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 1.30 2004/05/19 22:52:58 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);
195 extern int dev_ref_debug;
198 * NOTE: the vnode interlock must be held on call.
201 vmaybefree(struct vnode *vp)
208 * Initialize the vnode management data structures.
215 * Desired vnodes is a result of the physical page count
216 * and the size of kernel's heap. It scales in proportion
217 * to the amount of available physical memory. This can
218 * cause trouble on 64-bit and large memory platforms.
220 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
222 min(maxproc + vmstats.v_page_count /4,
223 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
224 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
226 minvnodes = desiredvnodes / 4;
227 lwkt_token_init(&mountlist_token);
228 lwkt_token_init(&mntvnode_token);
229 lwkt_token_init(&mntid_token);
230 lwkt_token_init(&spechash_token);
231 TAILQ_INIT(&vnode_free_list);
232 lwkt_token_init(&vnode_free_list_token);
233 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
235 * Initialize the filesystem syncer.
237 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
239 syncer_maxdelay = syncer_mask + 1;
243 * Mark a mount point as busy. Used to synchronize access and to delay
244 * unmounting. Interlock is not released on failure.
247 vfs_busy(struct mount *mp, int flags, lwkt_tokref_t interlkp, struct thread *td)
251 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
252 if (flags & LK_NOWAIT)
254 mp->mnt_kern_flag |= MNTK_MWAIT;
256 * Since all busy locks are shared except the exclusive
257 * lock granted when unmounting, the only place that a
258 * wakeup needs to be done is at the release of the
259 * exclusive lock at the end of dounmount.
261 * note: interlkp is a serializer and thus can be safely
262 * held through any sleep
264 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
267 lkflags = LK_SHARED | LK_NOPAUSE;
269 lkflags |= LK_INTERLOCK;
270 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
271 panic("vfs_busy: unexpected lock failure");
276 * Free a busy filesystem.
279 vfs_unbusy(struct mount *mp, struct thread *td)
281 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
285 * Lookup a filesystem type, and if found allocate and initialize
286 * a mount structure for it.
288 * Devname is usually updated by mount(8) after booting.
291 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
293 struct thread *td = curthread; /* XXX */
294 struct vfsconf *vfsp;
297 if (fstypename == NULL)
299 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
300 if (!strcmp(vfsp->vfc_name, fstypename))
304 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
305 bzero((char *)mp, (u_long)sizeof(struct mount));
306 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
307 vfs_busy(mp, LK_NOWAIT, NULL, td);
308 TAILQ_INIT(&mp->mnt_nvnodelist);
309 TAILQ_INIT(&mp->mnt_reservedvnlist);
310 mp->mnt_nvnodelistsize = 0;
312 mp->mnt_op = vfsp->vfc_vfsops;
313 mp->mnt_flag = MNT_RDONLY;
314 mp->mnt_vnodecovered = NULLVP;
315 vfsp->vfc_refcount++;
316 mp->mnt_iosize_max = DFLTPHYS;
317 mp->mnt_stat.f_type = vfsp->vfc_typenum;
318 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
319 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
320 mp->mnt_stat.f_mntonname[0] = '/';
321 mp->mnt_stat.f_mntonname[1] = 0;
322 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
328 * Find an appropriate filesystem to use for the root. If a filesystem
329 * has not been preselected, walk through the list of known filesystems
330 * trying those that have mountroot routines, and try them until one
331 * works or we have tried them all.
333 #ifdef notdef /* XXX JH */
335 lite2_vfs_mountroot()
337 struct vfsconf *vfsp;
338 extern int (*lite2_mountroot) (void);
341 if (lite2_mountroot != NULL)
342 return ((*lite2_mountroot)());
343 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
344 if (vfsp->vfc_mountroot == NULL)
346 if ((error = (*vfsp->vfc_mountroot)()) == 0)
348 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
355 * Lookup a mount point by filesystem identifier.
364 lwkt_gettoken(&ilock, &mountlist_token);
365 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
366 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
367 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
371 lwkt_reltoken(&ilock);
376 * Get a new unique fsid. Try to make its val[0] unique, since this value
377 * will be used to create fake device numbers for stat(). Also try (but
378 * not so hard) make its val[0] unique mod 2^16, since some emulators only
379 * support 16-bit device numbers. We end up with unique val[0]'s for the
380 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
382 * Keep in mind that several mounts may be running in parallel. Starting
383 * the search one past where the previous search terminated is both a
384 * micro-optimization and a defense against returning the same fsid to
391 static u_int16_t mntid_base;
396 lwkt_gettoken(&ilock, &mntid_token);
397 mtype = mp->mnt_vfc->vfc_typenum;
398 tfsid.val[1] = mtype;
399 mtype = (mtype & 0xFF) << 24;
401 tfsid.val[0] = makeudev(255,
402 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
404 if (vfs_getvfs(&tfsid) == NULL)
407 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
408 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
409 lwkt_reltoken(&ilock);
413 * Knob to control the precision of file timestamps:
415 * 0 = seconds only; nanoseconds zeroed.
416 * 1 = seconds and nanoseconds, accurate within 1/HZ.
417 * 2 = seconds and nanoseconds, truncated to microseconds.
418 * >=3 = seconds and nanoseconds, maximum precision.
420 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
422 static int timestamp_precision = TSP_SEC;
423 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
424 ×tamp_precision, 0, "");
427 * Get a current timestamp.
431 struct timespec *tsp;
435 switch (timestamp_precision) {
437 tsp->tv_sec = time_second;
445 TIMEVAL_TO_TIMESPEC(&tv, tsp);
455 * Set vnode attributes to VNOVAL
463 vap->va_size = VNOVAL;
464 vap->va_bytes = VNOVAL;
465 vap->va_mode = VNOVAL;
466 vap->va_nlink = VNOVAL;
467 vap->va_uid = VNOVAL;
468 vap->va_gid = VNOVAL;
469 vap->va_fsid = VNOVAL;
470 vap->va_fileid = VNOVAL;
471 vap->va_blocksize = VNOVAL;
472 vap->va_rdev = VNOVAL;
473 vap->va_atime.tv_sec = VNOVAL;
474 vap->va_atime.tv_nsec = VNOVAL;
475 vap->va_mtime.tv_sec = VNOVAL;
476 vap->va_mtime.tv_nsec = VNOVAL;
477 vap->va_ctime.tv_sec = VNOVAL;
478 vap->va_ctime.tv_nsec = VNOVAL;
479 vap->va_flags = VNOVAL;
480 vap->va_gen = VNOVAL;
485 * This routine is called when we have too many vnodes. It attempts
486 * to free <count> vnodes and will potentially free vnodes that still
487 * have VM backing store (VM backing store is typically the cause
488 * of a vnode blowout so we want to do this). Therefore, this operation
489 * is not considered cheap.
491 * A number of conditions may prevent a vnode from being reclaimed.
492 * the buffer cache may have references on the vnode, a directory
493 * vnode may still have references due to the namei cache representing
494 * underlying files, or the vnode may be in active use. It is not
495 * desireable to reuse such vnodes. These conditions may cause the
496 * number of vnodes to reach some minimum value regardless of what
497 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
500 vlrureclaim(struct mount *mp)
511 * Calculate the trigger point, don't allow user
512 * screwups to blow us up. This prevents us from
513 * recycling vnodes with lots of resident pages. We
514 * aren't trying to free memory, we are trying to
517 usevnodes = desiredvnodes;
520 trigger = vmstats.v_page_count * 2 / usevnodes;
523 lwkt_gettoken(&ilock, &mntvnode_token);
524 count = mp->mnt_nvnodelistsize / 10 + 1;
525 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
529 * The VP will stick around while we hold mntvnode_token,
530 * at least until we block, so we can safely do an initial
531 * check. But we have to check again after obtaining
532 * the vnode interlock. vp->v_interlock points to stable
533 * storage so it's ok if the vp gets ripped out from
534 * under us while we are blocked.
536 if (vp->v_type == VNON ||
537 vp->v_type == VBAD ||
538 !VMIGHTFREE(vp) || /* critical path opt */
540 vp->v_object->resident_page_count >= trigger)
542 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
543 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
549 * Get the interlock, delay moving the node to the tail so
550 * we don't race against new additions to the mountlist.
552 lwkt_gettoken(&vlock, vp->v_interlock);
553 if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
554 lwkt_reltoken(&vlock);
557 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
558 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
563 if (vp->v_type == VNON ||
564 vp->v_type == VBAD ||
565 !VMIGHTFREE(vp) || /* critical path opt */
567 vp->v_object->resident_page_count >= trigger)
569 lwkt_reltoken(&vlock);
573 vgonel(vp, &vlock, curthread);
577 lwkt_reltoken(&ilock);
582 * Attempt to recycle vnodes in a context that is always safe to block.
583 * Calling vlrurecycle() from the bowels of file system code has some
584 * interesting deadlock problems.
586 static struct thread *vnlruthread;
587 static int vnlruproc_sig;
592 struct mount *mp, *nmp;
596 struct thread *td = curthread;
598 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
603 kproc_suspend_loop();
604 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
606 wakeup(&vnlruproc_sig);
607 tsleep(td, 0, "vlruwt", hz);
611 lwkt_gettoken(&ilock, &mountlist_token);
612 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
613 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
614 nmp = TAILQ_NEXT(mp, mnt_list);
617 done += vlrureclaim(mp);
618 lwkt_gettokref(&ilock);
619 nmp = TAILQ_NEXT(mp, mnt_list);
622 lwkt_reltoken(&ilock);
625 tsleep(td, 0, "vlrup", hz * 3);
631 static struct kproc_desc vnlru_kp = {
636 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
639 * Routines having to do with the management of the vnode table.
641 extern vop_t **dead_vnodeop_p;
644 * Return the next vnode from the free list.
647 getnewvnode(tag, mp, vops, vpp)
654 struct thread *td = curthread; /* XXX */
655 struct vnode *vp = NULL;
664 * Try to reuse vnodes if we hit the max. This situation only
665 * occurs in certain large-memory (2G+) situations. We cannot
666 * attempt to directly reclaim vnodes due to nasty recursion
669 while (numvnodes - freevnodes > desiredvnodes) {
670 if (vnlruproc_sig == 0) {
671 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
674 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
679 * Attempt to reuse a vnode already on the free list, allocating
680 * a new vnode if we can't find one or if we have not reached a
681 * good minimum for good LRU performance.
683 lwkt_gettoken(&ilock, &vnode_free_list_token);
684 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
687 for (count = 0; count < freevnodes; count++) {
691 * Pull the next vnode off the free list and do some
692 * sanity checks. Note that regardless of how we
693 * block, if freevnodes is non-zero there had better
694 * be something on the list.
696 vp = TAILQ_FIRST(&vnode_free_list);
698 panic("getnewvnode: free vnode isn't");
701 * Move the vnode to the end of the list so other
702 * processes do not double-block trying to recycle
703 * the same vnode (as an optimization), then get
706 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
707 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
710 * Skip vnodes that are in the process of being
711 * held or referenced. Since the act of adding or
712 * removing a vnode on the freelist requires a token
713 * and may block, the ref count may be adjusted
714 * prior to its addition or removal.
716 if (VSHOULDBUSY(vp)) {
723 * Obtain the vnode interlock and check that the
724 * vnode is still on the free list.
726 * This normally devolves into a degenerate case so
727 * it is optimal. Loop up if it isn't. Note that
728 * the vnode could be in the middle of being moved
729 * off the free list (the VSHOULDBUSY() check) and
730 * must be skipped if so.
732 lwkt_gettoken(&vlock, vp->v_interlock);
733 TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
734 freelst, v_freelist) {
738 if (vp != xvp || VSHOULDBUSY(vp)) {
744 * We now safely own the vnode. If the vnode has
745 * an object do not recycle it if its VM object
746 * has resident pages or references.
748 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
749 (object->resident_page_count || object->ref_count))
751 lwkt_reltoken(&vlock);
757 * We can almost reuse this vnode. But we don't want
758 * to recycle it if the vnode has children in the
759 * namecache because that breaks the namecache's
760 * path element chain. (YYY use nc_refs for the
763 KKASSERT(vp->v_flag & VFREE);
764 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
766 if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
767 cache_leaf_test(vp) >= 0) {
768 /* ok, we can reuse this vnode */
771 lwkt_reltoken(&vlock);
772 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
778 * If vp is non-NULL we hold it's interlock.
781 vp->v_flag |= VDOOMED;
782 vp->v_flag &= ~VFREE;
784 lwkt_reltoken(&ilock);
785 cache_purge(vp); /* YYY may block */
787 if (vp->v_type != VBAD) {
788 vgonel(vp, &vlock, td);
790 lwkt_reltoken(&vlock);
798 panic("cleaned vnode isn't");
801 panic("Clean vnode has pending I/O's");
811 vp->v_writecount = 0; /* XXX */
813 lwkt_reltoken(&ilock);
814 vp = zalloc(vnode_zone);
815 bzero(vp, sizeof(*vp));
816 vp->v_interlock = lwkt_token_pool_get(vp);
817 lwkt_token_init(&vp->v_pollinfo.vpi_token);
819 TAILQ_INIT(&vp->v_namecache);
823 TAILQ_INIT(&vp->v_cleanblkhd);
824 TAILQ_INIT(&vp->v_dirtyblkhd);
834 vfs_object_create(vp, td);
839 * Move a vnode from one mount queue to another.
848 lwkt_gettoken(&ilock, &mntvnode_token);
850 * Delete from old mount point vnode list, if on one.
852 if (vp->v_mount != NULL) {
853 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
854 ("bad mount point vnode list size"));
855 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
856 vp->v_mount->mnt_nvnodelistsize--;
859 * Insert into list of vnodes for the new mount point, if available.
861 if ((vp->v_mount = mp) == NULL) {
862 lwkt_reltoken(&ilock);
865 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
866 mp->mnt_nvnodelistsize++;
867 lwkt_reltoken(&ilock);
871 * Update outstanding I/O count and do wakeup if requested.
879 bp->b_flags &= ~B_WRITEINPROG;
880 if ((vp = bp->b_vp)) {
882 if (vp->v_numoutput < 0)
883 panic("vwakeup: neg numoutput");
884 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
885 vp->v_flag &= ~VBWAIT;
886 wakeup((caddr_t) &vp->v_numoutput);
892 * Flush out and invalidate all buffers associated with a vnode.
893 * Called with the underlying object locked.
896 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
897 int slpflag, int slptimeo)
900 struct buf *nbp, *blist;
905 if (flags & V_SAVE) {
907 while (vp->v_numoutput) {
908 vp->v_flag |= VBWAIT;
909 error = tsleep((caddr_t)&vp->v_numoutput,
910 slpflag, "vinvlbuf", slptimeo);
916 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
918 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
921 if (vp->v_numoutput > 0 ||
922 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
923 panic("vinvalbuf: dirty bufs");
929 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
931 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
935 for (bp = blist; bp; bp = nbp) {
936 nbp = TAILQ_NEXT(bp, b_vnbufs);
937 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
938 error = BUF_TIMELOCK(bp,
939 LK_EXCLUSIVE | LK_SLEEPFAIL,
940 "vinvalbuf", slpflag, slptimeo);
947 * XXX Since there are no node locks for NFS, I
948 * believe there is a slight chance that a delayed
949 * write will occur while sleeping just above, so
950 * check for it. Note that vfs_bio_awrite expects
951 * buffers to reside on a queue, while VOP_BWRITE and
954 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
957 if (bp->b_vp == vp) {
958 if (bp->b_flags & B_CLUSTEROK) {
963 bp->b_flags |= B_ASYNC;
964 VOP_BWRITE(bp->b_vp, bp);
968 (void) VOP_BWRITE(bp->b_vp, bp);
973 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
974 bp->b_flags &= ~B_ASYNC;
980 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
981 * have write I/O in-progress but if there is a VM object then the
982 * VM object can also have read-I/O in-progress.
985 while (vp->v_numoutput > 0) {
986 vp->v_flag |= VBWAIT;
987 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
989 if (VOP_GETVOBJECT(vp, &object) == 0) {
990 while (object->paging_in_progress)
991 vm_object_pip_sleep(object, "vnvlbx");
993 } while (vp->v_numoutput > 0);
998 * Destroy the copy in the VM cache, too.
1000 lwkt_gettoken(&vlock, vp->v_interlock);
1001 if (VOP_GETVOBJECT(vp, &object) == 0) {
1002 vm_object_page_remove(object, 0, 0,
1003 (flags & V_SAVE) ? TRUE : FALSE);
1005 lwkt_reltoken(&vlock);
1007 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1008 panic("vinvalbuf: flush failed");
1013 * Truncate a file's buffer and pages to a specified length. This
1014 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1018 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
1026 * Round up to the *next* lbn.
1028 trunclbn = (length + blksize - 1) / blksize;
1035 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1036 nbp = TAILQ_NEXT(bp, b_vnbufs);
1037 if (bp->b_lblkno >= trunclbn) {
1038 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1039 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1043 bp->b_flags |= (B_INVAL | B_RELBUF);
1044 bp->b_flags &= ~B_ASYNC;
1049 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1050 (nbp->b_vp != vp) ||
1051 (nbp->b_flags & B_DELWRI))) {
1057 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1058 nbp = TAILQ_NEXT(bp, b_vnbufs);
1059 if (bp->b_lblkno >= trunclbn) {
1060 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1061 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1065 bp->b_flags |= (B_INVAL | B_RELBUF);
1066 bp->b_flags &= ~B_ASYNC;
1071 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1072 (nbp->b_vp != vp) ||
1073 (nbp->b_flags & B_DELWRI) == 0)) {
1082 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1083 nbp = TAILQ_NEXT(bp, b_vnbufs);
1084 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1085 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1086 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1090 if (bp->b_vp == vp) {
1091 bp->b_flags |= B_ASYNC;
1093 bp->b_flags &= ~B_ASYNC;
1095 VOP_BWRITE(bp->b_vp, bp);
1103 while (vp->v_numoutput > 0) {
1104 vp->v_flag |= VBWAIT;
1105 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1110 vnode_pager_setsize(vp, length);
1116 * Associate a buffer with a vnode.
1125 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1129 bp->b_dev = vn_todev(vp);
1131 * Insert onto list for new vnode.
1134 bp->b_xflags |= BX_VNCLEAN;
1135 bp->b_xflags &= ~BX_VNDIRTY;
1136 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1141 * Disassociate a buffer from a vnode.
1148 struct buflists *listheadp;
1151 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1154 * Delete from old vnode list, if on one.
1158 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1159 if (bp->b_xflags & BX_VNDIRTY)
1160 listheadp = &vp->v_dirtyblkhd;
1162 listheadp = &vp->v_cleanblkhd;
1163 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1164 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1166 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1167 vp->v_flag &= ~VONWORKLST;
1168 LIST_REMOVE(vp, v_synclist);
1171 bp->b_vp = (struct vnode *) 0;
1176 * The workitem queue.
1178 * It is useful to delay writes of file data and filesystem metadata
1179 * for tens of seconds so that quickly created and deleted files need
1180 * not waste disk bandwidth being created and removed. To realize this,
1181 * we append vnodes to a "workitem" queue. When running with a soft
1182 * updates implementation, most pending metadata dependencies should
1183 * not wait for more than a few seconds. Thus, mounted on block devices
1184 * are delayed only about a half the time that file data is delayed.
1185 * Similarly, directory updates are more critical, so are only delayed
1186 * about a third the time that file data is delayed. Thus, there are
1187 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1188 * one each second (driven off the filesystem syncer process). The
1189 * syncer_delayno variable indicates the next queue that is to be processed.
1190 * Items that need to be processed soon are placed in this queue:
1192 * syncer_workitem_pending[syncer_delayno]
1194 * A delay of fifteen seconds is done by placing the request fifteen
1195 * entries later in the queue:
1197 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1202 * Add an item to the syncer work queue.
1205 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1211 if (vp->v_flag & VONWORKLST) {
1212 LIST_REMOVE(vp, v_synclist);
1215 if (delay > syncer_maxdelay - 2)
1216 delay = syncer_maxdelay - 2;
1217 slot = (syncer_delayno + delay) & syncer_mask;
1219 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1220 vp->v_flag |= VONWORKLST;
1224 struct thread *updatethread;
1225 static void sched_sync (void);
1226 static struct kproc_desc up_kp = {
1231 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1234 * System filesystem synchronizer daemon.
1239 struct synclist *slp;
1243 struct thread *td = curthread;
1245 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1249 kproc_suspend_loop();
1251 starttime = time_second;
1254 * Push files whose dirty time has expired. Be careful
1255 * of interrupt race on slp queue.
1258 slp = &syncer_workitem_pending[syncer_delayno];
1259 syncer_delayno += 1;
1260 if (syncer_delayno == syncer_maxdelay)
1264 while ((vp = LIST_FIRST(slp)) != NULL) {
1265 if (VOP_ISLOCKED(vp, NULL) == 0) {
1266 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1267 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1268 VOP_UNLOCK(vp, NULL, 0, td);
1271 if (LIST_FIRST(slp) == vp) {
1273 * Note: v_tag VT_VFS vps can remain on the
1274 * worklist too with no dirty blocks, but
1275 * since sync_fsync() moves it to a different
1278 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1279 !vn_isdisk(vp, NULL))
1280 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1282 * Put us back on the worklist. The worklist
1283 * routine will remove us from our current
1284 * position and then add us back in at a later
1287 vn_syncer_add_to_worklist(vp, syncdelay);
1293 * Do soft update processing.
1296 (*bioops.io_sync)(NULL);
1299 * The variable rushjob allows the kernel to speed up the
1300 * processing of the filesystem syncer process. A rushjob
1301 * value of N tells the filesystem syncer to process the next
1302 * N seconds worth of work on its queue ASAP. Currently rushjob
1303 * is used by the soft update code to speed up the filesystem
1304 * syncer process when the incore state is getting so far
1305 * ahead of the disk that the kernel memory pool is being
1306 * threatened with exhaustion.
1313 * If it has taken us less than a second to process the
1314 * current work, then wait. Otherwise start right over
1315 * again. We can still lose time if any single round
1316 * takes more than two seconds, but it does not really
1317 * matter as we are just trying to generally pace the
1318 * filesystem activity.
1320 if (time_second == starttime)
1321 tsleep(&lbolt, 0, "syncer", 0);
1326 * Request the syncer daemon to speed up its work.
1327 * We never push it to speed up more than half of its
1328 * normal turn time, otherwise it could take over the cpu.
1330 * YYY wchan field protected by the BGL.
1336 if (updatethread->td_wchan == &lbolt) { /* YYY */
1337 unsleep(updatethread);
1338 lwkt_schedule(updatethread);
1341 if (rushjob < syncdelay / 2) {
1343 stat_rush_requests += 1;
1350 * Associate a p-buffer with a vnode.
1352 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1353 * with the buffer. i.e. the bp has not been linked into the vnode or
1362 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1365 bp->b_flags |= B_PAGING;
1366 bp->b_dev = vn_todev(vp);
1370 * Disassociate a p-buffer from a vnode.
1377 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1380 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1382 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1387 bp->b_vp = (struct vnode *) 0;
1388 bp->b_flags &= ~B_PAGING;
1392 pbreassignbuf(bp, newvp)
1394 struct vnode *newvp;
1396 if ((bp->b_flags & B_PAGING) == 0) {
1398 "pbreassignbuf() on non phys bp %p",
1406 * Reassign a buffer from one vnode to another.
1407 * Used to assign file specific control information
1408 * (indirect blocks) to the vnode to which they belong.
1411 reassignbuf(bp, newvp)
1413 struct vnode *newvp;
1415 struct buflists *listheadp;
1419 if (newvp == NULL) {
1420 printf("reassignbuf: NULL");
1426 * B_PAGING flagged buffers cannot be reassigned because their vp
1427 * is not fully linked in.
1429 if (bp->b_flags & B_PAGING)
1430 panic("cannot reassign paging buffer");
1434 * Delete from old vnode list, if on one.
1436 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1437 if (bp->b_xflags & BX_VNDIRTY)
1438 listheadp = &bp->b_vp->v_dirtyblkhd;
1440 listheadp = &bp->b_vp->v_cleanblkhd;
1441 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1442 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1443 if (bp->b_vp != newvp) {
1445 bp->b_vp = NULL; /* for clarification */
1449 * If dirty, put on list of dirty buffers; otherwise insert onto list
1452 if (bp->b_flags & B_DELWRI) {
1455 listheadp = &newvp->v_dirtyblkhd;
1456 if ((newvp->v_flag & VONWORKLST) == 0) {
1457 switch (newvp->v_type) {
1463 if (newvp->v_rdev &&
1464 newvp->v_rdev->si_mountpoint != NULL) {
1472 vn_syncer_add_to_worklist(newvp, delay);
1474 bp->b_xflags |= BX_VNDIRTY;
1475 tbp = TAILQ_FIRST(listheadp);
1477 bp->b_lblkno == 0 ||
1478 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1479 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1480 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1481 ++reassignbufsortgood;
1482 } else if (bp->b_lblkno < 0) {
1483 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1484 ++reassignbufsortgood;
1485 } else if (reassignbufmethod == 1) {
1487 * New sorting algorithm, only handle sequential case,
1488 * otherwise append to end (but before metadata)
1490 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1491 (tbp->b_xflags & BX_VNDIRTY)) {
1493 * Found the best place to insert the buffer
1495 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1496 ++reassignbufsortgood;
1499 * Missed, append to end, but before meta-data.
1500 * We know that the head buffer in the list is
1501 * not meta-data due to prior conditionals.
1503 * Indirect effects: NFS second stage write
1504 * tends to wind up here, giving maximum
1505 * distance between the unstable write and the
1508 tbp = TAILQ_LAST(listheadp, buflists);
1509 while (tbp && tbp->b_lblkno < 0)
1510 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1511 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1512 ++reassignbufsortbad;
1516 * Old sorting algorithm, scan queue and insert
1519 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1520 (ttbp->b_lblkno < bp->b_lblkno)) {
1524 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1527 bp->b_xflags |= BX_VNCLEAN;
1528 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1529 if ((newvp->v_flag & VONWORKLST) &&
1530 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1531 newvp->v_flag &= ~VONWORKLST;
1532 LIST_REMOVE(newvp, v_synclist);
1535 if (bp->b_vp != newvp) {
1543 * Create a vnode for a block device.
1544 * Used for mounting the root file system.
1547 bdevvp(dev_t dev, struct vnode **vpp)
1557 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1564 vp->v_udev = dev->si_udev;
1570 v_associate_rdev(struct vnode *vp, dev_t dev)
1574 if (dev == NULL || dev == NODEV)
1576 if (dev_is_good(dev) == 0)
1578 KKASSERT(vp->v_rdev == NULL);
1581 vp->v_rdev = reference_dev(dev);
1582 lwkt_gettoken(&ilock, &spechash_token);
1583 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
1584 lwkt_reltoken(&ilock);
1589 v_release_rdev(struct vnode *vp)
1594 if ((dev = vp->v_rdev) != NULL) {
1595 lwkt_gettoken(&ilock, &spechash_token);
1596 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1601 lwkt_reltoken(&ilock);
1606 * Add a vnode to the alias list hung off the dev_t. We only associate
1607 * the device number with the vnode. The actual device is not associated
1608 * until the vnode is opened (usually in spec_open()), and will be
1609 * disassociated on last close.
1612 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1614 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1615 panic("addaliasu on non-special vnode");
1616 nvp->v_udev = nvp_udev;
1620 * Grab a particular vnode from the free list, increment its
1621 * reference count and lock it. The vnode lock bit is set if the
1622 * vnode is being eliminated in vgone. The process is awakened
1623 * when the transition is completed, and an error returned to
1624 * indicate that the vnode is no longer usable (possibly having
1625 * been changed to a new file system type).
1627 * This code is very sensitive. We are depending on the vnode interlock
1628 * to be maintained through to the vn_lock() call, which means that we
1629 * cannot block which means that we cannot call vbusy() until after vn_lock().
1630 * If the interlock is not maintained, the VXLOCK check will not properly
1631 * interlock against a vclean()'s LK_DRAIN operation on the lock.
1634 vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
1640 * We need the interlock to safely modify the v_ fields. ZZZ it is
1641 * only legal to pass (1) the vnode's interlock and (2) only pass
1642 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1645 if ((flags & LK_INTERLOCK) == 0) {
1646 lwkt_gettoken(&vvlock, vp->v_interlock);
1651 * If the vnode is in the process of being cleaned out for
1652 * another use, we wait for the cleaning to finish and then
1653 * return failure. Cleaning is determined by checking that
1654 * the VXLOCK flag is set. It is possible for the vnode to be
1655 * self-referenced during the cleaning operation.
1657 if (vp->v_flag & VXLOCK) {
1658 if (vp->v_vxthread == curthread) {
1660 /* this can now occur in normal operation */
1661 log(LOG_INFO, "VXLOCK interlock avoided\n");
1664 vp->v_flag |= VXWANT;
1665 lwkt_reltoken(vlock);
1666 tsleep((caddr_t)vp, 0, "vget", 0);
1672 * Bump v_usecount to prevent the vnode from being recycled. The
1673 * usecount needs to be bumped before we successfully get our lock.
1676 if (flags & LK_TYPE_MASK) {
1677 if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
1679 * must expand vrele here because we do not want
1680 * to call VOP_INACTIVE if the reference count
1681 * drops back to zero since it was never really
1682 * active. We must remove it from the free list
1683 * before sleeping so that multiple processes do
1684 * not try to recycle it.
1686 lwkt_gettokref(vlock);
1689 lwkt_reltoken(vlock);
1693 if (VSHOULDBUSY(vp))
1694 vbusy(vp); /* interlock must be held on call */
1695 lwkt_reltoken(vlock);
1700 vref(struct vnode *vp)
1702 crit_enter(); /* YYY use crit section for moment / BGL protected */
1708 * Vnode put/release.
1709 * If count drops to zero, call inactive routine and return to freelist.
1712 vrele(struct vnode *vp)
1714 struct thread *td = curthread; /* XXX */
1717 KASSERT(vp != NULL && vp->v_usecount >= 0,
1718 ("vrele: null vp or <=0 v_usecount"));
1720 lwkt_gettoken(&vlock, vp->v_interlock);
1722 if (vp->v_usecount > 1) {
1724 lwkt_reltoken(&vlock);
1728 if (vp->v_usecount == 1) {
1731 * We must call VOP_INACTIVE with the node locked and the
1732 * usecount 0. If we are doing a vpu, the node is already
1733 * locked, but, in the case of vrele, we must explicitly lock
1734 * the vnode before calling VOP_INACTIVE.
1737 if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
1738 VOP_INACTIVE(vp, td);
1740 lwkt_reltoken(&vlock);
1743 vprint("vrele: negative ref count", vp);
1745 lwkt_reltoken(&vlock);
1746 panic("vrele: negative ref cnt");
1751 vput(struct vnode *vp)
1753 struct thread *td = curthread; /* XXX */
1756 KASSERT(vp != NULL, ("vput: null vp"));
1758 lwkt_gettoken(&vlock, vp->v_interlock);
1760 if (vp->v_usecount > 1) {
1762 VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
1766 if (vp->v_usecount == 1) {
1769 * We must call VOP_INACTIVE with the node locked.
1770 * If we are doing a vpu, the node is already locked,
1771 * so we just need to release the vnode mutex.
1773 VOP_INACTIVE(vp, td);
1775 lwkt_reltoken(&vlock);
1778 vprint("vput: negative ref count", vp);
1780 lwkt_reltoken(&vlock);
1781 panic("vput: negative ref cnt");
1786 * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1787 * be held but isn't.
1797 if (VSHOULDBUSY(vp))
1798 vbusy(vp); /* interlock must be held on call */
1803 * One less who cares about this vnode.
1811 lwkt_gettoken(&vlock, vp->v_interlock);
1812 if (vp->v_holdcnt <= 0)
1813 panic("vdrop: holdcnt");
1816 lwkt_reltoken(&vlock);
1822 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1823 int (*slowfunc)(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data),
1833 * Scan the vnodes on the mount's vnode list. Use a placemarker
1835 pvp = zalloc(vnode_zone);
1836 pvp->v_flag |= VPLACEMARKER;
1838 lwkt_gettoken(&ilock, &mntvnode_token);
1839 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1841 while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1843 * Move the placemarker and skip other placemarkers we
1844 * encounter. The nothing can get in our way so the
1845 * mount point on the vp must be valid.
1847 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1848 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1849 if (vp->v_flag & VPLACEMARKER)
1851 KKASSERT(vp->v_mount == mp);
1857 if ((r = fastfunc(mp, vp, data)) < 0)
1864 * Get the vnodes interlock and make sure it is still on the
1865 * mount list. Skip it if it has moved (we may encounter it
1866 * later). Then do the with-interlock test. The callback
1867 * is responsible for releasing the vnode interlock.
1869 * The interlock is type-stable.
1872 lwkt_gettoken(&vlock, vp->v_interlock);
1873 if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1874 printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1875 lwkt_reltoken(&vlock);
1878 if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1879 KKASSERT(lwkt_havetokref(&vlock) == 0);
1882 KKASSERT(lwkt_havetokref(&vlock) == 0);
1885 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1886 zfree(vnode_zone, pvp);
1887 lwkt_reltoken(&ilock);
1892 * Remove any vnodes in the vnode table belonging to mount point mp.
1894 * If FORCECLOSE is not specified, there should not be any active ones,
1895 * return error if any are found (nb: this is a user error, not a
1896 * system error). If FORCECLOSE is specified, detach any active vnodes
1899 * If WRITECLOSE is set, only flush out regular file vnodes open for
1902 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1904 * `rootrefs' specifies the base reference count for the root vnode
1905 * of this filesystem. The root vnode is considered busy if its
1906 * v_usecount exceeds this value. On a successful return, vflush()
1907 * will call vrele() on the root vnode exactly rootrefs times.
1908 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1912 static int busyprt = 0; /* print out busy vnodes */
1913 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1916 static int vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data);
1918 struct vflush_info {
1925 vflush(mp, rootrefs, flags)
1930 struct thread *td = curthread; /* XXX */
1931 struct vnode *rootvp = NULL;
1934 struct vflush_info vflush_info;
1937 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1938 ("vflush: bad args"));
1940 * Get the filesystem root vnode. We can vput() it
1941 * immediately, since with rootrefs > 0, it won't go away.
1943 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1948 vflush_info.busy = 0;
1949 vflush_info.flags = flags;
1950 vflush_info.td = td;
1951 vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
1953 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1955 * If just the root vnode is busy, and if its refcount
1956 * is equal to `rootrefs', then go ahead and kill it.
1958 lwkt_gettoken(&vlock, rootvp->v_interlock);
1959 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1960 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1961 if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1962 vgonel(rootvp, &vlock, td);
1963 vflush_info.busy = 0;
1965 lwkt_reltoken(&vlock);
1968 if (vflush_info.busy)
1970 for (; rootrefs > 0; rootrefs--)
1976 * The scan callback is made with an interlocked vnode.
1979 vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
1981 struct vflush_info *info = data;
1985 * Skip over a vnodes marked VSYSTEM.
1987 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1988 lwkt_reltoken(vlock);
1993 * If WRITECLOSE is set, flush out unlinked but still open
1994 * files (even if open only for reading) and regular file
1995 * vnodes open for writing.
1997 if ((info->flags & WRITECLOSE) &&
1998 (vp->v_type == VNON ||
1999 (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
2000 vattr.va_nlink > 0)) &&
2001 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2002 lwkt_reltoken(vlock);
2007 * With v_usecount == 0, all we need to do is clear out the
2008 * vnode data structures and we are done.
2010 if (vp->v_usecount == 0) {
2011 vgonel(vp, vlock, info->td);
2016 * If FORCECLOSE is set, forcibly close the vnode. For block
2017 * or character devices, revert to an anonymous device. For
2018 * all other files, just kill them.
2020 if (info->flags & FORCECLOSE) {
2021 if (vp->v_type != VBLK && vp->v_type != VCHR) {
2022 vgonel(vp, vlock, info->td);
2024 vclean(vp, vlock, 0, info->td);
2025 vp->v_op = spec_vnodeop_p;
2026 insmntque(vp, (struct mount *) 0);
2032 vprint("vflush: busy vnode", vp);
2034 lwkt_reltoken(vlock);
2040 * Disassociate the underlying file system from a vnode.
2043 vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
2048 * Check to see if the vnode is in use. If so we have to reference it
2049 * before we clean it out so that its count cannot fall to zero and
2050 * generate a race against ourselves to recycle it.
2052 if ((active = vp->v_usecount))
2056 * Prevent the vnode from being recycled or brought into use while we
2059 if (vp->v_flag & VXLOCK)
2060 panic("vclean: deadlock");
2061 vp->v_flag |= VXLOCK;
2062 vp->v_vxthread = curthread;
2065 * Even if the count is zero, the VOP_INACTIVE routine may still
2066 * have the object locked while it cleans it out. The VOP_LOCK
2067 * ensures that the VOP_INACTIVE routine is done with its work.
2068 * For active vnodes, it ensures that no other activity can
2069 * occur while the underlying object is being cleaned out.
2071 * NOTE: we continue to hold the vnode interlock through to the
2074 VOP_LOCK(vp, NULL, LK_DRAIN, td);
2077 * Clean out any buffers associated with the vnode.
2079 vinvalbuf(vp, V_SAVE, td, 0, 0);
2080 VOP_DESTROYVOBJECT(vp);
2083 * If purging an active vnode, it must be closed and
2084 * deactivated before being reclaimed. Note that the
2085 * VOP_INACTIVE will unlock the vnode.
2088 if (flags & DOCLOSE)
2089 VOP_CLOSE(vp, FNONBLOCK, td);
2090 VOP_INACTIVE(vp, td);
2093 * Any other processes trying to obtain this lock must first
2094 * wait for VXLOCK to clear, then call the new lock operation.
2096 VOP_UNLOCK(vp, NULL, 0, td);
2099 * Reclaim the vnode.
2101 if (VOP_RECLAIM(vp, td))
2102 panic("vclean: cannot reclaim");
2106 * Inline copy of vrele() since VOP_INACTIVE
2107 * has already been called.
2109 if (--vp->v_usecount <= 0) {
2111 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2112 vprint("vclean: bad ref count", vp);
2113 panic("vclean: ref cnt");
2121 vp->v_vnlock = NULL;
2125 * Done with purge, notify sleepers of the grim news.
2127 vp->v_op = dead_vnodeop_p;
2130 vp->v_flag &= ~VXLOCK;
2131 vp->v_vxthread = NULL;
2132 if (vp->v_flag & VXWANT) {
2133 vp->v_flag &= ~VXWANT;
2134 wakeup((caddr_t) vp);
2136 lwkt_reltoken(vlock);
2140 * Eliminate all activity associated with the requested vnode
2141 * and with all vnodes aliased to the requested vnode.
2145 struct vop_revoke_args /* {
2150 struct vnode *vp, *vq;
2154 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2158 * If a vgone (or vclean) is already in progress,
2159 * wait until it is done and return.
2161 if (vp->v_flag & VXLOCK) {
2162 vp->v_flag |= VXWANT;
2163 /*lwkt_reltoken(vlock); ZZZ */
2164 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2169 * If the vnode has a device association, scrap all vnodes associated
2170 * with the device. Don't let the device disappear on us while we
2171 * are scrapping the vnodes.
2173 if (vp->v_type != VCHR && vp->v_type != VBLK)
2175 if ((dev = vp->v_rdev) == NULL) {
2176 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
2181 lwkt_gettoken(&ilock, &spechash_token);
2182 vq = SLIST_FIRST(&dev->si_hlist);
2183 lwkt_reltoken(&ilock);
2193 * Recycle an unused vnode to the front of the free list.
2194 * Release the passed interlock if the vnode will be recycled.
2197 vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
2201 lwkt_gettoken(&vlock, vp->v_interlock);
2202 if (vp->v_usecount == 0) {
2204 lwkt_reltoken(inter_lkp);
2205 vgonel(vp, &vlock, td);
2208 lwkt_reltoken(&vlock);
2213 * Eliminate all activity associated with a vnode
2214 * in preparation for reuse.
2217 vgone(struct vnode *vp)
2219 struct thread *td = curthread; /* XXX */
2222 lwkt_gettoken(&vlock, vp->v_interlock);
2223 vgonel(vp, &vlock, td);
2227 * vgone, with the vp interlock held.
2230 vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
2236 * If a vgone (or vclean) is already in progress,
2237 * wait until it is done and return.
2239 if (vp->v_flag & VXLOCK) {
2240 vp->v_flag |= VXWANT;
2241 lwkt_reltoken(vlock);
2242 tsleep((caddr_t)vp, 0, "vgone", 0);
2247 * Clean out the filesystem specific data.
2249 vclean(vp, vlock, DOCLOSE, td);
2250 lwkt_gettokref(vlock);
2253 * Delete from old mount point vnode list, if on one.
2255 if (vp->v_mount != NULL)
2256 insmntque(vp, (struct mount *)0);
2258 * If special device, remove it from special device alias list
2261 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2266 * If it is on the freelist and not already at the head,
2267 * move it to the head of the list. The test of the
2268 * VDOOMED flag and the reference count of zero is because
2269 * it will be removed from the free list by getnewvnode,
2270 * but will not have its reference count incremented until
2271 * after calling vgone. If the reference count were
2272 * incremented first, vgone would (incorrectly) try to
2273 * close the previous instance of the underlying object.
2275 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2277 lwkt_gettoken(&ilock, &vnode_free_list_token);
2278 if (vp->v_flag & VFREE)
2279 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2282 vp->v_flag |= VFREE;
2283 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2284 lwkt_reltoken(&ilock);
2288 lwkt_reltoken(vlock);
2292 * Lookup a vnode by device number.
2295 vfinddev(dev, type, vpp)
2303 lwkt_gettoken(&ilock, &spechash_token);
2304 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2305 if (type == vp->v_type) {
2307 lwkt_reltoken(&ilock);
2311 lwkt_reltoken(&ilock);
2316 * Calculate the total number of references to a special device. This
2317 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
2318 * an overloaded field. Since udev2dev can now return NODEV, we have
2319 * to check for a NULL v_rdev.
2322 count_dev(dev_t dev)
2328 if (SLIST_FIRST(&dev->si_hlist)) {
2329 lwkt_gettoken(&ilock, &spechash_token);
2330 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2331 count += vp->v_usecount;
2333 lwkt_reltoken(&ilock);
2339 count_udev(udev_t udev)
2343 if ((dev = udev2dev(udev, 0)) == NODEV)
2345 return(count_dev(dev));
2349 vcount(struct vnode *vp)
2351 if (vp->v_rdev == NULL)
2353 return(count_dev(vp->v_rdev));
2357 * Print out a description of a vnode.
2359 static char *typename[] =
2360 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2370 printf("%s: %p: ", label, (void *)vp);
2372 printf("%p: ", (void *)vp);
2373 printf("type %s, usecount %d, writecount %d, refcount %d,",
2374 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2377 if (vp->v_flag & VROOT)
2378 strcat(buf, "|VROOT");
2379 if (vp->v_flag & VTEXT)
2380 strcat(buf, "|VTEXT");
2381 if (vp->v_flag & VSYSTEM)
2382 strcat(buf, "|VSYSTEM");
2383 if (vp->v_flag & VXLOCK)
2384 strcat(buf, "|VXLOCK");
2385 if (vp->v_flag & VXWANT)
2386 strcat(buf, "|VXWANT");
2387 if (vp->v_flag & VBWAIT)
2388 strcat(buf, "|VBWAIT");
2389 if (vp->v_flag & VDOOMED)
2390 strcat(buf, "|VDOOMED");
2391 if (vp->v_flag & VFREE)
2392 strcat(buf, "|VFREE");
2393 if (vp->v_flag & VOBJBUF)
2394 strcat(buf, "|VOBJBUF");
2396 printf(" flags (%s)", &buf[1]);
2397 if (vp->v_data == NULL) {
2406 #include <ddb/ddb.h>
2408 * List all of the locked vnodes in the system.
2409 * Called when debugging the kernel.
2411 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2413 struct thread *td = curthread; /* XXX */
2415 struct mount *mp, *nmp;
2418 printf("Locked vnodes\n");
2419 lwkt_gettoken(&ilock, &mountlist_token);
2420 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2421 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
2422 nmp = TAILQ_NEXT(mp, mnt_list);
2425 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2426 if (VOP_ISLOCKED(vp, NULL))
2427 vprint((char *)0, vp);
2429 lwkt_gettokref(&ilock);
2430 nmp = TAILQ_NEXT(mp, mnt_list);
2433 lwkt_reltoken(&ilock);
2438 * Top level filesystem related information gathering.
2440 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2443 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2445 int *name = (int *)arg1 - 1; /* XXX */
2446 u_int namelen = arg2 + 1; /* XXX */
2447 struct vfsconf *vfsp;
2449 #if 1 || defined(COMPAT_PRELITE2)
2450 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2452 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2456 /* all sysctl names at this level are at least name and field */
2458 return (ENOTDIR); /* overloaded */
2459 if (name[0] != VFS_GENERIC) {
2460 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2461 if (vfsp->vfc_typenum == name[0])
2464 return (EOPNOTSUPP);
2465 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2466 oldp, oldlenp, newp, newlen, p));
2470 case VFS_MAXTYPENUM:
2473 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2476 return (ENOTDIR); /* overloaded */
2477 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2478 if (vfsp->vfc_typenum == name[2])
2481 return (EOPNOTSUPP);
2482 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2484 return (EOPNOTSUPP);
2487 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2488 "Generic filesystem");
2490 #if 1 || defined(COMPAT_PRELITE2)
2493 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2496 struct vfsconf *vfsp;
2497 struct ovfsconf ovfs;
2499 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2500 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2501 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2502 ovfs.vfc_index = vfsp->vfc_typenum;
2503 ovfs.vfc_refcount = vfsp->vfc_refcount;
2504 ovfs.vfc_flags = vfsp->vfc_flags;
2505 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2512 #endif /* 1 || COMPAT_PRELITE2 */
2515 #define KINFO_VNODESLOP 10
2517 * Dump vnode list (via sysctl).
2518 * Copyout address of vnode followed by vnode.
2522 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2524 struct proc *p = curproc; /* XXX */
2525 struct mount *mp, *nmp;
2526 struct vnode *nvp, *vp;
2531 #define VPTRSZ sizeof (struct vnode *)
2532 #define VNODESZ sizeof (struct vnode)
2535 if (!req->oldptr) /* Make an estimate */
2536 return (SYSCTL_OUT(req, 0,
2537 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2539 lwkt_gettoken(&ilock, &mountlist_token);
2540 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2541 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
2542 nmp = TAILQ_NEXT(mp, mnt_list);
2545 lwkt_gettoken(&jlock, &mntvnode_token);
2547 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2551 * Check that the vp is still associated with
2552 * this filesystem. RACE: could have been
2553 * recycled onto the same filesystem.
2555 if (vp->v_mount != mp)
2557 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2558 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2559 (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2560 lwkt_reltoken(&jlock);
2564 lwkt_reltoken(&jlock);
2565 lwkt_gettokref(&ilock);
2566 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */
2569 lwkt_reltoken(&ilock);
2577 * Exporting the vnode list on large systems causes them to crash.
2578 * Exporting the vnode list on medium systems causes sysctl to coredump.
2581 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2582 0, 0, sysctl_vnode, "S,vnode", "");
2586 * Check to see if a filesystem is mounted on a block device.
2589 vfs_mountedon(struct vnode *vp)
2593 if ((dev = vp->v_rdev) == NULL)
2594 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
2595 if (dev != NODEV && dev->si_mountpoint)
2601 * Unmount all filesystems. The list is traversed in reverse order
2602 * of mounting to avoid dependencies.
2608 struct thread *td = curthread;
2611 if (td->td_proc == NULL)
2612 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2615 * Since this only runs when rebooting, it is not interlocked.
2617 while(!TAILQ_EMPTY(&mountlist)) {
2618 mp = TAILQ_LAST(&mountlist, mntlist);
2619 error = dounmount(mp, MNT_FORCE, td);
2621 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2622 printf("unmount of %s failed (",
2623 mp->mnt_stat.f_mntonname);
2627 printf("%d)\n", error);
2629 /* The unmount has removed mp from the mountlist */
2635 * Build hash lists of net addresses and hang them off the mount point.
2636 * Called by ufs_mount() to set up the lists of export addresses.
2639 vfs_hang_addrlist(mp, nep, argp)
2641 struct netexport *nep;
2642 struct export_args *argp;
2645 struct radix_node_head *rnh;
2647 struct radix_node *rn;
2648 struct sockaddr *saddr, *smask = 0;
2652 if (argp->ex_addrlen == 0) {
2653 if (mp->mnt_flag & MNT_DEFEXPORTED)
2655 np = &nep->ne_defexported;
2656 np->netc_exflags = argp->ex_flags;
2657 np->netc_anon = argp->ex_anon;
2658 np->netc_anon.cr_ref = 1;
2659 mp->mnt_flag |= MNT_DEFEXPORTED;
2663 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2665 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2668 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2669 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2670 bzero((caddr_t) np, i);
2671 saddr = (struct sockaddr *) (np + 1);
2672 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2674 if (saddr->sa_len > argp->ex_addrlen)
2675 saddr->sa_len = argp->ex_addrlen;
2676 if (argp->ex_masklen) {
2677 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2678 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2681 if (smask->sa_len > argp->ex_masklen)
2682 smask->sa_len = argp->ex_masklen;
2684 i = saddr->sa_family;
2685 if ((rnh = nep->ne_rtable[i]) == 0) {
2687 * Seems silly to initialize every AF when most are not used,
2688 * do so on demand here
2690 for (dom = domains; dom; dom = dom->dom_next)
2691 if (dom->dom_family == i && dom->dom_rtattach) {
2692 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2696 if ((rnh = nep->ne_rtable[i]) == 0) {
2701 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2703 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2707 np->netc_exflags = argp->ex_flags;
2708 np->netc_anon = argp->ex_anon;
2709 np->netc_anon.cr_ref = 1;
2712 free(np, M_NETADDR);
2718 vfs_free_netcred(rn, w)
2719 struct radix_node *rn;
2722 struct radix_node_head *rnh = (struct radix_node_head *) w;
2724 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2725 free((caddr_t) rn, M_NETADDR);
2730 * Free the net address hash lists that are hanging off the mount points.
2733 vfs_free_addrlist(nep)
2734 struct netexport *nep;
2737 struct radix_node_head *rnh;
2739 for (i = 0; i <= AF_MAX; i++)
2740 if ((rnh = nep->ne_rtable[i])) {
2741 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2743 free((caddr_t) rnh, M_RTABLE);
2744 nep->ne_rtable[i] = 0;
2749 vfs_export(mp, nep, argp)
2751 struct netexport *nep;
2752 struct export_args *argp;
2756 if (argp->ex_flags & MNT_DELEXPORT) {
2757 if (mp->mnt_flag & MNT_EXPUBLIC) {
2758 vfs_setpublicfs(NULL, NULL, NULL);
2759 mp->mnt_flag &= ~MNT_EXPUBLIC;
2761 vfs_free_addrlist(nep);
2762 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2764 if (argp->ex_flags & MNT_EXPORTED) {
2765 if (argp->ex_flags & MNT_EXPUBLIC) {
2766 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2768 mp->mnt_flag |= MNT_EXPUBLIC;
2770 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2772 mp->mnt_flag |= MNT_EXPORTED;
2779 * Set the publicly exported filesystem (WebNFS). Currently, only
2780 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2783 vfs_setpublicfs(mp, nep, argp)
2785 struct netexport *nep;
2786 struct export_args *argp;
2793 * mp == NULL -> invalidate the current info, the FS is
2794 * no longer exported. May be called from either vfs_export
2795 * or unmount, so check if it hasn't already been done.
2798 if (nfs_pub.np_valid) {
2799 nfs_pub.np_valid = 0;
2800 if (nfs_pub.np_index != NULL) {
2801 FREE(nfs_pub.np_index, M_TEMP);
2802 nfs_pub.np_index = NULL;
2809 * Only one allowed at a time.
2811 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2815 * Get real filehandle for root of exported FS.
2817 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2818 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2820 if ((error = VFS_ROOT(mp, &rvp)))
2823 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2829 * If an indexfile was specified, pull it in.
2831 if (argp->ex_indexfile != NULL) {
2832 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2834 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2835 MAXNAMLEN, (size_t *)0);
2838 * Check for illegal filenames.
2840 for (cp = nfs_pub.np_index; *cp; cp++) {
2848 FREE(nfs_pub.np_index, M_TEMP);
2853 nfs_pub.np_mount = mp;
2854 nfs_pub.np_valid = 1;
2859 vfs_export_lookup(mp, nep, nam)
2861 struct netexport *nep;
2862 struct sockaddr *nam;
2865 struct radix_node_head *rnh;
2866 struct sockaddr *saddr;
2869 if (mp->mnt_flag & MNT_EXPORTED) {
2871 * Lookup in the export list first.
2875 rnh = nep->ne_rtable[saddr->sa_family];
2877 np = (struct netcred *)
2878 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2880 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2885 * If no address match, use the default if it exists.
2887 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2888 np = &nep->ne_defexported;
2894 * perform msync on all vnodes under a mount point. The mount point must
2895 * be locked. This code is also responsible for lazy-freeing unreferenced
2896 * vnodes whos VM objects no longer contain pages.
2898 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2900 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2901 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2902 lwkt_tokref_t vlock, void *data);
2905 vfs_msync(struct mount *mp, int flags)
2907 vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2911 * scan1 is a fast pre-check. There could be hundreds of thousands of
2912 * vnodes, we cannot afford to do anything heavy weight until we have a
2913 * fairly good indication that there is work to do.
2917 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2919 int flags = (int)data;
2921 if ((vp->v_flag & VXLOCK) == 0) {
2922 if (VSHOULDFREE(vp))
2924 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2925 (vp->v_flag & VOBJDIRTY) &&
2926 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2935 vfs_msync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
2939 int flags = (int)data;
2941 if (vp->v_flag & VXLOCK)
2944 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2945 (vp->v_flag & VOBJDIRTY) &&
2946 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2947 error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2949 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2950 vm_object_page_clean(obj, 0, 0,
2951 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2958 lwkt_reltoken(vlock);
2963 * Create the VM object needed for VMIO and mmap support. This
2964 * is done for all VREG files in the system. Some filesystems might
2965 * afford the additional metadata buffering capability of the
2966 * VMIO code by making the device node be VMIO mode also.
2968 * vp must be locked when vfs_object_create is called.
2971 vfs_object_create(struct vnode *vp, struct thread *td)
2973 return (VOP_CREATEVOBJECT(vp, td));
2977 * NOTE: the vnode interlock must be held during the call. We have to recheck
2978 * the VFREE flag since the vnode may have been removed from the free list
2979 * while we were blocked on vnode_free_list_token. The use or hold count
2980 * must have already been bumped by the caller.
2983 vbusy(struct vnode *vp)
2987 lwkt_gettoken(&ilock, &vnode_free_list_token);
2988 if ((vp->v_flag & VFREE) != 0) {
2989 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2991 vp->v_flag &= ~(VFREE|VAGE);
2993 lwkt_reltoken(&ilock);
2997 * NOTE: the vnode interlock must be held during the call. The use or hold
2998 * count must have already been bumped by the caller. We use a VINFREE to
2999 * interlock against other calls to vfree() which might occur while we
3000 * are blocked. The vnode cannot be reused until it has actually been
3001 * placed on the free list, so there are no other races even though the
3002 * use and hold counts are 0.
3005 vfree(struct vnode *vp)
3009 if ((vp->v_flag & VINFREE) == 0) {
3010 vp->v_flag |= VINFREE;
3011 lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
3012 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
3013 if (vp->v_flag & VAGE) {
3014 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3016 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3019 vp->v_flag &= ~(VAGE|VINFREE);
3020 vp->v_flag |= VFREE;
3021 lwkt_reltoken(&ilock); /* can block */
3027 * Record a process's interest in events which might happen to
3028 * a vnode. Because poll uses the historic select-style interface
3029 * internally, this routine serves as both the ``check for any
3030 * pending events'' and the ``record my interest in future events''
3031 * functions. (These are done together, while the lock is held,
3032 * to avoid race conditions.)
3035 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3039 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3040 if (vp->v_pollinfo.vpi_revents & events) {
3042 * This leaves events we are not interested
3043 * in available for the other process which
3044 * which presumably had requested them
3045 * (otherwise they would never have been
3048 events &= vp->v_pollinfo.vpi_revents;
3049 vp->v_pollinfo.vpi_revents &= ~events;
3051 lwkt_reltoken(&ilock);
3054 vp->v_pollinfo.vpi_events |= events;
3055 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
3056 lwkt_reltoken(&ilock);
3061 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3062 * it is possible for us to miss an event due to race conditions, but
3063 * that condition is expected to be rare, so for the moment it is the
3064 * preferred interface.
3067 vn_pollevent(vp, events)
3073 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3074 if (vp->v_pollinfo.vpi_events & events) {
3076 * We clear vpi_events so that we don't
3077 * call selwakeup() twice if two events are
3078 * posted before the polling process(es) is
3079 * awakened. This also ensures that we take at
3080 * most one selwakeup() if the polling process
3081 * is no longer interested. However, it does
3082 * mean that only one event can be noticed at
3083 * a time. (Perhaps we should only clear those
3084 * event bits which we note?) XXX
3086 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
3087 vp->v_pollinfo.vpi_revents |= events;
3088 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3090 lwkt_reltoken(&ilock);
3094 * Wake up anyone polling on vp because it is being revoked.
3095 * This depends on dead_poll() returning POLLHUP for correct
3104 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3105 if (vp->v_pollinfo.vpi_events) {
3106 vp->v_pollinfo.vpi_events = 0;
3107 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3109 lwkt_reltoken(&ilock);
3115 * Routine to create and manage a filesystem syncer vnode.
3117 #define sync_close ((int (*) (struct vop_close_args *))nullop)
3118 static int sync_fsync (struct vop_fsync_args *);
3119 static int sync_inactive (struct vop_inactive_args *);
3120 static int sync_reclaim (struct vop_reclaim_args *);
3121 #define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
3122 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
3123 static int sync_print (struct vop_print_args *);
3124 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
3126 static vop_t **sync_vnodeop_p;
3127 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3128 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3129 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3130 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3131 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3132 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3133 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
3134 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
3135 { &vop_print_desc, (vop_t *) sync_print }, /* print */
3136 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
3139 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3140 { &sync_vnodeop_p, sync_vnodeop_entries };
3142 VNODEOP_SET(sync_vnodeop_opv_desc);
3145 * Create a new filesystem syncer vnode for the specified mount point.
3146 * This vnode is placed on the worklist and is responsible for sync'ing
3149 * NOTE: read-only mounts are also placed on the worklist. The filesystem
3150 * sync code is also responsible for cleaning up vnodes.
3153 vfs_allocate_syncvnode(struct mount *mp)
3156 static long start, incr, next;
3159 /* Allocate a new vnode */
3160 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3161 mp->mnt_syncer = NULL;
3166 * Place the vnode onto the syncer worklist. We attempt to
3167 * scatter them about on the list so that they will go off
3168 * at evenly distributed times even if all the filesystems
3169 * are mounted at once.
3172 if (next == 0 || next > syncer_maxdelay) {
3176 start = syncer_maxdelay / 2;
3177 incr = syncer_maxdelay;
3181 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3182 mp->mnt_syncer = vp;
3187 * Do a lazy sync of the filesystem.
3191 struct vop_fsync_args /* {
3193 struct ucred *a_cred;
3195 struct thread *a_td;
3198 struct vnode *syncvp = ap->a_vp;
3199 struct mount *mp = syncvp->v_mount;
3200 struct thread *td = ap->a_td;
3205 * We only need to do something if this is a lazy evaluation.
3207 if (ap->a_waitfor != MNT_LAZY)
3211 * Move ourselves to the back of the sync list.
3213 vn_syncer_add_to_worklist(syncvp, syncdelay);
3216 * Walk the list of vnodes pushing all that are dirty and
3217 * not already on the sync list, and freeing vnodes which have
3218 * no refs and whos VM objects are empty. vfs_msync() handles
3219 * the VM issues and must be called whether the mount is readonly
3222 lwkt_gettoken(&ilock, &mountlist_token);
3223 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3224 lwkt_reltoken(&ilock);
3227 if (mp->mnt_flag & MNT_RDONLY) {
3228 vfs_msync(mp, MNT_NOWAIT);
3230 asyncflag = mp->mnt_flag & MNT_ASYNC;
3231 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */
3232 vfs_msync(mp, MNT_NOWAIT);
3233 VFS_SYNC(mp, MNT_LAZY, td);
3235 mp->mnt_flag |= MNT_ASYNC;
3242 * The syncer vnode is no referenced.
3246 struct vop_inactive_args /* {
3257 * The syncer vnode is no longer needed and is being decommissioned.
3259 * Modifications to the worklist must be protected at splbio().
3263 struct vop_reclaim_args /* {
3267 struct vnode *vp = ap->a_vp;
3271 vp->v_mount->mnt_syncer = NULL;
3272 if (vp->v_flag & VONWORKLST) {
3273 LIST_REMOVE(vp, v_synclist);
3274 vp->v_flag &= ~VONWORKLST;
3282 * Print out a syncer vnode.
3286 struct vop_print_args /* {
3290 struct vnode *vp = ap->a_vp;
3292 printf("syncer vnode");
3293 if (vp->v_vnlock != NULL)
3294 lockmgr_printinfo(vp->v_vnlock);
3300 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
3301 * (or v_rdev might be NULL).
3304 vn_todev(struct vnode *vp)
3306 if (vp->v_type != VBLK && vp->v_type != VCHR)
3308 KKASSERT(vp->v_rdev != NULL);
3309 return (vp->v_rdev);
3313 * Check if vnode represents a disk device. The vnode does not need to be
3317 vn_isdisk(struct vnode *vp, int *errp)
3321 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3327 if ((dev = vp->v_rdev) == NULL)
3328 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
3329 if (dev == NULL || dev == NODEV) {
3334 if (dev_is_good(dev) == 0) {
3339 if ((dev_dflags(dev) & D_DISK) == 0) {
3351 struct nameidata *ndp;
3354 if (!(flags & NDF_NO_FREE_PNBUF) &&
3355 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3356 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3357 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3359 if (!(flags & NDF_NO_DNCP_RELE) &&
3360 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3362 cache_drop(ndp->ni_dncp);
3363 ndp->ni_dncp = NULL;
3365 if (!(flags & NDF_NO_NCP_RELE) &&
3366 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3368 cache_drop(ndp->ni_ncp);
3371 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3372 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3373 ndp->ni_dvp != ndp->ni_vp) {
3374 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
3376 if (!(flags & NDF_NO_DVP_RELE) &&
3377 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3381 if (!(flags & NDF_NO_VP_UNLOCK) &&
3382 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3383 VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
3385 if (!(flags & NDF_NO_VP_RELE) &&
3390 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3391 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3392 vrele(ndp->ni_startdir);
3393 ndp->ni_startdir = NULL;
3397 #ifdef DEBUG_VFS_LOCKS
3400 assert_vop_locked(struct vnode *vp, const char *str)
3403 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
3404 panic("%s: %p is not locked shared but should be", str, vp);
3409 assert_vop_unlocked(struct vnode *vp, const char *str)
3412 if (vp && IS_LOCKING_VFS(vp)) {
3413 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
3414 panic("%s: %p is locked but should not be", str, vp);