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.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.28 2004/03/28 07:54:00 dillon Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/namei.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
71 #include <machine/limits.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
85 #include <sys/thread2.h>
87 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
89 static void insmntque (struct vnode *vp, struct mount *mp);
90 static void vclean (struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td);
91 static unsigned long numvnodes;
92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
94 enum vtype iftovt_tab[16] = {
95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
100 S_IFSOCK, S_IFIFO, S_IFMT,
103 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
105 static u_long wantfreevnodes = 25;
106 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
107 static u_long freevnodes = 0;
108 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
110 static int reassignbufcalls;
111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
112 static int reassignbufloops;
113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
114 static int reassignbufsortgood;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
116 static int reassignbufsortbad;
117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
118 static int reassignbufmethod = 1;
119 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
121 #ifdef ENABLE_VFS_IOOPT
123 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
126 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
127 struct lwkt_token mountlist_token;
128 struct lwkt_token mntvnode_token;
129 int nfs_mount_type = -1;
130 static struct lwkt_token mntid_token;
131 static struct lwkt_token vnode_free_list_token;
132 static struct lwkt_token spechash_token;
133 struct nfs_public nfs_pub; /* publicly exported FS */
134 static vm_zone_t vnode_zone;
137 * The workitem queue.
139 #define SYNCER_MAXDELAY 32
140 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
141 time_t syncdelay = 30; /* max time to delay syncing data */
142 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
143 "VFS data synchronization delay");
144 time_t filedelay = 30; /* time to delay syncing files */
145 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
146 "File synchronization delay");
147 time_t dirdelay = 29; /* time to delay syncing directories */
148 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
149 "Directory synchronization delay");
150 time_t metadelay = 28; /* time to delay syncing metadata */
151 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
152 "VFS metadata synchronization delay");
153 static int rushjob; /* number of slots to run ASAP */
154 static int stat_rush_requests; /* number of times I/O speeded up */
155 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
157 static int syncer_delayno = 0;
158 static long syncer_mask;
159 LIST_HEAD(synclist, vnode);
160 static struct synclist *syncer_workitem_pending;
163 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
164 &desiredvnodes, 0, "Maximum number of vnodes");
165 static int minvnodes;
166 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
167 &minvnodes, 0, "Minimum number of vnodes");
168 static int vnlru_nowhere = 0;
169 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
170 "Number of times the vnlru process ran without success");
172 static void vfs_free_addrlist (struct netexport *nep);
173 static int vfs_free_netcred (struct radix_node *rn, void *w);
174 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
175 struct export_args *argp);
177 #define VSHOULDFREE(vp) \
178 (!((vp)->v_flag & (VFREE|VDOOMED)) && \
179 !(vp)->v_holdcnt && !(vp)->v_usecount && \
180 (!(vp)->v_object || \
181 !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
183 #define VMIGHTFREE(vp) \
184 (((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) == 0 && \
185 cache_leaf_test(vp) == 0 && (vp)->v_usecount == 0)
187 #define VSHOULDBUSY(vp) \
188 (((vp)->v_flag & VFREE) && \
189 ((vp)->v_holdcnt || (vp)->v_usecount))
191 static void vbusy(struct vnode *vp);
192 static void vfree(struct vnode *vp);
193 static void vmaybefree(struct vnode *vp);
196 * NOTE: the vnode interlock must be held on call.
199 vmaybefree(struct vnode *vp)
206 * Initialize the vnode management data structures.
213 * Desired vnodes is a result of the physical page count
214 * and the size of kernel's heap. It scales in proportion
215 * to the amount of available physical memory. This can
216 * cause trouble on 64-bit and large memory platforms.
218 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
220 min(maxproc + vmstats.v_page_count /4,
221 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
222 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
224 minvnodes = desiredvnodes / 4;
225 lwkt_token_init(&mountlist_token);
226 lwkt_token_init(&mntvnode_token);
227 lwkt_token_init(&mntid_token);
228 lwkt_token_init(&spechash_token);
229 TAILQ_INIT(&vnode_free_list);
230 lwkt_token_init(&vnode_free_list_token);
231 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
233 * Initialize the filesystem syncer.
235 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
237 syncer_maxdelay = syncer_mask + 1;
241 * Mark a mount point as busy. Used to synchronize access and to delay
242 * unmounting. Interlock is not released on failure.
245 vfs_busy(struct mount *mp, int flags, lwkt_tokref_t interlkp, struct thread *td)
249 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
250 if (flags & LK_NOWAIT)
252 mp->mnt_kern_flag |= MNTK_MWAIT;
254 * Since all busy locks are shared except the exclusive
255 * lock granted when unmounting, the only place that a
256 * wakeup needs to be done is at the release of the
257 * exclusive lock at the end of dounmount.
259 * note: interlkp is a serializer and thus can be safely
260 * held through any sleep
262 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
265 lkflags = LK_SHARED | LK_NOPAUSE;
267 lkflags |= LK_INTERLOCK;
268 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
269 panic("vfs_busy: unexpected lock failure");
274 * Free a busy filesystem.
277 vfs_unbusy(struct mount *mp, struct thread *td)
279 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
283 * Lookup a filesystem type, and if found allocate and initialize
284 * a mount structure for it.
286 * Devname is usually updated by mount(8) after booting.
289 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
291 struct thread *td = curthread; /* XXX */
292 struct vfsconf *vfsp;
295 if (fstypename == NULL)
297 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
298 if (!strcmp(vfsp->vfc_name, fstypename))
302 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
303 bzero((char *)mp, (u_long)sizeof(struct mount));
304 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
305 vfs_busy(mp, LK_NOWAIT, NULL, td);
306 TAILQ_INIT(&mp->mnt_nvnodelist);
307 TAILQ_INIT(&mp->mnt_reservedvnlist);
308 mp->mnt_nvnodelistsize = 0;
310 mp->mnt_op = vfsp->vfc_vfsops;
311 mp->mnt_flag = MNT_RDONLY;
312 mp->mnt_vnodecovered = NULLVP;
313 vfsp->vfc_refcount++;
314 mp->mnt_iosize_max = DFLTPHYS;
315 mp->mnt_stat.f_type = vfsp->vfc_typenum;
316 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
317 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
318 mp->mnt_stat.f_mntonname[0] = '/';
319 mp->mnt_stat.f_mntonname[1] = 0;
320 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
326 * Find an appropriate filesystem to use for the root. If a filesystem
327 * has not been preselected, walk through the list of known filesystems
328 * trying those that have mountroot routines, and try them until one
329 * works or we have tried them all.
331 #ifdef notdef /* XXX JH */
333 lite2_vfs_mountroot()
335 struct vfsconf *vfsp;
336 extern int (*lite2_mountroot) (void);
339 if (lite2_mountroot != NULL)
340 return ((*lite2_mountroot)());
341 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
342 if (vfsp->vfc_mountroot == NULL)
344 if ((error = (*vfsp->vfc_mountroot)()) == 0)
346 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
353 * Lookup a mount point by filesystem identifier.
362 lwkt_gettoken(&ilock, &mountlist_token);
363 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
364 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
365 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
369 lwkt_reltoken(&ilock);
374 * Get a new unique fsid. Try to make its val[0] unique, since this value
375 * will be used to create fake device numbers for stat(). Also try (but
376 * not so hard) make its val[0] unique mod 2^16, since some emulators only
377 * support 16-bit device numbers. We end up with unique val[0]'s for the
378 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
380 * Keep in mind that several mounts may be running in parallel. Starting
381 * the search one past where the previous search terminated is both a
382 * micro-optimization and a defense against returning the same fsid to
389 static u_int16_t mntid_base;
394 lwkt_gettoken(&ilock, &mntid_token);
395 mtype = mp->mnt_vfc->vfc_typenum;
396 tfsid.val[1] = mtype;
397 mtype = (mtype & 0xFF) << 24;
399 tfsid.val[0] = makeudev(255,
400 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
402 if (vfs_getvfs(&tfsid) == NULL)
405 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
406 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
407 lwkt_reltoken(&ilock);
411 * Knob to control the precision of file timestamps:
413 * 0 = seconds only; nanoseconds zeroed.
414 * 1 = seconds and nanoseconds, accurate within 1/HZ.
415 * 2 = seconds and nanoseconds, truncated to microseconds.
416 * >=3 = seconds and nanoseconds, maximum precision.
418 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
420 static int timestamp_precision = TSP_SEC;
421 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
422 ×tamp_precision, 0, "");
425 * Get a current timestamp.
429 struct timespec *tsp;
433 switch (timestamp_precision) {
435 tsp->tv_sec = time_second;
443 TIMEVAL_TO_TIMESPEC(&tv, tsp);
453 * Set vnode attributes to VNOVAL
461 vap->va_size = VNOVAL;
462 vap->va_bytes = VNOVAL;
463 vap->va_mode = VNOVAL;
464 vap->va_nlink = VNOVAL;
465 vap->va_uid = VNOVAL;
466 vap->va_gid = VNOVAL;
467 vap->va_fsid = VNOVAL;
468 vap->va_fileid = VNOVAL;
469 vap->va_blocksize = VNOVAL;
470 vap->va_rdev = VNOVAL;
471 vap->va_atime.tv_sec = VNOVAL;
472 vap->va_atime.tv_nsec = VNOVAL;
473 vap->va_mtime.tv_sec = VNOVAL;
474 vap->va_mtime.tv_nsec = VNOVAL;
475 vap->va_ctime.tv_sec = VNOVAL;
476 vap->va_ctime.tv_nsec = VNOVAL;
477 vap->va_flags = VNOVAL;
478 vap->va_gen = VNOVAL;
483 * This routine is called when we have too many vnodes. It attempts
484 * to free <count> vnodes and will potentially free vnodes that still
485 * have VM backing store (VM backing store is typically the cause
486 * of a vnode blowout so we want to do this). Therefore, this operation
487 * is not considered cheap.
489 * A number of conditions may prevent a vnode from being reclaimed.
490 * the buffer cache may have references on the vnode, a directory
491 * vnode may still have references due to the namei cache representing
492 * underlying files, or the vnode may be in active use. It is not
493 * desireable to reuse such vnodes. These conditions may cause the
494 * number of vnodes to reach some minimum value regardless of what
495 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
498 vlrureclaim(struct mount *mp)
509 * Calculate the trigger point, don't allow user
510 * screwups to blow us up. This prevents us from
511 * recycling vnodes with lots of resident pages. We
512 * aren't trying to free memory, we are trying to
515 usevnodes = desiredvnodes;
518 trigger = vmstats.v_page_count * 2 / usevnodes;
521 lwkt_gettoken(&ilock, &mntvnode_token);
522 count = mp->mnt_nvnodelistsize / 10 + 1;
523 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
527 * The VP will stick around while we hold mntvnode_token,
528 * at least until we block, so we can safely do an initial
529 * check. But we have to check again after obtaining
530 * the vnode interlock. vp->v_interlock points to stable
531 * storage so it's ok if the vp gets ripped out from
532 * under us while we are blocked.
534 if (vp->v_type == VNON ||
535 vp->v_type == VBAD ||
536 !VMIGHTFREE(vp) || /* critical path opt */
538 vp->v_object->resident_page_count >= trigger)
540 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
541 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
547 * Get the interlock, delay moving the node to the tail so
548 * we don't race against new additions to the mountlist.
550 lwkt_gettoken(&vlock, vp->v_interlock);
551 if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
552 lwkt_reltoken(&vlock);
555 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
556 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
561 if (vp->v_type == VNON ||
562 vp->v_type == VBAD ||
563 !VMIGHTFREE(vp) || /* critical path opt */
565 vp->v_object->resident_page_count >= trigger)
567 lwkt_reltoken(&vlock);
571 vgonel(vp, &vlock, curthread);
575 lwkt_reltoken(&ilock);
580 * Attempt to recycle vnodes in a context that is always safe to block.
581 * Calling vlrurecycle() from the bowels of file system code has some
582 * interesting deadlock problems.
584 static struct thread *vnlruthread;
585 static int vnlruproc_sig;
590 struct mount *mp, *nmp;
594 struct thread *td = curthread;
596 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
601 kproc_suspend_loop();
602 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
604 wakeup(&vnlruproc_sig);
605 tsleep(td, 0, "vlruwt", hz);
609 lwkt_gettoken(&ilock, &mountlist_token);
610 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
611 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
612 nmp = TAILQ_NEXT(mp, mnt_list);
615 done += vlrureclaim(mp);
616 lwkt_gettokref(&ilock);
617 nmp = TAILQ_NEXT(mp, mnt_list);
620 lwkt_reltoken(&ilock);
623 tsleep(td, 0, "vlrup", hz * 3);
629 static struct kproc_desc vnlru_kp = {
634 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
637 * Routines having to do with the management of the vnode table.
639 extern vop_t **dead_vnodeop_p;
642 * Return the next vnode from the free list.
645 getnewvnode(tag, mp, vops, vpp)
652 struct thread *td = curthread; /* XXX */
653 struct vnode *vp = NULL;
662 * Try to reuse vnodes if we hit the max. This situation only
663 * occurs in certain large-memory (2G+) situations. We cannot
664 * attempt to directly reclaim vnodes due to nasty recursion
667 while (numvnodes - freevnodes > desiredvnodes) {
668 if (vnlruproc_sig == 0) {
669 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
672 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
677 * Attempt to reuse a vnode already on the free list, allocating
678 * a new vnode if we can't find one or if we have not reached a
679 * good minimum for good LRU performance.
681 lwkt_gettoken(&ilock, &vnode_free_list_token);
682 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
685 for (count = 0; count < freevnodes; count++) {
689 * Pull the next vnode off the free list and do some
690 * sanity checks. Note that regardless of how we
691 * block, if freevnodes is non-zero there had better
692 * be something on the list.
694 vp = TAILQ_FIRST(&vnode_free_list);
696 panic("getnewvnode: free vnode isn't");
699 * Move the vnode to the end of the list so other
700 * processes do not double-block trying to recycle
701 * the same vnode (as an optimization), then get
704 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
705 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
708 * Skip vnodes that are in the process of being
709 * held or referenced. Since the act of adding or
710 * removing a vnode on the freelist requires a token
711 * and may block, the ref count may be adjusted
712 * prior to its addition or removal.
714 if (VSHOULDBUSY(vp)) {
721 * Obtain the vnode interlock and check that the
722 * vnode is still on the free list.
724 * This normally devolves into a degenerate case so
725 * it is optimal. Loop up if it isn't. Note that
726 * the vnode could be in the middle of being moved
727 * off the free list (the VSHOULDBUSY() check) and
728 * must be skipped if so.
730 lwkt_gettoken(&vlock, vp->v_interlock);
731 TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
732 freelst, v_freelist) {
736 if (vp != xvp || VSHOULDBUSY(vp)) {
742 * We now safely own the vnode. If the vnode has
743 * an object do not recycle it if its VM object
744 * has resident pages or references.
746 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
747 (object->resident_page_count || object->ref_count))
749 lwkt_reltoken(&vlock);
755 * We can almost reuse this vnode. But we don't want
756 * to recycle it if the vnode has children in the
757 * namecache because that breaks the namecache's
758 * path element chain. (YYY use nc_refs for the
761 KKASSERT(vp->v_flag & VFREE);
762 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
764 if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
765 cache_leaf_test(vp) >= 0) {
766 /* ok, we can reuse this vnode */
769 lwkt_reltoken(&vlock);
770 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
776 * If vp is non-NULL we hold it's interlock.
779 vp->v_flag |= VDOOMED;
780 vp->v_flag &= ~VFREE;
782 lwkt_reltoken(&ilock);
783 cache_purge(vp); /* YYY may block */
785 if (vp->v_type != VBAD) {
786 vgonel(vp, &vlock, td);
788 lwkt_reltoken(&vlock);
796 panic("cleaned vnode isn't");
799 panic("Clean vnode has pending I/O's");
809 vp->v_writecount = 0; /* XXX */
811 lwkt_reltoken(&ilock);
812 vp = zalloc(vnode_zone);
813 bzero(vp, sizeof(*vp));
814 vp->v_interlock = lwkt_token_pool_get(vp);
815 lwkt_token_init(&vp->v_pollinfo.vpi_token);
818 TAILQ_INIT(&vp->v_namecache);
822 TAILQ_INIT(&vp->v_cleanblkhd);
823 TAILQ_INIT(&vp->v_dirtyblkhd);
833 vfs_object_create(vp, td);
838 * Move a vnode from one mount queue to another.
847 lwkt_gettoken(&ilock, &mntvnode_token);
849 * Delete from old mount point vnode list, if on one.
851 if (vp->v_mount != NULL) {
852 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
853 ("bad mount point vnode list size"));
854 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
855 vp->v_mount->mnt_nvnodelistsize--;
858 * Insert into list of vnodes for the new mount point, if available.
860 if ((vp->v_mount = mp) == NULL) {
861 lwkt_reltoken(&ilock);
864 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
865 mp->mnt_nvnodelistsize++;
866 lwkt_reltoken(&ilock);
870 * Update outstanding I/O count and do wakeup if requested.
878 bp->b_flags &= ~B_WRITEINPROG;
879 if ((vp = bp->b_vp)) {
881 if (vp->v_numoutput < 0)
882 panic("vwakeup: neg numoutput");
883 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
884 vp->v_flag &= ~VBWAIT;
885 wakeup((caddr_t) &vp->v_numoutput);
891 * Flush out and invalidate all buffers associated with a vnode.
892 * Called with the underlying object locked.
895 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
896 int slpflag, int slptimeo)
899 struct buf *nbp, *blist;
904 if (flags & V_SAVE) {
906 while (vp->v_numoutput) {
907 vp->v_flag |= VBWAIT;
908 error = tsleep((caddr_t)&vp->v_numoutput,
909 slpflag, "vinvlbuf", slptimeo);
915 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
917 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
920 if (vp->v_numoutput > 0 ||
921 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
922 panic("vinvalbuf: dirty bufs");
928 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
930 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
934 for (bp = blist; bp; bp = nbp) {
935 nbp = TAILQ_NEXT(bp, b_vnbufs);
936 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
937 error = BUF_TIMELOCK(bp,
938 LK_EXCLUSIVE | LK_SLEEPFAIL,
939 "vinvalbuf", slpflag, slptimeo);
946 * XXX Since there are no node locks for NFS, I
947 * believe there is a slight chance that a delayed
948 * write will occur while sleeping just above, so
949 * check for it. Note that vfs_bio_awrite expects
950 * buffers to reside on a queue, while VOP_BWRITE and
953 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
956 if (bp->b_vp == vp) {
957 if (bp->b_flags & B_CLUSTEROK) {
962 bp->b_flags |= B_ASYNC;
963 VOP_BWRITE(bp->b_vp, bp);
967 (void) VOP_BWRITE(bp->b_vp, bp);
972 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
973 bp->b_flags &= ~B_ASYNC;
979 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
980 * have write I/O in-progress but if there is a VM object then the
981 * VM object can also have read-I/O in-progress.
984 while (vp->v_numoutput > 0) {
985 vp->v_flag |= VBWAIT;
986 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
988 if (VOP_GETVOBJECT(vp, &object) == 0) {
989 while (object->paging_in_progress)
990 vm_object_pip_sleep(object, "vnvlbx");
992 } while (vp->v_numoutput > 0);
997 * Destroy the copy in the VM cache, too.
999 lwkt_gettoken(&vlock, vp->v_interlock);
1000 if (VOP_GETVOBJECT(vp, &object) == 0) {
1001 vm_object_page_remove(object, 0, 0,
1002 (flags & V_SAVE) ? TRUE : FALSE);
1004 lwkt_reltoken(&vlock);
1006 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1007 panic("vinvalbuf: flush failed");
1012 * Truncate a file's buffer and pages to a specified length. This
1013 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1017 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
1025 * Round up to the *next* lbn.
1027 trunclbn = (length + blksize - 1) / blksize;
1034 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1035 nbp = TAILQ_NEXT(bp, b_vnbufs);
1036 if (bp->b_lblkno >= trunclbn) {
1037 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1038 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1042 bp->b_flags |= (B_INVAL | B_RELBUF);
1043 bp->b_flags &= ~B_ASYNC;
1048 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1049 (nbp->b_vp != vp) ||
1050 (nbp->b_flags & B_DELWRI))) {
1056 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1057 nbp = TAILQ_NEXT(bp, b_vnbufs);
1058 if (bp->b_lblkno >= trunclbn) {
1059 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1060 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1064 bp->b_flags |= (B_INVAL | B_RELBUF);
1065 bp->b_flags &= ~B_ASYNC;
1070 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1071 (nbp->b_vp != vp) ||
1072 (nbp->b_flags & B_DELWRI) == 0)) {
1081 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1082 nbp = TAILQ_NEXT(bp, b_vnbufs);
1083 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1084 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1085 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1089 if (bp->b_vp == vp) {
1090 bp->b_flags |= B_ASYNC;
1092 bp->b_flags &= ~B_ASYNC;
1094 VOP_BWRITE(bp->b_vp, bp);
1102 while (vp->v_numoutput > 0) {
1103 vp->v_flag |= VBWAIT;
1104 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1109 vnode_pager_setsize(vp, length);
1115 * Associate a buffer with a vnode.
1124 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1128 bp->b_dev = vn_todev(vp);
1130 * Insert onto list for new vnode.
1133 bp->b_xflags |= BX_VNCLEAN;
1134 bp->b_xflags &= ~BX_VNDIRTY;
1135 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1140 * Disassociate a buffer from a vnode.
1147 struct buflists *listheadp;
1150 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1153 * Delete from old vnode list, if on one.
1157 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1158 if (bp->b_xflags & BX_VNDIRTY)
1159 listheadp = &vp->v_dirtyblkhd;
1161 listheadp = &vp->v_cleanblkhd;
1162 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1163 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1165 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1166 vp->v_flag &= ~VONWORKLST;
1167 LIST_REMOVE(vp, v_synclist);
1170 bp->b_vp = (struct vnode *) 0;
1175 * The workitem queue.
1177 * It is useful to delay writes of file data and filesystem metadata
1178 * for tens of seconds so that quickly created and deleted files need
1179 * not waste disk bandwidth being created and removed. To realize this,
1180 * we append vnodes to a "workitem" queue. When running with a soft
1181 * updates implementation, most pending metadata dependencies should
1182 * not wait for more than a few seconds. Thus, mounted on block devices
1183 * are delayed only about a half the time that file data is delayed.
1184 * Similarly, directory updates are more critical, so are only delayed
1185 * about a third the time that file data is delayed. Thus, there are
1186 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1187 * one each second (driven off the filesystem syncer process). The
1188 * syncer_delayno variable indicates the next queue that is to be processed.
1189 * Items that need to be processed soon are placed in this queue:
1191 * syncer_workitem_pending[syncer_delayno]
1193 * A delay of fifteen seconds is done by placing the request fifteen
1194 * entries later in the queue:
1196 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1201 * Add an item to the syncer work queue.
1204 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1210 if (vp->v_flag & VONWORKLST) {
1211 LIST_REMOVE(vp, v_synclist);
1214 if (delay > syncer_maxdelay - 2)
1215 delay = syncer_maxdelay - 2;
1216 slot = (syncer_delayno + delay) & syncer_mask;
1218 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1219 vp->v_flag |= VONWORKLST;
1223 struct thread *updatethread;
1224 static void sched_sync (void);
1225 static struct kproc_desc up_kp = {
1230 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1233 * System filesystem synchronizer daemon.
1238 struct synclist *slp;
1242 struct thread *td = curthread;
1244 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1248 kproc_suspend_loop();
1250 starttime = time_second;
1253 * Push files whose dirty time has expired. Be careful
1254 * of interrupt race on slp queue.
1257 slp = &syncer_workitem_pending[syncer_delayno];
1258 syncer_delayno += 1;
1259 if (syncer_delayno == syncer_maxdelay)
1263 while ((vp = LIST_FIRST(slp)) != NULL) {
1264 if (VOP_ISLOCKED(vp, NULL) == 0) {
1265 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1266 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1267 VOP_UNLOCK(vp, NULL, 0, td);
1270 if (LIST_FIRST(slp) == vp) {
1272 * Note: v_tag VT_VFS vps can remain on the
1273 * worklist too with no dirty blocks, but
1274 * since sync_fsync() moves it to a different
1277 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1278 !vn_isdisk(vp, NULL))
1279 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1281 * Put us back on the worklist. The worklist
1282 * routine will remove us from our current
1283 * position and then add us back in at a later
1286 vn_syncer_add_to_worklist(vp, syncdelay);
1292 * Do soft update processing.
1295 (*bioops.io_sync)(NULL);
1298 * The variable rushjob allows the kernel to speed up the
1299 * processing of the filesystem syncer process. A rushjob
1300 * value of N tells the filesystem syncer to process the next
1301 * N seconds worth of work on its queue ASAP. Currently rushjob
1302 * is used by the soft update code to speed up the filesystem
1303 * syncer process when the incore state is getting so far
1304 * ahead of the disk that the kernel memory pool is being
1305 * threatened with exhaustion.
1312 * If it has taken us less than a second to process the
1313 * current work, then wait. Otherwise start right over
1314 * again. We can still lose time if any single round
1315 * takes more than two seconds, but it does not really
1316 * matter as we are just trying to generally pace the
1317 * filesystem activity.
1319 if (time_second == starttime)
1320 tsleep(&lbolt, 0, "syncer", 0);
1325 * Request the syncer daemon to speed up its work.
1326 * We never push it to speed up more than half of its
1327 * normal turn time, otherwise it could take over the cpu.
1329 * YYY wchan field protected by the BGL.
1335 if (updatethread->td_wchan == &lbolt) { /* YYY */
1336 unsleep(updatethread);
1337 lwkt_schedule(updatethread);
1340 if (rushjob < syncdelay / 2) {
1342 stat_rush_requests += 1;
1349 * Associate a p-buffer with a vnode.
1351 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1352 * with the buffer. i.e. the bp has not been linked into the vnode or
1361 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1364 bp->b_flags |= B_PAGING;
1365 bp->b_dev = vn_todev(vp);
1369 * Disassociate a p-buffer from a vnode.
1376 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1379 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1381 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1386 bp->b_vp = (struct vnode *) 0;
1387 bp->b_flags &= ~B_PAGING;
1391 pbreassignbuf(bp, newvp)
1393 struct vnode *newvp;
1395 if ((bp->b_flags & B_PAGING) == 0) {
1397 "pbreassignbuf() on non phys bp %p",
1405 * Reassign a buffer from one vnode to another.
1406 * Used to assign file specific control information
1407 * (indirect blocks) to the vnode to which they belong.
1410 reassignbuf(bp, newvp)
1412 struct vnode *newvp;
1414 struct buflists *listheadp;
1418 if (newvp == NULL) {
1419 printf("reassignbuf: NULL");
1425 * B_PAGING flagged buffers cannot be reassigned because their vp
1426 * is not fully linked in.
1428 if (bp->b_flags & B_PAGING)
1429 panic("cannot reassign paging buffer");
1433 * Delete from old vnode list, if on one.
1435 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1436 if (bp->b_xflags & BX_VNDIRTY)
1437 listheadp = &bp->b_vp->v_dirtyblkhd;
1439 listheadp = &bp->b_vp->v_cleanblkhd;
1440 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1441 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1442 if (bp->b_vp != newvp) {
1444 bp->b_vp = NULL; /* for clarification */
1448 * If dirty, put on list of dirty buffers; otherwise insert onto list
1451 if (bp->b_flags & B_DELWRI) {
1454 listheadp = &newvp->v_dirtyblkhd;
1455 if ((newvp->v_flag & VONWORKLST) == 0) {
1456 switch (newvp->v_type) {
1462 if (newvp->v_specmountpoint != NULL) {
1470 vn_syncer_add_to_worklist(newvp, delay);
1472 bp->b_xflags |= BX_VNDIRTY;
1473 tbp = TAILQ_FIRST(listheadp);
1475 bp->b_lblkno == 0 ||
1476 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1477 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1478 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1479 ++reassignbufsortgood;
1480 } else if (bp->b_lblkno < 0) {
1481 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1482 ++reassignbufsortgood;
1483 } else if (reassignbufmethod == 1) {
1485 * New sorting algorithm, only handle sequential case,
1486 * otherwise append to end (but before metadata)
1488 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1489 (tbp->b_xflags & BX_VNDIRTY)) {
1491 * Found the best place to insert the buffer
1493 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1494 ++reassignbufsortgood;
1497 * Missed, append to end, but before meta-data.
1498 * We know that the head buffer in the list is
1499 * not meta-data due to prior conditionals.
1501 * Indirect effects: NFS second stage write
1502 * tends to wind up here, giving maximum
1503 * distance between the unstable write and the
1506 tbp = TAILQ_LAST(listheadp, buflists);
1507 while (tbp && tbp->b_lblkno < 0)
1508 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1509 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1510 ++reassignbufsortbad;
1514 * Old sorting algorithm, scan queue and insert
1517 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1518 (ttbp->b_lblkno < bp->b_lblkno)) {
1522 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1525 bp->b_xflags |= BX_VNCLEAN;
1526 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1527 if ((newvp->v_flag & VONWORKLST) &&
1528 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1529 newvp->v_flag &= ~VONWORKLST;
1530 LIST_REMOVE(newvp, v_synclist);
1533 if (bp->b_vp != newvp) {
1541 * Create a vnode for a block device.
1542 * Used for mounting the root file system.
1557 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1570 * Add a vnode to the alias list hung off the dev_t.
1572 * The reason for this gunk is that multiple vnodes can reference
1573 * the same physical device, so checking vp->v_usecount to see
1574 * how many users there are is inadequate; the v_usecount for
1575 * the vnodes need to be accumulated. vcount() does that.
1578 addaliasu(struct vnode *nvp, udev_t nvp_rdev)
1582 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1583 panic("addaliasu on non-special vnode");
1584 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
1593 addalias(struct vnode *nvp, dev_t dev)
1597 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1598 panic("addalias on non-special vnode");
1601 lwkt_gettoken(&ilock, &spechash_token);
1602 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1603 lwkt_reltoken(&ilock);
1607 * Grab a particular vnode from the free list, increment its
1608 * reference count and lock it. The vnode lock bit is set if the
1609 * vnode is being eliminated in vgone. The process is awakened
1610 * when the transition is completed, and an error returned to
1611 * indicate that the vnode is no longer usable (possibly having
1612 * been changed to a new file system type).
1614 * This code is very sensitive. We are depending on the vnode interlock
1615 * to be maintained through to the vn_lock() call, which means that we
1616 * cannot block which means that we cannot call vbusy() until after vn_lock().
1617 * If the interlock is not maintained, the VXLOCK check will not properly
1618 * interlock against a vclean()'s LK_DRAIN operation on the lock.
1621 vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
1627 * We need the interlock to safely modify the v_ fields. ZZZ it is
1628 * only legal to pass (1) the vnode's interlock and (2) only pass
1629 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1632 if ((flags & LK_INTERLOCK) == 0) {
1633 lwkt_gettoken(&vvlock, vp->v_interlock);
1638 * If the vnode is in the process of being cleaned out for
1639 * another use, we wait for the cleaning to finish and then
1640 * return failure. Cleaning is determined by checking that
1641 * the VXLOCK flag is set. It is possible for the vnode to be
1642 * self-referenced during the cleaning operation.
1644 if (vp->v_flag & VXLOCK) {
1645 if (vp->v_vxthread == curthread) {
1647 /* this can now occur in normal operation */
1648 log(LOG_INFO, "VXLOCK interlock avoided\n");
1651 vp->v_flag |= VXWANT;
1652 lwkt_reltoken(vlock);
1653 tsleep((caddr_t)vp, 0, "vget", 0);
1659 * Bump v_usecount to prevent the vnode from being recycled. The
1660 * usecount needs to be bumped before we successfully get our lock.
1663 if (flags & LK_TYPE_MASK) {
1664 if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
1666 * must expand vrele here because we do not want
1667 * to call VOP_INACTIVE if the reference count
1668 * drops back to zero since it was never really
1669 * active. We must remove it from the free list
1670 * before sleeping so that multiple processes do
1671 * not try to recycle it.
1673 lwkt_gettokref(vlock);
1676 lwkt_reltoken(vlock);
1680 if (VSHOULDBUSY(vp))
1681 vbusy(vp); /* interlock must be held on call */
1682 lwkt_reltoken(vlock);
1687 vref(struct vnode *vp)
1689 crit_enter(); /* YYY use crit section for moment / BGL protected */
1695 * Vnode put/release.
1696 * If count drops to zero, call inactive routine and return to freelist.
1699 vrele(struct vnode *vp)
1701 struct thread *td = curthread; /* XXX */
1704 KASSERT(vp != NULL && vp->v_usecount >= 0,
1705 ("vrele: null vp or <=0 v_usecount"));
1707 lwkt_gettoken(&vlock, vp->v_interlock);
1709 if (vp->v_usecount > 1) {
1711 lwkt_reltoken(&vlock);
1715 if (vp->v_usecount == 1) {
1718 * We must call VOP_INACTIVE with the node locked and the
1719 * usecount 0. If we are doing a vpu, the node is already
1720 * locked, but, in the case of vrele, we must explicitly lock
1721 * the vnode before calling VOP_INACTIVE.
1724 if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
1725 VOP_INACTIVE(vp, td);
1727 lwkt_reltoken(&vlock);
1730 vprint("vrele: negative ref count", vp);
1732 lwkt_reltoken(&vlock);
1733 panic("vrele: negative ref cnt");
1738 vput(struct vnode *vp)
1740 struct thread *td = curthread; /* XXX */
1743 KASSERT(vp != NULL, ("vput: null vp"));
1745 lwkt_gettoken(&vlock, vp->v_interlock);
1747 if (vp->v_usecount > 1) {
1749 VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
1753 if (vp->v_usecount == 1) {
1756 * We must call VOP_INACTIVE with the node locked.
1757 * If we are doing a vpu, the node is already locked,
1758 * so we just need to release the vnode mutex.
1760 VOP_INACTIVE(vp, td);
1762 lwkt_reltoken(&vlock);
1765 vprint("vput: negative ref count", vp);
1767 lwkt_reltoken(&vlock);
1768 panic("vput: negative ref cnt");
1773 * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1774 * be held but isn't.
1784 if (VSHOULDBUSY(vp))
1785 vbusy(vp); /* interlock must be held on call */
1790 * One less who cares about this vnode.
1798 lwkt_gettoken(&vlock, vp->v_interlock);
1799 if (vp->v_holdcnt <= 0)
1800 panic("vdrop: holdcnt");
1803 lwkt_reltoken(&vlock);
1809 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1810 int (*slowfunc)(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data),
1820 * Scan the vnodes on the mount's vnode list. Use a placemarker
1822 pvp = zalloc(vnode_zone);
1823 pvp->v_flag |= VPLACEMARKER;
1825 lwkt_gettoken(&ilock, &mntvnode_token);
1826 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1828 while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1830 * Move the placemarker and skip other placemarkers we
1831 * encounter. The nothing can get in our way so the
1832 * mount point on the vp must be valid.
1834 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1835 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1836 if (vp->v_flag & VPLACEMARKER)
1838 KKASSERT(vp->v_mount == mp);
1844 if ((r = fastfunc(mp, vp, data)) < 0)
1851 * Get the vnodes interlock and make sure it is still on the
1852 * mount list. Skip it if it has moved (we may encounter it
1853 * later). Then do the with-interlock test. The callback
1854 * is responsible for releasing the vnode interlock.
1856 * The interlock is type-stable.
1859 lwkt_gettoken(&vlock, vp->v_interlock);
1860 if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1861 printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1862 lwkt_reltoken(&vlock);
1865 if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1866 KKASSERT(lwkt_havetokref(&vlock) == 0);
1869 KKASSERT(lwkt_havetokref(&vlock) == 0);
1872 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1873 zfree(vnode_zone, pvp);
1874 lwkt_reltoken(&ilock);
1879 * Remove any vnodes in the vnode table belonging to mount point mp.
1881 * If FORCECLOSE is not specified, there should not be any active ones,
1882 * return error if any are found (nb: this is a user error, not a
1883 * system error). If FORCECLOSE is specified, detach any active vnodes
1886 * If WRITECLOSE is set, only flush out regular file vnodes open for
1889 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1891 * `rootrefs' specifies the base reference count for the root vnode
1892 * of this filesystem. The root vnode is considered busy if its
1893 * v_usecount exceeds this value. On a successful return, vflush()
1894 * will call vrele() on the root vnode exactly rootrefs times.
1895 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1899 static int busyprt = 0; /* print out busy vnodes */
1900 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1903 static int vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data);
1905 struct vflush_info {
1912 vflush(mp, rootrefs, flags)
1917 struct thread *td = curthread; /* XXX */
1918 struct vnode *rootvp = NULL;
1921 struct vflush_info vflush_info;
1924 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1925 ("vflush: bad args"));
1927 * Get the filesystem root vnode. We can vput() it
1928 * immediately, since with rootrefs > 0, it won't go away.
1930 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1935 vflush_info.busy = 0;
1936 vflush_info.flags = flags;
1937 vflush_info.td = td;
1938 vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
1940 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1942 * If just the root vnode is busy, and if its refcount
1943 * is equal to `rootrefs', then go ahead and kill it.
1945 lwkt_gettoken(&vlock, rootvp->v_interlock);
1946 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1947 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1948 if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1949 vgonel(rootvp, &vlock, td);
1950 vflush_info.busy = 0;
1952 lwkt_reltoken(&vlock);
1955 if (vflush_info.busy)
1957 for (; rootrefs > 0; rootrefs--)
1963 * The scan callback is made with an interlocked vnode.
1966 vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
1968 struct vflush_info *info = data;
1972 * Skip over a vnodes marked VSYSTEM.
1974 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1975 lwkt_reltoken(vlock);
1980 * If WRITECLOSE is set, flush out unlinked but still open
1981 * files (even if open only for reading) and regular file
1982 * vnodes open for writing.
1984 if ((info->flags & WRITECLOSE) &&
1985 (vp->v_type == VNON ||
1986 (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
1987 vattr.va_nlink > 0)) &&
1988 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1989 lwkt_reltoken(vlock);
1994 * With v_usecount == 0, all we need to do is clear out the
1995 * vnode data structures and we are done.
1997 if (vp->v_usecount == 0) {
1998 vgonel(vp, vlock, info->td);
2003 * If FORCECLOSE is set, forcibly close the vnode. For block
2004 * or character devices, revert to an anonymous device. For
2005 * all other files, just kill them.
2007 if (info->flags & FORCECLOSE) {
2008 if (vp->v_type != VBLK && vp->v_type != VCHR) {
2009 vgonel(vp, vlock, info->td);
2011 vclean(vp, vlock, 0, info->td);
2012 vp->v_op = spec_vnodeop_p;
2013 insmntque(vp, (struct mount *) 0);
2019 vprint("vflush: busy vnode", vp);
2021 lwkt_reltoken(vlock);
2027 * Disassociate the underlying file system from a vnode.
2030 vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
2035 * Check to see if the vnode is in use. If so we have to reference it
2036 * before we clean it out so that its count cannot fall to zero and
2037 * generate a race against ourselves to recycle it.
2039 if ((active = vp->v_usecount))
2043 * Prevent the vnode from being recycled or brought into use while we
2046 if (vp->v_flag & VXLOCK)
2047 panic("vclean: deadlock");
2048 vp->v_flag |= VXLOCK;
2049 vp->v_vxthread = curthread;
2052 * Even if the count is zero, the VOP_INACTIVE routine may still
2053 * have the object locked while it cleans it out. The VOP_LOCK
2054 * ensures that the VOP_INACTIVE routine is done with its work.
2055 * For active vnodes, it ensures that no other activity can
2056 * occur while the underlying object is being cleaned out.
2058 * NOTE: we continue to hold the vnode interlock through to the
2061 VOP_LOCK(vp, NULL, LK_DRAIN, td);
2064 * Clean out any buffers associated with the vnode.
2066 vinvalbuf(vp, V_SAVE, td, 0, 0);
2067 VOP_DESTROYVOBJECT(vp);
2070 * If purging an active vnode, it must be closed and
2071 * deactivated before being reclaimed. Note that the
2072 * VOP_INACTIVE will unlock the vnode.
2075 if (flags & DOCLOSE)
2076 VOP_CLOSE(vp, FNONBLOCK, td);
2077 VOP_INACTIVE(vp, td);
2080 * Any other processes trying to obtain this lock must first
2081 * wait for VXLOCK to clear, then call the new lock operation.
2083 VOP_UNLOCK(vp, NULL, 0, td);
2086 * Reclaim the vnode.
2088 if (VOP_RECLAIM(vp, td))
2089 panic("vclean: cannot reclaim");
2093 * Inline copy of vrele() since VOP_INACTIVE
2094 * has already been called.
2096 if (--vp->v_usecount <= 0) {
2098 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2099 vprint("vclean: bad ref count", vp);
2100 panic("vclean: ref cnt");
2108 vp->v_vnlock = NULL;
2112 * Done with purge, notify sleepers of the grim news.
2114 vp->v_op = dead_vnodeop_p;
2117 vp->v_flag &= ~VXLOCK;
2118 vp->v_vxthread = NULL;
2119 if (vp->v_flag & VXWANT) {
2120 vp->v_flag &= ~VXWANT;
2121 wakeup((caddr_t) vp);
2123 lwkt_reltoken(vlock);
2127 * Eliminate all activity associated with the requested vnode
2128 * and with all vnodes aliased to the requested vnode.
2132 struct vop_revoke_args /* {
2137 struct vnode *vp, *vq;
2141 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2145 * If a vgone (or vclean) is already in progress,
2146 * wait until it is done and return.
2148 if (vp->v_flag & VXLOCK) {
2149 vp->v_flag |= VXWANT;
2150 /*lwkt_reltoken(vlock); ZZZ */
2151 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2156 lwkt_gettoken(&ilock, &spechash_token);
2157 vq = SLIST_FIRST(&dev->si_hlist);
2158 lwkt_reltoken(&ilock);
2167 * Recycle an unused vnode to the front of the free list.
2168 * Release the passed interlock if the vnode will be recycled.
2171 vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
2175 lwkt_gettoken(&vlock, vp->v_interlock);
2176 if (vp->v_usecount == 0) {
2178 lwkt_reltoken(inter_lkp);
2179 vgonel(vp, &vlock, td);
2182 lwkt_reltoken(&vlock);
2187 * Eliminate all activity associated with a vnode
2188 * in preparation for reuse.
2191 vgone(struct vnode *vp)
2193 struct thread *td = curthread; /* XXX */
2196 lwkt_gettoken(&vlock, vp->v_interlock);
2197 vgonel(vp, &vlock, td);
2201 * vgone, with the vp interlock held.
2204 vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
2210 * If a vgone (or vclean) is already in progress,
2211 * wait until it is done and return.
2213 if (vp->v_flag & VXLOCK) {
2214 vp->v_flag |= VXWANT;
2215 lwkt_reltoken(vlock);
2216 tsleep((caddr_t)vp, 0, "vgone", 0);
2221 * Clean out the filesystem specific data.
2223 vclean(vp, vlock, DOCLOSE, td);
2224 lwkt_gettokref(vlock);
2227 * Delete from old mount point vnode list, if on one.
2229 if (vp->v_mount != NULL)
2230 insmntque(vp, (struct mount *)0);
2232 * If special device, remove it from special device alias list
2235 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2236 lwkt_gettoken(&ilock, &spechash_token);
2237 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2238 freedev(vp->v_rdev);
2239 lwkt_reltoken(&ilock);
2244 * If it is on the freelist and not already at the head,
2245 * move it to the head of the list. The test of the
2246 * VDOOMED flag and the reference count of zero is because
2247 * it will be removed from the free list by getnewvnode,
2248 * but will not have its reference count incremented until
2249 * after calling vgone. If the reference count were
2250 * incremented first, vgone would (incorrectly) try to
2251 * close the previous instance of the underlying object.
2253 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2255 lwkt_gettoken(&ilock, &vnode_free_list_token);
2256 if (vp->v_flag & VFREE)
2257 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2260 vp->v_flag |= VFREE;
2261 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2262 lwkt_reltoken(&ilock);
2266 lwkt_reltoken(vlock);
2270 * Lookup a vnode by device number.
2273 vfinddev(dev, type, vpp)
2281 lwkt_gettoken(&ilock, &spechash_token);
2282 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2283 if (type == vp->v_type) {
2285 lwkt_reltoken(&ilock);
2289 lwkt_reltoken(&ilock);
2294 * Calculate the total number of references to a special device.
2305 lwkt_gettoken(&ilock, &spechash_token);
2306 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2307 count += vq->v_usecount;
2308 lwkt_reltoken(&ilock);
2313 * Same as above, but using the dev_t as argument
2322 vp = SLIST_FIRST(&dev->si_hlist);
2329 * Print out a description of a vnode.
2331 static char *typename[] =
2332 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2342 printf("%s: %p: ", label, (void *)vp);
2344 printf("%p: ", (void *)vp);
2345 printf("type %s, usecount %d, writecount %d, refcount %d,",
2346 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2349 if (vp->v_flag & VROOT)
2350 strcat(buf, "|VROOT");
2351 if (vp->v_flag & VTEXT)
2352 strcat(buf, "|VTEXT");
2353 if (vp->v_flag & VSYSTEM)
2354 strcat(buf, "|VSYSTEM");
2355 if (vp->v_flag & VXLOCK)
2356 strcat(buf, "|VXLOCK");
2357 if (vp->v_flag & VXWANT)
2358 strcat(buf, "|VXWANT");
2359 if (vp->v_flag & VBWAIT)
2360 strcat(buf, "|VBWAIT");
2361 if (vp->v_flag & VDOOMED)
2362 strcat(buf, "|VDOOMED");
2363 if (vp->v_flag & VFREE)
2364 strcat(buf, "|VFREE");
2365 if (vp->v_flag & VOBJBUF)
2366 strcat(buf, "|VOBJBUF");
2368 printf(" flags (%s)", &buf[1]);
2369 if (vp->v_data == NULL) {
2378 #include <ddb/ddb.h>
2380 * List all of the locked vnodes in the system.
2381 * Called when debugging the kernel.
2383 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2385 struct thread *td = curthread; /* XXX */
2387 struct mount *mp, *nmp;
2390 printf("Locked vnodes\n");
2391 lwkt_gettoken(&ilock, &mountlist_token);
2392 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2393 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
2394 nmp = TAILQ_NEXT(mp, mnt_list);
2397 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2398 if (VOP_ISLOCKED(vp, NULL))
2399 vprint((char *)0, vp);
2401 lwkt_gettokref(&ilock);
2402 nmp = TAILQ_NEXT(mp, mnt_list);
2405 lwkt_reltoken(&ilock);
2410 * Top level filesystem related information gathering.
2412 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2415 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2417 int *name = (int *)arg1 - 1; /* XXX */
2418 u_int namelen = arg2 + 1; /* XXX */
2419 struct vfsconf *vfsp;
2421 #if 1 || defined(COMPAT_PRELITE2)
2422 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2424 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2428 /* all sysctl names at this level are at least name and field */
2430 return (ENOTDIR); /* overloaded */
2431 if (name[0] != VFS_GENERIC) {
2432 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2433 if (vfsp->vfc_typenum == name[0])
2436 return (EOPNOTSUPP);
2437 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2438 oldp, oldlenp, newp, newlen, p));
2442 case VFS_MAXTYPENUM:
2445 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2448 return (ENOTDIR); /* overloaded */
2449 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2450 if (vfsp->vfc_typenum == name[2])
2453 return (EOPNOTSUPP);
2454 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2456 return (EOPNOTSUPP);
2459 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2460 "Generic filesystem");
2462 #if 1 || defined(COMPAT_PRELITE2)
2465 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2468 struct vfsconf *vfsp;
2469 struct ovfsconf ovfs;
2471 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2472 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2473 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2474 ovfs.vfc_index = vfsp->vfc_typenum;
2475 ovfs.vfc_refcount = vfsp->vfc_refcount;
2476 ovfs.vfc_flags = vfsp->vfc_flags;
2477 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2484 #endif /* 1 || COMPAT_PRELITE2 */
2487 #define KINFO_VNODESLOP 10
2489 * Dump vnode list (via sysctl).
2490 * Copyout address of vnode followed by vnode.
2494 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2496 struct proc *p = curproc; /* XXX */
2497 struct mount *mp, *nmp;
2498 struct vnode *nvp, *vp;
2503 #define VPTRSZ sizeof (struct vnode *)
2504 #define VNODESZ sizeof (struct vnode)
2507 if (!req->oldptr) /* Make an estimate */
2508 return (SYSCTL_OUT(req, 0,
2509 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2511 lwkt_gettoken(&ilock, &mountlist_token);
2512 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2513 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
2514 nmp = TAILQ_NEXT(mp, mnt_list);
2517 lwkt_gettoken(&jlock, &mntvnode_token);
2519 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2523 * Check that the vp is still associated with
2524 * this filesystem. RACE: could have been
2525 * recycled onto the same filesystem.
2527 if (vp->v_mount != mp)
2529 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2530 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2531 (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2532 lwkt_reltoken(&jlock);
2536 lwkt_reltoken(&jlock);
2537 lwkt_gettokref(&ilock);
2538 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */
2541 lwkt_reltoken(&ilock);
2549 * Exporting the vnode list on large systems causes them to crash.
2550 * Exporting the vnode list on medium systems causes sysctl to coredump.
2553 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2554 0, 0, sysctl_vnode, "S,vnode", "");
2558 * Check to see if a filesystem is mounted on a block device.
2565 if (vp->v_specmountpoint != NULL)
2571 * Unmount all filesystems. The list is traversed in reverse order
2572 * of mounting to avoid dependencies.
2578 struct thread *td = curthread;
2581 if (td->td_proc == NULL)
2582 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2585 * Since this only runs when rebooting, it is not interlocked.
2587 while(!TAILQ_EMPTY(&mountlist)) {
2588 mp = TAILQ_LAST(&mountlist, mntlist);
2589 error = dounmount(mp, MNT_FORCE, td);
2591 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2592 printf("unmount of %s failed (",
2593 mp->mnt_stat.f_mntonname);
2597 printf("%d)\n", error);
2599 /* The unmount has removed mp from the mountlist */
2605 * Build hash lists of net addresses and hang them off the mount point.
2606 * Called by ufs_mount() to set up the lists of export addresses.
2609 vfs_hang_addrlist(mp, nep, argp)
2611 struct netexport *nep;
2612 struct export_args *argp;
2615 struct radix_node_head *rnh;
2617 struct radix_node *rn;
2618 struct sockaddr *saddr, *smask = 0;
2622 if (argp->ex_addrlen == 0) {
2623 if (mp->mnt_flag & MNT_DEFEXPORTED)
2625 np = &nep->ne_defexported;
2626 np->netc_exflags = argp->ex_flags;
2627 np->netc_anon = argp->ex_anon;
2628 np->netc_anon.cr_ref = 1;
2629 mp->mnt_flag |= MNT_DEFEXPORTED;
2633 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2635 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2638 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2639 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2640 bzero((caddr_t) np, i);
2641 saddr = (struct sockaddr *) (np + 1);
2642 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2644 if (saddr->sa_len > argp->ex_addrlen)
2645 saddr->sa_len = argp->ex_addrlen;
2646 if (argp->ex_masklen) {
2647 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2648 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2651 if (smask->sa_len > argp->ex_masklen)
2652 smask->sa_len = argp->ex_masklen;
2654 i = saddr->sa_family;
2655 if ((rnh = nep->ne_rtable[i]) == 0) {
2657 * Seems silly to initialize every AF when most are not used,
2658 * do so on demand here
2660 for (dom = domains; dom; dom = dom->dom_next)
2661 if (dom->dom_family == i && dom->dom_rtattach) {
2662 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2666 if ((rnh = nep->ne_rtable[i]) == 0) {
2671 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2673 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2677 np->netc_exflags = argp->ex_flags;
2678 np->netc_anon = argp->ex_anon;
2679 np->netc_anon.cr_ref = 1;
2682 free(np, M_NETADDR);
2688 vfs_free_netcred(rn, w)
2689 struct radix_node *rn;
2692 struct radix_node_head *rnh = (struct radix_node_head *) w;
2694 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2695 free((caddr_t) rn, M_NETADDR);
2700 * Free the net address hash lists that are hanging off the mount points.
2703 vfs_free_addrlist(nep)
2704 struct netexport *nep;
2707 struct radix_node_head *rnh;
2709 for (i = 0; i <= AF_MAX; i++)
2710 if ((rnh = nep->ne_rtable[i])) {
2711 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2713 free((caddr_t) rnh, M_RTABLE);
2714 nep->ne_rtable[i] = 0;
2719 vfs_export(mp, nep, argp)
2721 struct netexport *nep;
2722 struct export_args *argp;
2726 if (argp->ex_flags & MNT_DELEXPORT) {
2727 if (mp->mnt_flag & MNT_EXPUBLIC) {
2728 vfs_setpublicfs(NULL, NULL, NULL);
2729 mp->mnt_flag &= ~MNT_EXPUBLIC;
2731 vfs_free_addrlist(nep);
2732 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2734 if (argp->ex_flags & MNT_EXPORTED) {
2735 if (argp->ex_flags & MNT_EXPUBLIC) {
2736 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2738 mp->mnt_flag |= MNT_EXPUBLIC;
2740 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2742 mp->mnt_flag |= MNT_EXPORTED;
2749 * Set the publicly exported filesystem (WebNFS). Currently, only
2750 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2753 vfs_setpublicfs(mp, nep, argp)
2755 struct netexport *nep;
2756 struct export_args *argp;
2763 * mp == NULL -> invalidate the current info, the FS is
2764 * no longer exported. May be called from either vfs_export
2765 * or unmount, so check if it hasn't already been done.
2768 if (nfs_pub.np_valid) {
2769 nfs_pub.np_valid = 0;
2770 if (nfs_pub.np_index != NULL) {
2771 FREE(nfs_pub.np_index, M_TEMP);
2772 nfs_pub.np_index = NULL;
2779 * Only one allowed at a time.
2781 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2785 * Get real filehandle for root of exported FS.
2787 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2788 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2790 if ((error = VFS_ROOT(mp, &rvp)))
2793 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2799 * If an indexfile was specified, pull it in.
2801 if (argp->ex_indexfile != NULL) {
2802 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2804 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2805 MAXNAMLEN, (size_t *)0);
2808 * Check for illegal filenames.
2810 for (cp = nfs_pub.np_index; *cp; cp++) {
2818 FREE(nfs_pub.np_index, M_TEMP);
2823 nfs_pub.np_mount = mp;
2824 nfs_pub.np_valid = 1;
2829 vfs_export_lookup(mp, nep, nam)
2831 struct netexport *nep;
2832 struct sockaddr *nam;
2835 struct radix_node_head *rnh;
2836 struct sockaddr *saddr;
2839 if (mp->mnt_flag & MNT_EXPORTED) {
2841 * Lookup in the export list first.
2845 rnh = nep->ne_rtable[saddr->sa_family];
2847 np = (struct netcred *)
2848 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2850 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2855 * If no address match, use the default if it exists.
2857 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2858 np = &nep->ne_defexported;
2864 * perform msync on all vnodes under a mount point. The mount point must
2865 * be locked. This code is also responsible for lazy-freeing unreferenced
2866 * vnodes whos VM objects no longer contain pages.
2868 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2870 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2871 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2872 lwkt_tokref_t vlock, void *data);
2875 vfs_msync(struct mount *mp, int flags)
2877 vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2881 * scan1 is a fast pre-check. There could be hundreds of thousands of
2882 * vnodes, we cannot afford to do anything heavy weight until we have a
2883 * fairly good indication that there is work to do.
2887 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2889 int flags = (int)data;
2891 if ((vp->v_flag & VXLOCK) == 0) {
2892 if (VSHOULDFREE(vp))
2894 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2895 (vp->v_flag & VOBJDIRTY) &&
2896 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2905 vfs_msync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
2909 int flags = (int)data;
2911 if (vp->v_flag & VXLOCK)
2914 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2915 (vp->v_flag & VOBJDIRTY) &&
2916 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2917 error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2919 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2920 vm_object_page_clean(obj, 0, 0,
2921 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2928 lwkt_reltoken(vlock);
2933 * Create the VM object needed for VMIO and mmap support. This
2934 * is done for all VREG files in the system. Some filesystems might
2935 * afford the additional metadata buffering capability of the
2936 * VMIO code by making the device node be VMIO mode also.
2938 * vp must be locked when vfs_object_create is called.
2941 vfs_object_create(struct vnode *vp, struct thread *td)
2943 return (VOP_CREATEVOBJECT(vp, td));
2947 * NOTE: the vnode interlock must be held during the call. We have to recheck
2948 * the VFREE flag since the vnode may have been removed from the free list
2949 * while we were blocked on vnode_free_list_token. The use or hold count
2950 * must have already been bumped by the caller.
2953 vbusy(struct vnode *vp)
2957 lwkt_gettoken(&ilock, &vnode_free_list_token);
2958 if ((vp->v_flag & VFREE) != 0) {
2959 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2961 vp->v_flag &= ~(VFREE|VAGE);
2963 lwkt_reltoken(&ilock);
2967 * NOTE: the vnode interlock must be held during the call. The use or hold
2968 * count must have already been bumped by the caller. We use a VINFREE to
2969 * interlock against other calls to vfree() which might occur while we
2970 * are blocked. The vnode cannot be reused until it has actually been
2971 * placed on the free list, so there are no other races even though the
2972 * use and hold counts are 0.
2975 vfree(struct vnode *vp)
2979 if ((vp->v_flag & VINFREE) == 0) {
2980 vp->v_flag |= VINFREE;
2981 lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
2982 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2983 if (vp->v_flag & VAGE) {
2984 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2986 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2989 vp->v_flag &= ~(VAGE|VINFREE);
2990 vp->v_flag |= VFREE;
2991 lwkt_reltoken(&ilock); /* can block */
2997 * Record a process's interest in events which might happen to
2998 * a vnode. Because poll uses the historic select-style interface
2999 * internally, this routine serves as both the ``check for any
3000 * pending events'' and the ``record my interest in future events''
3001 * functions. (These are done together, while the lock is held,
3002 * to avoid race conditions.)
3005 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3009 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3010 if (vp->v_pollinfo.vpi_revents & events) {
3012 * This leaves events we are not interested
3013 * in available for the other process which
3014 * which presumably had requested them
3015 * (otherwise they would never have been
3018 events &= vp->v_pollinfo.vpi_revents;
3019 vp->v_pollinfo.vpi_revents &= ~events;
3021 lwkt_reltoken(&ilock);
3024 vp->v_pollinfo.vpi_events |= events;
3025 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
3026 lwkt_reltoken(&ilock);
3031 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3032 * it is possible for us to miss an event due to race conditions, but
3033 * that condition is expected to be rare, so for the moment it is the
3034 * preferred interface.
3037 vn_pollevent(vp, events)
3043 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3044 if (vp->v_pollinfo.vpi_events & events) {
3046 * We clear vpi_events so that we don't
3047 * call selwakeup() twice if two events are
3048 * posted before the polling process(es) is
3049 * awakened. This also ensures that we take at
3050 * most one selwakeup() if the polling process
3051 * is no longer interested. However, it does
3052 * mean that only one event can be noticed at
3053 * a time. (Perhaps we should only clear those
3054 * event bits which we note?) XXX
3056 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
3057 vp->v_pollinfo.vpi_revents |= events;
3058 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3060 lwkt_reltoken(&ilock);
3064 * Wake up anyone polling on vp because it is being revoked.
3065 * This depends on dead_poll() returning POLLHUP for correct
3074 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3075 if (vp->v_pollinfo.vpi_events) {
3076 vp->v_pollinfo.vpi_events = 0;
3077 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3079 lwkt_reltoken(&ilock);
3085 * Routine to create and manage a filesystem syncer vnode.
3087 #define sync_close ((int (*) (struct vop_close_args *))nullop)
3088 static int sync_fsync (struct vop_fsync_args *);
3089 static int sync_inactive (struct vop_inactive_args *);
3090 static int sync_reclaim (struct vop_reclaim_args *);
3091 #define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
3092 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
3093 static int sync_print (struct vop_print_args *);
3094 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
3096 static vop_t **sync_vnodeop_p;
3097 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3098 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3099 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3100 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3101 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3102 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3103 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
3104 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
3105 { &vop_print_desc, (vop_t *) sync_print }, /* print */
3106 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
3109 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3110 { &sync_vnodeop_p, sync_vnodeop_entries };
3112 VNODEOP_SET(sync_vnodeop_opv_desc);
3115 * Create a new filesystem syncer vnode for the specified mount point.
3116 * This vnode is placed on the worklist and is responsible for sync'ing
3119 * NOTE: read-only mounts are also placed on the worklist. The filesystem
3120 * sync code is also responsible for cleaning up vnodes.
3123 vfs_allocate_syncvnode(struct mount *mp)
3126 static long start, incr, next;
3129 /* Allocate a new vnode */
3130 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3131 mp->mnt_syncer = NULL;
3136 * Place the vnode onto the syncer worklist. We attempt to
3137 * scatter them about on the list so that they will go off
3138 * at evenly distributed times even if all the filesystems
3139 * are mounted at once.
3142 if (next == 0 || next > syncer_maxdelay) {
3146 start = syncer_maxdelay / 2;
3147 incr = syncer_maxdelay;
3151 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3152 mp->mnt_syncer = vp;
3157 * Do a lazy sync of the filesystem.
3161 struct vop_fsync_args /* {
3163 struct ucred *a_cred;
3165 struct thread *a_td;
3168 struct vnode *syncvp = ap->a_vp;
3169 struct mount *mp = syncvp->v_mount;
3170 struct thread *td = ap->a_td;
3175 * We only need to do something if this is a lazy evaluation.
3177 if (ap->a_waitfor != MNT_LAZY)
3181 * Move ourselves to the back of the sync list.
3183 vn_syncer_add_to_worklist(syncvp, syncdelay);
3186 * Walk the list of vnodes pushing all that are dirty and
3187 * not already on the sync list, and freeing vnodes which have
3188 * no refs and whos VM objects are empty. vfs_msync() handles
3189 * the VM issues and must be called whether the mount is readonly
3192 lwkt_gettoken(&ilock, &mountlist_token);
3193 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3194 lwkt_reltoken(&ilock);
3197 if (mp->mnt_flag & MNT_RDONLY) {
3198 vfs_msync(mp, MNT_NOWAIT);
3200 asyncflag = mp->mnt_flag & MNT_ASYNC;
3201 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */
3202 vfs_msync(mp, MNT_NOWAIT);
3203 VFS_SYNC(mp, MNT_LAZY, td);
3205 mp->mnt_flag |= MNT_ASYNC;
3212 * The syncer vnode is no referenced.
3216 struct vop_inactive_args /* {
3227 * The syncer vnode is no longer needed and is being decommissioned.
3229 * Modifications to the worklist must be protected at splbio().
3233 struct vop_reclaim_args /* {
3237 struct vnode *vp = ap->a_vp;
3241 vp->v_mount->mnt_syncer = NULL;
3242 if (vp->v_flag & VONWORKLST) {
3243 LIST_REMOVE(vp, v_synclist);
3244 vp->v_flag &= ~VONWORKLST;
3252 * Print out a syncer vnode.
3256 struct vop_print_args /* {
3260 struct vnode *vp = ap->a_vp;
3262 printf("syncer vnode");
3263 if (vp->v_vnlock != NULL)
3264 lockmgr_printinfo(vp->v_vnlock);
3270 * extract the dev_t from a VBLK or VCHR
3276 if (vp->v_type != VBLK && vp->v_type != VCHR)
3278 return (vp->v_rdev);
3282 * Check if vnode represents a disk device
3289 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3294 if (vp->v_rdev == NULL) {
3299 if (!dev_dport(vp->v_rdev)) {
3304 if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
3316 struct nameidata *ndp;
3319 if (!(flags & NDF_NO_FREE_PNBUF) &&
3320 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3321 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3322 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3324 if (!(flags & NDF_NO_DNCP_RELE) &&
3325 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3327 cache_drop(ndp->ni_dncp);
3328 ndp->ni_dncp = NULL;
3330 if (!(flags & NDF_NO_NCP_RELE) &&
3331 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3333 cache_drop(ndp->ni_ncp);
3336 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3337 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3338 ndp->ni_dvp != ndp->ni_vp) {
3339 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
3341 if (!(flags & NDF_NO_DVP_RELE) &&
3342 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3346 if (!(flags & NDF_NO_VP_UNLOCK) &&
3347 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3348 VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
3350 if (!(flags & NDF_NO_VP_RELE) &&
3355 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3356 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3357 vrele(ndp->ni_startdir);
3358 ndp->ni_startdir = NULL;
3362 #ifdef DEBUG_VFS_LOCKS
3365 assert_vop_locked(struct vnode *vp, const char *str)
3368 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
3369 panic("%s: %p is not locked shared but should be", str, vp);
3374 assert_vop_unlocked(struct vnode *vp, const char *str)
3377 if (vp && IS_LOCKING_VFS(vp)) {
3378 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
3379 panic("%s: %p is locked but should not be", str, vp);