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 1.35 2004/07/10 16:29:45 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,
91 int flags, struct thread *td);
93 static unsigned long numvnodes;
94 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
96 enum vtype iftovt_tab[16] = {
97 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
100 int vttoif_tab[9] = {
101 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 S_IFSOCK, S_IFIFO, S_IFMT,
105 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */
107 static u_long wantfreevnodes = 25;
108 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW,
109 &wantfreevnodes, 0, "");
110 static u_long freevnodes = 0;
111 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD,
114 static int reassignbufcalls;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
116 &reassignbufcalls, 0, "");
117 static int reassignbufloops;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
119 &reassignbufloops, 0, "");
120 static int reassignbufsortgood;
121 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
122 &reassignbufsortgood, 0, "");
123 static int reassignbufsortbad;
124 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
125 &reassignbufsortbad, 0, "");
126 static int reassignbufmethod = 1;
127 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
128 &reassignbufmethod, 0, "");
130 #ifdef ENABLE_VFS_IOOPT
132 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
135 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
136 struct lwkt_token mountlist_token;
137 struct lwkt_token mntvnode_token;
138 int nfs_mount_type = -1;
139 static struct lwkt_token mntid_token;
140 static struct lwkt_token vnode_free_list_token;
141 static struct lwkt_token spechash_token;
142 struct nfs_public nfs_pub; /* publicly exported FS */
143 static vm_zone_t vnode_zone;
146 * The workitem queue.
148 #define SYNCER_MAXDELAY 32
149 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
150 time_t syncdelay = 30; /* max time to delay syncing data */
151 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW,
152 &syncdelay, 0, "VFS data synchronization delay");
153 time_t filedelay = 30; /* time to delay syncing files */
154 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW,
155 &filedelay, 0, "File synchronization delay");
156 time_t dirdelay = 29; /* time to delay syncing directories */
157 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW,
158 &dirdelay, 0, "Directory synchronization delay");
159 time_t metadelay = 28; /* time to delay syncing metadata */
160 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW,
161 &metadelay, 0, "VFS metadata synchronization delay");
162 static int rushjob; /* number of slots to run ASAP */
163 static int stat_rush_requests; /* number of times I/O speeded up */
164 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW,
165 &stat_rush_requests, 0, "");
167 static int syncer_delayno = 0;
168 static long syncer_mask;
169 LIST_HEAD(synclist, vnode);
170 static struct synclist *syncer_workitem_pending;
173 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
174 &desiredvnodes, 0, "Maximum number of vnodes");
175 static int minvnodes;
176 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
177 &minvnodes, 0, "Minimum number of vnodes");
178 static int vnlru_nowhere = 0;
179 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
181 "Number of times the vnlru process ran without success");
183 static void vfs_free_addrlist (struct netexport *nep);
184 static int vfs_free_netcred (struct radix_node *rn, void *w);
185 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
186 struct export_args *argp);
188 #define VSHOULDFREE(vp) \
189 (!((vp)->v_flag & (VFREE|VDOOMED)) && \
190 !(vp)->v_holdcnt && !(vp)->v_usecount && \
191 (!(vp)->v_object || \
192 !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
194 #define VMIGHTFREE(vp) \
195 (((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) == 0 && \
196 cache_leaf_test(vp) == 0 && (vp)->v_usecount == 0)
198 #define VSHOULDBUSY(vp) \
199 (((vp)->v_flag & VFREE) && \
200 ((vp)->v_holdcnt || (vp)->v_usecount))
202 static void vbusy(struct vnode *vp);
203 static void vfree(struct vnode *vp);
204 static void vmaybefree(struct vnode *vp);
206 extern int dev_ref_debug;
209 * NOTE: the vnode interlock must be held on call.
212 vmaybefree(struct vnode *vp)
219 * Initialize the vnode management data structures.
226 * Desired vnodes is a result of the physical page count
227 * and the size of kernel's heap. It scales in proportion
228 * to the amount of available physical memory. This can
229 * cause trouble on 64-bit and large memory platforms.
231 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
233 min(maxproc + vmstats.v_page_count /4,
234 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
235 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
237 minvnodes = desiredvnodes / 4;
238 lwkt_token_init(&mountlist_token);
239 lwkt_token_init(&mntvnode_token);
240 lwkt_token_init(&mntid_token);
241 lwkt_token_init(&spechash_token);
242 TAILQ_INIT(&vnode_free_list);
243 lwkt_token_init(&vnode_free_list_token);
244 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
246 * Initialize the filesystem syncer.
248 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
250 syncer_maxdelay = syncer_mask + 1;
254 * Mark a mount point as busy. Used to synchronize access and to delay
255 * unmounting. Interlock is not released on failure.
258 vfs_busy(struct mount *mp, int flags,
259 lwkt_tokref_t interlkp, struct thread *td)
263 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
264 if (flags & LK_NOWAIT)
266 mp->mnt_kern_flag |= MNTK_MWAIT;
268 * Since all busy locks are shared except the exclusive
269 * lock granted when unmounting, the only place that a
270 * wakeup needs to be done is at the release of the
271 * exclusive lock at the end of dounmount.
273 * note: interlkp is a serializer and thus can be safely
274 * held through any sleep
276 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
279 lkflags = LK_SHARED | LK_NOPAUSE;
281 lkflags |= LK_INTERLOCK;
282 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
283 panic("vfs_busy: unexpected lock failure");
288 * Free a busy filesystem.
291 vfs_unbusy(struct mount *mp, struct thread *td)
293 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
297 * Lookup a filesystem type, and if found allocate and initialize
298 * a mount structure for it.
300 * Devname is usually updated by mount(8) after booting.
303 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
305 struct thread *td = curthread; /* XXX */
306 struct vfsconf *vfsp;
309 if (fstypename == NULL)
311 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
312 if (!strcmp(vfsp->vfc_name, fstypename))
317 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
318 bzero((char *)mp, (u_long)sizeof(struct mount));
319 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
320 vfs_busy(mp, LK_NOWAIT, NULL, td);
321 TAILQ_INIT(&mp->mnt_nvnodelist);
322 TAILQ_INIT(&mp->mnt_reservedvnlist);
323 mp->mnt_nvnodelistsize = 0;
325 mp->mnt_op = vfsp->vfc_vfsops;
326 mp->mnt_flag = MNT_RDONLY;
327 mp->mnt_vnodecovered = NULLVP;
328 vfsp->vfc_refcount++;
329 mp->mnt_iosize_max = DFLTPHYS;
330 mp->mnt_stat.f_type = vfsp->vfc_typenum;
331 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
332 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
333 mp->mnt_stat.f_mntonname[0] = '/';
334 mp->mnt_stat.f_mntonname[1] = 0;
335 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
341 * Lookup a mount point by filesystem identifier.
344 vfs_getvfs(fsid_t *fsid)
349 lwkt_gettoken(&ilock, &mountlist_token);
350 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
351 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
352 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
356 lwkt_reltoken(&ilock);
361 * Get a new unique fsid. Try to make its val[0] unique, since this value
362 * will be used to create fake device numbers for stat(). Also try (but
363 * not so hard) make its val[0] unique mod 2^16, since some emulators only
364 * support 16-bit device numbers. We end up with unique val[0]'s for the
365 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
367 * Keep in mind that several mounts may be running in parallel. Starting
368 * the search one past where the previous search terminated is both a
369 * micro-optimization and a defense against returning the same fsid to
373 vfs_getnewfsid(struct mount *mp)
375 static u_int16_t mntid_base;
380 lwkt_gettoken(&ilock, &mntid_token);
381 mtype = mp->mnt_vfc->vfc_typenum;
382 tfsid.val[1] = mtype;
383 mtype = (mtype & 0xFF) << 24;
385 tfsid.val[0] = makeudev(255,
386 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
388 if (vfs_getvfs(&tfsid) == NULL)
391 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
392 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
393 lwkt_reltoken(&ilock);
397 * Knob to control the precision of file timestamps:
399 * 0 = seconds only; nanoseconds zeroed.
400 * 1 = seconds and nanoseconds, accurate within 1/HZ.
401 * 2 = seconds and nanoseconds, truncated to microseconds.
402 * >=3 = seconds and nanoseconds, maximum precision.
404 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
406 static int timestamp_precision = TSP_SEC;
407 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
408 ×tamp_precision, 0, "");
411 * Get a current timestamp.
414 vfs_timestamp(struct timespec *tsp)
418 switch (timestamp_precision) {
420 tsp->tv_sec = time_second;
428 TIMEVAL_TO_TIMESPEC(&tv, tsp);
438 * Set vnode attributes to VNOVAL
441 vattr_null(struct vattr *vap)
444 vap->va_size = VNOVAL;
445 vap->va_bytes = VNOVAL;
446 vap->va_mode = VNOVAL;
447 vap->va_nlink = VNOVAL;
448 vap->va_uid = VNOVAL;
449 vap->va_gid = VNOVAL;
450 vap->va_fsid = VNOVAL;
451 vap->va_fileid = VNOVAL;
452 vap->va_blocksize = VNOVAL;
453 vap->va_rdev = VNOVAL;
454 vap->va_atime.tv_sec = VNOVAL;
455 vap->va_atime.tv_nsec = VNOVAL;
456 vap->va_mtime.tv_sec = VNOVAL;
457 vap->va_mtime.tv_nsec = VNOVAL;
458 vap->va_ctime.tv_sec = VNOVAL;
459 vap->va_ctime.tv_nsec = VNOVAL;
460 vap->va_flags = VNOVAL;
461 vap->va_gen = VNOVAL;
466 * This routine is called when we have too many vnodes. It attempts
467 * to free <count> vnodes and will potentially free vnodes that still
468 * have VM backing store (VM backing store is typically the cause
469 * of a vnode blowout so we want to do this). Therefore, this operation
470 * is not considered cheap.
472 * A number of conditions may prevent a vnode from being reclaimed.
473 * the buffer cache may have references on the vnode, a directory
474 * vnode may still have references due to the namei cache representing
475 * underlying files, or the vnode may be in active use. It is not
476 * desireable to reuse such vnodes. These conditions may cause the
477 * number of vnodes to reach some minimum value regardless of what
478 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
481 vlrureclaim(struct mount *mp)
492 * Calculate the trigger point, don't allow user
493 * screwups to blow us up. This prevents us from
494 * recycling vnodes with lots of resident pages. We
495 * aren't trying to free memory, we are trying to
498 usevnodes = desiredvnodes;
501 trigger = vmstats.v_page_count * 2 / usevnodes;
504 lwkt_gettoken(&ilock, &mntvnode_token);
505 count = mp->mnt_nvnodelistsize / 10 + 1;
506 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
510 * The VP will stick around while we hold mntvnode_token,
511 * at least until we block, so we can safely do an initial
512 * check. But we have to check again after obtaining
513 * the vnode interlock. vp->v_interlock points to stable
514 * storage so it's ok if the vp gets ripped out from
515 * under us while we are blocked.
517 if (vp->v_type == VNON ||
518 vp->v_type == VBAD ||
519 !VMIGHTFREE(vp) || /* critical path opt */
521 vp->v_object->resident_page_count >= trigger)
523 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
524 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
530 * Get the interlock, delay moving the node to the tail so
531 * we don't race against new additions to the mountlist.
533 lwkt_gettoken(&vlock, vp->v_interlock);
534 if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
535 lwkt_reltoken(&vlock);
538 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
539 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
544 if (vp->v_type == VNON ||
545 vp->v_type == VBAD ||
546 !VMIGHTFREE(vp) || /* critical path opt */
548 vp->v_object->resident_page_count >= trigger)
550 lwkt_reltoken(&vlock);
554 vgonel(vp, &vlock, curthread);
558 lwkt_reltoken(&ilock);
563 * Attempt to recycle vnodes in a context that is always safe to block.
564 * Calling vlrurecycle() from the bowels of file system code has some
565 * interesting deadlock problems.
567 static struct thread *vnlruthread;
568 static int vnlruproc_sig;
573 struct mount *mp, *nmp;
577 struct thread *td = curthread;
579 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
584 kproc_suspend_loop();
585 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
587 wakeup(&vnlruproc_sig);
588 tsleep(td, 0, "vlruwt", hz);
592 lwkt_gettoken(&ilock, &mountlist_token);
593 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
594 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
595 nmp = TAILQ_NEXT(mp, mnt_list);
598 done += vlrureclaim(mp);
599 lwkt_gettokref(&ilock);
600 nmp = TAILQ_NEXT(mp, mnt_list);
603 lwkt_reltoken(&ilock);
606 tsleep(td, 0, "vlrup", hz * 3);
612 static struct kproc_desc vnlru_kp = {
617 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
620 * Routines having to do with the management of the vnode table.
622 extern vop_t **dead_vnodeop_p;
625 * Return the next vnode from the free list.
628 getnewvnode(enum vtagtype tag, struct mount *mp,
629 vop_t **vops, struct vnode **vpp)
632 struct thread *td = curthread; /* XXX */
633 struct vnode *vp = NULL;
639 s = splbio(); /* YYY remove me */
642 * Try to reuse vnodes if we hit the max. This situation only
643 * occurs in certain large-memory (2G+) situations. We cannot
644 * attempt to directly reclaim vnodes due to nasty recursion
647 while (numvnodes - freevnodes > desiredvnodes) {
648 if (vnlruproc_sig == 0) {
649 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
652 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
657 * Attempt to reuse a vnode already on the free list, allocating
658 * a new vnode if we can't find one or if we have not reached a
659 * good minimum for good LRU performance.
661 lwkt_gettoken(&ilock, &vnode_free_list_token);
662 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
665 for (count = 0; count < freevnodes; count++) {
669 * Pull the next vnode off the free list and do some
670 * sanity checks. Note that regardless of how we
671 * block, if freevnodes is non-zero there had better
672 * be something on the list.
674 vp = TAILQ_FIRST(&vnode_free_list);
676 panic("getnewvnode: free vnode isn't");
679 * Move the vnode to the end of the list so other
680 * processes do not double-block trying to recycle
681 * the same vnode (as an optimization), then get
684 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
685 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
688 * Skip vnodes that are in the process of being
689 * held or referenced. Since the act of adding or
690 * removing a vnode on the freelist requires a token
691 * and may block, the ref count may be adjusted
692 * prior to its addition or removal.
694 if (VSHOULDBUSY(vp)) {
701 * Obtain the vnode interlock and check that the
702 * vnode is still on the free list.
704 * This normally devolves into a degenerate case so
705 * it is optimal. Loop up if it isn't. Note that
706 * the vnode could be in the middle of being moved
707 * off the free list (the VSHOULDBUSY() check) and
708 * must be skipped if so.
710 lwkt_gettoken(&vlock, vp->v_interlock);
711 TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
712 freelst, v_freelist) {
716 if (vp != xvp || VSHOULDBUSY(vp)) {
722 * We now safely own the vnode. If the vnode has
723 * an object do not recycle it if its VM object
724 * has resident pages or references.
726 if ((VOP_GETVOBJECT(vp, &object) == 0 &&
727 (object->resident_page_count || object->ref_count))
729 lwkt_reltoken(&vlock);
735 * We can almost reuse this vnode. But we don't want
736 * to recycle it if the vnode has children in the
737 * namecache because that breaks the namecache's
738 * path element chain. (YYY use nc_refs for the
741 KKASSERT(vp->v_flag & VFREE);
742 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
744 if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
745 cache_leaf_test(vp) >= 0) {
746 /* ok, we can reuse this vnode */
749 lwkt_reltoken(&vlock);
750 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
756 * If vp is non-NULL we hold it's interlock.
759 vp->v_flag |= VDOOMED;
760 vp->v_flag &= ~VFREE;
762 lwkt_reltoken(&ilock);
763 cache_purge(vp); /* YYY may block */
765 if (vp->v_type != VBAD) {
766 vgonel(vp, &vlock, td);
768 lwkt_reltoken(&vlock);
776 panic("cleaned vnode isn't");
779 panic("Clean vnode has pending I/O's");
789 vp->v_writecount = 0; /* XXX */
791 lwkt_reltoken(&ilock);
792 vp = zalloc(vnode_zone);
793 bzero(vp, sizeof(*vp));
794 vp->v_interlock = lwkt_token_pool_get(vp);
795 lwkt_token_init(&vp->v_pollinfo.vpi_token);
797 TAILQ_INIT(&vp->v_namecache);
801 TAILQ_INIT(&vp->v_cleanblkhd);
802 TAILQ_INIT(&vp->v_dirtyblkhd);
812 * Placing the vnode on the mount point's queue makes it visible.
813 * We had better already have a ref on it.
817 vfs_object_create(vp, td);
822 * Move a vnode from one mount queue to another.
825 insmntque(struct vnode *vp, struct mount *mp)
829 lwkt_gettoken(&ilock, &mntvnode_token);
831 * Delete from old mount point vnode list, if on one.
833 if (vp->v_mount != NULL) {
834 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
835 ("bad mount point vnode list size"));
836 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
837 vp->v_mount->mnt_nvnodelistsize--;
840 * Insert into list of vnodes for the new mount point, if available.
842 if ((vp->v_mount = mp) == NULL) {
843 lwkt_reltoken(&ilock);
846 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
847 mp->mnt_nvnodelistsize++;
848 lwkt_reltoken(&ilock);
852 * Update outstanding I/O count and do wakeup if requested.
855 vwakeup(struct buf *bp)
859 bp->b_flags &= ~B_WRITEINPROG;
860 if ((vp = bp->b_vp)) {
862 if (vp->v_numoutput < 0)
863 panic("vwakeup: neg numoutput");
864 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
865 vp->v_flag &= ~VBWAIT;
866 wakeup((caddr_t) &vp->v_numoutput);
872 * Flush out and invalidate all buffers associated with a vnode.
873 * Called with the underlying object locked.
876 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
877 int slpflag, int slptimeo)
880 struct buf *nbp, *blist;
885 if (flags & V_SAVE) {
887 while (vp->v_numoutput) {
888 vp->v_flag |= VBWAIT;
889 error = tsleep((caddr_t)&vp->v_numoutput,
890 slpflag, "vinvlbuf", slptimeo);
896 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
898 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
901 if (vp->v_numoutput > 0 ||
902 !TAILQ_EMPTY(&vp->v_dirtyblkhd))
903 panic("vinvalbuf: dirty bufs");
909 blist = TAILQ_FIRST(&vp->v_cleanblkhd);
911 blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
915 for (bp = blist; bp; bp = nbp) {
916 nbp = TAILQ_NEXT(bp, b_vnbufs);
917 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
918 error = BUF_TIMELOCK(bp,
919 LK_EXCLUSIVE | LK_SLEEPFAIL,
920 "vinvalbuf", slpflag, slptimeo);
927 * XXX Since there are no node locks for NFS, I
928 * believe there is a slight chance that a delayed
929 * write will occur while sleeping just above, so
930 * check for it. Note that vfs_bio_awrite expects
931 * buffers to reside on a queue, while VOP_BWRITE and
934 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
937 if (bp->b_vp == vp) {
938 if (bp->b_flags & B_CLUSTEROK) {
943 bp->b_flags |= B_ASYNC;
944 VOP_BWRITE(bp->b_vp, bp);
948 (void) VOP_BWRITE(bp->b_vp, bp);
953 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
954 bp->b_flags &= ~B_ASYNC;
960 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
961 * have write I/O in-progress but if there is a VM object then the
962 * VM object can also have read-I/O in-progress.
965 while (vp->v_numoutput > 0) {
966 vp->v_flag |= VBWAIT;
967 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
969 if (VOP_GETVOBJECT(vp, &object) == 0) {
970 while (object->paging_in_progress)
971 vm_object_pip_sleep(object, "vnvlbx");
973 } while (vp->v_numoutput > 0);
978 * Destroy the copy in the VM cache, too.
980 lwkt_gettoken(&vlock, vp->v_interlock);
981 if (VOP_GETVOBJECT(vp, &object) == 0) {
982 vm_object_page_remove(object, 0, 0,
983 (flags & V_SAVE) ? TRUE : FALSE);
985 lwkt_reltoken(&vlock);
987 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
988 panic("vinvalbuf: flush failed");
993 * Truncate a file's buffer and pages to a specified length. This
994 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
998 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
1006 * Round up to the *next* lbn.
1008 trunclbn = (length + blksize - 1) / blksize;
1015 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1016 nbp = TAILQ_NEXT(bp, b_vnbufs);
1017 if (bp->b_lblkno >= trunclbn) {
1018 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1019 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1023 bp->b_flags |= (B_INVAL | B_RELBUF);
1024 bp->b_flags &= ~B_ASYNC;
1029 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1030 (nbp->b_vp != vp) ||
1031 (nbp->b_flags & B_DELWRI))) {
1037 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1038 nbp = TAILQ_NEXT(bp, b_vnbufs);
1039 if (bp->b_lblkno >= trunclbn) {
1040 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1041 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1045 bp->b_flags |= (B_INVAL | B_RELBUF);
1046 bp->b_flags &= ~B_ASYNC;
1051 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1052 (nbp->b_vp != vp) ||
1053 (nbp->b_flags & B_DELWRI) == 0)) {
1062 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1063 nbp = TAILQ_NEXT(bp, b_vnbufs);
1064 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1065 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1066 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1070 if (bp->b_vp == vp) {
1071 bp->b_flags |= B_ASYNC;
1073 bp->b_flags &= ~B_ASYNC;
1075 VOP_BWRITE(bp->b_vp, bp);
1083 while (vp->v_numoutput > 0) {
1084 vp->v_flag |= VBWAIT;
1085 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1090 vnode_pager_setsize(vp, length);
1096 * Associate a buffer with a vnode.
1099 bgetvp(struct vnode *vp, struct buf *bp)
1103 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1107 bp->b_dev = vn_todev(vp);
1109 * Insert onto list for new vnode.
1112 bp->b_xflags |= BX_VNCLEAN;
1113 bp->b_xflags &= ~BX_VNDIRTY;
1114 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1119 * Disassociate a buffer from a vnode.
1122 brelvp(struct buf *bp)
1125 struct buflists *listheadp;
1128 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1131 * Delete from old vnode list, if on one.
1135 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1136 if (bp->b_xflags & BX_VNDIRTY)
1137 listheadp = &vp->v_dirtyblkhd;
1139 listheadp = &vp->v_cleanblkhd;
1140 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1141 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1143 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1144 vp->v_flag &= ~VONWORKLST;
1145 LIST_REMOVE(vp, v_synclist);
1148 bp->b_vp = (struct vnode *) 0;
1153 * The workitem queue.
1155 * It is useful to delay writes of file data and filesystem metadata
1156 * for tens of seconds so that quickly created and deleted files need
1157 * not waste disk bandwidth being created and removed. To realize this,
1158 * we append vnodes to a "workitem" queue. When running with a soft
1159 * updates implementation, most pending metadata dependencies should
1160 * not wait for more than a few seconds. Thus, mounted on block devices
1161 * are delayed only about a half the time that file data is delayed.
1162 * Similarly, directory updates are more critical, so are only delayed
1163 * about a third the time that file data is delayed. Thus, there are
1164 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1165 * one each second (driven off the filesystem syncer process). The
1166 * syncer_delayno variable indicates the next queue that is to be processed.
1167 * Items that need to be processed soon are placed in this queue:
1169 * syncer_workitem_pending[syncer_delayno]
1171 * A delay of fifteen seconds is done by placing the request fifteen
1172 * entries later in the queue:
1174 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1179 * Add an item to the syncer work queue.
1182 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1188 if (vp->v_flag & VONWORKLST) {
1189 LIST_REMOVE(vp, v_synclist);
1192 if (delay > syncer_maxdelay - 2)
1193 delay = syncer_maxdelay - 2;
1194 slot = (syncer_delayno + delay) & syncer_mask;
1196 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1197 vp->v_flag |= VONWORKLST;
1201 struct thread *updatethread;
1202 static void sched_sync (void);
1203 static struct kproc_desc up_kp = {
1208 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1211 * System filesystem synchronizer daemon.
1216 struct synclist *slp;
1220 struct thread *td = curthread;
1222 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1226 kproc_suspend_loop();
1228 starttime = time_second;
1231 * Push files whose dirty time has expired. Be careful
1232 * of interrupt race on slp queue.
1235 slp = &syncer_workitem_pending[syncer_delayno];
1236 syncer_delayno += 1;
1237 if (syncer_delayno == syncer_maxdelay)
1241 while ((vp = LIST_FIRST(slp)) != NULL) {
1242 if (VOP_ISLOCKED(vp, NULL) == 0) {
1243 vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1244 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1245 VOP_UNLOCK(vp, NULL, 0, td);
1248 if (LIST_FIRST(slp) == vp) {
1250 * Note: v_tag VT_VFS vps can remain on the
1251 * worklist too with no dirty blocks, but
1252 * since sync_fsync() moves it to a different
1255 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1256 !vn_isdisk(vp, NULL))
1257 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1259 * Put us back on the worklist. The worklist
1260 * routine will remove us from our current
1261 * position and then add us back in at a later
1264 vn_syncer_add_to_worklist(vp, syncdelay);
1270 * Do soft update processing.
1273 (*bioops.io_sync)(NULL);
1276 * The variable rushjob allows the kernel to speed up the
1277 * processing of the filesystem syncer process. A rushjob
1278 * value of N tells the filesystem syncer to process the next
1279 * N seconds worth of work on its queue ASAP. Currently rushjob
1280 * is used by the soft update code to speed up the filesystem
1281 * syncer process when the incore state is getting so far
1282 * ahead of the disk that the kernel memory pool is being
1283 * threatened with exhaustion.
1290 * If it has taken us less than a second to process the
1291 * current work, then wait. Otherwise start right over
1292 * again. We can still lose time if any single round
1293 * takes more than two seconds, but it does not really
1294 * matter as we are just trying to generally pace the
1295 * filesystem activity.
1297 if (time_second == starttime)
1298 tsleep(&lbolt, 0, "syncer", 0);
1303 * Request the syncer daemon to speed up its work.
1304 * We never push it to speed up more than half of its
1305 * normal turn time, otherwise it could take over the cpu.
1307 * YYY wchan field protected by the BGL.
1310 speedup_syncer(void)
1313 if (updatethread->td_wchan == &lbolt) { /* YYY */
1314 unsleep(updatethread);
1315 lwkt_schedule(updatethread);
1318 if (rushjob < syncdelay / 2) {
1320 stat_rush_requests += 1;
1327 * Associate a p-buffer with a vnode.
1329 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1330 * with the buffer. i.e. the bp has not been linked into the vnode or
1334 pbgetvp(struct vnode *vp, struct buf *bp)
1336 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1339 bp->b_flags |= B_PAGING;
1340 bp->b_dev = vn_todev(vp);
1344 * Disassociate a p-buffer from a vnode.
1347 pbrelvp(struct buf *bp)
1349 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1352 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1354 "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1359 bp->b_vp = (struct vnode *) 0;
1360 bp->b_flags &= ~B_PAGING;
1364 pbreassignbuf(struct buf *bp, struct vnode *newvp)
1366 if ((bp->b_flags & B_PAGING) == 0) {
1368 "pbreassignbuf() on non phys bp %p",
1376 * Reassign a buffer from one vnode to another.
1377 * Used to assign file specific control information
1378 * (indirect blocks) to the vnode to which they belong.
1381 reassignbuf(struct buf *bp, struct vnode *newvp)
1383 struct buflists *listheadp;
1387 if (newvp == NULL) {
1388 printf("reassignbuf: NULL");
1394 * B_PAGING flagged buffers cannot be reassigned because their vp
1395 * is not fully linked in.
1397 if (bp->b_flags & B_PAGING)
1398 panic("cannot reassign paging buffer");
1402 * Delete from old vnode list, if on one.
1404 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1405 if (bp->b_xflags & BX_VNDIRTY)
1406 listheadp = &bp->b_vp->v_dirtyblkhd;
1408 listheadp = &bp->b_vp->v_cleanblkhd;
1409 TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1410 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1411 if (bp->b_vp != newvp) {
1413 bp->b_vp = NULL; /* for clarification */
1417 * If dirty, put on list of dirty buffers; otherwise insert onto list
1420 if (bp->b_flags & B_DELWRI) {
1423 listheadp = &newvp->v_dirtyblkhd;
1424 if ((newvp->v_flag & VONWORKLST) == 0) {
1425 switch (newvp->v_type) {
1431 if (newvp->v_rdev &&
1432 newvp->v_rdev->si_mountpoint != NULL) {
1440 vn_syncer_add_to_worklist(newvp, delay);
1442 bp->b_xflags |= BX_VNDIRTY;
1443 tbp = TAILQ_FIRST(listheadp);
1445 bp->b_lblkno == 0 ||
1446 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1447 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1448 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1449 ++reassignbufsortgood;
1450 } else if (bp->b_lblkno < 0) {
1451 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1452 ++reassignbufsortgood;
1453 } else if (reassignbufmethod == 1) {
1455 * New sorting algorithm, only handle sequential case,
1456 * otherwise append to end (but before metadata)
1458 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1459 (tbp->b_xflags & BX_VNDIRTY)) {
1461 * Found the best place to insert the buffer
1463 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1464 ++reassignbufsortgood;
1467 * Missed, append to end, but before meta-data.
1468 * We know that the head buffer in the list is
1469 * not meta-data due to prior conditionals.
1471 * Indirect effects: NFS second stage write
1472 * tends to wind up here, giving maximum
1473 * distance between the unstable write and the
1476 tbp = TAILQ_LAST(listheadp, buflists);
1477 while (tbp && tbp->b_lblkno < 0)
1478 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1479 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1480 ++reassignbufsortbad;
1484 * Old sorting algorithm, scan queue and insert
1487 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1488 (ttbp->b_lblkno < bp->b_lblkno)) {
1492 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1495 bp->b_xflags |= BX_VNCLEAN;
1496 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1497 if ((newvp->v_flag & VONWORKLST) &&
1498 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1499 newvp->v_flag &= ~VONWORKLST;
1500 LIST_REMOVE(newvp, v_synclist);
1503 if (bp->b_vp != newvp) {
1511 * Create a vnode for a block device.
1512 * Used for mounting the root file system.
1515 bdevvp(dev_t dev, struct vnode **vpp)
1525 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1532 vp->v_udev = dev->si_udev;
1538 v_associate_rdev(struct vnode *vp, dev_t dev)
1542 if (dev == NULL || dev == NODEV)
1544 if (dev_is_good(dev) == 0)
1546 KKASSERT(vp->v_rdev == NULL);
1549 vp->v_rdev = reference_dev(dev);
1550 lwkt_gettoken(&ilock, &spechash_token);
1551 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
1552 lwkt_reltoken(&ilock);
1557 v_release_rdev(struct vnode *vp)
1562 if ((dev = vp->v_rdev) != NULL) {
1563 lwkt_gettoken(&ilock, &spechash_token);
1564 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1565 if (dev_ref_debug && vp->v_opencount != 0) {
1566 printf("releasing rdev with non-0 "
1567 "v_opencount(%d) (revoked?)\n",
1571 vp->v_opencount = 0;
1573 lwkt_reltoken(&ilock);
1578 * Add a vnode to the alias list hung off the dev_t. We only associate
1579 * the device number with the vnode. The actual device is not associated
1580 * until the vnode is opened (usually in spec_open()), and will be
1581 * disassociated on last close.
1584 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1586 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1587 panic("addaliasu on non-special vnode");
1588 nvp->v_udev = nvp_udev;
1592 * Grab a particular vnode from the free list, increment its
1593 * reference count and lock it. The vnode lock bit is set if the
1594 * vnode is being eliminated in vgone. The process is awakened
1595 * when the transition is completed, and an error returned to
1596 * indicate that the vnode is no longer usable (possibly having
1597 * been changed to a new file system type).
1599 * This code is very sensitive. We are depending on the vnode interlock
1600 * to be maintained through to the vn_lock() call, which means that we
1601 * cannot block which means that we cannot call vbusy() until after vn_lock().
1602 * If the interlock is not maintained, the VXLOCK check will not properly
1603 * interlock against a vclean()'s LK_DRAIN operation on the lock.
1606 vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
1612 * We need the interlock to safely modify the v_ fields. ZZZ it is
1613 * only legal to pass (1) the vnode's interlock and (2) only pass
1614 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1617 if ((flags & LK_INTERLOCK) == 0) {
1618 lwkt_gettoken(&vvlock, vp->v_interlock);
1623 * If the vnode is in the process of being cleaned out for
1624 * another use, we wait for the cleaning to finish and then
1625 * return failure. Cleaning is determined by checking that
1626 * the VXLOCK flag is set. It is possible for the vnode to be
1627 * self-referenced during the cleaning operation.
1629 if (vp->v_flag & VXLOCK) {
1630 if (vp->v_vxthread == curthread) {
1632 /* this can now occur in normal operation */
1633 log(LOG_INFO, "VXLOCK interlock avoided\n");
1636 vp->v_flag |= VXWANT;
1637 lwkt_reltoken(vlock);
1638 tsleep((caddr_t)vp, 0, "vget", 0);
1644 * Bump v_usecount to prevent the vnode from being recycled. The
1645 * usecount needs to be bumped before we successfully get our lock.
1648 if (flags & LK_TYPE_MASK) {
1649 if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
1651 * must expand vrele here because we do not want
1652 * to call VOP_INACTIVE if the reference count
1653 * drops back to zero since it was never really
1654 * active. We must remove it from the free list
1655 * before sleeping so that multiple processes do
1656 * not try to recycle it.
1658 lwkt_gettokref(vlock);
1661 lwkt_reltoken(vlock);
1665 if (VSHOULDBUSY(vp))
1666 vbusy(vp); /* interlock must be held on call */
1667 lwkt_reltoken(vlock);
1672 vref(struct vnode *vp)
1674 crit_enter(); /* YYY use crit section for moment / BGL protected */
1680 * Vnode put/release.
1681 * If count drops to zero, call inactive routine and return to freelist.
1684 vrele(struct vnode *vp)
1686 struct thread *td = curthread; /* XXX */
1689 KASSERT(vp != NULL && vp->v_usecount >= 0,
1690 ("vrele: null vp or <=0 v_usecount"));
1692 lwkt_gettoken(&vlock, vp->v_interlock);
1694 if (vp->v_usecount > 1) {
1696 lwkt_reltoken(&vlock);
1700 if (vp->v_usecount == 1) {
1703 * We must call VOP_INACTIVE with the node locked and the
1704 * usecount 0. If we are doing a vpu, the node is already
1705 * locked, but, in the case of vrele, we must explicitly lock
1706 * the vnode before calling VOP_INACTIVE.
1709 if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
1710 VOP_INACTIVE(vp, td);
1712 lwkt_reltoken(&vlock);
1715 vprint("vrele: negative ref count", vp);
1717 lwkt_reltoken(&vlock);
1718 panic("vrele: negative ref cnt");
1723 vput(struct vnode *vp)
1725 struct thread *td = curthread; /* XXX */
1728 KASSERT(vp != NULL, ("vput: null vp"));
1730 lwkt_gettoken(&vlock, vp->v_interlock);
1732 if (vp->v_usecount > 1) {
1734 VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
1738 if (vp->v_usecount == 1) {
1741 * We must call VOP_INACTIVE with the node locked.
1742 * If we are doing a vpu, the node is already locked,
1743 * so we just need to release the vnode mutex.
1745 VOP_INACTIVE(vp, td);
1747 lwkt_reltoken(&vlock);
1750 vprint("vput: negative ref count", vp);
1752 lwkt_reltoken(&vlock);
1753 panic("vput: negative ref cnt");
1758 * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1759 * be held but isn't.
1762 vhold(struct vnode *vp)
1768 if (VSHOULDBUSY(vp))
1769 vbusy(vp); /* interlock must be held on call */
1774 * One less who cares about this vnode.
1777 vdrop(struct vnode *vp)
1781 lwkt_gettoken(&vlock, vp->v_interlock);
1782 if (vp->v_holdcnt <= 0)
1783 panic("vdrop: holdcnt");
1786 lwkt_reltoken(&vlock);
1792 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1793 int (*slowfunc)(struct mount *mp, struct vnode *vp,
1794 lwkt_tokref_t vlock, void *data),
1804 * Scan the vnodes on the mount's vnode list. Use a placemarker
1806 pvp = zalloc(vnode_zone);
1807 pvp->v_flag |= VPLACEMARKER;
1809 lwkt_gettoken(&ilock, &mntvnode_token);
1810 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1812 while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1814 * Move the placemarker and skip other placemarkers we
1815 * encounter. The nothing can get in our way so the
1816 * mount point on the vp must be valid.
1818 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1819 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1820 if (vp->v_flag & VPLACEMARKER)
1822 KKASSERT(vp->v_mount == mp);
1828 if ((r = fastfunc(mp, vp, data)) < 0)
1835 * Get the vnodes interlock and make sure it is still on the
1836 * mount list. Skip it if it has moved (we may encounter it
1837 * later). Then do the with-interlock test. The callback
1838 * is responsible for releasing the vnode interlock.
1840 * The interlock is type-stable.
1843 lwkt_gettoken(&vlock, vp->v_interlock);
1844 if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1845 printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1846 lwkt_reltoken(&vlock);
1849 if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1850 KKASSERT(lwkt_havetokref(&vlock) == 0);
1853 KKASSERT(lwkt_havetokref(&vlock) == 0);
1856 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1857 zfree(vnode_zone, pvp);
1858 lwkt_reltoken(&ilock);
1863 * Remove any vnodes in the vnode table belonging to mount point mp.
1865 * If FORCECLOSE is not specified, there should not be any active ones,
1866 * return error if any are found (nb: this is a user error, not a
1867 * system error). If FORCECLOSE is specified, detach any active vnodes
1870 * If WRITECLOSE is set, only flush out regular file vnodes open for
1873 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1875 * `rootrefs' specifies the base reference count for the root vnode
1876 * of this filesystem. The root vnode is considered busy if its
1877 * v_usecount exceeds this value. On a successful return, vflush()
1878 * will call vrele() on the root vnode exactly rootrefs times.
1879 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1883 static int busyprt = 0; /* print out busy vnodes */
1884 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1887 static int vflush_scan(struct mount *mp, struct vnode *vp,
1888 lwkt_tokref_t vlock, void *data);
1890 struct vflush_info {
1897 vflush(struct mount *mp, int rootrefs, int flags)
1899 struct thread *td = curthread; /* XXX */
1900 struct vnode *rootvp = NULL;
1903 struct vflush_info vflush_info;
1906 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1907 ("vflush: bad args"));
1909 * Get the filesystem root vnode. We can vput() it
1910 * immediately, since with rootrefs > 0, it won't go away.
1912 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1917 vflush_info.busy = 0;
1918 vflush_info.flags = flags;
1919 vflush_info.td = td;
1920 vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
1922 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1924 * If just the root vnode is busy, and if its refcount
1925 * is equal to `rootrefs', then go ahead and kill it.
1927 lwkt_gettoken(&vlock, rootvp->v_interlock);
1928 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1929 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1930 if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1931 vgonel(rootvp, &vlock, td);
1932 vflush_info.busy = 0;
1934 lwkt_reltoken(&vlock);
1937 if (vflush_info.busy)
1939 for (; rootrefs > 0; rootrefs--)
1945 * The scan callback is made with an interlocked vnode.
1948 vflush_scan(struct mount *mp, struct vnode *vp,
1949 lwkt_tokref_t vlock, void *data)
1951 struct vflush_info *info = data;
1955 * Skip over a vnodes marked VSYSTEM.
1957 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1958 lwkt_reltoken(vlock);
1963 * If WRITECLOSE is set, flush out unlinked but still open
1964 * files (even if open only for reading) and regular file
1965 * vnodes open for writing.
1967 if ((info->flags & WRITECLOSE) &&
1968 (vp->v_type == VNON ||
1969 (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
1970 vattr.va_nlink > 0)) &&
1971 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1972 lwkt_reltoken(vlock);
1977 * With v_usecount == 0, all we need to do is clear out the
1978 * vnode data structures and we are done.
1980 if (vp->v_usecount == 0) {
1981 vgonel(vp, vlock, info->td);
1986 * If FORCECLOSE is set, forcibly close the vnode. For block
1987 * or character devices, revert to an anonymous device. For
1988 * all other files, just kill them.
1990 if (info->flags & FORCECLOSE) {
1991 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1992 vgonel(vp, vlock, info->td);
1994 vclean(vp, vlock, 0, info->td);
1995 vp->v_op = spec_vnodeop_p;
1996 insmntque(vp, (struct mount *) 0);
2002 vprint("vflush: busy vnode", vp);
2004 lwkt_reltoken(vlock);
2010 * Disassociate the underlying file system from a vnode.
2013 vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
2018 * Check to see if the vnode is in use. If so we have to reference it
2019 * before we clean it out so that its count cannot fall to zero and
2020 * generate a race against ourselves to recycle it.
2022 if ((active = vp->v_usecount))
2026 * Prevent the vnode from being recycled or brought into use while we
2029 if (vp->v_flag & VXLOCK)
2030 panic("vclean: deadlock");
2031 vp->v_flag |= VXLOCK;
2032 vp->v_vxthread = curthread;
2035 * Even if the count is zero, the VOP_INACTIVE routine may still
2036 * have the object locked while it cleans it out. The VOP_LOCK
2037 * ensures that the VOP_INACTIVE routine is done with its work.
2038 * For active vnodes, it ensures that no other activity can
2039 * occur while the underlying object is being cleaned out.
2041 * NOTE: we continue to hold the vnode interlock through to the
2044 VOP_LOCK(vp, NULL, LK_DRAIN, td);
2047 * Clean out any buffers associated with the vnode.
2049 vinvalbuf(vp, V_SAVE, td, 0, 0);
2050 VOP_DESTROYVOBJECT(vp);
2053 * If purging an active vnode, it must be closed and
2054 * deactivated before being reclaimed. Note that the
2055 * VOP_INACTIVE will unlock the vnode.
2058 if (flags & DOCLOSE)
2059 VOP_CLOSE(vp, FNONBLOCK, td);
2060 VOP_INACTIVE(vp, td);
2063 * Any other processes trying to obtain this lock must first
2064 * wait for VXLOCK to clear, then call the new lock operation.
2066 VOP_UNLOCK(vp, NULL, 0, td);
2069 * Reclaim the vnode.
2071 if (VOP_RECLAIM(vp, td))
2072 panic("vclean: cannot reclaim");
2076 * Inline copy of vrele() since VOP_INACTIVE
2077 * has already been called.
2079 if (--vp->v_usecount <= 0) {
2081 if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2082 vprint("vclean: bad ref count", vp);
2083 panic("vclean: ref cnt");
2091 vp->v_vnlock = NULL;
2095 * Done with purge, notify sleepers of the grim news.
2097 vp->v_op = dead_vnodeop_p;
2100 vp->v_flag &= ~VXLOCK;
2101 vp->v_vxthread = NULL;
2102 if (vp->v_flag & VXWANT) {
2103 vp->v_flag &= ~VXWANT;
2104 wakeup((caddr_t) vp);
2106 lwkt_reltoken(vlock);
2110 * Eliminate all activity associated with the requested vnode
2111 * and with all vnodes aliased to the requested vnode.
2113 * revoke { struct vnode *a_vp, int a_flags }
2116 vop_revoke(struct vop_revoke_args *ap)
2118 struct vnode *vp, *vq;
2122 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2126 * If a vgone (or vclean) is already in progress,
2127 * wait until it is done and return.
2129 if (vp->v_flag & VXLOCK) {
2130 vp->v_flag |= VXWANT;
2131 /*lwkt_reltoken(vlock); ZZZ */
2132 tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2137 * If the vnode has a device association, scrap all vnodes associated
2138 * with the device. Don't let the device disappear on us while we
2139 * are scrapping the vnodes.
2141 if (vp->v_type != VCHR && vp->v_type != VBLK)
2143 if ((dev = vp->v_rdev) == NULL) {
2144 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
2149 lwkt_gettoken(&ilock, &spechash_token);
2150 vq = SLIST_FIRST(&dev->si_hlist);
2151 lwkt_reltoken(&ilock);
2161 * Recycle an unused vnode to the front of the free list.
2162 * Release the passed interlock if the vnode will be recycled.
2165 vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
2169 lwkt_gettoken(&vlock, vp->v_interlock);
2170 if (vp->v_usecount == 0) {
2172 lwkt_reltoken(inter_lkp);
2173 vgonel(vp, &vlock, td);
2176 lwkt_reltoken(&vlock);
2181 * Eliminate all activity associated with a vnode
2182 * in preparation for reuse.
2185 vgone(struct vnode *vp)
2187 struct thread *td = curthread; /* XXX */
2190 lwkt_gettoken(&vlock, vp->v_interlock);
2191 vgonel(vp, &vlock, td);
2195 * vgone, with the vp interlock held.
2198 vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
2204 * If a vgone (or vclean) is already in progress,
2205 * wait until it is done and return.
2207 if (vp->v_flag & VXLOCK) {
2208 vp->v_flag |= VXWANT;
2209 lwkt_reltoken(vlock);
2210 tsleep((caddr_t)vp, 0, "vgone", 0);
2215 * Clean out the filesystem specific data.
2217 vclean(vp, vlock, DOCLOSE, td);
2218 lwkt_gettokref(vlock);
2221 * Delete from old mount point vnode list, if on one.
2223 if (vp->v_mount != NULL)
2224 insmntque(vp, (struct mount *)0);
2227 * If special device, remove it from special device alias list
2228 * if it is on one. This should normally only occur if a vnode is
2229 * being revoked as the device should otherwise have been released
2232 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2237 * If it is on the freelist and not already at the head,
2238 * move it to the head of the list. The test of the
2239 * VDOOMED flag and the reference count of zero is because
2240 * it will be removed from the free list by getnewvnode,
2241 * but will not have its reference count incremented until
2242 * after calling vgone. If the reference count were
2243 * incremented first, vgone would (incorrectly) try to
2244 * close the previous instance of the underlying object.
2246 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2248 lwkt_gettoken(&ilock, &vnode_free_list_token);
2249 if (vp->v_flag & VFREE)
2250 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2253 vp->v_flag |= VFREE;
2254 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2255 lwkt_reltoken(&ilock);
2259 lwkt_reltoken(vlock);
2263 * Lookup a vnode by device number.
2266 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
2271 lwkt_gettoken(&ilock, &spechash_token);
2272 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2273 if (type == vp->v_type) {
2275 lwkt_reltoken(&ilock);
2279 lwkt_reltoken(&ilock);
2284 * Calculate the total number of references to a special device. This
2285 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
2286 * an overloaded field. Since udev2dev can now return NODEV, we have
2287 * to check for a NULL v_rdev.
2290 count_dev(dev_t dev)
2296 if (SLIST_FIRST(&dev->si_hlist)) {
2297 lwkt_gettoken(&ilock, &spechash_token);
2298 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2299 count += vp->v_usecount;
2301 lwkt_reltoken(&ilock);
2307 count_udev(udev_t udev)
2311 if ((dev = udev2dev(udev, 0)) == NODEV)
2313 return(count_dev(dev));
2317 vcount(struct vnode *vp)
2319 if (vp->v_rdev == NULL)
2321 return(count_dev(vp->v_rdev));
2325 * Print out a description of a vnode.
2327 static char *typename[] =
2328 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2331 vprint(char *label, struct vnode *vp)
2336 printf("%s: %p: ", label, (void *)vp);
2338 printf("%p: ", (void *)vp);
2339 printf("type %s, usecount %d, writecount %d, refcount %d,",
2340 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2343 if (vp->v_flag & VROOT)
2344 strcat(buf, "|VROOT");
2345 if (vp->v_flag & VTEXT)
2346 strcat(buf, "|VTEXT");
2347 if (vp->v_flag & VSYSTEM)
2348 strcat(buf, "|VSYSTEM");
2349 if (vp->v_flag & VXLOCK)
2350 strcat(buf, "|VXLOCK");
2351 if (vp->v_flag & VXWANT)
2352 strcat(buf, "|VXWANT");
2353 if (vp->v_flag & VBWAIT)
2354 strcat(buf, "|VBWAIT");
2355 if (vp->v_flag & VDOOMED)
2356 strcat(buf, "|VDOOMED");
2357 if (vp->v_flag & VFREE)
2358 strcat(buf, "|VFREE");
2359 if (vp->v_flag & VOBJBUF)
2360 strcat(buf, "|VOBJBUF");
2362 printf(" flags (%s)", &buf[1]);
2363 if (vp->v_data == NULL) {
2372 #include <ddb/ddb.h>
2374 * List all of the locked vnodes in the system.
2375 * Called when debugging the kernel.
2377 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2379 struct thread *td = curthread; /* XXX */
2381 struct mount *mp, *nmp;
2384 printf("Locked vnodes\n");
2385 lwkt_gettoken(&ilock, &mountlist_token);
2386 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2387 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
2388 nmp = TAILQ_NEXT(mp, mnt_list);
2391 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2392 if (VOP_ISLOCKED(vp, NULL))
2393 vprint((char *)0, vp);
2395 lwkt_gettokref(&ilock);
2396 nmp = TAILQ_NEXT(mp, mnt_list);
2399 lwkt_reltoken(&ilock);
2404 * Top level filesystem related information gathering.
2406 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2409 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2411 int *name = (int *)arg1 - 1; /* XXX */
2412 u_int namelen = arg2 + 1; /* XXX */
2413 struct vfsconf *vfsp;
2415 #if 1 || defined(COMPAT_PRELITE2)
2416 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2418 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2422 /* all sysctl names at this level are at least name and field */
2424 return (ENOTDIR); /* overloaded */
2425 if (name[0] != VFS_GENERIC) {
2426 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2427 if (vfsp->vfc_typenum == name[0])
2430 return (EOPNOTSUPP);
2431 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2432 oldp, oldlenp, newp, newlen, p));
2436 case VFS_MAXTYPENUM:
2439 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2442 return (ENOTDIR); /* overloaded */
2443 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2444 if (vfsp->vfc_typenum == name[2])
2447 return (EOPNOTSUPP);
2448 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2450 return (EOPNOTSUPP);
2453 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2454 "Generic filesystem");
2456 #if 1 || defined(COMPAT_PRELITE2)
2459 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2462 struct vfsconf *vfsp;
2463 struct ovfsconf ovfs;
2465 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2466 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2467 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2468 ovfs.vfc_index = vfsp->vfc_typenum;
2469 ovfs.vfc_refcount = vfsp->vfc_refcount;
2470 ovfs.vfc_flags = vfsp->vfc_flags;
2471 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2478 #endif /* 1 || COMPAT_PRELITE2 */
2481 #define KINFO_VNODESLOP 10
2483 * Dump vnode list (via sysctl).
2484 * Copyout address of vnode followed by vnode.
2488 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2490 struct proc *p = curproc; /* XXX */
2491 struct mount *mp, *nmp;
2492 struct vnode *nvp, *vp;
2497 #define VPTRSZ sizeof (struct vnode *)
2498 #define VNODESZ sizeof (struct vnode)
2501 if (!req->oldptr) /* Make an estimate */
2502 return (SYSCTL_OUT(req, 0,
2503 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2505 lwkt_gettoken(&ilock, &mountlist_token);
2506 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2507 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
2508 nmp = TAILQ_NEXT(mp, mnt_list);
2511 lwkt_gettoken(&jlock, &mntvnode_token);
2513 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2517 * Check that the vp is still associated with
2518 * this filesystem. RACE: could have been
2519 * recycled onto the same filesystem.
2521 if (vp->v_mount != mp)
2523 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2524 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2525 (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2526 lwkt_reltoken(&jlock);
2530 lwkt_reltoken(&jlock);
2531 lwkt_gettokref(&ilock);
2532 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */
2535 lwkt_reltoken(&ilock);
2543 * Exporting the vnode list on large systems causes them to crash.
2544 * Exporting the vnode list on medium systems causes sysctl to coredump.
2547 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2548 0, 0, sysctl_vnode, "S,vnode", "");
2552 * Check to see if a filesystem is mounted on a block device.
2555 vfs_mountedon(struct vnode *vp)
2559 if ((dev = vp->v_rdev) == NULL)
2560 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
2561 if (dev != NODEV && dev->si_mountpoint)
2567 * Unmount all filesystems. The list is traversed in reverse order
2568 * of mounting to avoid dependencies.
2571 vfs_unmountall(void)
2574 struct thread *td = curthread;
2577 if (td->td_proc == NULL)
2578 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
2581 * Since this only runs when rebooting, it is not interlocked.
2583 while(!TAILQ_EMPTY(&mountlist)) {
2584 mp = TAILQ_LAST(&mountlist, mntlist);
2585 error = dounmount(mp, MNT_FORCE, td);
2587 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2588 printf("unmount of %s failed (",
2589 mp->mnt_stat.f_mntonname);
2593 printf("%d)\n", error);
2595 /* The unmount has removed mp from the mountlist */
2601 * Build hash lists of net addresses and hang them off the mount point.
2602 * Called by ufs_mount() to set up the lists of export addresses.
2605 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
2606 struct export_args *argp)
2609 struct radix_node_head *rnh;
2611 struct radix_node *rn;
2612 struct sockaddr *saddr, *smask = 0;
2616 if (argp->ex_addrlen == 0) {
2617 if (mp->mnt_flag & MNT_DEFEXPORTED)
2619 np = &nep->ne_defexported;
2620 np->netc_exflags = argp->ex_flags;
2621 np->netc_anon = argp->ex_anon;
2622 np->netc_anon.cr_ref = 1;
2623 mp->mnt_flag |= MNT_DEFEXPORTED;
2627 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2629 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2632 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2633 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2634 bzero((caddr_t) np, i);
2635 saddr = (struct sockaddr *) (np + 1);
2636 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2638 if (saddr->sa_len > argp->ex_addrlen)
2639 saddr->sa_len = argp->ex_addrlen;
2640 if (argp->ex_masklen) {
2641 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2642 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2645 if (smask->sa_len > argp->ex_masklen)
2646 smask->sa_len = argp->ex_masklen;
2648 i = saddr->sa_family;
2649 if ((rnh = nep->ne_rtable[i]) == 0) {
2651 * Seems silly to initialize every AF when most are not used,
2652 * do so on demand here
2654 for (dom = domains; dom; dom = dom->dom_next)
2655 if (dom->dom_family == i && dom->dom_rtattach) {
2656 dom->dom_rtattach((void **) &nep->ne_rtable[i],
2660 if ((rnh = nep->ne_rtable[i]) == 0) {
2665 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2667 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
2671 np->netc_exflags = argp->ex_flags;
2672 np->netc_anon = argp->ex_anon;
2673 np->netc_anon.cr_ref = 1;
2676 free(np, M_NETADDR);
2682 vfs_free_netcred(struct radix_node *rn, void *w)
2684 struct radix_node_head *rnh = (struct radix_node_head *) w;
2686 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2687 free((caddr_t) rn, M_NETADDR);
2692 * Free the net address hash lists that are hanging off the mount points.
2695 vfs_free_addrlist(struct netexport *nep)
2698 struct radix_node_head *rnh;
2700 for (i = 0; i <= AF_MAX; i++)
2701 if ((rnh = nep->ne_rtable[i])) {
2702 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2704 free((caddr_t) rnh, M_RTABLE);
2705 nep->ne_rtable[i] = 0;
2710 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
2714 if (argp->ex_flags & MNT_DELEXPORT) {
2715 if (mp->mnt_flag & MNT_EXPUBLIC) {
2716 vfs_setpublicfs(NULL, NULL, NULL);
2717 mp->mnt_flag &= ~MNT_EXPUBLIC;
2719 vfs_free_addrlist(nep);
2720 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2722 if (argp->ex_flags & MNT_EXPORTED) {
2723 if (argp->ex_flags & MNT_EXPUBLIC) {
2724 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2726 mp->mnt_flag |= MNT_EXPUBLIC;
2728 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2730 mp->mnt_flag |= MNT_EXPORTED;
2737 * Set the publicly exported filesystem (WebNFS). Currently, only
2738 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2741 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2742 struct export_args *argp)
2749 * mp == NULL -> invalidate the current info, the FS is
2750 * no longer exported. May be called from either vfs_export
2751 * or unmount, so check if it hasn't already been done.
2754 if (nfs_pub.np_valid) {
2755 nfs_pub.np_valid = 0;
2756 if (nfs_pub.np_index != NULL) {
2757 FREE(nfs_pub.np_index, M_TEMP);
2758 nfs_pub.np_index = NULL;
2765 * Only one allowed at a time.
2767 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2771 * Get real filehandle for root of exported FS.
2773 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2774 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2776 if ((error = VFS_ROOT(mp, &rvp)))
2779 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2785 * If an indexfile was specified, pull it in.
2787 if (argp->ex_indexfile != NULL) {
2788 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2790 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2791 MAXNAMLEN, (size_t *)0);
2794 * Check for illegal filenames.
2796 for (cp = nfs_pub.np_index; *cp; cp++) {
2804 FREE(nfs_pub.np_index, M_TEMP);
2809 nfs_pub.np_mount = mp;
2810 nfs_pub.np_valid = 1;
2815 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2816 struct sockaddr *nam)
2819 struct radix_node_head *rnh;
2820 struct sockaddr *saddr;
2823 if (mp->mnt_flag & MNT_EXPORTED) {
2825 * Lookup in the export list first.
2829 rnh = nep->ne_rtable[saddr->sa_family];
2831 np = (struct netcred *)
2832 (*rnh->rnh_matchaddr)((caddr_t)saddr,
2834 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2839 * If no address match, use the default if it exists.
2841 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2842 np = &nep->ne_defexported;
2848 * perform msync on all vnodes under a mount point. The mount point must
2849 * be locked. This code is also responsible for lazy-freeing unreferenced
2850 * vnodes whos VM objects no longer contain pages.
2852 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2854 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2855 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2856 lwkt_tokref_t vlock, void *data);
2859 vfs_msync(struct mount *mp, int flags)
2861 vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2865 * scan1 is a fast pre-check. There could be hundreds of thousands of
2866 * vnodes, we cannot afford to do anything heavy weight until we have a
2867 * fairly good indication that there is work to do.
2871 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2873 int flags = (int)data;
2875 if ((vp->v_flag & VXLOCK) == 0) {
2876 if (VSHOULDFREE(vp))
2878 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2879 (vp->v_flag & VOBJDIRTY) &&
2880 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2889 vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2890 lwkt_tokref_t vlock, void *data)
2894 int flags = (int)data;
2896 if (vp->v_flag & VXLOCK)
2899 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2900 (vp->v_flag & VOBJDIRTY) &&
2901 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2902 error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2904 if (VOP_GETVOBJECT(vp, &obj) == 0) {
2905 vm_object_page_clean(obj, 0, 0,
2906 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2913 lwkt_reltoken(vlock);
2918 * Create the VM object needed for VMIO and mmap support. This
2919 * is done for all VREG files in the system. Some filesystems might
2920 * afford the additional metadata buffering capability of the
2921 * VMIO code by making the device node be VMIO mode also.
2923 * vp must be locked when vfs_object_create is called.
2926 vfs_object_create(struct vnode *vp, struct thread *td)
2928 return (VOP_CREATEVOBJECT(vp, td));
2932 * NOTE: the vnode interlock must be held during the call. We have to recheck
2933 * the VFREE flag since the vnode may have been removed from the free list
2934 * while we were blocked on vnode_free_list_token. The use or hold count
2935 * must have already been bumped by the caller.
2938 vbusy(struct vnode *vp)
2942 lwkt_gettoken(&ilock, &vnode_free_list_token);
2943 if ((vp->v_flag & VFREE) != 0) {
2944 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2946 vp->v_flag &= ~(VFREE|VAGE);
2948 lwkt_reltoken(&ilock);
2952 * NOTE: the vnode interlock must be held during the call. The use or hold
2953 * count must have already been bumped by the caller. We use a VINFREE to
2954 * interlock against other calls to vfree() which might occur while we
2955 * are blocked. The vnode cannot be reused until it has actually been
2956 * placed on the free list, so there are no other races even though the
2957 * use and hold counts are 0.
2960 vfree(struct vnode *vp)
2964 if ((vp->v_flag & VINFREE) == 0) {
2965 vp->v_flag |= VINFREE;
2966 lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
2967 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2968 if (vp->v_flag & VAGE) {
2969 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2971 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2974 vp->v_flag &= ~(VAGE|VINFREE);
2975 vp->v_flag |= VFREE;
2976 lwkt_reltoken(&ilock); /* can block */
2982 * Record a process's interest in events which might happen to
2983 * a vnode. Because poll uses the historic select-style interface
2984 * internally, this routine serves as both the ``check for any
2985 * pending events'' and the ``record my interest in future events''
2986 * functions. (These are done together, while the lock is held,
2987 * to avoid race conditions.)
2990 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
2994 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2995 if (vp->v_pollinfo.vpi_revents & events) {
2997 * This leaves events we are not interested
2998 * in available for the other process which
2999 * which presumably had requested them
3000 * (otherwise they would never have been
3003 events &= vp->v_pollinfo.vpi_revents;
3004 vp->v_pollinfo.vpi_revents &= ~events;
3006 lwkt_reltoken(&ilock);
3009 vp->v_pollinfo.vpi_events |= events;
3010 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
3011 lwkt_reltoken(&ilock);
3016 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
3017 * it is possible for us to miss an event due to race conditions, but
3018 * that condition is expected to be rare, so for the moment it is the
3019 * preferred interface.
3022 vn_pollevent(struct vnode *vp, int events)
3026 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3027 if (vp->v_pollinfo.vpi_events & events) {
3029 * We clear vpi_events so that we don't
3030 * call selwakeup() twice if two events are
3031 * posted before the polling process(es) is
3032 * awakened. This also ensures that we take at
3033 * most one selwakeup() if the polling process
3034 * is no longer interested. However, it does
3035 * mean that only one event can be noticed at
3036 * a time. (Perhaps we should only clear those
3037 * event bits which we note?) XXX
3039 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
3040 vp->v_pollinfo.vpi_revents |= events;
3041 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3043 lwkt_reltoken(&ilock);
3047 * Wake up anyone polling on vp because it is being revoked.
3048 * This depends on dead_poll() returning POLLHUP for correct
3052 vn_pollgone(struct vnode *vp)
3056 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3057 if (vp->v_pollinfo.vpi_events) {
3058 vp->v_pollinfo.vpi_events = 0;
3059 selwakeup(&vp->v_pollinfo.vpi_selinfo);
3061 lwkt_reltoken(&ilock);
3067 * Routine to create and manage a filesystem syncer vnode.
3069 #define sync_close ((int (*) (struct vop_close_args *))nullop)
3070 static int sync_fsync (struct vop_fsync_args *);
3071 static int sync_inactive (struct vop_inactive_args *);
3072 static int sync_reclaim (struct vop_reclaim_args *);
3073 #define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock)
3074 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock)
3075 static int sync_print (struct vop_print_args *);
3076 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
3078 static vop_t **sync_vnodeop_p;
3079 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3080 { &vop_default_desc, (vop_t *) vop_eopnotsupp },
3081 { &vop_close_desc, (vop_t *) sync_close }, /* close */
3082 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */
3083 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */
3084 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */
3085 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */
3086 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */
3087 { &vop_print_desc, (vop_t *) sync_print }, /* print */
3088 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */
3091 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3092 { &sync_vnodeop_p, sync_vnodeop_entries };
3094 VNODEOP_SET(sync_vnodeop_opv_desc);
3097 * Create a new filesystem syncer vnode for the specified mount point.
3098 * This vnode is placed on the worklist and is responsible for sync'ing
3101 * NOTE: read-only mounts are also placed on the worklist. The filesystem
3102 * sync code is also responsible for cleaning up vnodes.
3105 vfs_allocate_syncvnode(struct mount *mp)
3108 static long start, incr, next;
3111 /* Allocate a new vnode */
3112 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3113 mp->mnt_syncer = NULL;
3118 * Place the vnode onto the syncer worklist. We attempt to
3119 * scatter them about on the list so that they will go off
3120 * at evenly distributed times even if all the filesystems
3121 * are mounted at once.
3124 if (next == 0 || next > syncer_maxdelay) {
3128 start = syncer_maxdelay / 2;
3129 incr = syncer_maxdelay;
3133 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3134 mp->mnt_syncer = vp;
3139 * Do a lazy sync of the filesystem.
3141 * sync_fsync { struct vnode *a_vp, struct ucred *a_cred, int a_waitfor,
3142 * struct thread *a_td }
3145 sync_fsync(struct vop_fsync_args *ap)
3147 struct vnode *syncvp = ap->a_vp;
3148 struct mount *mp = syncvp->v_mount;
3149 struct thread *td = ap->a_td;
3154 * We only need to do something if this is a lazy evaluation.
3156 if (ap->a_waitfor != MNT_LAZY)
3160 * Move ourselves to the back of the sync list.
3162 vn_syncer_add_to_worklist(syncvp, syncdelay);
3165 * Walk the list of vnodes pushing all that are dirty and
3166 * not already on the sync list, and freeing vnodes which have
3167 * no refs and whos VM objects are empty. vfs_msync() handles
3168 * the VM issues and must be called whether the mount is readonly
3171 lwkt_gettoken(&ilock, &mountlist_token);
3172 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3173 lwkt_reltoken(&ilock);
3176 if (mp->mnt_flag & MNT_RDONLY) {
3177 vfs_msync(mp, MNT_NOWAIT);
3179 asyncflag = mp->mnt_flag & MNT_ASYNC;
3180 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */
3181 vfs_msync(mp, MNT_NOWAIT);
3182 VFS_SYNC(mp, MNT_LAZY, td);
3184 mp->mnt_flag |= MNT_ASYNC;
3191 * The syncer vnode is no referenced.
3193 * sync_inactive { struct vnode *a_vp, struct proc *a_p }
3196 sync_inactive(struct vop_inactive_args *ap)
3203 * The syncer vnode is no longer needed and is being decommissioned.
3205 * Modifications to the worklist must be protected at splbio().
3207 * sync_reclaim { struct vnode *a_vp }
3210 sync_reclaim(struct vop_reclaim_args *ap)
3212 struct vnode *vp = ap->a_vp;
3216 vp->v_mount->mnt_syncer = NULL;
3217 if (vp->v_flag & VONWORKLST) {
3218 LIST_REMOVE(vp, v_synclist);
3219 vp->v_flag &= ~VONWORKLST;
3227 * Print out a syncer vnode.
3229 * sync_print { struct vnode *a_vp }
3232 sync_print(struct vop_print_args *ap)
3234 struct vnode *vp = ap->a_vp;
3236 printf("syncer vnode");
3237 if (vp->v_vnlock != NULL)
3238 lockmgr_printinfo(vp->v_vnlock);
3244 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
3245 * (or v_rdev might be NULL).
3248 vn_todev(struct vnode *vp)
3250 if (vp->v_type != VBLK && vp->v_type != VCHR)
3252 KKASSERT(vp->v_rdev != NULL);
3253 return (vp->v_rdev);
3257 * Check if vnode represents a disk device. The vnode does not need to be
3261 vn_isdisk(struct vnode *vp, int *errp)
3265 if (vp->v_type != VBLK && vp->v_type != VCHR) {
3271 if ((dev = vp->v_rdev) == NULL)
3272 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
3273 if (dev == NULL || dev == NODEV) {
3278 if (dev_is_good(dev) == 0) {
3283 if ((dev_dflags(dev) & D_DISK) == 0) {
3294 NDFREE(struct nameidata *ndp, const uint flags)
3296 if (!(flags & NDF_NO_FREE_PNBUF) &&
3297 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3298 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3299 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3301 if (!(flags & NDF_NO_DNCP_RELE) &&
3302 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3304 cache_drop(ndp->ni_dncp);
3305 ndp->ni_dncp = NULL;
3307 if (!(flags & NDF_NO_NCP_RELE) &&
3308 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3310 cache_drop(ndp->ni_ncp);
3313 if (!(flags & NDF_NO_DVP_UNLOCK) &&
3314 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3315 ndp->ni_dvp != ndp->ni_vp) {
3316 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
3318 if (!(flags & NDF_NO_DVP_RELE) &&
3319 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3323 if (!(flags & NDF_NO_VP_UNLOCK) &&
3324 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3325 VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
3327 if (!(flags & NDF_NO_VP_RELE) &&
3332 if (!(flags & NDF_NO_STARTDIR_RELE) &&
3333 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3334 vrele(ndp->ni_startdir);
3335 ndp->ni_startdir = NULL;
3339 #ifdef DEBUG_VFS_LOCKS
3342 assert_vop_locked(struct vnode *vp, const char *str)
3344 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
3345 panic("%s: %p is not locked shared but should be", str, vp);
3350 assert_vop_unlocked(struct vnode *vp, const char *str)
3352 if (vp && IS_LOCKING_VFS(vp)) {
3353 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
3354 panic("%s: %p is locked but should not be", str, vp);