);
2848}
2849
2850/*
2851 * Same as above, but using the dev_t as argument
2852 */
2853int
2854count_dev(dev)
2855 dev_t dev;
2856{
2857 struct vnode *vp;
2858
2859 vp = SLIST_FIRST(&dev->si_hlist);
2860 if (vp == NULL)
2861 return (0);
2862 return(vcount(vp));
2863}
2864
2865/*
2866 * Print out a description of a vnode.
2867 */
2868static char *typename[] =
2869{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2870
2871void
2872vprint(label, vp)
2873 char *label;
2874 struct vnode *vp;
2875{
2876 char buf[96];
2877
2878 if (label != NULL)
2879 printf("%s: %p: ", label, (void *)vp);
2880 else
2881 printf("%p: ", (void *)vp);
2882 printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2883 vp->v_tag, typename[vp->v_type], vp->v_usecount,
2884 vp->v_writecount, vp->v_holdcnt);
2885 buf[0] = '\0';
2886 if (vp->v_vflag & VV_ROOT)
2887 strcat(buf, "|VV_ROOT");
2888 if (vp->v_vflag & VV_TEXT)
2889 strcat(buf, "|VV_TEXT");
2890 if (vp->v_vflag & VV_SYSTEM)
2891 strcat(buf, "|VV_SYSTEM");
2892 if (vp->v_iflag & VI_XLOCK)
2893 strcat(buf, "|VI_XLOCK");
2894 if (vp->v_iflag & VI_XWANT)
2895 strcat(buf, "|VI_XWANT");
2896 if (vp->v_iflag & VI_BWAIT)
2897 strcat(buf, "|VI_BWAIT");
2898 if (vp->v_iflag & VI_DOOMED)
2899 strcat(buf, "|VI_DOOMED");
2900 if (vp->v_iflag & VI_FREE)
2901 strcat(buf, "|VI_FREE");
2902 if (vp->v_vflag & VV_OBJBUF)
2903 strcat(buf, "|VV_OBJBUF");
2904 if (buf[0] != '\0')
2905 printf(" flags (%s),", &buf[1]);
2906 lockmgr_printinfo(vp->v_vnlock);
2907 printf("\n");
2908 if (vp->v_data != NULL)
2909 VOP_PRINT(vp);
2910}
2911
2912#ifdef DDB
2913#include <ddb/ddb.h>
2914/*
2915 * List all of the locked vnodes in the system.
2916 * Called when debugging the kernel.
2917 */
2918DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2919{
2920 struct mount *mp, *nmp;
2921 struct vnode *vp;
2922
2923 /*
2924 * Note: because this is DDB, we can't obey the locking semantics
2925 * for these structures, which means we could catch an inconsistent
2926 * state and dereference a nasty pointer. Not much to be done
2927 * about that.
2928 */
2929 printf("Locked vnodes\n");
2930 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2931 nmp = TAILQ_NEXT(mp, mnt_list);
2932 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2933 if (VOP_ISLOCKED(vp, NULL))
2934 vprint(NULL, vp);
2935 }
2936 nmp = TAILQ_NEXT(mp, mnt_list);
2937 }
2938}
2939#endif
2940
2941/*
2942 * Fill in a struct xvfsconf based on a struct vfsconf.
2943 */
2944static void
2945vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2946{
2947
2948 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2949 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2950 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2951 xvfsp->vfc_flags = vfsp->vfc_flags;
2952 /*
2953 * These are unused in userland, we keep them
2954 * to not break binary compatibility.
2955 */
2956 xvfsp->vfc_vfsops = NULL;
2957 xvfsp->vfc_next = NULL;
2958}
2959
2960static int
2961sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2962{
2963 struct vfsconf *vfsp;
2964 struct xvfsconf *xvfsp;
2965 int cnt, error, i;
2966
2967 cnt = 0;
2968 for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2969 cnt++;
2970 xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2971 /*
2972 * Handle the race that we will have here when struct vfsconf
2973 * will be locked down by using both cnt and checking vfc_next
2974 * against NULL to determine the end of the loop. The race will
2975 * happen because we will have to unlock before calling malloc().
2976 * We are protected by Giant for now.
2977 */
2978 i = 0;
2979 for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2980 vfsconf2x(vfsp, xvfsp + i);
2981 i++;
2982 }
2983 error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2984 free(xvfsp, M_TEMP);
2985 return (error);
2986}
2987
2988SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2989 "S,xvfsconf", "List of all configured filesystems");
2990
2991/*
2992 * Top level filesystem related information gathering.
2993 */
2994static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2995
2996static int
2997vfs_sysctl(SYSCTL_HANDLER_ARGS)
2998{
2999 int *name = (int *)arg1 - 1; /* XXX */
3000 u_int namelen = arg2 + 1; /* XXX */
3001 struct vfsconf *vfsp;
3002 struct xvfsconf xvfsp;
3003
3004 printf("WARNING: userland calling deprecated sysctl, "
3005 "please rebuild world\n");
3006
3007#if 1 || defined(COMPAT_PRELITE2)
3008 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3009 if (namelen == 1)
3010 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3011#endif
3012
3013 switch (name[1]) {
3014 case VFS_MAXTYPENUM:
3015 if (namelen != 2)
3016 return (ENOTDIR);
3017 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3018 case VFS_CONF:
3019 if (namelen != 3)
3020 return (ENOTDIR); /* overloaded */
3021 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
3022 if (vfsp->vfc_typenum == name[2])
3023 break;
3024 if (vfsp == NULL)
3025 return (EOPNOTSUPP);
3026 vfsconf2x(vfsp, &xvfsp);
3027 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3028 }
3029 return (EOPNOTSUPP);
3030}
3031
3032SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
3033 "Generic filesystem");
3034
3035#if 1 || defined(COMPAT_PRELITE2)
3036
3037static int
3038sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3039{
3040 int error;
3041 struct vfsconf *vfsp;
3042 struct ovfsconf ovfs;
3043
3044 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
3045 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3046 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3047 ovfs.vfc_index = vfsp->vfc_typenum;
3048 ovfs.vfc_refcount = vfsp->vfc_refcount;
3049 ovfs.vfc_flags = vfsp->vfc_flags;
3050 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3051 if (error)
3052 return error;
3053 }
3054 return 0;
3055}
3056
3057#endif /* 1 || COMPAT_PRELITE2 */
3058
3059#define KINFO_VNODESLOP 10
3060#ifdef notyet
3061/*
3062 * Dump vnode list (via sysctl).
3063 */
3064/* ARGSUSED */
3065static int
3066sysctl_vnode(SYSCTL_HANDLER_ARGS)
3067{
3068 struct xvnode *xvn;
3069 struct thread *td = req->td;
3070 struct mount *mp;
3071 struct vnode *vp;
3072 int error, len, n;
3073
3074 /*
3075 * Stale numvnodes access is not fatal here.
3076 */
3077 req->lock = 0;
3078 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3079 if (!req->oldptr)
3080 /* Make an estimate */
3081 return (SYSCTL_OUT(req, 0, len));
3082
3083 sysctl_wire_old_buffer(req, 0);
3084 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3085 n = 0;
3086 mtx_lock(&mountlist_mtx);
3087 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3088 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3089 continue;
3090 mtx_lock(&mntvnode_mtx);
3091 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3092 if (n == len)
3093 break;
3094 vref(vp);
3095 xvn[n].xv_size = sizeof *xvn;
3096 xvn[n].xv_vnode = vp;
3097#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3098 XV_COPY(usecount);
3099 XV_COPY(writecount);
3100 XV_COPY(holdcnt);
3101 XV_COPY(id);
3102 XV_COPY(mount);
3103 XV_COPY(numoutput);
3104 XV_COPY(type);
3105#undef XV_COPY
3106 xvn[n].xv_flag = vp->v_vflag;
3107
3108 switch (vp->v_type) {
3109 case VREG:
3110 case VDIR:
3111 case VLNK:
3112 xvn[n].xv_dev = vp->v_cachedfs;
3113 xvn[n].xv_ino = vp->v_cachedid;
3114 break;
3115 case VBLK:
3116 case VCHR:
3117 if (vp->v_rdev == NULL) {
3118 vrele(vp);
3119 continue;
3120 }
3121 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3122 break;
3123 case VSOCK:
3124 xvn[n].xv_socket = vp->v_socket;
3125 break;
3126 case VFIFO:
3127 xvn[n].xv_fifo = vp->v_fifoinfo;
3128 break;
3129 case VNON:
3130 case VBAD:
3131 default:
3132 /* shouldn't happen? */
3133 vrele(vp);
3134 continue;
3135 }
3136 vrele(vp);
3137 ++n;
3138 }
3139 mtx_unlock(&mntvnode_mtx);
3140 mtx_lock(&mountlist_mtx);
3141 vfs_unbusy(mp, td);
3142 if (n == len)
3143 break;
3144 }
3145 mtx_unlock(&mountlist_mtx);
3146
3147 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3148 free(xvn, M_TEMP);
3149 return (error);
3150}
3151
3152SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3153 0, 0, sysctl_vnode, "S,xvnode", "");
3154#endif
3155
3156/*
3157 * Check to see if a filesystem is mounted on a block device.
3158 */
3159int
3160vfs_mountedon(vp)
3161 struct vnode *vp;
3162{
3163
3164 if (vp->v_rdev->si_mountpoint != NULL)
3165 return (EBUSY);
3166 return (0);
3167}
3168
3169/*
3170 * Unmount all filesystems. The list is traversed in reverse order
3171 * of mounting to avoid dependencies.
3172 */
3173void
3174vfs_unmountall()
3175{
3176 struct mount *mp;
3177 struct thread *td;
3178 int error;
3179
3180 if (curthread != NULL)
3181 td = curthread;
3182 else
3183 td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3184 /*
3185 * Since this only runs when rebooting, it is not interlocked.
3186 */
3187 while(!TAILQ_EMPTY(&mountlist)) {
3188 mp = TAILQ_LAST(&mountlist, mntlist);
3189 error = dounmount(mp, MNT_FORCE, td);
3190 if (error) {
3191 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3192 printf("unmount of %s failed (",
3193 mp->mnt_stat.f_mntonname);
3194 if (error == EBUSY)
3195 printf("BUSY)\n");
3196 else
3197 printf("%d)\n", error);
3198 } else {
3199 /* The unmount has removed mp from the mountlist */
3200 }
3201 }
3202}
3203
3204/*
3205 * perform msync on all vnodes under a mount point
3206 * the mount point must be locked.
3207 */
3208void
3209vfs_msync(struct mount *mp, int flags)
3210{
3211 struct vnode *vp, *nvp;
3212 struct vm_object *obj;
3213 int tries;
3214
3215 GIANT_REQUIRED;
3216
3217 tries = 5;
3218 mtx_lock(&mntvnode_mtx);
3219loop:
3220 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3221 if (vp->v_mount != mp) {
3222 if (--tries > 0)
3223 goto loop;
3224 break;
3225 }
3226 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3227
3228 VI_LOCK(vp);
3229 if (vp->v_iflag & VI_XLOCK) { /* XXX: what if MNT_WAIT? */
3230 VI_UNLOCK(vp);
3231 continue;
3232 }
3233
3234 if ((vp->v_iflag & VI_OBJDIRTY) &&
3235 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3236 mtx_unlock(&mntvnode_mtx);
3237 if (!vget(vp,
3238 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3239 curthread)) {
3240 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3241 vput(vp);
3242 mtx_lock(&mntvnode_mtx);
3243 continue;
3244 }
3245
3246 if (VOP_GETVOBJECT(vp, &obj) == 0) {
3247 VM_OBJECT_LOCK(obj);
3248 vm_object_page_clean(obj, 0, 0,
3249 flags == MNT_WAIT ?
3250 OBJPC_SYNC : OBJPC_NOSYNC);
3251 VM_OBJECT_UNLOCK(obj);
3252 }
3253 vput(vp);
3254 }
3255 mtx_lock(&mntvnode_mtx);
3256 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3257 if (--tries > 0)
3258 goto loop;
3259 break;
3260 }
3261 } else
3262 VI_UNLOCK(vp);
3263 }
3264 mtx_unlock(&mntvnode_mtx);
3265}
3266
3267/*
3268 * Create the VM object needed for VMIO and mmap support. This
3269 * is done for all VREG files in the system. Some filesystems might
3270 * afford the additional metadata buffering capability of the
3271 * VMIO code by making the device node be VMIO mode also.
3272 *
3273 * vp must be locked when vfs_object_create is called.
3274 */
3275int
3276vfs_object_create(vp, td, cred)
3277 struct vnode *vp;
3278 struct thread *td;
3279 struct ucred *cred;
3280{
3281 GIANT_REQUIRED;
3282 return (VOP_CREATEVOBJECT(vp, cred, td));
3283}
3284
3285/*
3286 * Mark a vnode as free, putting it up for recycling.
3287 */
3288void
3289vfree(vp)
3290 struct vnode *vp;
3291{
3292 int s;
3293
3294 ASSERT_VI_LOCKED(vp, "vfree");
3295 s = splbio();
3296 mtx_lock(&vnode_free_list_mtx);
3297 KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3298 if (vp->v_iflag & VI_AGE) {
3299 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3300 } else {
3301 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3302 }
3303 freevnodes++;
3304 mtx_unlock(&vnode_free_list_mtx);
3305 vp->v_iflag &= ~VI_AGE;
3306 vp->v_iflag |= VI_FREE;
3307 splx(s);
3308}
3309
3310/*
3311 * Opposite of vfree() - mark a vnode as in use.
3312 */
3313void
3314vbusy(vp)
3315 struct vnode *vp;
3316{
3317 int s;
3318
3319 s = splbio();
3320 ASSERT_VI_LOCKED(vp, "vbusy");
3321 KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3322
3323 mtx_lock(&vnode_free_list_mtx);
3324 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3325 freevnodes--;
3326 mtx_unlock(&vnode_free_list_mtx);
3327
3328 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3329 splx(s);
3330}
3331
3332/*
3333 * Record a process's interest in events which might happen to
3334 * a vnode. Because poll uses the historic select-style interface
3335 * internally, this routine serves as both the ``check for any
3336 * pending events'' and the ``record my interest in future events''
3337 * functions. (These are done together, while the lock is held,
3338 * to avoid race conditions.)
3339 */
3340int
3341vn_pollrecord(vp, td, events)
3342 struct vnode *vp;
3343 struct thread *td;
3344 short events;
3345{
3346
3347 if (vp->v_pollinfo == NULL)
3348 v_addpollinfo(vp);
3349 mtx_lock(&vp->v_pollinfo->vpi_lock);
3350 if (vp->v_pollinfo->vpi_revents & events) {
3351 /*
3352 * This leaves events we are not interested
3353 * in available for the other process which
3354 * which presumably had requested them
3355 * (otherwise they would never have been
3356 * recorded).
3357 */
3358 events &= vp->v_pollinfo->vpi_revents;
3359 vp->v_pollinfo->vpi_revents &= ~events;
3360
3361 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3362 return events;
3363 }
3364 vp->v_pollinfo->vpi_events |= events;
3365 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3366 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3367 return 0;
3368}
3369
3370/*
3371 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3372 * it is possible for us to miss an event due to race conditions, but
3373 * that condition is expected to be rare, so for the moment it is the
3374 * preferred interface.
3375 */
3376void
3377vn_pollevent(vp, events)
3378 struct vnode *vp;
3379 short events;
3380{
3381
3382 if (vp->v_pollinfo == NULL)
3383 v_addpollinfo(vp);
3384 mtx_lock(&vp->v_pollinfo->vpi_lock);
3385 if (vp->v_pollinfo->vpi_events & events) {
3386 /*
3387 * We clear vpi_events so that we don't
3388 * call selwakeup() twice if two events are
3389 * posted before the polling process(es) is
3390 * awakened. This also ensures that we take at
3391 * most one selwakeup() if the polling process
3392 * is no longer interested. However, it does
3393 * mean that only one event can be noticed at
3394 * a time. (Perhaps we should only clear those
3395 * event bits which we note?) XXX
3396 */
3397 vp->v_pollinfo->vpi_events = 0; /* &= ~events ??? */
3398 vp->v_pollinfo->vpi_revents |= events;
3399 selwakeup(&vp->v_pollinfo->vpi_selinfo);
3400 }
3401 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3402}
3403
3404/*
3405 * Wake up anyone polling on vp because it is being revoked.
3406 * This depends on dead_poll() returning POLLHUP for correct
3407 * behavior.
3408 */
3409void
3410vn_pollgone(vp)
3411 struct vnode *vp;
3412{
3413
3414 mtx_lock(&vp->v_pollinfo->vpi_lock);
3415 VN_KNOTE(vp, NOTE_REVOKE);
3416 if (vp->v_pollinfo->vpi_events) {
3417 vp->v_pollinfo->vpi_events = 0;
3418 selwakeup(&vp->v_pollinfo->vpi_selinfo);
3419 }
3420 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3421}
3422
3423
3424
3425/*
3426 * Routine to create and manage a filesystem syncer vnode.
3427 */
3428#define sync_close ((int (*)(struct vop_close_args *))nullop)
3429static int sync_fsync(struct vop_fsync_args *);
3430static int sync_inactive(struct vop_inactive_args *);
3431static int sync_reclaim(struct vop_reclaim_args *);
3432
3433static vop_t **sync_vnodeop_p;
3434static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3435 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3436 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3437 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3438 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3439 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3440 { &vop_lock_desc, (vop_t *) vop_stdlock }, /* lock */
3441 { &vop_unlock_desc, (vop_t *) vop_stdunlock }, /* unlock */
3442 { &vop_islocked_desc, (vop_t *) vop_stdislocked }, /* islocked */
3443 { NULL, NULL }
3444};
3445static struct vnodeopv_desc sync_vnodeop_opv_desc =
3446 { &sync_vnodeop_p, sync_vnodeop_entries };
3447
3448VNODEOP_SET(sync_vnodeop_opv_desc);
3449
3450/*
3451 * Create a new filesystem syncer vnode for the specified mount point.
3452 */
3453int
3454vfs_allocate_syncvnode(mp)
3455 struct mount *mp;
3456{
3457 struct vnode *vp;
3458 static long start, incr, next;
3459 int error;
3460
3461 /* Allocate a new vnode */
3462 if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3463 mp->mnt_syncer = NULL;
3464 return (error);
3465 }
3466 vp->v_type = VNON;
3467 /*
3468 * Place the vnode onto the syncer worklist. We attempt to
3469 * scatter them about on the list so that they will go off
3470 * at evenly distributed times even if all the filesystems
3471 * are mounted at once.
3472 */
3473 next += incr;
3474 if (next == 0 || next > syncer_maxdelay) {
3475 start /= 2;
3476 incr /= 2;
3477 if (start == 0) {
3478 start = syncer_maxdelay / 2;
3479 incr = syncer_maxdelay;
3480 }
3481 next = start;
3482 }
3483 VI_LOCK(vp);
3484 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3485 VI_UNLOCK(vp);
3486 mp->mnt_syncer = vp;
3487 return (0);
3488}
3489
3490/*
3491 * Do a lazy sync of the filesystem.
3492 */
3493static int
3494sync_fsync(ap)
3495 struct vop_fsync_args /* {
3496 struct vnode *a_vp;
3497 struct ucred *a_cred;
3498 int a_waitfor;
3499 struct thread *a_td;
3500 } */ *ap;
3501{
3502 struct vnode *syncvp = ap->a_vp;
3503 struct mount *mp = syncvp->v_mount;
3504 struct thread *td = ap->a_td;
3505 int error, asyncflag;
3506
3507 /*
3508 * We only need to do something if this is a lazy evaluation.
3509 */
3510 if (ap->a_waitfor != MNT_LAZY)
3511 return (0);
3512
3513 /*
3514 * Move ourselves to the back of the sync list.
3515 */
3516 VI_LOCK(syncvp);
3517 vn_syncer_add_to_worklist(syncvp, syncdelay);
3518 VI_UNLOCK(syncvp);
3519
3520 /*
3521 * Walk the list of vnodes pushing all that are dirty and
3522 * not already on the sync list.
3523 */
3524 mtx_lock(&mountlist_mtx);
3525 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3526 mtx_unlock(&mountlist_mtx);
3527 return (0);
3528 }
3529 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3530 vfs_unbusy(mp, td);
3531 return (0);
3532 }
3533 asyncflag = mp->mnt_flag & MNT_ASYNC;
3534 mp->mnt_flag &= ~MNT_ASYNC;
3535 vfs_msync(mp, MNT_NOWAIT);
3536 error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3537 if (asyncflag)
3538 mp->mnt_flag |= MNT_ASYNC;
3539 vn_finished_write(mp);
3540 vfs_unbusy(mp, td);
3541 return (error);
3542}
3543
3544/*
3545 * The syncer vnode is no referenced.
3546 */
3547static int
3548sync_inactive(ap)
3549 struct vop_inactive_args /* {
3550 struct vnode *a_vp;
3551 struct thread *a_td;
3552 } */ *ap;
3553{
3554
3555 VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3556 vgone(ap->a_vp);
3557 return (0);
3558}
3559
3560/*
3561 * The syncer vnode is no longer needed and is being decommissioned.
3562 *
3563 * Modifications to the worklist must be protected at splbio().
3564 */
3565static int
3566sync_reclaim(ap)
3567 struct vop_reclaim_args /* {
3568 struct vnode *a_vp;
3569 } */ *ap;
3570{
3571 struct vnode *vp = ap->a_vp;
3572 int s;
3573
3574 s = splbio();
3575 vp->v_mount->mnt_syncer = NULL;
3576 VI_LOCK(vp);
3577 if (vp->v_iflag & VI_ONWORKLST) {
3578 mtx_lock(&sync_mtx);
3579 LIST_REMOVE(vp, v_synclist);
3580 mtx_unlock(&sync_mtx);
3581 vp->v_iflag &= ~VI_ONWORKLST;
3582 }
3583 VI_UNLOCK(vp);
3584 splx(s);
3585
3586 return (0);
3587}
3588
3589/*
3590 * extract the dev_t from a VCHR
3591 */
3592dev_t
3593vn_todev(vp)
3594 struct vnode *vp;
3595{
3596 if (vp->v_type != VCHR)
3597 return (NODEV);
3598 return (vp->v_rdev);
3599}
3600
3601/*
3602 * Check if vnode represents a disk device
3603 */
3604int
3605vn_isdisk(vp, errp)
3606 struct vnode *vp;
3607 int *errp;
3608{
3609 struct cdevsw *cdevsw;
3610
3611 if (vp->v_type != VCHR) {
3612 if (errp != NULL)
3613 *errp = ENOTBLK;
3614 return (0);
3615 }
3616 if (vp->v_rdev == NULL) {
3617 if (errp != NULL)
3618 *errp = ENXIO;
3619 return (0);
3620 }
3621 cdevsw = devsw(vp->v_rdev);
3622 if (cdevsw == NULL) {
3623 if (errp != NULL)
3624 *errp = ENXIO;
3625 return (0);
3626 }
3627 if (!(cdevsw->d_flags & D_DISK)) {
3628 if (errp != NULL)
3629 *errp = ENOTBLK;
3630 return (0);
3631 }
3632 if (errp != NULL)
3633 *errp = 0;
3634 return (1);
3635}
3636
3637/*
3638 * Free data allocated by namei(); see namei(9) for details.
3639 */
3640void
3641NDFREE(ndp, flags)
3642 struct nameidata *ndp;
3643 const uint flags;
3644{
3645 if (!(flags & NDF_NO_FREE_PNBUF) &&
3646 (ndp->ni_cnd.cn_flags & HASBUF)) {
3647 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3648 ndp->ni_cnd.cn_flags &= ~HASBUF;
3649 }
3650 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3651 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3652 ndp->ni_dvp != ndp->ni_vp)
3653 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3654 if (!(flags & NDF_NO_DVP_RELE) &&
3655 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3656 vrele(ndp->ni_dvp);
3657 ndp->ni_dvp = NULL;
3658 }
3659 if (!(flags & NDF_NO_VP_UNLOCK) &&
3660 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3661 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3662 if (!(flags & NDF_NO_VP_RELE) &&
3663 ndp->ni_vp) {
3664 vrele(ndp->ni_vp);
3665 ndp->ni_vp = NULL;
3666 }
3667 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3668 (ndp->ni_cnd.cn_flags & SAVESTART)) {
3669 vrele(ndp->ni_startdir);
3670 ndp->ni_startdir = NULL;
3671 }
3672}
3673
3674/*
3675 * Common filesystem object access control check routine. Accepts a
3676 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3677 * and optional call-by-reference privused argument allowing vaccess()
3678 * to indicate to the caller whether privilege was used to satisfy the
3679 * request (obsoleted). Returns 0 on success, or an errno on failure.
3680 */
3681int
3682vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3683 enum vtype type;
3684 mode_t file_mode;
3685 uid_t file_uid;
3686 gid_t file_gid;
3687 mode_t acc_mode;
3688 struct ucred *cred;
3689 int *privused;
3690{
3691 mode_t dac_granted;
3692#ifdef CAPABILITIES
3693 mode_t cap_granted;
3694#endif
3695
3696 /*
3697 * Look for a normal, non-privileged way to access the file/directory
3698 * as requested. If it exists, go with that.
3699 */
3700
3701 if (privused != NULL)
3702 *privused = 0;
3703
3704 dac_granted = 0;
3705
3706 /* Check the owner. */
3707 if (cred->cr_uid == file_uid) {
3708 dac_granted |= VADMIN;
3709 if (file_mode & S_IXUSR)
3710 dac_granted |= VEXEC;
3711 if (file_mode & S_IRUSR)
3712 dac_granted |= VREAD;
3713 if (file_mode & S_IWUSR)
3714 dac_granted |= (VWRITE | VAPPEND);
3715
3716 if ((acc_mode & dac_granted) == acc_mode)
3717 return (0);
3718
3719 goto privcheck;
3720 }
3721
3722 /* Otherwise, check the groups (first match) */
3723 if (groupmember(file_gid, cred)) {
3724 if (file_mode & S_IXGRP)
3725 dac_granted |= VEXEC;
3726 if (file_mode & S_IRGRP)
3727 dac_granted |= VREAD;
3728 if (file_mode & S_IWGRP)
3729 dac_granted |= (VWRITE | VAPPEND);
3730
3731 if ((acc_mode & dac_granted) == acc_mode)
3732 return (0);
3733
3734 goto privcheck;
3735 }
3736
3737 /* Otherwise, check everyone else. */
3738 if (file_mode & S_IXOTH)
3739 dac_granted |= VEXEC;
3740 if (file_mode & S_IROTH)
3741 dac_granted |= VREAD;
3742 if (file_mode & S_IWOTH)
3743 dac_granted |= (VWRITE | VAPPEND);
3744 if ((acc_mode & dac_granted) == acc_mode)
3745 return (0);
3746
3747privcheck:
3748 if (!suser_cred(cred, PRISON_ROOT)) {
3749 /* XXX audit: privilege used */
3750 if (privused != NULL)
3751 *privused = 1;
3752 return (0);
3753 }
3754
3755#ifdef CAPABILITIES
3756 /*
3757 * Build a capability mask to determine if the set of capabilities
3758 * satisfies the requirements when combined with the granted mask
3759 * from above.
3760 * For each capability, if the capability is required, bitwise
3761 * or the request type onto the cap_granted mask.
3762 */
3763 cap_granted = 0;
3764
3765 if (type == VDIR) {
3766 /*
3767 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3768 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3769 */
3770 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3771 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3772 cap_granted |= VEXEC;
3773 } else {
3774 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3775 !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3776 cap_granted |= VEXEC;
3777 }
3778
3779 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3780 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3781 cap_granted |= VREAD;
3782
3783 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3784 !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3785 cap_granted |= (VWRITE | VAPPEND);
3786
3787 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3788 !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3789 cap_granted |= VADMIN;
3790
3791 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3792 /* XXX audit: privilege used */
3793 if (privused != NULL)
3794 *privused = 1;
3795 return (0);
3796 }
3797#endif
3798
3799 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3800}
3801
3802/*
3803 * Credential check based on process requesting service, and per-attribute
3804 * permissions.
3805 */
3806int
3807extattr_check_cred(struct vnode *vp, int attrnamespace,
3808 struct ucred *cred, struct thread *td, int access)
3809{
3810
3811 /*
3812 * Kernel-invoked always succeeds.
3813 */
3814 if (cred == NOCRED)
3815 return (0);
3816
3817 /*
3818 * Do not allow privileged processes in jail to directly
3819 * manipulate system attributes.
3820 *
3821 * XXX What capability should apply here?
3822 * Probably CAP_SYS_SETFFLAG.
3823 */
3824 switch (attrnamespace) {
3825 case EXTATTR_NAMESPACE_SYSTEM:
3826 /* Potentially should be: return (EPERM); */
3827 return (suser_cred(cred, 0));
3828 case EXTATTR_NAMESPACE_USER:
3829 return (VOP_ACCESS(vp, access, cred, td));
3830 default:
3831 return (EPERM);
3832 }
3833}
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 */
40
41/*
42 * External virtual filesystem routines
43 */
44
45#include <sys/cdefs.h>
46__FBSDID("$FreeBSD: head/sys/kern/vfs_subr.c 117879 2003-07-22 10:36:36Z phk $");
47
48#include "opt_ddb.h"
49#include "opt_mac.h"
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/bio.h>
54#include <sys/buf.h>
55#include <sys/conf.h>
56#include <sys/eventhandler.h>
57#include <sys/extattr.h>
58#include <sys/fcntl.h>
59#include <sys/kernel.h>
60#include <sys/kthread.h>
61#include <sys/mac.h>
62#include <sys/malloc.h>
63#include <sys/mount.h>
64#include <sys/namei.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 <vm/vm.h>
72#include <vm/vm_object.h>
73#include <vm/vm_extern.h>
74#include <vm/pmap.h>
75#include <vm/vm_map.h>
76#include <vm/vm_page.h>
77#include <vm/vm_kern.h>
78#include <vm/uma.h>
79
80static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
81
82static void addalias(struct vnode *vp, dev_t nvp_rdev);
83static void insmntque(struct vnode *vp, struct mount *mp);
84static void vclean(struct vnode *vp, int flags, struct thread *td);
85static void vlruvp(struct vnode *vp);
86static int flushbuflist(struct buf *blist, int flags, struct vnode *vp,
87 int slpflag, int slptimeo, int *errorp);
88static int vcanrecycle(struct vnode *vp, struct mount **vnmpp);
89
90
91/*
92 * Number of vnodes in existence. Increased whenever getnewvnode()
93 * allocates a new vnode, never decreased.
94 */
95static unsigned long numvnodes;
96
97SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
98
99/*
100 * Conversion tables for conversion from vnode types to inode formats
101 * and back.
102 */
103enum vtype iftovt_tab[16] = {
104 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
105 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
106};
107int vttoif_tab[9] = {
108 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
109 S_IFSOCK, S_IFIFO, S_IFMT,
110};
111
112/*
113 * List of vnodes that are ready for recycling.
114 */
115static TAILQ_HEAD(freelst, vnode) vnode_free_list;
116
117/*
118 * Minimum number of free vnodes. If there are fewer than this free vnodes,
119 * getnewvnode() will return a newly allocated vnode.
120 */
121static u_long wantfreevnodes = 25;
122SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
123/* Number of vnodes in the free list. */
124static u_long freevnodes;
125SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
126
127/*
128 * Various variables used for debugging the new implementation of
129 * reassignbuf().
130 * XXX these are probably of (very) limited utility now.
131 */
132static int reassignbufcalls;
133SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
134static int nameileafonly;
135SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
136
137/*
138 * Cache for the mount type id assigned to NFS. This is used for
139 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
140 */
141int nfs_mount_type = -1;
142
143/* To keep more than one thread at a time from running vfs_getnewfsid */
144static struct mtx mntid_mtx;
145
146/*
147 * Lock for any access to the following:
148 * vnode_free_list
149 * numvnodes
150 * freevnodes
151 */
152static struct mtx vnode_free_list_mtx;
153
154/*
155 * For any iteration/modification of dev->si_hlist (linked through
156 * v_specnext)
157 */
158static struct mtx spechash_mtx;
159
160/* Publicly exported FS */
161struct nfs_public nfs_pub;
162
163/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
164static uma_zone_t vnode_zone;
165static uma_zone_t vnodepoll_zone;
166
167/* Set to 1 to print out reclaim of active vnodes */
168int prtactive;
169
170/*
171 * The workitem queue.
172 *
173 * It is useful to delay writes of file data and filesystem metadata
174 * for tens of seconds so that quickly created and deleted files need
175 * not waste disk bandwidth being created and removed. To realize this,
176 * we append vnodes to a "workitem" queue. When running with a soft
177 * updates implementation, most pending metadata dependencies should
178 * not wait for more than a few seconds. Thus, mounted on block devices
179 * are delayed only about a half the time that file data is delayed.
180 * Similarly, directory updates are more critical, so are only delayed
181 * about a third the time that file data is delayed. Thus, there are
182 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
183 * one each second (driven off the filesystem syncer process). The
184 * syncer_delayno variable indicates the next queue that is to be processed.
185 * Items that need to be processed soon are placed in this queue:
186 *
187 * syncer_workitem_pending[syncer_delayno]
188 *
189 * A delay of fifteen seconds is done by placing the request fifteen
190 * entries later in the queue:
191 *
192 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
193 *
194 */
195static int syncer_delayno;
196static long syncer_mask;
197LIST_HEAD(synclist, vnode);
198static struct synclist *syncer_workitem_pending;
199/*
200 * The sync_mtx protects:
201 * vp->v_synclist
202 * syncer_delayno
203 * syncer_workitem_pending
204 * rushjob
205 */
206static struct mtx sync_mtx;
207
208#define SYNCER_MAXDELAY 32
209static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
210static int syncdelay = 30; /* max time to delay syncing data */
211static int filedelay = 30; /* time to delay syncing files */
212SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
213static int dirdelay = 29; /* time to delay syncing directories */
214SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
215static int metadelay = 28; /* time to delay syncing metadata */
216SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
217static int rushjob; /* number of slots to run ASAP */
218static int stat_rush_requests; /* number of times I/O speeded up */
219SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
220
221/*
222 * Number of vnodes we want to exist at any one time. This is mostly used
223 * to size hash tables in vnode-related code. It is normally not used in
224 * getnewvnode(), as wantfreevnodes is normally nonzero.)
225 *
226 * XXX desiredvnodes is historical cruft and should not exist.
227 */
228int desiredvnodes;
229SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
230 &desiredvnodes, 0, "Maximum number of vnodes");
231static int minvnodes;
232SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
233 &minvnodes, 0, "Minimum number of vnodes");
234static int vnlru_nowhere;
235SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
236 "Number of times the vnlru process ran without success");
237
238/* Hook for calling soft updates */
239int (*softdep_process_worklist_hook)(struct mount *);
240
241/*
242 * This only exists to supress warnings from unlocked specfs accesses. It is
243 * no longer ok to have an unlocked VFS.
244 */
245#define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
246
247/* Print lock violations */
248int vfs_badlock_print = 1;
249
250/* Panic on violation */
251int vfs_badlock_panic = 1;
252
253/* Check for interlock across VOPs */
254int vfs_badlock_mutex = 1;
255
256static void
257vfs_badlock(char *msg, char *str, struct vnode *vp)
258{
259 if (vfs_badlock_print)
260 printf("%s: %p %s\n", str, vp, msg);
261 if (vfs_badlock_panic)
262 Debugger("Lock violation.\n");
263}
264
265void
266assert_vi_unlocked(struct vnode *vp, char *str)
267{
268 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
269 vfs_badlock("interlock is locked but should not be", str, vp);
270}
271
272void
273assert_vi_locked(struct vnode *vp, char *str)
274{
275 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
276 vfs_badlock("interlock is not locked but should be", str, vp);
277}
278
279void
280assert_vop_locked(struct vnode *vp, char *str)
281{
282 if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
283 vfs_badlock("is not locked but should be", str, vp);
284}
285
286void
287assert_vop_unlocked(struct vnode *vp, char *str)
288{
289 if (vp && !IGNORE_LOCK(vp) &&
290 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
291 vfs_badlock("is locked but should not be", str, vp);
292}
293
294void
295assert_vop_elocked(struct vnode *vp, char *str)
296{
297 if (vp && !IGNORE_LOCK(vp) &&
298 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
299 vfs_badlock("is not exclusive locked but should be", str, vp);
300}
301
302void
303assert_vop_elocked_other(struct vnode *vp, char *str)
304{
305 if (vp && !IGNORE_LOCK(vp) &&
306 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
307 vfs_badlock("is not exclusive locked by another thread",
308 str, vp);
309}
310
311void
312assert_vop_slocked(struct vnode *vp, char *str)
313{
314 if (vp && !IGNORE_LOCK(vp) &&
315 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
316 vfs_badlock("is not locked shared but should be", str, vp);
317}
318
319void
320vop_rename_pre(void *ap)
321{
322 struct vop_rename_args *a = ap;
323
324 if (a->a_tvp)
325 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
326 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
327 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
328 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
329
330 /* Check the source (from) */
331 if (a->a_tdvp != a->a_fdvp)
332 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
333 if (a->a_tvp != a->a_fvp)
334 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
335
336 /* Check the target */
337 if (a->a_tvp)
338 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
339
340 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
341}
342
343void
344vop_strategy_pre(void *ap)
345{
346 struct vop_strategy_args *a = ap;
347 struct buf *bp;
348
349 bp = a->a_bp;
350
351 /*
352 * Cluster ops lock their component buffers but not the IO container.
353 */
354 if ((bp->b_flags & B_CLUSTER) != 0)
355 return;
356
357 if (BUF_REFCNT(bp) < 1) {
358 if (vfs_badlock_print)
359 printf("VOP_STRATEGY: bp is not locked but should be.\n");
360 if (vfs_badlock_panic)
361 Debugger("Lock violation.\n");
362 }
363}
364
365void
366vop_lookup_pre(void *ap)
367{
368 struct vop_lookup_args *a = ap;
369 struct vnode *dvp;
370
371 dvp = a->a_dvp;
372
373 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
374 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
375}
376
377void
378vop_lookup_post(void *ap, int rc)
379{
380 struct vop_lookup_args *a = ap;
381 struct componentname *cnp;
382 struct vnode *dvp;
383 struct vnode *vp;
384 int flags;
385
386 dvp = a->a_dvp;
387 cnp = a->a_cnp;
388 vp = *(a->a_vpp);
389 flags = cnp->cn_flags;
390
391
392 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
393 /*
394 * If this is the last path component for this lookup and LOCPARENT
395 * is set, OR if there is an error the directory has to be locked.
396 */
397 if ((flags & LOCKPARENT) && (flags & ISLASTCN))
398 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
399 else if (rc != 0)
400 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
401 else if (dvp != vp)
402 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
403
404 if (flags & PDIRUNLOCK)
405 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
406}
407
408void
409vop_unlock_pre(void *ap)
410{
411 struct vop_unlock_args *a = ap;
412
413 if (a->a_flags & LK_INTERLOCK)
414 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
415
416 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
417}
418
419void
420vop_unlock_post(void *ap, int rc)
421{
422 struct vop_unlock_args *a = ap;
423
424 if (a->a_flags & LK_INTERLOCK)
425 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
426}
427
428void
429vop_lock_pre(void *ap)
430{
431 struct vop_lock_args *a = ap;
432
433 if ((a->a_flags & LK_INTERLOCK) == 0)
434 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
435 else
436 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
437}
438
439void
440vop_lock_post(void *ap, int rc)
441{
442 struct vop_lock_args *a;
443
444 a = ap;
445
446 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
447 if (rc == 0)
448 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
449}
450
451void
452v_addpollinfo(struct vnode *vp)
453{
454 vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
455 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
456}
457
458/*
459 * Initialize the vnode management data structures.
460 */
461static void
462vntblinit(void *dummy __unused)
463{
464
465 /*
466 * Desiredvnodes is a function of the physical memory size and
467 * the kernel's heap size. Specifically, desiredvnodes scales
468 * in proportion to the physical memory size until two fifths
469 * of the kernel's heap size is consumed by vnodes and vm
470 * objects.
471 */
472 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
473 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
474 minvnodes = desiredvnodes / 4;
475 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
476 mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
477 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
478 mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
479 TAILQ_INIT(&vnode_free_list);
480 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
481 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
482 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
483 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
484 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
485 /*
486 * Initialize the filesystem syncer.
487 */
488 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
489 &syncer_mask);
490 syncer_maxdelay = syncer_mask + 1;
491 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
492}
493SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
494
495
496/*
497 * Mark a mount point as busy. Used to synchronize access and to delay
498 * unmounting. Interlock is not released on failure.
499 */
500int
501vfs_busy(mp, flags, interlkp, td)
502 struct mount *mp;
503 int flags;
504 struct mtx *interlkp;
505 struct thread *td;
506{
507 int lkflags;
508
509 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
510 if (flags & LK_NOWAIT)
511 return (ENOENT);
512 mp->mnt_kern_flag |= MNTK_MWAIT;
513 /*
514 * Since all busy locks are shared except the exclusive
515 * lock granted when unmounting, the only place that a
516 * wakeup needs to be done is at the release of the
517 * exclusive lock at the end of dounmount.
518 */
519 msleep(mp, interlkp, PVFS, "vfs_busy", 0);
520 return (ENOENT);
521 }
522 lkflags = LK_SHARED | LK_NOPAUSE;
523 if (interlkp)
524 lkflags |= LK_INTERLOCK;
525 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
526 panic("vfs_busy: unexpected lock failure");
527 return (0);
528}
529
530/*
531 * Free a busy filesystem.
532 */
533void
534vfs_unbusy(mp, td)
535 struct mount *mp;
536 struct thread *td;
537{
538
539 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
540}
541
542/*
543 * Lookup a mount point by filesystem identifier.
544 */
545struct mount *
546vfs_getvfs(fsid)
547 fsid_t *fsid;
548{
549 register struct mount *mp;
550
551 mtx_lock(&mountlist_mtx);
552 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
553 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
554 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
555 mtx_unlock(&mountlist_mtx);
556 return (mp);
557 }
558 }
559 mtx_unlock(&mountlist_mtx);
560 return ((struct mount *) 0);
561}
562
563/*
564 * Get a new unique fsid. Try to make its val[0] unique, since this value
565 * will be used to create fake device numbers for stat(). Also try (but
566 * not so hard) make its val[0] unique mod 2^16, since some emulators only
567 * support 16-bit device numbers. We end up with unique val[0]'s for the
568 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
569 *
570 * Keep in mind that several mounts may be running in parallel. Starting
571 * the search one past where the previous search terminated is both a
572 * micro-optimization and a defense against returning the same fsid to
573 * different mounts.
574 */
575void
576vfs_getnewfsid(mp)
577 struct mount *mp;
578{
579 static u_int16_t mntid_base;
580 fsid_t tfsid;
581 int mtype;
582
583 mtx_lock(&mntid_mtx);
584 mtype = mp->mnt_vfc->vfc_typenum;
585 tfsid.val[1] = mtype;
586 mtype = (mtype & 0xFF) << 24;
587 for (;;) {
588 tfsid.val[0] = makeudev(255,
589 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
590 mntid_base++;
591 if (vfs_getvfs(&tfsid) == NULL)
592 break;
593 }
594 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
595 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
596 mtx_unlock(&mntid_mtx);
597}
598
599/*
600 * Knob to control the precision of file timestamps:
601 *
602 * 0 = seconds only; nanoseconds zeroed.
603 * 1 = seconds and nanoseconds, accurate within 1/HZ.
604 * 2 = seconds and nanoseconds, truncated to microseconds.
605 * >=3 = seconds and nanoseconds, maximum precision.
606 */
607enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
608
609static int timestamp_precision = TSP_SEC;
610SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
611 ×tamp_precision, 0, "");
612
613/*
614 * Get a current timestamp.
615 */
616void
617vfs_timestamp(tsp)
618 struct timespec *tsp;
619{
620 struct timeval tv;
621
622 switch (timestamp_precision) {
623 case TSP_SEC:
624 tsp->tv_sec = time_second;
625 tsp->tv_nsec = 0;
626 break;
627 case TSP_HZ:
628 getnanotime(tsp);
629 break;
630 case TSP_USEC:
631 microtime(&tv);
632 TIMEVAL_TO_TIMESPEC(&tv, tsp);
633 break;
634 case TSP_NSEC:
635 default:
636 nanotime(tsp);
637 break;
638 }
639}
640
641/*
642 * Set vnode attributes to VNOVAL
643 */
644void
645vattr_null(vap)
646 register struct vattr *vap;
647{
648
649 vap->va_type = VNON;
650 vap->va_size = VNOVAL;
651 vap->va_bytes = VNOVAL;
652 vap->va_mode = VNOVAL;
653 vap->va_nlink = VNOVAL;
654 vap->va_uid = VNOVAL;
655 vap->va_gid = VNOVAL;
656 vap->va_fsid = VNOVAL;
657 vap->va_fileid = VNOVAL;
658 vap->va_blocksize = VNOVAL;
659 vap->va_rdev = VNOVAL;
660 vap->va_atime.tv_sec = VNOVAL;
661 vap->va_atime.tv_nsec = VNOVAL;
662 vap->va_mtime.tv_sec = VNOVAL;
663 vap->va_mtime.tv_nsec = VNOVAL;
664 vap->va_ctime.tv_sec = VNOVAL;
665 vap->va_ctime.tv_nsec = VNOVAL;
666 vap->va_birthtime.tv_sec = VNOVAL;
667 vap->va_birthtime.tv_nsec = VNOVAL;
668 vap->va_flags = VNOVAL;
669 vap->va_gen = VNOVAL;
670 vap->va_vaflags = 0;
671}
672
673/*
674 * This routine is called when we have too many vnodes. It attempts
675 * to free <count> vnodes and will potentially free vnodes that still
676 * have VM backing store (VM backing store is typically the cause
677 * of a vnode blowout so we want to do this). Therefore, this operation
678 * is not considered cheap.
679 *
680 * A number of conditions may prevent a vnode from being reclaimed.
681 * the buffer cache may have references on the vnode, a directory
682 * vnode may still have references due to the namei cache representing
683 * underlying files, or the vnode may be in active use. It is not
684 * desireable to reuse such vnodes. These conditions may cause the
685 * number of vnodes to reach some minimum value regardless of what
686 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
687 */
688static int
689vlrureclaim(struct mount *mp)
690{
691 struct vnode *vp;
692 int done;
693 int trigger;
694 int usevnodes;
695 int count;
696
697 /*
698 * Calculate the trigger point, don't allow user
699 * screwups to blow us up. This prevents us from
700 * recycling vnodes with lots of resident pages. We
701 * aren't trying to free memory, we are trying to
702 * free vnodes.
703 */
704 usevnodes = desiredvnodes;
705 if (usevnodes <= 0)
706 usevnodes = 1;
707 trigger = cnt.v_page_count * 2 / usevnodes;
708
709 done = 0;
710 mtx_lock(&mntvnode_mtx);
711 count = mp->mnt_nvnodelistsize / 10 + 1;
712 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
713 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
714 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
715
716 if (vp->v_type != VNON &&
717 vp->v_type != VBAD &&
718 VI_TRYLOCK(vp)) {
719 if (VMIGHTFREE(vp) && /* critical path opt */
720 (vp->v_object == NULL ||
721 vp->v_object->resident_page_count < trigger)) {
722 mtx_unlock(&mntvnode_mtx);
723 vgonel(vp, curthread);
724 done++;
725 mtx_lock(&mntvnode_mtx);
726 } else
727 VI_UNLOCK(vp);
728 }
729 --count;
730 }
731 mtx_unlock(&mntvnode_mtx);
732 return done;
733}
734
735/*
736 * Attempt to recycle vnodes in a context that is always safe to block.
737 * Calling vlrurecycle() from the bowels of filesystem code has some
738 * interesting deadlock problems.
739 */
740static struct proc *vnlruproc;
741static int vnlruproc_sig;
742
743static void
744vnlru_proc(void)
745{
746 struct mount *mp, *nmp;
747 int s;
748 int done;
749 struct proc *p = vnlruproc;
750 struct thread *td = FIRST_THREAD_IN_PROC(p); /* XXXKSE */
751
752 mtx_lock(&Giant);
753
754 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
755 SHUTDOWN_PRI_FIRST);
756
757 s = splbio();
758 for (;;) {
759 kthread_suspend_check(p);
760 mtx_lock(&vnode_free_list_mtx);
761 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
762 mtx_unlock(&vnode_free_list_mtx);
763 vnlruproc_sig = 0;
764 wakeup(&vnlruproc_sig);
765 tsleep(vnlruproc, PVFS, "vlruwt", hz);
766 continue;
767 }
768 mtx_unlock(&vnode_free_list_mtx);
769 done = 0;
770 mtx_lock(&mountlist_mtx);
771 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
772 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
773 nmp = TAILQ_NEXT(mp, mnt_list);
774 continue;
775 }
776 done += vlrureclaim(mp);
777 mtx_lock(&mountlist_mtx);
778 nmp = TAILQ_NEXT(mp, mnt_list);
779 vfs_unbusy(mp, td);
780 }
781 mtx_unlock(&mountlist_mtx);
782 if (done == 0) {
783#if 0
784 /* These messages are temporary debugging aids */
785 if (vnlru_nowhere < 5)
786 printf("vnlru process getting nowhere..\n");
787 else if (vnlru_nowhere == 5)
788 printf("vnlru process messages stopped.\n");
789#endif
790 vnlru_nowhere++;
791 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
792 }
793 }
794 splx(s);
795}
796
797static struct kproc_desc vnlru_kp = {
798 "vnlru",
799 vnlru_proc,
800 &vnlruproc
801};
802SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
803
804
805/*
806 * Routines having to do with the management of the vnode table.
807 */
808
809/*
810 * Check to see if a free vnode can be recycled. If it can,
811 * return it locked with the vn lock, but not interlock. Also
812 * get the vn_start_write lock. Otherwise indicate the error.
813 */
814static int
815vcanrecycle(struct vnode *vp, struct mount **vnmpp)
816{
817 struct thread *td = curthread;
818 vm_object_t object;
819 int error;
820
821 /* Don't recycle if we can't get the interlock */
822 if (!VI_TRYLOCK(vp))
823 return (EWOULDBLOCK);
824
825 /* We should be able to immediately acquire this */
826 /* XXX This looks like it should panic if it fails */
827 if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
828 if (VOP_ISLOCKED(vp, td))
829 panic("vcanrecycle: locked vnode");
830 return (EWOULDBLOCK);
831 }
832
833 /*
834 * Don't recycle if its filesystem is being suspended.
835 */
836 if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
837 error = EBUSY;
838 goto done;
839 }
840
841 /*
842 * Don't recycle if we still have cached pages.
843 */
844 if (VOP_GETVOBJECT(vp, &object) == 0) {
845 VM_OBJECT_LOCK(object);
846 if (object->resident_page_count ||
847 object->ref_count) {
848 VM_OBJECT_UNLOCK(object);
849 error = EBUSY;
850 goto done;
851 }
852 VM_OBJECT_UNLOCK(object);
853 }
854 if (LIST_FIRST(&vp->v_cache_src)) {
855 /*
856 * note: nameileafonly sysctl is temporary,
857 * for debugging only, and will eventually be
858 * removed.
859 */
860 if (nameileafonly > 0) {
861 /*
862 * Do not reuse namei-cached directory
863 * vnodes that have cached
864 * subdirectories.
865 */
866 if (cache_leaf_test(vp) < 0) {
867 error = EISDIR;
868 goto done;
869 }
870 } else if (nameileafonly < 0 ||
871 vmiodirenable == 0) {
872 /*
873 * Do not reuse namei-cached directory
874 * vnodes if nameileafonly is -1 or
875 * if VMIO backing for directories is
876 * turned off (otherwise we reuse them
877 * too quickly).
878 */
879 error = EBUSY;
880 goto done;
881 }
882 }
883 return (0);
884done:
885 VOP_UNLOCK(vp, 0, td);
886 return (error);
887}
888
889/*
890 * Return the next vnode from the free list.
891 */
892int
893getnewvnode(tag, mp, vops, vpp)
894 const char *tag;
895 struct mount *mp;
896 vop_t **vops;
897 struct vnode **vpp;
898{
899 struct thread *td = curthread; /* XXX */
900 struct vnode *vp = NULL;
901 struct vpollinfo *pollinfo = NULL;
902 struct mount *vnmp;
903
904 mtx_lock(&vnode_free_list_mtx);
905
906 /*
907 * Try to reuse vnodes if we hit the max. This situation only
908 * occurs in certain large-memory (2G+) situations. We cannot
909 * attempt to directly reclaim vnodes due to nasty recursion
910 * problems.
911 */
912 while (numvnodes - freevnodes > desiredvnodes) {
913 if (vnlruproc_sig == 0) {
914 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
915 wakeup(vnlruproc);
916 }
917 mtx_unlock(&vnode_free_list_mtx);
918 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
919 mtx_lock(&vnode_free_list_mtx);
920 }
921
922 /*
923 * Attempt to reuse a vnode already on the free list, allocating
924 * a new vnode if we can't find one or if we have not reached a
925 * good minimum for good LRU performance.
926 */
927
928 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
929 int error;
930 int count;
931
932 for (count = 0; count < freevnodes; count++) {
933 vp = TAILQ_FIRST(&vnode_free_list);
934
935 KASSERT(vp->v_usecount == 0 &&
936 (vp->v_iflag & VI_DOINGINACT) == 0,
937 ("getnewvnode: free vnode isn't"));
938
939 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
940 /*
941 * We have to drop the free list mtx to avoid lock
942 * order reversals with interlock.
943 */
944 mtx_unlock(&vnode_free_list_mtx);
945 error = vcanrecycle(vp, &vnmp);
946 mtx_lock(&vnode_free_list_mtx);
947 if (error == 0)
948 break;
949 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
950 vp = NULL;
951 }
952 }
953 if (vp) {
954 freevnodes--;
955 mtx_unlock(&vnode_free_list_mtx);
956
957 cache_purge(vp);
958 VI_LOCK(vp);
959 vp->v_iflag |= VI_DOOMED;
960 vp->v_iflag &= ~VI_FREE;
961 if (vp->v_type != VBAD) {
962 VOP_UNLOCK(vp, 0, td);
963 vgonel(vp, td);
964 VI_LOCK(vp);
965 } else {
966 VOP_UNLOCK(vp, 0, td);
967 }
968 vn_finished_write(vnmp);
969
970#ifdef INVARIANTS
971 {
972 if (vp->v_data)
973 panic("cleaned vnode isn't");
974 if (vp->v_numoutput)
975 panic("Clean vnode has pending I/O's");
976 if (vp->v_writecount != 0)
977 panic("Non-zero write count");
978 }
979#endif
980 if ((pollinfo = vp->v_pollinfo) != NULL) {
981 /*
982 * To avoid lock order reversals, the call to
983 * uma_zfree() must be delayed until the vnode
984 * interlock is released.
985 */
986 vp->v_pollinfo = NULL;
987 }
988#ifdef MAC
989 mac_destroy_vnode(vp);
990#endif
991 vp->v_iflag = 0;
992 vp->v_vflag = 0;
993 vp->v_lastw = 0;
994 vp->v_lasta = 0;
995 vp->v_cstart = 0;
996 vp->v_clen = 0;
997 vp->v_socket = 0;
998 lockdestroy(vp->v_vnlock);
999 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1000 KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
1001 KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
1002 KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
1003 KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
1004 } else {
1005 numvnodes++;
1006 mtx_unlock(&vnode_free_list_mtx);
1007
1008 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1009 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1010 VI_LOCK(vp);
1011 vp->v_dd = vp;
1012 vp->v_vnlock = &vp->v_lock;
1013 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1014 cache_purge(vp);
1015 LIST_INIT(&vp->v_cache_src);
1016 TAILQ_INIT(&vp->v_cache_dst);
1017 }
1018
1019 TAILQ_INIT(&vp->v_cleanblkhd);
1020 TAILQ_INIT(&vp->v_dirtyblkhd);
1021 vp->v_type = VNON;
1022 vp->v_tag = tag;
1023 vp->v_op = vops;
1024 *vpp = vp;
1025 vp->v_usecount = 1;
1026 vp->v_data = 0;
1027 vp->v_cachedid = -1;
1028 VI_UNLOCK(vp);
1029 if (pollinfo != NULL) {
1030 mtx_destroy(&pollinfo->vpi_lock);
1031 uma_zfree(vnodepoll_zone, pollinfo);
1032 }
1033#ifdef MAC
1034 mac_init_vnode(vp);
1035 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1036 mac_associate_vnode_singlelabel(mp, vp);
1037#endif
1038 insmntque(vp, mp);
1039
1040 return (0);
1041}
1042
1043/*
1044 * Move a vnode from one mount queue to another.
1045 */
1046static void
1047insmntque(vp, mp)
1048 register struct vnode *vp;
1049 register struct mount *mp;
1050{
1051
1052 mtx_lock(&mntvnode_mtx);
1053 /*
1054 * Delete from old mount point vnode list, if on one.
1055 */
1056 if (vp->v_mount != NULL) {
1057 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
1058 ("bad mount point vnode list size"));
1059 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1060 vp->v_mount->mnt_nvnodelistsize--;
1061 }
1062 /*
1063 * Insert into list of vnodes for the new mount point, if available.
1064 */
1065 if ((vp->v_mount = mp) == NULL) {
1066 mtx_unlock(&mntvnode_mtx);
1067 return;
1068 }
1069 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1070 mp->mnt_nvnodelistsize++;
1071 mtx_unlock(&mntvnode_mtx);
1072}
1073
1074/*
1075 * Update outstanding I/O count and do wakeup if requested.
1076 */
1077void
1078vwakeup(bp)
1079 register struct buf *bp;
1080{
1081 register struct vnode *vp;
1082
1083 bp->b_flags &= ~B_WRITEINPROG;
1084 if ((vp = bp->b_vp)) {
1085 VI_LOCK(vp);
1086 vp->v_numoutput--;
1087 if (vp->v_numoutput < 0)
1088 panic("vwakeup: neg numoutput");
1089 if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1090 vp->v_iflag &= ~VI_BWAIT;
1091 wakeup(&vp->v_numoutput);
1092 }
1093 VI_UNLOCK(vp);
1094 }
1095}
1096
1097/*
1098 * Flush out and invalidate all buffers associated with a vnode.
1099 * Called with the underlying object locked.
1100 */
1101int
1102vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1103 struct vnode *vp;
1104 int flags;
1105 struct ucred *cred;
1106 struct thread *td;
1107 int slpflag, slptimeo;
1108{
1109 struct buf *blist;
1110 int s, error;
1111 vm_object_t object;
1112
1113 GIANT_REQUIRED;
1114
1115 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1116
1117 VI_LOCK(vp);
1118 if (flags & V_SAVE) {
1119 s = splbio();
1120 while (vp->v_numoutput) {
1121 vp->v_iflag |= VI_BWAIT;
1122 error = msleep(&vp->v_numoutput, VI_MTX(vp),
1123 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1124 if (error) {
1125 VI_UNLOCK(vp);
1126 splx(s);
1127 return (error);
1128 }
1129 }
1130 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1131 splx(s);
1132 VI_UNLOCK(vp);
1133 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1134 return (error);
1135 /*
1136 * XXX We could save a lock/unlock if this was only
1137 * enabled under INVARIANTS
1138 */
1139 VI_LOCK(vp);
1140 s = splbio();
1141 if (vp->v_numoutput > 0 ||
1142 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1143 panic("vinvalbuf: dirty bufs");
1144 }
1145 splx(s);
1146 }
1147 s = splbio();
1148 /*
1149 * If you alter this loop please notice that interlock is dropped and
1150 * reacquired in flushbuflist. Special care is needed to ensure that
1151 * no race conditions occur from this.
1152 */
1153 for (error = 0;;) {
1154 if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1155 flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1156 if (error)
1157 break;
1158 continue;
1159 }
1160 if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1161 flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1162 if (error)
1163 break;
1164 continue;
1165 }
1166 break;
1167 }
1168 if (error) {
1169 splx(s);
1170 VI_UNLOCK(vp);
1171 return (error);
1172 }
1173
1174 /*
1175 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1176 * have write I/O in-progress but if there is a VM object then the
1177 * VM object can also have read-I/O in-progress.
1178 */
1179 do {
1180 while (vp->v_numoutput > 0) {
1181 vp->v_iflag |= VI_BWAIT;
1182 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1183 }
1184 VI_UNLOCK(vp);
1185 if (VOP_GETVOBJECT(vp, &object) == 0) {
1186 VM_OBJECT_LOCK(object);
1187 vm_object_pip_wait(object, "vnvlbx");
1188 VM_OBJECT_UNLOCK(object);
1189 }
1190 VI_LOCK(vp);
1191 } while (vp->v_numoutput > 0);
1192 VI_UNLOCK(vp);
1193
1194 splx(s);
1195
1196 /*
1197 * Destroy the copy in the VM cache, too.
1198 */
1199 if (VOP_GETVOBJECT(vp, &object) == 0) {
1200 VM_OBJECT_LOCK(object);
1201 vm_object_page_remove(object, 0, 0,
1202 (flags & V_SAVE) ? TRUE : FALSE);
1203 VM_OBJECT_UNLOCK(object);
1204 }
1205
1206#ifdef INVARIANTS
1207 VI_LOCK(vp);
1208 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1209 (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1210 !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1211 panic("vinvalbuf: flush failed");
1212 VI_UNLOCK(vp);
1213#endif
1214 return (0);
1215}
1216
1217/*
1218 * Flush out buffers on the specified list.
1219 *
1220 */
1221static int
1222flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1223 struct buf *blist;
1224 int flags;
1225 struct vnode *vp;
1226 int slpflag, slptimeo;
1227 int *errorp;
1228{
1229 struct buf *bp, *nbp;
1230 int found, error;
1231
1232 ASSERT_VI_LOCKED(vp, "flushbuflist");
1233
1234 for (found = 0, bp = blist; bp; bp = nbp) {
1235 nbp = TAILQ_NEXT(bp, b_vnbufs);
1236 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1237 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1238 continue;
1239 }
1240 found += 1;
1241 error = BUF_TIMELOCK(bp,
1242 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1243 "flushbuf", slpflag, slptimeo);
1244 if (error) {
1245 if (error != ENOLCK)
1246 *errorp = error;
1247 goto done;
1248 }
1249 /*
1250 * XXX Since there are no node locks for NFS, I
1251 * believe there is a slight chance that a delayed
1252 * write will occur while sleeping just above, so
1253 * check for it. Note that vfs_bio_awrite expects
1254 * buffers to reside on a queue, while BUF_WRITE and
1255 * brelse do not.
1256 */
1257 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1258 (flags & V_SAVE)) {
1259
1260 if (bp->b_vp == vp) {
1261 if (bp->b_flags & B_CLUSTEROK) {
1262 vfs_bio_awrite(bp);
1263 } else {
1264 bremfree(bp);
1265 bp->b_flags |= B_ASYNC;
1266 BUF_WRITE(bp);
1267 }
1268 } else {
1269 bremfree(bp);
1270 (void) BUF_WRITE(bp);
1271 }
1272 goto done;
1273 }
1274 bremfree(bp);
1275 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1276 bp->b_flags &= ~B_ASYNC;
1277 brelse(bp);
1278 VI_LOCK(vp);
1279 }
1280 return (found);
1281done:
1282 VI_LOCK(vp);
1283 return (found);
1284}
1285
1286/*
1287 * Truncate a file's buffer and pages to a specified length. This
1288 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1289 * sync activity.
1290 */
1291int
1292vtruncbuf(vp, cred, td, length, blksize)
1293 register struct vnode *vp;
1294 struct ucred *cred;
1295 struct thread *td;
1296 off_t length;
1297 int blksize;
1298{
1299 register struct buf *bp;
1300 struct buf *nbp;
1301 int s, anyfreed;
1302 int trunclbn;
1303
1304 /*
1305 * Round up to the *next* lbn.
1306 */
1307 trunclbn = (length + blksize - 1) / blksize;
1308
1309 s = splbio();
1310 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1311restart:
1312 VI_LOCK(vp);
1313 anyfreed = 1;
1314 for (;anyfreed;) {
1315 anyfreed = 0;
1316 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1317 nbp = TAILQ_NEXT(bp, b_vnbufs);
1318 if (bp->b_lblkno >= trunclbn) {
1319 if (BUF_LOCK(bp,
1320 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1321 VI_MTX(vp)) == ENOLCK)
1322 goto restart;
1323
1324 bremfree(bp);
1325 bp->b_flags |= (B_INVAL | B_RELBUF);
1326 bp->b_flags &= ~B_ASYNC;
1327 brelse(bp);
1328 anyfreed = 1;
1329
1330 if (nbp &&
1331 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1332 (nbp->b_vp != vp) ||
1333 (nbp->b_flags & B_DELWRI))) {
1334 goto restart;
1335 }
1336 VI_LOCK(vp);
1337 }
1338 }
1339
1340 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1341 nbp = TAILQ_NEXT(bp, b_vnbufs);
1342 if (bp->b_lblkno >= trunclbn) {
1343 if (BUF_LOCK(bp,
1344 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1345 VI_MTX(vp)) == ENOLCK)
1346 goto restart;
1347 bremfree(bp);
1348 bp->b_flags |= (B_INVAL | B_RELBUF);
1349 bp->b_flags &= ~B_ASYNC;
1350 brelse(bp);
1351 anyfreed = 1;
1352 if (nbp &&
1353 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1354 (nbp->b_vp != vp) ||
1355 (nbp->b_flags & B_DELWRI) == 0)) {
1356 goto restart;
1357 }
1358 VI_LOCK(vp);
1359 }
1360 }
1361 }
1362
1363 if (length > 0) {
1364restartsync:
1365 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1366 nbp = TAILQ_NEXT(bp, b_vnbufs);
1367 if (bp->b_lblkno > 0)
1368 continue;
1369 /*
1370 * Since we hold the vnode lock this should only
1371 * fail if we're racing with the buf daemon.
1372 */
1373 if (BUF_LOCK(bp,
1374 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1375 VI_MTX(vp)) == ENOLCK) {
1376 goto restart;
1377 }
1378 KASSERT((bp->b_flags & B_DELWRI),
1379 ("buf(%p) on dirty queue without DELWRI.", bp));
1380
1381 bremfree(bp);
1382 bawrite(bp);
1383 VI_LOCK(vp);
1384 goto restartsync;
1385 }
1386 }
1387
1388 while (vp->v_numoutput > 0) {
1389 vp->v_iflag |= VI_BWAIT;
1390 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1391 }
1392 VI_UNLOCK(vp);
1393 splx(s);
1394
1395 vnode_pager_setsize(vp, length);
1396
1397 return (0);
1398}
1399
1400/*
1401 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1402 * a vnode.
1403 *
1404 * NOTE: We have to deal with the special case of a background bitmap
1405 * buffer, a situation where two buffers will have the same logical
1406 * block offset. We want (1) only the foreground buffer to be accessed
1407 * in a lookup and (2) must differentiate between the foreground and
1408 * background buffer in the splay tree algorithm because the splay
1409 * tree cannot normally handle multiple entities with the same 'index'.
1410 * We accomplish this by adding differentiating flags to the splay tree's
1411 * numerical domain.
1412 */
1413static
1414struct buf *
1415buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1416{
1417 struct buf dummy;
1418 struct buf *lefttreemax, *righttreemin, *y;
1419
1420 if (root == NULL)
1421 return (NULL);
1422 lefttreemax = righttreemin = &dummy;
1423 for (;;) {
1424 if (lblkno < root->b_lblkno ||
1425 (lblkno == root->b_lblkno &&
1426 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1427 if ((y = root->b_left) == NULL)
1428 break;
1429 if (lblkno < y->b_lblkno) {
1430 /* Rotate right. */
1431 root->b_left = y->b_right;
1432 y->b_right = root;
1433 root = y;
1434 if ((y = root->b_left) == NULL)
1435 break;
1436 }
1437 /* Link into the new root's right tree. */
1438 righttreemin->b_left = root;
1439 righttreemin = root;
1440 } else if (lblkno > root->b_lblkno ||
1441 (lblkno == root->b_lblkno &&
1442 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1443 if ((y = root->b_right) == NULL)
1444 break;
1445 if (lblkno > y->b_lblkno) {
1446 /* Rotate left. */
1447 root->b_right = y->b_left;
1448 y->b_left = root;
1449 root = y;
1450 if ((y = root->b_right) == NULL)
1451 break;
1452 }
1453 /* Link into the new root's left tree. */
1454 lefttreemax->b_right = root;
1455 lefttreemax = root;
1456 } else {
1457 break;
1458 }
1459 root = y;
1460 }
1461 /* Assemble the new root. */
1462 lefttreemax->b_right = root->b_left;
1463 righttreemin->b_left = root->b_right;
1464 root->b_left = dummy.b_right;
1465 root->b_right = dummy.b_left;
1466 return (root);
1467}
1468
1469static
1470void
1471buf_vlist_remove(struct buf *bp)
1472{
1473 struct vnode *vp = bp->b_vp;
1474 struct buf *root;
1475
1476 ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1477 if (bp->b_xflags & BX_VNDIRTY) {
1478 if (bp != vp->v_dirtyblkroot) {
1479 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1480 KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1481 }
1482 if (bp->b_left == NULL) {
1483 root = bp->b_right;
1484 } else {
1485 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1486 root->b_right = bp->b_right;
1487 }
1488 vp->v_dirtyblkroot = root;
1489 TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1490 vp->v_dirtybufcnt--;
1491 } else {
1492 /* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1493 if (bp != vp->v_cleanblkroot) {
1494 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1495 KASSERT(root == bp, ("splay lookup failed during clean remove"));
1496 }
1497 if (bp->b_left == NULL) {
1498 root = bp->b_right;
1499 } else {
1500 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1501 root->b_right = bp->b_right;
1502 }
1503 vp->v_cleanblkroot = root;
1504 TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1505 vp->v_cleanbufcnt--;
1506 }
1507 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1508}
1509
1510/*
1511 * Add the buffer to the sorted clean or dirty block list using a
1512 * splay tree algorithm.
1513 *
1514 * NOTE: xflags is passed as a constant, optimizing this inline function!
1515 */
1516static
1517void
1518buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1519{
1520 struct buf *root;
1521
1522 ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1523 bp->b_xflags |= xflags;
1524 if (xflags & BX_VNDIRTY) {
1525 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1526 if (root == NULL) {
1527 bp->b_left = NULL;
1528 bp->b_right = NULL;
1529 TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1530 } else if (bp->b_lblkno < root->b_lblkno ||
1531 (bp->b_lblkno == root->b_lblkno &&
1532 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1533 bp->b_left = root->b_left;
1534 bp->b_right = root;
1535 root->b_left = NULL;
1536 TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1537 } else {
1538 bp->b_right = root->b_right;
1539 bp->b_left = root;
1540 root->b_right = NULL;
1541 TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1542 root, bp, b_vnbufs);
1543 }
1544 vp->v_dirtybufcnt++;
1545 vp->v_dirtyblkroot = bp;
1546 } else {
1547 /* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1548 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1549 if (root == NULL) {
1550 bp->b_left = NULL;
1551 bp->b_right = NULL;
1552 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1553 } else if (bp->b_lblkno < root->b_lblkno ||
1554 (bp->b_lblkno == root->b_lblkno &&
1555 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1556 bp->b_left = root->b_left;
1557 bp->b_right = root;
1558 root->b_left = NULL;
1559 TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1560 } else {
1561 bp->b_right = root->b_right;
1562 bp->b_left = root;
1563 root->b_right = NULL;
1564 TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1565 root, bp, b_vnbufs);
1566 }
1567 vp->v_cleanbufcnt++;
1568 vp->v_cleanblkroot = bp;
1569 }
1570}
1571
1572/*
1573 * Lookup a buffer using the splay tree. Note that we specifically avoid
1574 * shadow buffers used in background bitmap writes.
1575 *
1576 * This code isn't quite efficient as it could be because we are maintaining
1577 * two sorted lists and do not know which list the block resides in.
1578 *
1579 * During a "make buildworld" the desired buffer is found at one of
1580 * the roots more than 60% of the time. Thus, checking both roots
1581 * before performing either splay eliminates unnecessary splays on the
1582 * first tree splayed.
1583 */
1584struct buf *
1585gbincore(struct vnode *vp, daddr_t lblkno)
1586{
1587 struct buf *bp;
1588
1589 GIANT_REQUIRED;
1590
1591 ASSERT_VI_LOCKED(vp, "gbincore");
1592 if ((bp = vp->v_cleanblkroot) != NULL &&
1593 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1594 return (bp);
1595 if ((bp = vp->v_dirtyblkroot) != NULL &&
1596 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1597 return (bp);
1598 if ((bp = vp->v_cleanblkroot) != NULL) {
1599 vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
1600 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1601 return (bp);
1602 }
1603 if ((bp = vp->v_dirtyblkroot) != NULL) {
1604 vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
1605 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1606 return (bp);
1607 }
1608 return (NULL);
1609}
1610
1611/*
1612 * Associate a buffer with a vnode.
1613 */
1614void
1615bgetvp(vp, bp)
1616 register struct vnode *vp;
1617 register struct buf *bp;
1618{
1619 int s;
1620
1621 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1622
1623 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1624 ("bgetvp: bp already attached! %p", bp));
1625
1626 ASSERT_VI_LOCKED(vp, "bgetvp");
1627 vholdl(vp);
1628 bp->b_vp = vp;
1629 bp->b_dev = vn_todev(vp);
1630 /*
1631 * Insert onto list for new vnode.
1632 */
1633 s = splbio();
1634 buf_vlist_add(bp, vp, BX_VNCLEAN);
1635 splx(s);
1636}
1637
1638/*
1639 * Disassociate a buffer from a vnode.
1640 */
1641void
1642brelvp(bp)
1643 register struct buf *bp;
1644{
1645 struct vnode *vp;
1646 int s;
1647
1648 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1649
1650 /*
1651 * Delete from old vnode list, if on one.
1652 */
1653 vp = bp->b_vp;
1654 s = splbio();
1655 VI_LOCK(vp);
1656 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1657 buf_vlist_remove(bp);
1658 if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1659 vp->v_iflag &= ~VI_ONWORKLST;
1660 mtx_lock(&sync_mtx);
1661 LIST_REMOVE(vp, v_synclist);
1662 mtx_unlock(&sync_mtx);
1663 }
1664 vdropl(vp);
1665 VI_UNLOCK(vp);
1666 bp->b_vp = (struct vnode *) 0;
1667 if (bp->b_object)
1668 bp->b_object = NULL;
1669 splx(s);
1670}
1671
1672/*
1673 * Add an item to the syncer work queue.
1674 */
1675static void
1676vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1677{
1678 int s, slot;
1679
1680 s = splbio();
1681 ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1682
1683 mtx_lock(&sync_mtx);
1684 if (vp->v_iflag & VI_ONWORKLST)
1685 LIST_REMOVE(vp, v_synclist);
1686 else
1687 vp->v_iflag |= VI_ONWORKLST;
1688
1689 if (delay > syncer_maxdelay - 2)
1690 delay = syncer_maxdelay - 2;
1691 slot = (syncer_delayno + delay) & syncer_mask;
1692
1693 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1694 mtx_unlock(&sync_mtx);
1695
1696 splx(s);
1697}
1698
1699struct proc *updateproc;
1700static void sched_sync(void);
1701static struct kproc_desc up_kp = {
1702 "syncer",
1703 sched_sync,
1704 &updateproc
1705};
1706SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1707
1708/*
1709 * System filesystem synchronizer daemon.
1710 */
1711static void
1712sched_sync(void)
1713{
1714 struct synclist *slp;
1715 struct vnode *vp;
1716 struct mount *mp;
1717 long starttime;
1718 int s;
1719 struct thread *td = FIRST_THREAD_IN_PROC(updateproc); /* XXXKSE */
1720
1721 mtx_lock(&Giant);
1722
1723 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1724 SHUTDOWN_PRI_LAST);
1725
1726 for (;;) {
1727 kthread_suspend_check(td->td_proc);
1728
1729 starttime = time_second;
1730
1731 /*
1732 * Push files whose dirty time has expired. Be careful
1733 * of interrupt race on slp queue.
1734 */
1735 s = splbio();
1736 mtx_lock(&sync_mtx);
1737 slp = &syncer_workitem_pending[syncer_delayno];
1738 syncer_delayno += 1;
1739 if (syncer_delayno == syncer_maxdelay)
1740 syncer_delayno = 0;
1741 splx(s);
1742
1743 while ((vp = LIST_FIRST(slp)) != NULL) {
1744 mtx_unlock(&sync_mtx);
1745 if (VOP_ISLOCKED(vp, NULL) == 0 &&
1746 vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1747 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1748 (void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1749 VOP_UNLOCK(vp, 0, td);
1750 vn_finished_write(mp);
1751 }
1752 s = splbio();
1753 mtx_lock(&sync_mtx);
1754 if (LIST_FIRST(slp) == vp) {
1755 mtx_unlock(&sync_mtx);
1756 /*
1757 * Note: VFS vnodes can remain on the
1758 * worklist too with no dirty blocks, but
1759 * since sync_fsync() moves it to a different
1760 * slot we are safe.
1761 */
1762 VI_LOCK(vp);
1763 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1764 !vn_isdisk(vp, NULL)) {
1765 panic("sched_sync: fsync failed "
1766 "vp %p tag %s", vp, vp->v_tag);
1767 }
1768 /*
1769 * Put us back on the worklist. The worklist
1770 * routine will remove us from our current
1771 * position and then add us back in at a later
1772 * position.
1773 */
1774 vn_syncer_add_to_worklist(vp, syncdelay);
1775 VI_UNLOCK(vp);
1776 mtx_lock(&sync_mtx);
1777 }
1778 splx(s);
1779 }
1780 mtx_unlock(&sync_mtx);
1781
1782 /*
1783 * Do soft update processing.
1784 */
1785 if (softdep_process_worklist_hook != NULL)
1786 (*softdep_process_worklist_hook)(NULL);
1787
1788 /*
1789 * The variable rushjob allows the kernel to speed up the
1790 * processing of the filesystem syncer process. A rushjob
1791 * value of N tells the filesystem syncer to process the next
1792 * N seconds worth of work on its queue ASAP. Currently rushjob
1793 * is used by the soft update code to speed up the filesystem
1794 * syncer process when the incore state is getting so far
1795 * ahead of the disk that the kernel memory pool is being
1796 * threatened with exhaustion.
1797 */
1798 mtx_lock(&sync_mtx);
1799 if (rushjob > 0) {
1800 rushjob -= 1;
1801 mtx_unlock(&sync_mtx);
1802 continue;
1803 }
1804 mtx_unlock(&sync_mtx);
1805 /*
1806 * If it has taken us less than a second to process the
1807 * current work, then wait. Otherwise start right over
1808 * again. We can still lose time if any single round
1809 * takes more than two seconds, but it does not really
1810 * matter as we are just trying to generally pace the
1811 * filesystem activity.
1812 */
1813 if (time_second == starttime)
1814 tsleep(&lbolt, PPAUSE, "syncer", 0);
1815 }
1816}
1817
1818/*
1819 * Request the syncer daemon to speed up its work.
1820 * We never push it to speed up more than half of its
1821 * normal turn time, otherwise it could take over the cpu.
1822 * XXXKSE only one update?
1823 */
1824int
1825speedup_syncer()
1826{
1827 struct thread *td;
1828 int ret = 0;
1829
1830 td = FIRST_THREAD_IN_PROC(updateproc);
1831 mtx_lock_spin(&sched_lock);
1832 if (td->td_wchan == &lbolt) {
1833 unsleep(td);
1834 TD_CLR_SLEEPING(td);
1835 setrunnable(td);
1836 }
1837 mtx_unlock_spin(&sched_lock);
1838 mtx_lock(&sync_mtx);
1839 if (rushjob < syncdelay / 2) {
1840 rushjob += 1;
1841 stat_rush_requests += 1;
1842 ret = 1;
1843 }
1844 mtx_unlock(&sync_mtx);
1845 return (ret);
1846}
1847
1848/*
1849 * Associate a p-buffer with a vnode.
1850 *
1851 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1852 * with the buffer. i.e. the bp has not been linked into the vnode or
1853 * ref-counted.
1854 */
1855void
1856pbgetvp(vp, bp)
1857 register struct vnode *vp;
1858 register struct buf *bp;
1859{
1860
1861 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1862
1863 bp->b_vp = vp;
1864 bp->b_flags |= B_PAGING;
1865 bp->b_dev = vn_todev(vp);
1866}
1867
1868/*
1869 * Disassociate a p-buffer from a vnode.
1870 */
1871void
1872pbrelvp(bp)
1873 register struct buf *bp;
1874{
1875
1876 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1877
1878 /* XXX REMOVE ME */
1879 VI_LOCK(bp->b_vp);
1880 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1881 panic(
1882 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1883 bp,
1884 (int)bp->b_flags
1885 );
1886 }
1887 VI_UNLOCK(bp->b_vp);
1888 bp->b_vp = (struct vnode *) 0;
1889 bp->b_flags &= ~B_PAGING;
1890}
1891
1892/*
1893 * Reassign a buffer from one vnode to another.
1894 * Used to assign file specific control information
1895 * (indirect blocks) to the vnode to which they belong.
1896 */
1897void
1898reassignbuf(bp, newvp)
1899 register struct buf *bp;
1900 register struct vnode *newvp;
1901{
1902 int delay;
1903 int s;
1904
1905 if (newvp == NULL) {
1906 printf("reassignbuf: NULL");
1907 return;
1908 }
1909 ++reassignbufcalls;
1910
1911 /*
1912 * B_PAGING flagged buffers cannot be reassigned because their vp
1913 * is not fully linked in.
1914 */
1915 if (bp->b_flags & B_PAGING)
1916 panic("cannot reassign paging buffer");
1917
1918 s = splbio();
1919 /*
1920 * Delete from old vnode list, if on one.
1921 */
1922 VI_LOCK(bp->b_vp);
1923 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1924 buf_vlist_remove(bp);
1925 if (bp->b_vp != newvp) {
1926 vdropl(bp->b_vp);
1927 bp->b_vp = NULL; /* for clarification */
1928 }
1929 }
1930 VI_UNLOCK(bp->b_vp);
1931 /*
1932 * If dirty, put on list of dirty buffers; otherwise insert onto list
1933 * of clean buffers.
1934 */
1935 VI_LOCK(newvp);
1936 if (bp->b_flags & B_DELWRI) {
1937 if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1938 switch (newvp->v_type) {
1939 case VDIR:
1940 delay = dirdelay;
1941 break;
1942 case VCHR:
1943 if (newvp->v_rdev->si_mountpoint != NULL) {
1944 delay = metadelay;
1945 break;
1946 }
1947 /* FALLTHROUGH */
1948 default:
1949 delay = filedelay;
1950 }
1951 vn_syncer_add_to_worklist(newvp, delay);
1952 }
1953 buf_vlist_add(bp, newvp, BX_VNDIRTY);
1954 } else {
1955 buf_vlist_add(bp, newvp, BX_VNCLEAN);
1956
1957 if ((newvp->v_iflag & VI_ONWORKLST) &&
1958 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1959 mtx_lock(&sync_mtx);
1960 LIST_REMOVE(newvp, v_synclist);
1961 mtx_unlock(&sync_mtx);
1962 newvp->v_iflag &= ~VI_ONWORKLST;
1963 }
1964 }
1965 if (bp->b_vp != newvp) {
1966 bp->b_vp = newvp;
1967 vholdl(bp->b_vp);
1968 }
1969 VI_UNLOCK(newvp);
1970 splx(s);
1971}
1972
1973/*
1974 * Create a vnode for a device.
1975 * Used for mounting the root filesystem.
1976 */
1977int
1978bdevvp(dev, vpp)
1979 dev_t dev;
1980 struct vnode **vpp;
1981{
1982 register struct vnode *vp;
1983 struct vnode *nvp;
1984 int error;
1985
1986 if (dev == NODEV) {
1987 *vpp = NULLVP;
1988 return (ENXIO);
1989 }
1990 if (vfinddev(dev, VCHR, vpp))
1991 return (0);
1992 error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1993 if (error) {
1994 *vpp = NULLVP;
1995 return (error);
1996 }
1997 vp = nvp;
1998 vp->v_type = VCHR;
1999 addalias(vp, dev);
2000 *vpp = vp;
2001 return (0);
2002}
2003
2004static void
2005v_incr_usecount(struct vnode *vp, int delta)
2006{
2007 vp->v_usecount += delta;
2008 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2009 mtx_lock(&spechash_mtx);
2010 vp->v_rdev->si_usecount += delta;
2011 mtx_unlock(&spechash_mtx);
2012 }
2013}
2014
2015/*
2016 * Add vnode to the alias list hung off the dev_t.
2017 *
2018 * The reason for this gunk is that multiple vnodes can reference
2019 * the same physical device, so checking vp->v_usecount to see
2020 * how many users there are is inadequate; the v_usecount for
2021 * the vnodes need to be accumulated. vcount() does that.
2022 */
2023struct vnode *
2024addaliasu(nvp, nvp_rdev)
2025 struct vnode *nvp;
2026 udev_t nvp_rdev;
2027{
2028 struct vnode *ovp;
2029 vop_t **ops;
2030 dev_t dev;
2031
2032 if (nvp->v_type == VBLK)
2033 return (nvp);
2034 if (nvp->v_type != VCHR)
2035 panic("addaliasu on non-special vnode");
2036 dev = udev2dev(nvp_rdev, 0);
2037 /*
2038 * Check to see if we have a bdevvp vnode with no associated
2039 * filesystem. If so, we want to associate the filesystem of
2040 * the new newly instigated vnode with the bdevvp vnode and
2041 * discard the newly created vnode rather than leaving the
2042 * bdevvp vnode lying around with no associated filesystem.
2043 */
2044 if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
2045 addalias(nvp, dev);
2046 return (nvp);
2047 }
2048 /*
2049 * Discard unneeded vnode, but save its node specific data.
2050 * Note that if there is a lock, it is carried over in the
2051 * node specific data to the replacement vnode.
2052 */
2053 vref(ovp);
2054 ovp->v_data = nvp->v_data;
2055 ovp->v_tag = nvp->v_tag;
2056 nvp->v_data = NULL;
2057 lockdestroy(ovp->v_vnlock);
2058 lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2059 nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2060 ops = ovp->v_op;
2061 ovp->v_op = nvp->v_op;
2062 if (VOP_ISLOCKED(nvp, curthread)) {
2063 VOP_UNLOCK(nvp, 0, curthread);
2064 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2065 }
2066 nvp->v_op = ops;
2067 insmntque(ovp, nvp->v_mount);
2068 vrele(nvp);
2069 vgone(nvp);
2070 return (ovp);
2071}
2072
2073/* This is a local helper function that do the same as addaliasu, but for a
2074 * dev_t instead of an udev_t. */
2075static void
2076addalias(nvp, dev)
2077 struct vnode *nvp;
2078 dev_t dev;
2079{
2080
2081 KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2082 nvp->v_rdev = dev;
2083 VI_LOCK(nvp);
2084 mtx_lock(&spechash_mtx);
2085 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2086 dev->si_usecount += nvp->v_usecount;
2087 mtx_unlock(&spechash_mtx);
2088 VI_UNLOCK(nvp);
2089}
2090
2091/*
2092 * Grab a particular vnode from the free list, increment its
2093 * reference count and lock it. The vnode lock bit is set if the
2094 * vnode is being eliminated in vgone. The process is awakened
2095 * when the transition is completed, and an error returned to
2096 * indicate that the vnode is no longer usable (possibly having
2097 * been changed to a new filesystem type).
2098 */
2099int
2100vget(vp, flags, td)
2101 register struct vnode *vp;
2102 int flags;
2103 struct thread *td;
2104{
2105 int error;
2106
2107 /*
2108 * If the vnode is in the process of being cleaned out for
2109 * another use, we wait for the cleaning to finish and then
2110 * return failure. Cleaning is determined by checking that
2111 * the VI_XLOCK flag is set.
2112 */
2113 if ((flags & LK_INTERLOCK) == 0)
2114 VI_LOCK(vp);
2115 if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2116 vp->v_iflag |= VI_XWANT;
2117 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2118 return (ENOENT);
2119 }
2120
2121 v_incr_usecount(vp, 1);
2122
2123 if (VSHOULDBUSY(vp))
2124 vbusy(vp);
2125 if (flags & LK_TYPE_MASK) {
2126 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2127 /*
2128 * must expand vrele here because we do not want
2129 * to call VOP_INACTIVE if the reference count
2130 * drops back to zero since it was never really
2131 * active. We must remove it from the free list
2132 * before sleeping so that multiple processes do
2133 * not try to recycle it.
2134 */
2135 VI_LOCK(vp);
2136 v_incr_usecount(vp, -1);
2137 if (VSHOULDFREE(vp))
2138 vfree(vp);
2139 else
2140 vlruvp(vp);
2141 VI_UNLOCK(vp);
2142 }
2143 return (error);
2144 }
2145 VI_UNLOCK(vp);
2146 return (0);
2147}
2148
2149/*
2150 * Increase the reference count of a vnode.
2151 */
2152void
2153vref(struct vnode *vp)
2154{
2155 VI_LOCK(vp);
2156 v_incr_usecount(vp, 1);
2157 VI_UNLOCK(vp);
2158}
2159
2160/*
2161 * Return reference count of a vnode.
2162 *
2163 * The results of this call are only guaranteed when some mechanism other
2164 * than the VI lock is used to stop other processes from gaining references
2165 * to the vnode. This may be the case if the caller holds the only reference.
2166 * This is also useful when stale data is acceptable as race conditions may
2167 * be accounted for by some other means.
2168 */
2169int
2170vrefcnt(struct vnode *vp)
2171{
2172 int usecnt;
2173
2174 VI_LOCK(vp);
2175 usecnt = vp->v_usecount;
2176 VI_UNLOCK(vp);
2177
2178 return (usecnt);
2179}
2180
2181
2182/*
2183 * Vnode put/release.
2184 * If count drops to zero, call inactive routine and return to freelist.
2185 */
2186void
2187vrele(vp)
2188 struct vnode *vp;
2189{
2190 struct thread *td = curthread; /* XXX */
2191
2192 KASSERT(vp != NULL, ("vrele: null vp"));
2193
2194 VI_LOCK(vp);
2195
2196 /* Skip this v_writecount check if we're going to panic below. */
2197 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2198 ("vrele: missed vn_close"));
2199
2200 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2201 vp->v_usecount == 1)) {
2202 v_incr_usecount(vp, -1);
2203 VI_UNLOCK(vp);
2204
2205 return;
2206 }
2207
2208 if (vp->v_usecount == 1) {
2209 v_incr_usecount(vp, -1);
2210 /*
2211 * We must call VOP_INACTIVE with the node locked. Mark
2212 * as VI_DOINGINACT to avoid recursion.
2213 */
2214 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2215 VI_LOCK(vp);
2216 vp->v_iflag |= VI_DOINGINACT;
2217 VI_UNLOCK(vp);
2218 VOP_INACTIVE(vp, td);
2219 VI_LOCK(vp);
2220 KASSERT(vp->v_iflag & VI_DOINGINACT,
2221 ("vrele: lost VI_DOINGINACT"));
2222 vp->v_iflag &= ~VI_DOINGINACT;
2223 VI_UNLOCK(vp);
2224 }
2225 VI_LOCK(vp);
2226 if (VSHOULDFREE(vp))
2227 vfree(vp);
2228 else
2229 vlruvp(vp);
2230 VI_UNLOCK(vp);
2231
2232 } else {
2233#ifdef DIAGNOSTIC
2234 vprint("vrele: negative ref count", vp);
2235#endif
2236 VI_UNLOCK(vp);
2237 panic("vrele: negative ref cnt");
2238 }
2239}
2240
2241/*
2242 * Release an already locked vnode. This give the same effects as
2243 * unlock+vrele(), but takes less time and avoids releasing and
2244 * re-aquiring the lock (as vrele() aquires the lock internally.)
2245 */
2246void
2247vput(vp)
2248 struct vnode *vp;
2249{
2250 struct thread *td = curthread; /* XXX */
2251
2252 GIANT_REQUIRED;
2253
2254 KASSERT(vp != NULL, ("vput: null vp"));
2255 VI_LOCK(vp);
2256 /* Skip this v_writecount check if we're going to panic below. */
2257 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2258 ("vput: missed vn_close"));
2259
2260 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2261 vp->v_usecount == 1)) {
2262 v_incr_usecount(vp, -1);
2263 VOP_UNLOCK(vp, LK_INTERLOCK, td);
2264 return;
2265 }
2266
2267 if (vp->v_usecount == 1) {
2268 v_incr_usecount(vp, -1);
2269 /*
2270 * We must call VOP_INACTIVE with the node locked, so
2271 * we just need to release the vnode mutex. Mark as
2272 * as VI_DOINGINACT to avoid recursion.
2273 */
2274 vp->v_iflag |= VI_DOINGINACT;
2275 VI_UNLOCK(vp);
2276 VOP_INACTIVE(vp, td);
2277 VI_LOCK(vp);
2278 KASSERT(vp->v_iflag & VI_DOINGINACT,
2279 ("vput: lost VI_DOINGINACT"));
2280 vp->v_iflag &= ~VI_DOINGINACT;
2281 if (VSHOULDFREE(vp))
2282 vfree(vp);
2283 else
2284 vlruvp(vp);
2285 VI_UNLOCK(vp);
2286
2287 } else {
2288#ifdef DIAGNOSTIC
2289 vprint("vput: negative ref count", vp);
2290#endif
2291 panic("vput: negative ref cnt");
2292 }
2293}
2294
2295/*
2296 * Somebody doesn't want the vnode recycled.
2297 */
2298void
2299vhold(struct vnode *vp)
2300{
2301 VI_LOCK(vp);
2302 vholdl(vp);
2303 VI_UNLOCK(vp);
2304}
2305
2306void
2307vholdl(vp)
2308 register struct vnode *vp;
2309{
2310 int s;
2311
2312 s = splbio();
2313 vp->v_holdcnt++;
2314 if (VSHOULDBUSY(vp))
2315 vbusy(vp);
2316 splx(s);
2317}
2318
2319/*
2320 * Note that there is one less who cares about this vnode. vdrop() is the
2321 * opposite of vhold().
2322 */
2323void
2324vdrop(struct vnode *vp)
2325{
2326 VI_LOCK(vp);
2327 vdropl(vp);
2328 VI_UNLOCK(vp);
2329}
2330
2331void
2332vdropl(vp)
2333 register struct vnode *vp;
2334{
2335 int s;
2336
2337 s = splbio();
2338 if (vp->v_holdcnt <= 0)
2339 panic("vdrop: holdcnt");
2340 vp->v_holdcnt--;
2341 if (VSHOULDFREE(vp))
2342 vfree(vp);
2343 else
2344 vlruvp(vp);
2345 splx(s);
2346}
2347
2348/*
2349 * Remove any vnodes in the vnode table belonging to mount point mp.
2350 *
2351 * If FORCECLOSE is not specified, there should not be any active ones,
2352 * return error if any are found (nb: this is a user error, not a
2353 * system error). If FORCECLOSE is specified, detach any active vnodes
2354 * that are found.
2355 *
2356 * If WRITECLOSE is set, only flush out regular file vnodes open for
2357 * writing.
2358 *
2359 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2360 *
2361 * `rootrefs' specifies the base reference count for the root vnode
2362 * of this filesystem. The root vnode is considered busy if its
2363 * v_usecount exceeds this value. On a successful return, vflush()
2364 * will call vrele() on the root vnode exactly rootrefs times.
2365 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2366 * be zero.
2367 */
2368#ifdef DIAGNOSTIC
2369static int busyprt = 0; /* print out busy vnodes */
2370SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2371#endif
2372
2373int
2374vflush(mp, rootrefs, flags)
2375 struct mount *mp;
2376 int rootrefs;
2377 int flags;
2378{
2379 struct thread *td = curthread; /* XXX */
2380 struct vnode *vp, *nvp, *rootvp = NULL;
2381 struct vattr vattr;
2382 int busy = 0, error;
2383
2384 if (rootrefs > 0) {
2385 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2386 ("vflush: bad args"));
2387 /*
2388 * Get the filesystem root vnode. We can vput() it
2389 * immediately, since with rootrefs > 0, it won't go away.
2390 */
2391 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2392 return (error);
2393 vput(rootvp);
2394
2395 }
2396 mtx_lock(&mntvnode_mtx);
2397loop:
2398 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2399 /*
2400 * Make sure this vnode wasn't reclaimed in getnewvnode().
2401 * Start over if it has (it won't be on the list anymore).
2402 */
2403 if (vp->v_mount != mp)
2404 goto loop;
2405 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2406
2407 VI_LOCK(vp);
2408 mtx_unlock(&mntvnode_mtx);
2409 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
2410 /*
2411 * This vnode could have been reclaimed while we were
2412 * waiting for the lock since we are not holding a
2413 * reference.
2414 * Start over if the vnode was reclaimed.
2415 */
2416 if (vp->v_mount != mp) {
2417 VOP_UNLOCK(vp, 0, td);
2418 mtx_lock(&mntvnode_mtx);
2419 goto loop;
2420 }
2421 /*
2422 * Skip over a vnodes marked VV_SYSTEM.
2423 */
2424 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2425 VOP_UNLOCK(vp, 0, td);
2426 mtx_lock(&mntvnode_mtx);
2427 continue;
2428 }
2429 /*
2430 * If WRITECLOSE is set, flush out unlinked but still open
2431 * files (even if open only for reading) and regular file
2432 * vnodes open for writing.
2433 */
2434 if (flags & WRITECLOSE) {
2435 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2436 VI_LOCK(vp);
2437
2438 if ((vp->v_type == VNON ||
2439 (error == 0 && vattr.va_nlink > 0)) &&
2440 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2441 VOP_UNLOCK(vp, LK_INTERLOCK, td);
2442 mtx_lock(&mntvnode_mtx);
2443 continue;
2444 }
2445 } else
2446 VI_LOCK(vp);
2447
2448 VOP_UNLOCK(vp, 0, td);
2449
2450 /*
2451 * With v_usecount == 0, all we need to do is clear out the
2452 * vnode data structures and we are done.
2453 */
2454 if (vp->v_usecount == 0) {
2455 vgonel(vp, td);
2456 mtx_lock(&mntvnode_mtx);
2457 continue;
2458 }
2459
2460 /*
2461 * If FORCECLOSE is set, forcibly close the vnode. For block
2462 * or character devices, revert to an anonymous device. For
2463 * all other files, just kill them.
2464 */
2465 if (flags & FORCECLOSE) {
2466 if (vp->v_type != VCHR) {
2467 vgonel(vp, td);
2468 } else {
2469 vclean(vp, 0, td);
2470 VI_UNLOCK(vp);
2471 vp->v_op = spec_vnodeop_p;
2472 insmntque(vp, (struct mount *) 0);
2473 }
2474 mtx_lock(&mntvnode_mtx);
2475 continue;
2476 }
2477#ifdef DIAGNOSTIC
2478 if (busyprt)
2479 vprint("vflush: busy vnode", vp);
2480#endif
2481 VI_UNLOCK(vp);
2482 mtx_lock(&mntvnode_mtx);
2483 busy++;
2484 }
2485 mtx_unlock(&mntvnode_mtx);
2486 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2487 /*
2488 * If just the root vnode is busy, and if its refcount
2489 * is equal to `rootrefs', then go ahead and kill it.
2490 */
2491 VI_LOCK(rootvp);
2492 KASSERT(busy > 0, ("vflush: not busy"));
2493 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2494 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2495 vgonel(rootvp, td);
2496 busy = 0;
2497 } else
2498 VI_UNLOCK(rootvp);
2499 }
2500 if (busy)
2501 return (EBUSY);
2502 for (; rootrefs > 0; rootrefs--)
2503 vrele(rootvp);
2504 return (0);
2505}
2506
2507/*
2508 * This moves a now (likely recyclable) vnode to the end of the
2509 * mountlist. XXX However, it is temporarily disabled until we
2510 * can clean up ffs_sync() and friends, which have loop restart
2511 * conditions which this code causes to operate O(N^2).
2512 */
2513static void
2514vlruvp(struct vnode *vp)
2515{
2516#if 0
2517 struct mount *mp;
2518
2519 if ((mp = vp->v_mount) != NULL) {
2520 mtx_lock(&mntvnode_mtx);
2521 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2522 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2523 mtx_unlock(&mntvnode_mtx);
2524 }
2525#endif
2526}
2527
2528/*
2529 * Disassociate the underlying filesystem from a vnode.
2530 */
2531static void
2532vclean(vp, flags, td)
2533 struct vnode *vp;
2534 int flags;
2535 struct thread *td;
2536{
2537 int active;
2538
2539 ASSERT_VI_LOCKED(vp, "vclean");
2540 /*
2541 * Check to see if the vnode is in use. If so we have to reference it
2542 * before we clean it out so that its count cannot fall to zero and
2543 * generate a race against ourselves to recycle it.
2544 */
2545 if ((active = vp->v_usecount))
2546 v_incr_usecount(vp, 1);
2547
2548 /*
2549 * Prevent the vnode from being recycled or brought into use while we
2550 * clean it out.
2551 */
2552 if (vp->v_iflag & VI_XLOCK)
2553 panic("vclean: deadlock");
2554 vp->v_iflag |= VI_XLOCK;
2555 vp->v_vxproc = curthread;
2556 /*
2557 * Even if the count is zero, the VOP_INACTIVE routine may still
2558 * have the object locked while it cleans it out. The VOP_LOCK
2559 * ensures that the VOP_INACTIVE routine is done with its work.
2560 * For active vnodes, it ensures that no other activity can
2561 * occur while the underlying object is being cleaned out.
2562 */
2563 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2564
2565 /*
2566 * Clean out any buffers associated with the vnode.
2567 * If the flush fails, just toss the buffers.
2568 */
2569 if (flags & DOCLOSE) {
2570 struct buf *bp;
2571 VI_LOCK(vp);
2572 bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2573 VI_UNLOCK(vp);
2574 if (bp != NULL)
2575 (void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2576 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2577 vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2578 }
2579
2580 VOP_DESTROYVOBJECT(vp);
2581
2582 /*
2583 * Any other processes trying to obtain this lock must first
2584 * wait for VXLOCK to clear, then call the new lock operation.
2585 */
2586 VOP_UNLOCK(vp, 0, td);
2587
2588 /*
2589 * If purging an active vnode, it must be closed and
2590 * deactivated before being reclaimed. Note that the
2591 * VOP_INACTIVE will unlock the vnode.
2592 */
2593 if (active) {
2594 if (flags & DOCLOSE)
2595 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2596 VI_LOCK(vp);
2597 if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2598 vp->v_iflag |= VI_DOINGINACT;
2599 VI_UNLOCK(vp);
2600 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2601 panic("vclean: cannot relock.");
2602 VOP_INACTIVE(vp, td);
2603 VI_LOCK(vp);
2604 KASSERT(vp->v_iflag & VI_DOINGINACT,
2605 ("vclean: lost VI_DOINGINACT"));
2606 vp->v_iflag &= ~VI_DOINGINACT;
2607 }
2608 VI_UNLOCK(vp);
2609 }
2610
2611 /*
2612 * Reclaim the vnode.
2613 */
2614 if (VOP_RECLAIM(vp, td))
2615 panic("vclean: cannot reclaim");
2616
2617 if (active) {
2618 /*
2619 * Inline copy of vrele() since VOP_INACTIVE
2620 * has already been called.
2621 */
2622 VI_LOCK(vp);
2623 v_incr_usecount(vp, -1);
2624 if (vp->v_usecount <= 0) {
2625#ifdef DIAGNOSTIC
2626 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2627 vprint("vclean: bad ref count", vp);
2628 panic("vclean: ref cnt");
2629 }
2630#endif
2631 vfree(vp);
2632 }
2633 VI_UNLOCK(vp);
2634 }
2635
2636 cache_purge(vp);
2637 VI_LOCK(vp);
2638 if (VSHOULDFREE(vp))
2639 vfree(vp);
2640
2641 /*
2642 * Done with purge, reset to the standard lock and
2643 * notify sleepers of the grim news.
2644 */
2645 vp->v_vnlock = &vp->v_lock;
2646 vp->v_op = dead_vnodeop_p;
2647 if (vp->v_pollinfo != NULL)
2648 vn_pollgone(vp);
2649 vp->v_tag = "none";
2650 vp->v_iflag &= ~VI_XLOCK;
2651 vp->v_vxproc = NULL;
2652 if (vp->v_iflag & VI_XWANT) {
2653 vp->v_iflag &= ~VI_XWANT;
2654 wakeup(vp);
2655 }
2656}
2657
2658/*
2659 * Eliminate all activity associated with the requested vnode
2660 * and with all vnodes aliased to the requested vnode.
2661 */
2662int
2663vop_revoke(ap)
2664 struct vop_revoke_args /* {
2665 struct vnode *a_vp;
2666 int a_flags;
2667 } */ *ap;
2668{
2669 struct vnode *vp, *vq;
2670 dev_t dev;
2671
2672 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2673 vp = ap->a_vp;
2674 KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2675
2676 VI_LOCK(vp);
2677 /*
2678 * If a vgone (or vclean) is already in progress,
2679 * wait until it is done and return.
2680 */
2681 if (vp->v_iflag & VI_XLOCK) {
2682 vp->v_iflag |= VI_XWANT;
2683 msleep(vp, VI_MTX(vp), PINOD | PDROP,
2684 "vop_revokeall", 0);
2685 return (0);
2686 }
2687 VI_UNLOCK(vp);
2688 dev = vp->v_rdev;
2689 for (;;) {
2690 mtx_lock(&spechash_mtx);
2691 vq = SLIST_FIRST(&dev->si_hlist);
2692 mtx_unlock(&spechash_mtx);
2693 if (!vq)
2694 break;
2695 vgone(vq);
2696 }
2697 return (0);
2698}
2699
2700/*
2701 * Recycle an unused vnode to the front of the free list.
2702 * Release the passed interlock if the vnode will be recycled.
2703 */
2704int
2705vrecycle(vp, inter_lkp, td)
2706 struct vnode *vp;
2707 struct mtx *inter_lkp;
2708 struct thread *td;
2709{
2710
2711 VI_LOCK(vp);
2712 if (vp->v_usecount == 0) {
2713 if (inter_lkp) {
2714 mtx_unlock(inter_lkp);
2715 }
2716 vgonel(vp, td);
2717 return (1);
2718 }
2719 VI_UNLOCK(vp);
2720 return (0);
2721}
2722
2723/*
2724 * Eliminate all activity associated with a vnode
2725 * in preparation for reuse.
2726 */
2727void
2728vgone(vp)
2729 register struct vnode *vp;
2730{
2731 struct thread *td = curthread; /* XXX */
2732
2733 VI_LOCK(vp);
2734 vgonel(vp, td);
2735}
2736
2737/*
2738 * vgone, with the vp interlock held.
2739 */
2740void
2741vgonel(vp, td)
2742 struct vnode *vp;
2743 struct thread *td;
2744{
2745 int s;
2746
2747 /*
2748 * If a vgone (or vclean) is already in progress,
2749 * wait until it is done and return.
2750 */
2751 ASSERT_VI_LOCKED(vp, "vgonel");
2752 if (vp->v_iflag & VI_XLOCK) {
2753 vp->v_iflag |= VI_XWANT;
2754 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2755 return;
2756 }
2757
2758 /*
2759 * Clean out the filesystem specific data.
2760 */
2761 vclean(vp, DOCLOSE, td);
2762 VI_UNLOCK(vp);
2763
2764 /*
2765 * Delete from old mount point vnode list, if on one.
2766 */
2767 if (vp->v_mount != NULL)
2768 insmntque(vp, (struct mount *)0);
2769 /*
2770 * If special device, remove it from special device alias list
2771 * if it is on one.
2772 */
2773 if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2774 VI_LOCK(vp);
2775 mtx_lock(&spechash_mtx);
2776 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2777 vp->v_rdev->si_usecount -= vp->v_usecount;
2778 mtx_unlock(&spechash_mtx);
2779 VI_UNLOCK(vp);
2780 vp->v_rdev = NULL;
2781 }
2782
2783 /*
2784 * If it is on the freelist and not already at the head,
2785 * move it to the head of the list. The test of the
2786 * VDOOMED flag and the reference count of zero is because
2787 * it will be removed from the free list by getnewvnode,
2788 * but will not have its reference count incremented until
2789 * after calling vgone. If the reference count were
2790 * incremented first, vgone would (incorrectly) try to
2791 * close the previous instance of the underlying object.
2792 */
2793 VI_LOCK(vp);
2794 if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2795 s = splbio();
2796 mtx_lock(&vnode_free_list_mtx);
2797 if (vp->v_iflag & VI_FREE) {
2798 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2799 } else {
2800 vp->v_iflag |= VI_FREE;
2801 freevnodes++;
2802 }
2803 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2804 mtx_unlock(&vnode_free_list_mtx);
2805 splx(s);
2806 }
2807
2808 vp->v_type = VBAD;
2809 VI_UNLOCK(vp);
2810}
2811
2812/*
2813 * Lookup a vnode by device number.
2814 */
2815int
2816vfinddev(dev, type, vpp)
2817 dev_t dev;
2818 enum vtype type;
2819 struct vnode **vpp;
2820{
2821 struct vnode *vp;
2822
2823 mtx_lock(&spechash_mtx);
2824 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2825 if (type == vp->v_type) {
2826 *vpp = vp;
2827 mtx_unlock(&spechash_mtx);
2828 return (1);
2829 }
2830 }
2831 mtx_unlock(&spechash_mtx);
2832 return (0);
2833}
2834
2835/*
2836 * Calculate the total number of references to a special device.
2837 */
2838int
2839vcount(vp)
2840 struct vnode *vp;
2841{
2842 int count;
2843
2844 mtx_lock(&spechash_mtx);
2845 count = vp->v_rdev->si_usecount;
2846 mtx_unlock(&spechash_mtx);
2847 return (count);
2848}
2849
2850/*
2851 * Same as above, but using the dev_t as argument
2852 */
2853int
2854count_dev(dev)
2855 dev_t dev;
2856{
2857 struct vnode *vp;
2858
2859 vp = SLIST_FIRST(&dev->si_hlist);
2860 if (vp == NULL)
2861 return (0);
2862 return(vcount(vp));
2863}
2864
2865/*
2866 * Print out a description of a vnode.
2867 */
2868static char *typename[] =
2869{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2870
2871void
2872vprint(label, vp)
2873 char *label;
2874 struct vnode *vp;
2875{
2876 char buf[96];
2877
2878 if (label != NULL)
2879 printf("%s: %p: ", label, (void *)vp);
2880 else
2881 printf("%p: ", (void *)vp);
2882 printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2883 vp->v_tag, typename[vp->v_type], vp->v_usecount,
2884 vp->v_writecount, vp->v_holdcnt);
2885 buf[0] = '\0';
2886 if (vp->v_vflag & VV_ROOT)
2887 strcat(buf, "|VV_ROOT");
2888 if (vp->v_vflag & VV_TEXT)
2889 strcat(buf, "|VV_TEXT");
2890 if (vp->v_vflag & VV_SYSTEM)
2891 strcat(buf, "|VV_SYSTEM");
2892 if (vp->v_iflag & VI_XLOCK)
2893 strcat(buf, "|VI_XLOCK");
2894 if (vp->v_iflag & VI_XWANT)
2895 strcat(buf, "|VI_XWANT");
2896 if (vp->v_iflag & VI_BWAIT)
2897 strcat(buf, "|VI_BWAIT");
2898 if (vp->v_iflag & VI_DOOMED)
2899 strcat(buf, "|VI_DOOMED");
2900 if (vp->v_iflag & VI_FREE)
2901 strcat(buf, "|VI_FREE");
2902 if (vp->v_vflag & VV_OBJBUF)
2903 strcat(buf, "|VV_OBJBUF");
2904 if (buf[0] != '\0')
2905 printf(" flags (%s),", &buf[1]);
2906 lockmgr_printinfo(vp->v_vnlock);
2907 printf("\n");
2908 if (vp->v_data != NULL)
2909 VOP_PRINT(vp);
2910}
2911
2912#ifdef DDB
2913#include <ddb/ddb.h>
2914/*
2915 * List all of the locked vnodes in the system.
2916 * Called when debugging the kernel.
2917 */
2918DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2919{
2920 struct mount *mp, *nmp;
2921 struct vnode *vp;
2922
2923 /*
2924 * Note: because this is DDB, we can't obey the locking semantics
2925 * for these structures, which means we could catch an inconsistent
2926 * state and dereference a nasty pointer. Not much to be done
2927 * about that.
2928 */
2929 printf("Locked vnodes\n");
2930 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2931 nmp = TAILQ_NEXT(mp, mnt_list);
2932 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2933 if (VOP_ISLOCKED(vp, NULL))
2934 vprint(NULL, vp);
2935 }
2936 nmp = TAILQ_NEXT(mp, mnt_list);
2937 }
2938}
2939#endif
2940
2941/*
2942 * Fill in a struct xvfsconf based on a struct vfsconf.
2943 */
2944static void
2945vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2946{
2947
2948 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2949 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2950 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2951 xvfsp->vfc_flags = vfsp->vfc_flags;
2952 /*
2953 * These are unused in userland, we keep them
2954 * to not break binary compatibility.
2955 */
2956 xvfsp->vfc_vfsops = NULL;
2957 xvfsp->vfc_next = NULL;
2958}
2959
2960static int
2961sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2962{
2963 struct vfsconf *vfsp;
2964 struct xvfsconf *xvfsp;
2965 int cnt, error, i;
2966
2967 cnt = 0;
2968 for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2969 cnt++;
2970 xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2971 /*
2972 * Handle the race that we will have here when struct vfsconf
2973 * will be locked down by using both cnt and checking vfc_next
2974 * against NULL to determine the end of the loop. The race will
2975 * happen because we will have to unlock before calling malloc().
2976 * We are protected by Giant for now.
2977 */
2978 i = 0;
2979 for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2980 vfsconf2x(vfsp, xvfsp + i);
2981 i++;
2982 }
2983 error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2984 free(xvfsp, M_TEMP);
2985 return (error);
2986}
2987
2988SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2989 "S,xvfsconf", "List of all configured filesystems");
2990
2991/*
2992 * Top level filesystem related information gathering.
2993 */
2994static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2995
2996static int
2997vfs_sysctl(SYSCTL_HANDLER_ARGS)
2998{
2999 int *name = (int *)arg1 - 1; /* XXX */
3000 u_int namelen = arg2 + 1; /* XXX */
3001 struct vfsconf *vfsp;
3002 struct xvfsconf xvfsp;
3003
3004 printf("WARNING: userland calling deprecated sysctl, "
3005 "please rebuild world\n");
3006
3007#if 1 || defined(COMPAT_PRELITE2)
3008 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3009 if (namelen == 1)
3010 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3011#endif
3012
3013 switch (name[1]) {
3014 case VFS_MAXTYPENUM:
3015 if (namelen != 2)
3016 return (ENOTDIR);
3017 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3018 case VFS_CONF:
3019 if (namelen != 3)
3020 return (ENOTDIR); /* overloaded */
3021 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
3022 if (vfsp->vfc_typenum == name[2])
3023 break;
3024 if (vfsp == NULL)
3025 return (EOPNOTSUPP);
3026 vfsconf2x(vfsp, &xvfsp);
3027 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3028 }
3029 return (EOPNOTSUPP);
3030}
3031
3032SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
3033 "Generic filesystem");
3034
3035#if 1 || defined(COMPAT_PRELITE2)
3036
3037static int
3038sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3039{
3040 int error;
3041 struct vfsconf *vfsp;
3042 struct ovfsconf ovfs;
3043
3044 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
3045 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3046 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3047 ovfs.vfc_index = vfsp->vfc_typenum;
3048 ovfs.vfc_refcount = vfsp->vfc_refcount;
3049 ovfs.vfc_flags = vfsp->vfc_flags;
3050 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3051 if (error)
3052 return error;
3053 }
3054 return 0;
3055}
3056
3057#endif /* 1 || COMPAT_PRELITE2 */
3058
3059#define KINFO_VNODESLOP 10
3060#ifdef notyet
3061/*
3062 * Dump vnode list (via sysctl).
3063 */
3064/* ARGSUSED */
3065static int
3066sysctl_vnode(SYSCTL_HANDLER_ARGS)
3067{
3068 struct xvnode *xvn;
3069 struct thread *td = req->td;
3070 struct mount *mp;
3071 struct vnode *vp;
3072 int error, len, n;
3073
3074 /*
3075 * Stale numvnodes access is not fatal here.
3076 */
3077 req->lock = 0;
3078 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3079 if (!req->oldptr)
3080 /* Make an estimate */
3081 return (SYSCTL_OUT(req, 0, len));
3082
3083 sysctl_wire_old_buffer(req, 0);
3084 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3085 n = 0;
3086 mtx_lock(&mountlist_mtx);
3087 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3088 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3089 continue;
3090 mtx_lock(&mntvnode_mtx);
3091 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3092 if (n == len)
3093 break;
3094 vref(vp);
3095 xvn[n].xv_size = sizeof *xvn;
3096 xvn[n].xv_vnode = vp;
3097#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3098 XV_COPY(usecount);
3099 XV_COPY(writecount);
3100 XV_COPY(holdcnt);
3101 XV_COPY(id);
3102 XV_COPY(mount);
3103 XV_COPY(numoutput);
3104 XV_COPY(type);
3105#undef XV_COPY
3106 xvn[n].xv_flag = vp->v_vflag;
3107
3108 switch (vp->v_type) {
3109 case VREG:
3110 case VDIR:
3111 case VLNK:
3112 xvn[n].xv_dev = vp->v_cachedfs;
3113 xvn[n].xv_ino = vp->v_cachedid;
3114 break;
3115 case VBLK:
3116 case VCHR:
3117 if (vp->v_rdev == NULL) {
3118 vrele(vp);
3119 continue;
3120 }
3121 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3122 break;
3123 case VSOCK:
3124 xvn[n].xv_socket = vp->v_socket;
3125 break;
3126 case VFIFO:
3127 xvn[n].xv_fifo = vp->v_fifoinfo;
3128 break;
3129 case VNON:
3130 case VBAD:
3131 default:
3132 /* shouldn't happen? */
3133 vrele(vp);
3134 continue;
3135 }
3136 vrele(vp);
3137 ++n;
3138 }
3139 mtx_unlock(&mntvnode_mtx);
3140 mtx_lock(&mountlist_mtx);
3141 vfs_unbusy(mp, td);
3142 if (n == len)
3143 break;
3144 }
3145 mtx_unlock(&mountlist_mtx);
3146
3147 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3148 free(xvn, M_TEMP);
3149 return (error);
3150}
3151
3152SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3153 0, 0, sysctl_vnode, "S,xvnode", "");
3154#endif
3155
3156/*
3157 * Check to see if a filesystem is mounted on a block device.
3158 */
3159int
3160vfs_mountedon(vp)
3161 struct vnode *vp;
3162{
3163
3164 if (vp->v_rdev->si_mountpoint != NULL)
3165 return (EBUSY);
3166 return (0);
3167}
3168
3169/*
3170 * Unmount all filesystems. The list is traversed in reverse order
3171 * of mounting to avoid dependencies.
3172 */
3173void
3174vfs_unmountall()
3175{
3176 struct mount *mp;
3177 struct thread *td;
3178 int error;
3179
3180 if (curthread != NULL)
3181 td = curthread;
3182 else
3183 td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3184 /*
3185 * Since this only runs when rebooting, it is not interlocked.
3186 */
3187 while(!TAILQ_EMPTY(&mountlist)) {
3188 mp = TAILQ_LAST(&mountlist, mntlist);
3189 error = dounmount(mp, MNT_FORCE, td);
3190 if (error) {
3191 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3192 printf("unmount of %s failed (",
3193 mp->mnt_stat.f_mntonname);
3194 if (error == EBUSY)
3195 printf("BUSY)\n");
3196 else
3197 printf("%d)\n", error);
3198 } else {
3199 /* The unmount has removed mp from the mountlist */
3200 }
3201 }
3202}
3203
3204/*
3205 * perform msync on all vnodes under a mount point
3206 * the mount point must be locked.
3207 */
3208void
3209vfs_msync(struct mount *mp, int flags)
3210{
3211 struct vnode *vp, *nvp;
3212 struct vm_object *obj;
3213 int tries;
3214
3215 GIANT_REQUIRED;
3216
3217 tries = 5;
3218 mtx_lock(&mntvnode_mtx);
3219loop:
3220 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3221 if (vp->v_mount != mp) {
3222 if (--tries > 0)
3223 goto loop;
3224 break;
3225 }
3226 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3227
3228 VI_LOCK(vp);
3229 if (vp->v_iflag & VI_XLOCK) { /* XXX: what if MNT_WAIT? */
3230 VI_UNLOCK(vp);
3231 continue;
3232 }
3233
3234 if ((vp->v_iflag & VI_OBJDIRTY) &&
3235 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3236 mtx_unlock(&mntvnode_mtx);
3237 if (!vget(vp,
3238 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3239 curthread)) {
3240 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3241 vput(vp);
3242 mtx_lock(&mntvnode_mtx);
3243 continue;
3244 }
3245
3246 if (VOP_GETVOBJECT(vp, &obj) == 0) {
3247 VM_OBJECT_LOCK(obj);
3248 vm_object_page_clean(obj, 0, 0,
3249 flags == MNT_WAIT ?
3250 OBJPC_SYNC : OBJPC_NOSYNC);
3251 VM_OBJECT_UNLOCK(obj);
3252 }
3253 vput(vp);
3254 }
3255 mtx_lock(&mntvnode_mtx);
3256 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3257 if (--tries > 0)
3258 goto loop;
3259 break;
3260 }
3261 } else
3262 VI_UNLOCK(vp);
3263 }
3264 mtx_unlock(&mntvnode_mtx);
3265}
3266
3267/*
3268 * Create the VM object needed for VMIO and mmap support. This
3269 * is done for all VREG files in the system. Some filesystems might
3270 * afford the additional metadata buffering capability of the
3271 * VMIO code by making the device node be VMIO mode also.
3272 *
3273 * vp must be locked when vfs_object_create is called.
3274 */
3275int
3276vfs_object_create(vp, td, cred)
3277 struct vnode *vp;
3278 struct thread *td;
3279 struct ucred *cred;
3280{
3281 GIANT_REQUIRED;
3282 return (VOP_CREATEVOBJECT(vp, cred, td));
3283}
3284
3285/*
3286 * Mark a vnode as free, putting it up for recycling.
3287 */
3288void
3289vfree(vp)
3290 struct vnode *vp;
3291{
3292 int s;
3293
3294 ASSERT_VI_LOCKED(vp, "vfree");
3295 s = splbio();
3296 mtx_lock(&vnode_free_list_mtx);
3297 KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3298 if (vp->v_iflag & VI_AGE) {
3299 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3300 } else {
3301 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3302 }
3303 freevnodes++;
3304 mtx_unlock(&vnode_free_list_mtx);
3305 vp->v_iflag &= ~VI_AGE;
3306 vp->v_iflag |= VI_FREE;
3307 splx(s);
3308}
3309
3310/*
3311 * Opposite of vfree() - mark a vnode as in use.
3312 */
3313void
3314vbusy(vp)
3315 struct vnode *vp;
3316{
3317 int s;
3318
3319 s = splbio();
3320 ASSERT_VI_LOCKED(vp, "vbusy");
3321 KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3322
3323 mtx_lock(&vnode_free_list_mtx);
3324 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3325 freevnodes--;
3326 mtx_unlock(&vnode_free_list_mtx);
3327
3328 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3329 splx(s);
3330}
3331
3332/*
3333 * Record a process's interest in events which might happen to
3334 * a vnode. Because poll uses the historic select-style interface
3335 * internally, this routine serves as both the ``check for any
3336 * pending events'' and the ``record my interest in future events''
3337 * functions. (These are done together, while the lock is held,
3338 * to avoid race conditions.)
3339 */
3340int
3341vn_pollrecord(vp, td, events)
3342 struct vnode *vp;
3343 struct thread *td;
3344 short events;
3345{
3346
3347 if (vp->v_pollinfo == NULL)
3348 v_addpollinfo(vp);
3349 mtx_lock(&vp->v_pollinfo->vpi_lock);
3350 if (vp->v_pollinfo->vpi_revents & events) {
3351 /*
3352 * This leaves events we are not interested
3353 * in available for the other process which
3354 * which presumably had requested them
3355 * (otherwise they would never have been
3356 * recorded).
3357 */
3358 events &= vp->v_pollinfo->vpi_revents;
3359 vp->v_pollinfo->vpi_revents &= ~events;
3360
3361 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3362 return events;
3363 }
3364 vp->v_pollinfo->vpi_events |= events;
3365 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3366 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3367 return 0;
3368}
3369
3370/*
3371 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3372 * it is possible for us to miss an event due to race conditions, but
3373 * that condition is expected to be rare, so for the moment it is the
3374 * preferred interface.
3375 */
3376void
3377vn_pollevent(vp, events)
3378 struct vnode *vp;
3379 short events;
3380{
3381
3382 if (vp->v_pollinfo == NULL)
3383 v_addpollinfo(vp);
3384 mtx_lock(&vp->v_pollinfo->vpi_lock);
3385 if (vp->v_pollinfo->vpi_events & events) {
3386 /*
3387 * We clear vpi_events so that we don't
3388 * call selwakeup() twice if two events are
3389 * posted before the polling process(es) is
3390 * awakened. This also ensures that we take at
3391 * most one selwakeup() if the polling process
3392 * is no longer interested. However, it does
3393 * mean that only one event can be noticed at
3394 * a time. (Perhaps we should only clear those
3395 * event bits which we note?) XXX
3396 */
3397 vp->v_pollinfo->vpi_events = 0; /* &= ~events ??? */
3398 vp->v_pollinfo->vpi_revents |= events;
3399 selwakeup(&vp->v_pollinfo->vpi_selinfo);
3400 }
3401 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3402}
3403
3404/*
3405 * Wake up anyone polling on vp because it is being revoked.
3406 * This depends on dead_poll() returning POLLHUP for correct
3407 * behavior.
3408 */
3409void
3410vn_pollgone(vp)
3411 struct vnode *vp;
3412{
3413
3414 mtx_lock(&vp->v_pollinfo->vpi_lock);
3415 VN_KNOTE(vp, NOTE_REVOKE);
3416 if (vp->v_pollinfo->vpi_events) {
3417 vp->v_pollinfo->vpi_events = 0;
3418 selwakeup(&vp->v_pollinfo->vpi_selinfo);
3419 }
3420 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3421}
3422
3423
3424
3425/*
3426 * Routine to create and manage a filesystem syncer vnode.
3427 */
3428#define sync_close ((int (*)(struct vop_close_args *))nullop)
3429static int sync_fsync(struct vop_fsync_args *);
3430static int sync_inactive(struct vop_inactive_args *);
3431static int sync_reclaim(struct vop_reclaim_args *);
3432
3433static vop_t **sync_vnodeop_p;
3434static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3435 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3436 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3437 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3438 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3439 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3440 { &vop_lock_desc, (vop_t *) vop_stdlock }, /* lock */
3441 { &vop_unlock_desc, (vop_t *) vop_stdunlock }, /* unlock */
3442 { &vop_islocked_desc, (vop_t *) vop_stdislocked }, /* islocked */
3443 { NULL, NULL }
3444};
3445static struct vnodeopv_desc sync_vnodeop_opv_desc =
3446 { &sync_vnodeop_p, sync_vnodeop_entries };
3447
3448VNODEOP_SET(sync_vnodeop_opv_desc);
3449
3450/*
3451 * Create a new filesystem syncer vnode for the specified mount point.
3452 */
3453int
3454vfs_allocate_syncvnode(mp)
3455 struct mount *mp;
3456{
3457 struct vnode *vp;
3458 static long start, incr, next;
3459 int error;
3460
3461 /* Allocate a new vnode */
3462 if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3463 mp->mnt_syncer = NULL;
3464 return (error);
3465 }
3466 vp->v_type = VNON;
3467 /*
3468 * Place the vnode onto the syncer worklist. We attempt to
3469 * scatter them about on the list so that they will go off
3470 * at evenly distributed times even if all the filesystems
3471 * are mounted at once.
3472 */
3473 next += incr;
3474 if (next == 0 || next > syncer_maxdelay) {
3475 start /= 2;
3476 incr /= 2;
3477 if (start == 0) {
3478 start = syncer_maxdelay / 2;
3479 incr = syncer_maxdelay;
3480 }
3481 next = start;
3482 }
3483 VI_LOCK(vp);
3484 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3485 VI_UNLOCK(vp);
3486 mp->mnt_syncer = vp;
3487 return (0);
3488}
3489
3490/*
3491 * Do a lazy sync of the filesystem.
3492 */
3493static int
3494sync_fsync(ap)
3495 struct vop_fsync_args /* {
3496 struct vnode *a_vp;
3497 struct ucred *a_cred;
3498 int a_waitfor;
3499 struct thread *a_td;
3500 } */ *ap;
3501{
3502 struct vnode *syncvp = ap->a_vp;
3503 struct mount *mp = syncvp->v_mount;
3504 struct thread *td = ap->a_td;
3505 int error, asyncflag;
3506
3507 /*
3508 * We only need to do something if this is a lazy evaluation.
3509 */
3510 if (ap->a_waitfor != MNT_LAZY)
3511 return (0);
3512
3513 /*
3514 * Move ourselves to the back of the sync list.
3515 */
3516 VI_LOCK(syncvp);
3517 vn_syncer_add_to_worklist(syncvp, syncdelay);
3518 VI_UNLOCK(syncvp);
3519
3520 /*
3521 * Walk the list of vnodes pushing all that are dirty and
3522 * not already on the sync list.
3523 */
3524 mtx_lock(&mountlist_mtx);
3525 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3526 mtx_unlock(&mountlist_mtx);
3527 return (0);
3528 }
3529 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3530 vfs_unbusy(mp, td);
3531 return (0);
3532 }
3533 asyncflag = mp->mnt_flag & MNT_ASYNC;
3534 mp->mnt_flag &= ~MNT_ASYNC;
3535 vfs_msync(mp, MNT_NOWAIT);
3536 error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3537 if (asyncflag)
3538 mp->mnt_flag |= MNT_ASYNC;
3539 vn_finished_write(mp);
3540 vfs_unbusy(mp, td);
3541 return (error);
3542}
3543
3544/*
3545 * The syncer vnode is no referenced.
3546 */
3547static int
3548sync_inactive(ap)
3549 struct vop_inactive_args /* {
3550 struct vnode *a_vp;
3551 struct thread *a_td;
3552 } */ *ap;
3553{
3554
3555 VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3556 vgone(ap->a_vp);
3557 return (0);
3558}
3559
3560/*
3561 * The syncer vnode is no longer needed and is being decommissioned.
3562 *
3563 * Modifications to the worklist must be protected at splbio().
3564 */
3565static int
3566sync_reclaim(ap)
3567 struct vop_reclaim_args /* {
3568 struct vnode *a_vp;
3569 } */ *ap;
3570{
3571 struct vnode *vp = ap->a_vp;
3572 int s;
3573
3574 s = splbio();
3575 vp->v_mount->mnt_syncer = NULL;
3576 VI_LOCK(vp);
3577 if (vp->v_iflag & VI_ONWORKLST) {
3578 mtx_lock(&sync_mtx);
3579 LIST_REMOVE(vp, v_synclist);
3580 mtx_unlock(&sync_mtx);
3581 vp->v_iflag &= ~VI_ONWORKLST;
3582 }
3583 VI_UNLOCK(vp);
3584 splx(s);
3585
3586 return (0);
3587}
3588
3589/*
3590 * extract the dev_t from a VCHR
3591 */
3592dev_t
3593vn_todev(vp)
3594 struct vnode *vp;
3595{
3596 if (vp->v_type != VCHR)
3597 return (NODEV);
3598 return (vp->v_rdev);
3599}
3600
3601/*
3602 * Check if vnode represents a disk device
3603 */
3604int
3605vn_isdisk(vp, errp)
3606 struct vnode *vp;
3607 int *errp;
3608{
3609 struct cdevsw *cdevsw;
3610
3611 if (vp->v_type != VCHR) {
3612 if (errp != NULL)
3613 *errp = ENOTBLK;
3614 return (0);
3615 }
3616 if (vp->v_rdev == NULL) {
3617 if (errp != NULL)
3618 *errp = ENXIO;
3619 return (0);
3620 }
3621 cdevsw = devsw(vp->v_rdev);
3622 if (cdevsw == NULL) {
3623 if (errp != NULL)
3624 *errp = ENXIO;
3625 return (0);
3626 }
3627 if (!(cdevsw->d_flags & D_DISK)) {
3628 if (errp != NULL)
3629 *errp = ENOTBLK;
3630 return (0);
3631 }
3632 if (errp != NULL)
3633 *errp = 0;
3634 return (1);
3635}
3636
3637/*
3638 * Free data allocated by namei(); see namei(9) for details.
3639 */
3640void
3641NDFREE(ndp, flags)
3642 struct nameidata *ndp;
3643 const uint flags;
3644{
3645 if (!(flags & NDF_NO_FREE_PNBUF) &&
3646 (ndp->ni_cnd.cn_flags & HASBUF)) {
3647 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3648 ndp->ni_cnd.cn_flags &= ~HASBUF;
3649 }
3650 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3651 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3652 ndp->ni_dvp != ndp->ni_vp)
3653 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3654 if (!(flags & NDF_NO_DVP_RELE) &&
3655 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3656 vrele(ndp->ni_dvp);
3657 ndp->ni_dvp = NULL;
3658 }
3659 if (!(flags & NDF_NO_VP_UNLOCK) &&
3660 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3661 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3662 if (!(flags & NDF_NO_VP_RELE) &&
3663 ndp->ni_vp) {
3664 vrele(ndp->ni_vp);
3665 ndp->ni_vp = NULL;
3666 }
3667 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3668 (ndp->ni_cnd.cn_flags & SAVESTART)) {
3669 vrele(ndp->ni_startdir);
3670 ndp->ni_startdir = NULL;
3671 }
3672}
3673
3674/*
3675 * Common filesystem object access control check routine. Accepts a
3676 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3677 * and optional call-by-reference privused argument allowing vaccess()
3678 * to indicate to the caller whether privilege was used to satisfy the
3679 * request (obsoleted). Returns 0 on success, or an errno on failure.
3680 */
3681int
3682vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3683 enum vtype type;
3684 mode_t file_mode;
3685 uid_t file_uid;
3686 gid_t file_gid;
3687 mode_t acc_mode;
3688 struct ucred *cred;
3689 int *privused;
3690{
3691 mode_t dac_granted;
3692#ifdef CAPABILITIES
3693 mode_t cap_granted;
3694#endif
3695
3696 /*
3697 * Look for a normal, non-privileged way to access the file/directory
3698 * as requested. If it exists, go with that.
3699 */
3700
3701 if (privused != NULL)
3702 *privused = 0;
3703
3704 dac_granted = 0;
3705
3706 /* Check the owner. */
3707 if (cred->cr_uid == file_uid) {
3708 dac_granted |= VADMIN;
3709 if (file_mode & S_IXUSR)
3710 dac_granted |= VEXEC;
3711 if (file_mode & S_IRUSR)
3712 dac_granted |= VREAD;
3713 if (file_mode & S_IWUSR)
3714 dac_granted |= (VWRITE | VAPPEND);
3715
3716 if ((acc_mode & dac_granted) == acc_mode)
3717 return (0);
3718
3719 goto privcheck;
3720 }
3721
3722 /* Otherwise, check the groups (first match) */
3723 if (groupmember(file_gid, cred)) {
3724 if (file_mode & S_IXGRP)
3725 dac_granted |= VEXEC;
3726 if (file_mode & S_IRGRP)
3727 dac_granted |= VREAD;
3728 if (file_mode & S_IWGRP)
3729 dac_granted |= (VWRITE | VAPPEND);
3730
3731 if ((acc_mode & dac_granted) == acc_mode)
3732 return (0);
3733
3734 goto privcheck;
3735 }
3736
3737 /* Otherwise, check everyone else. */
3738 if (file_mode & S_IXOTH)
3739 dac_granted |= VEXEC;
3740 if (file_mode & S_IROTH)
3741 dac_granted |= VREAD;
3742 if (file_mode & S_IWOTH)
3743 dac_granted |= (VWRITE | VAPPEND);
3744 if ((acc_mode & dac_granted) == acc_mode)
3745 return (0);
3746
3747privcheck:
3748 if (!suser_cred(cred, PRISON_ROOT)) {
3749 /* XXX audit: privilege used */
3750 if (privused != NULL)
3751 *privused = 1;
3752 return (0);
3753 }
3754
3755#ifdef CAPABILITIES
3756 /*
3757 * Build a capability mask to determine if the set of capabilities
3758 * satisfies the requirements when combined with the granted mask
3759 * from above.
3760 * For each capability, if the capability is required, bitwise
3761 * or the request type onto the cap_granted mask.
3762 */
3763 cap_granted = 0;
3764
3765 if (type == VDIR) {
3766 /*
3767 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3768 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3769 */
3770 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3771 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3772 cap_granted |= VEXEC;
3773 } else {
3774 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3775 !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3776 cap_granted |= VEXEC;
3777 }
3778
3779 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3780 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3781 cap_granted |= VREAD;
3782
3783 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3784 !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3785 cap_granted |= (VWRITE | VAPPEND);
3786
3787 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3788 !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3789 cap_granted |= VADMIN;
3790
3791 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3792 /* XXX audit: privilege used */
3793 if (privused != NULL)
3794 *privused = 1;
3795 return (0);
3796 }
3797#endif
3798
3799 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3800}
3801
3802/*
3803 * Credential check based on process requesting service, and per-attribute
3804 * permissions.
3805 */
3806int
3807extattr_check_cred(struct vnode *vp, int attrnamespace,
3808 struct ucred *cred, struct thread *td, int access)
3809{
3810
3811 /*
3812 * Kernel-invoked always succeeds.
3813 */
3814 if (cred == NOCRED)
3815 return (0);
3816
3817 /*
3818 * Do not allow privileged processes in jail to directly
3819 * manipulate system attributes.
3820 *
3821 * XXX What capability should apply here?
3822 * Probably CAP_SYS_SETFFLAG.
3823 */
3824 switch (attrnamespace) {
3825 case EXTATTR_NAMESPACE_SYSTEM:
3826 /* Potentially should be: return (EPERM); */
3827 return (suser_cred(cred, 0));
3828 case EXTATTR_NAMESPACE_USER:
3829 return (VOP_ACCESS(vp, access, cred, td));
3830 default:
3831 return (EPERM);
3832 }
3833}