2 * Copyright (c) 2004,2013 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * Copyright (c) 1989, 1993
35 * The Regents of the University of California. All rights reserved.
36 * (c) UNIX System Laboratories, Inc.
37 * All or some portions of this file are derived from material licensed
38 * to the University of California by American Telephone and Telegraph
39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40 * the permission of UNIX System Laboratories, Inc.
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68 * External virtual filesystem routines
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/mount.h>
77 #include <sys/vnode.h>
79 #include <sys/eventhandler.h>
80 #include <sys/kthread.h>
81 #include <sys/sysctl.h>
83 #include <machine/limits.h>
86 #include <sys/thread2.h>
87 #include <sys/sysref2.h>
90 #include <vm/vm_object.h>
92 struct mountscan_info {
93 TAILQ_ENTRY(mountscan_info) msi_entry;
95 struct mount *msi_node;
98 struct vmntvnodescan_info {
99 TAILQ_ENTRY(vmntvnodescan_info) entry;
107 static int vnlru_nowhere = 0;
108 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD,
110 "Number of times the vnlru process ran without success");
113 static struct lwkt_token mntid_token;
114 static struct mount dummymount;
116 /* note: mountlist exported to pstat */
117 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
118 static TAILQ_HEAD(,mountscan_info) mountscan_list;
119 static struct lwkt_token mountlist_token;
121 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list);
124 * Called from vfsinit()
129 lwkt_token_init(&mountlist_token, "mntlist");
130 lwkt_token_init(&mntid_token, "mntid");
131 TAILQ_INIT(&mountscan_list);
132 mount_init(&dummymount);
133 dummymount.mnt_flag |= MNT_RDONLY;
134 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE;
138 * Support function called to remove a vnode from the mountlist and
139 * deal with side effects for scans in progress.
141 * Target mnt_token is held on call.
144 vremovevnodemnt(struct vnode *vp)
146 struct vmntvnodescan_info *info;
147 struct mount *mp = vp->v_mount;
149 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) {
151 info->vp = TAILQ_NEXT(vp, v_nmntvnodes);
153 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
157 * Allocate a new vnode and associate it with a tag, mount point, and
160 * A VX locked and refd vnode is returned. The caller should setup the
161 * remaining fields and vx_put() or, if he wishes to leave a vref,
162 * vx_unlock() the vnode.
165 getnewvnode(enum vtagtype tag, struct mount *mp,
166 struct vnode **vpp, int lktimeout, int lkflags)
170 KKASSERT(mp != NULL);
172 vp = allocvnode(lktimeout, lkflags);
177 * By default the vnode is assigned the mount point's normal
180 vp->v_ops = &mp->mnt_vn_use_ops;
183 * Placing the vnode on the mount point's queue makes it visible.
184 * VNON prevents it from being messed with, however.
189 * A VX locked & refd vnode is returned.
196 * This function creates vnodes with special operations vectors. The
197 * mount point is optional.
199 * This routine is being phased out but is still used by vfs_conf to
200 * create vnodes for devices prior to the root mount (with mp == NULL).
203 getspecialvnode(enum vtagtype tag, struct mount *mp,
204 struct vop_ops **ops,
205 struct vnode **vpp, int lktimeout, int lkflags)
209 vp = allocvnode(lktimeout, lkflags);
218 * Placing the vnode on the mount point's queue makes it visible.
219 * VNON prevents it from being messed with, however.
224 * A VX locked & refd vnode is returned.
231 * Interlock against an unmount, return 0 on success, non-zero on failure.
233 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount
236 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits
237 * are used. A shared locked will be obtained and the filesystem will not
238 * be unmountable until the lock is released.
241 vfs_busy(struct mount *mp, int flags)
245 atomic_add_int(&mp->mnt_refs, 1);
246 lwkt_gettoken(&mp->mnt_token);
247 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
248 if (flags & LK_NOWAIT) {
249 lwkt_reltoken(&mp->mnt_token);
250 atomic_add_int(&mp->mnt_refs, -1);
253 /* XXX not MP safe */
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 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
262 lwkt_reltoken(&mp->mnt_token);
263 atomic_add_int(&mp->mnt_refs, -1);
267 if (lockmgr(&mp->mnt_lock, lkflags))
268 panic("vfs_busy: unexpected lock failure");
269 lwkt_reltoken(&mp->mnt_token);
274 * Free a busy filesystem.
276 * Decrement refs before releasing the lock so e.g. a pending umount
277 * doesn't give us an unexpected busy error.
280 vfs_unbusy(struct mount *mp)
282 atomic_add_int(&mp->mnt_refs, -1);
283 lockmgr(&mp->mnt_lock, LK_RELEASE);
287 * Lookup a filesystem type, and if found allocate and initialize
288 * a mount structure for it.
290 * Devname is usually updated by mount(8) after booting.
293 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
295 struct vfsconf *vfsp;
298 if (fstypename == NULL)
301 vfsp = vfsconf_find_by_name(fstypename);
304 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
306 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0);
310 mp->mnt_op = vfsp->vfc_vfsops;
311 vfsp->vfc_refcount++;
312 mp->mnt_stat.f_type = vfsp->vfc_typenum;
313 mp->mnt_flag |= MNT_RDONLY;
314 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
315 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
316 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
322 * Basic mount structure initialization
325 mount_init(struct mount *mp)
327 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0);
328 lwkt_token_init(&mp->mnt_token, "permnt");
330 TAILQ_INIT(&mp->mnt_vnodescan_list);
331 TAILQ_INIT(&mp->mnt_nvnodelist);
332 TAILQ_INIT(&mp->mnt_reservedvnlist);
333 TAILQ_INIT(&mp->mnt_jlist);
334 mp->mnt_nvnodelistsize = 0;
336 mp->mnt_iosize_max = MAXPHYS;
337 vn_syncer_thr_create(mp);
341 * Lookup a mount point by filesystem identifier.
344 vfs_getvfs(fsid_t *fsid)
348 lwkt_gettoken(&mountlist_token);
349 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
350 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
351 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
355 lwkt_reltoken(&mountlist_token);
360 * Get a new unique fsid. Try to make its val[0] unique, since this value
361 * will be used to create fake device numbers for stat(). Also try (but
362 * not so hard) make its val[0] unique mod 2^16, since some emulators only
363 * support 16-bit device numbers. We end up with unique val[0]'s for the
364 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
366 * Keep in mind that several mounts may be running in parallel. Starting
367 * the search one past where the previous search terminated is both a
368 * micro-optimization and a defense against returning the same fsid to
372 vfs_getnewfsid(struct mount *mp)
374 static u_int16_t mntid_base;
378 lwkt_gettoken(&mntid_token);
379 mtype = mp->mnt_vfc->vfc_typenum;
380 tfsid.val[1] = mtype;
381 mtype = (mtype & 0xFF) << 24;
383 tfsid.val[0] = makeudev(255,
384 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
386 if (vfs_getvfs(&tfsid) == NULL)
389 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
390 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
391 lwkt_reltoken(&mntid_token);
395 * Set the FSID for a new mount point to the template. Adjust
396 * the FSID to avoid collisions.
399 vfs_setfsid(struct mount *mp, fsid_t *template)
403 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid));
405 if (vfs_getvfs(template) == NULL)
410 mp->mnt_stat.f_fsid = *template;
415 * This routine is called when we have too many vnodes. It attempts
416 * to free <count> vnodes and will potentially free vnodes that still
417 * have VM backing store (VM backing store is typically the cause
418 * of a vnode blowout so we want to do this). Therefore, this operation
419 * is not considered cheap.
421 * A number of conditions may prevent a vnode from being reclaimed.
422 * the buffer cache may have references on the vnode, a directory
423 * vnode may still have references due to the namei cache representing
424 * underlying files, or the vnode may be in active use. It is not
425 * desireable to reuse such vnodes. These conditions may cause the
426 * number of vnodes to reach some minimum value regardless of what
427 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
431 * Attempt to recycle vnodes in a context that is always safe to block.
432 * Calling vlrurecycle() from the bowels of file system code has some
433 * interesting deadlock problems.
435 static struct thread *vnlruthread;
440 struct thread *td = curthread;
442 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
446 kproc_suspend_loop();
449 * Try to free some vnodes if we have too many. Trigger based
450 * on potentially freeable vnodes but calculate the count
451 * based on total vnodes.
453 * (long) -> deal with 64 bit machines, intermediate overflow
455 if (numvnodes >= desiredvnodes * 9 / 10 &&
456 cachedvnodes + inactivevnodes >= desiredvnodes * 5 / 10) {
457 int count = numvnodes - desiredvnodes * 9 / 10;
459 if (count > (cachedvnodes + inactivevnodes) / 100)
460 count = (cachedvnodes + inactivevnodes) / 100;
463 freesomevnodes(count);
467 * Do non-critical-path (more robust) cache cleaning,
468 * even if vnode counts are nominal, to try to avoid
469 * having to do it in the critical path.
474 * Nothing to do if most of our vnodes are already on
477 if (numvnodes <= desiredvnodes * 9 / 10 ||
478 cachedvnodes + inactivevnodes <= desiredvnodes * 5 / 10) {
479 tsleep(vnlruthread, 0, "vlruwt", hz);
486 * MOUNTLIST FUNCTIONS
490 * mountlist_insert (MP SAFE)
492 * Add a new mount point to the mount list.
495 mountlist_insert(struct mount *mp, int how)
497 lwkt_gettoken(&mountlist_token);
498 if (how == MNTINS_FIRST)
499 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
501 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
502 lwkt_reltoken(&mountlist_token);
506 * mountlist_interlock (MP SAFE)
508 * Execute the specified interlock function with the mountlist token
509 * held. The function will be called in a serialized fashion verses
510 * other functions called through this mechanism.
513 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp)
517 lwkt_gettoken(&mountlist_token);
518 error = callback(mp);
519 lwkt_reltoken(&mountlist_token);
524 * mountlist_boot_getfirst (DURING BOOT ONLY)
526 * This function returns the first mount on the mountlist, which is
527 * expected to be the root mount. Since no interlocks are obtained
528 * this function is only safe to use during booting.
532 mountlist_boot_getfirst(void)
534 return(TAILQ_FIRST(&mountlist));
538 * mountlist_remove (MP SAFE)
540 * Remove a node from the mountlist. If this node is the next scan node
541 * for any active mountlist scans, the active mountlist scan will be
542 * adjusted to skip the node, thus allowing removals during mountlist
546 mountlist_remove(struct mount *mp)
548 struct mountscan_info *msi;
550 lwkt_gettoken(&mountlist_token);
551 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) {
552 if (msi->msi_node == mp) {
553 if (msi->msi_how & MNTSCAN_FORWARD)
554 msi->msi_node = TAILQ_NEXT(mp, mnt_list);
556 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
559 TAILQ_REMOVE(&mountlist, mp, mnt_list);
560 lwkt_reltoken(&mountlist_token);
564 * mountlist_exists (MP SAFE)
566 * Checks if a node exists in the mountlist.
567 * This function is mainly used by VFS quota code to check if a
568 * cached nullfs struct mount pointer is still valid at use time
570 * FIXME: there is no warranty the mp passed to that function
571 * will be the same one used by VFS_ACCOUNT() later
574 mountlist_exists(struct mount *mp)
579 lwkt_gettoken(&mountlist_token);
580 TAILQ_FOREACH(lmp, &mountlist, mnt_list) {
586 lwkt_reltoken(&mountlist_token);
591 * mountlist_scan (MP SAFE)
593 * Safely scan the mount points on the mount list. Unless otherwise
594 * specified each mount point will be busied prior to the callback and
595 * unbusied afterwords. The callback may safely remove any mount point
596 * without interfering with the scan. If the current callback
597 * mount is removed the scanner will not attempt to unbusy it.
599 * If a mount node cannot be busied it is silently skipped.
601 * The callback return value is aggregated and a total is returned. A return
602 * value of < 0 is not aggregated and will terminate the scan.
604 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction
605 * MNTSCAN_REVERSE - the mountlist is scanned in reverse
606 * MNTSCAN_NOBUSY - the scanner will make the callback without busying
610 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how)
612 struct mountscan_info info;
617 lwkt_gettoken(&mountlist_token);
620 info.msi_node = NULL; /* paranoia */
621 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry);
625 if (how & MNTSCAN_FORWARD) {
626 info.msi_node = TAILQ_FIRST(&mountlist);
627 while ((mp = info.msi_node) != NULL) {
628 if (how & MNTSCAN_NOBUSY) {
629 count = callback(mp, data);
630 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
631 count = callback(mp, data);
632 if (mp == info.msi_node)
640 if (mp == info.msi_node)
641 info.msi_node = TAILQ_NEXT(mp, mnt_list);
643 } else if (how & MNTSCAN_REVERSE) {
644 info.msi_node = TAILQ_LAST(&mountlist, mntlist);
645 while ((mp = info.msi_node) != NULL) {
646 if (how & MNTSCAN_NOBUSY) {
647 count = callback(mp, data);
648 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
649 count = callback(mp, data);
650 if (mp == info.msi_node)
658 if (mp == info.msi_node)
659 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
662 TAILQ_REMOVE(&mountscan_list, &info, msi_entry);
663 lwkt_reltoken(&mountlist_token);
668 * MOUNT RELATED VNODE FUNCTIONS
671 static struct kproc_desc vnlru_kp = {
676 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
679 * Move a vnode from one mount queue to another.
682 insmntque(struct vnode *vp, struct mount *mp)
687 * Delete from old mount point vnode list, if on one.
689 if ((omp = vp->v_mount) != NULL) {
690 lwkt_gettoken(&omp->mnt_token);
691 KKASSERT(omp == vp->v_mount);
692 KASSERT(omp->mnt_nvnodelistsize > 0,
693 ("bad mount point vnode list size"));
695 omp->mnt_nvnodelistsize--;
696 lwkt_reltoken(&omp->mnt_token);
700 * Insert into list of vnodes for the new mount point, if available.
701 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer.
707 lwkt_gettoken(&mp->mnt_token);
709 if (mp->mnt_syncer) {
710 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes);
712 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
714 mp->mnt_nvnodelistsize++;
715 lwkt_reltoken(&mp->mnt_token);
720 * Scan the vnodes under a mount point and issue appropriate callbacks.
722 * The fastfunc() callback is called with just the mountlist token held
723 * (no vnode lock). It may not block and the vnode may be undergoing
724 * modifications while the caller is processing it. The vnode will
725 * not be entirely destroyed, however, due to the fact that the mountlist
726 * token is held. A return value < 0 skips to the next vnode without calling
727 * the slowfunc(), a return value > 0 terminates the loop.
729 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp
730 * data structure is unstable when called from fastfunc().
732 * The slowfunc() callback is called after the vnode has been successfully
733 * locked based on passed flags. The vnode is skipped if it gets rearranged
734 * or destroyed while blocking on the lock. A non-zero return value from
735 * the slow function terminates the loop. The slow function is allowed to
736 * arbitrarily block. The scanning code guarentees consistency of operation
737 * even if the slow function deletes or moves the node, or blocks and some
738 * other thread deletes or moves the node.
744 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
745 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
748 struct vmntvnodescan_info info;
751 int maxcount = mp->mnt_nvnodelistsize * 2;
755 lwkt_gettoken(&mp->mnt_token);
758 * If asked to do one pass stop after iterating available vnodes.
759 * Under heavy loads new vnodes can be added while we are scanning,
760 * so this isn't perfect. Create a slop factor of 2x.
762 if (flags & VMSC_ONEPASS)
763 stopcount = mp->mnt_nvnodelistsize;
765 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
766 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry);
768 while ((vp = info.vp) != NULL) {
769 if (--maxcount == 0) {
770 kprintf("Warning: excessive fssync iteration\n");
771 maxcount = mp->mnt_nvnodelistsize * 2;
775 * Skip if visible but not ready, or special (e.g.
778 if (vp->v_type == VNON)
780 KKASSERT(vp->v_mount == mp);
783 * Quick test. A negative return continues the loop without
784 * calling the slow test. 0 continues onto the slow test.
785 * A positive number aborts the loop.
788 if ((r = fastfunc(mp, vp, data)) < 0) {
797 * Get a vxlock on the vnode, retry if it has moved or isn't
798 * in the mountlist where we expect it.
803 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
805 error = vget(vp, LK_EXCLUSIVE);
807 case VMSC_GETVP|VMSC_NOWAIT:
808 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT);
821 * Do not call the slow function if the vnode is
822 * invalid or if it was ripped out from under us
823 * while we (potentially) blocked.
825 if (info.vp == vp && vp->v_type != VNON)
826 r = slowfunc(mp, vp, data);
831 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
833 case VMSC_GETVP|VMSC_NOWAIT:
848 * Yield after some processing. Depending on the number
849 * of vnodes, we might wind up running for a long time.
850 * Because threads are not preemptable, time critical
851 * userland processes might starve. Give them a chance
854 if (++count == 10000) {
856 * We really want to yield a bit, so we simply
859 tsleep(mp, 0, "vnodescn", 1);
864 * If doing one pass this decrements to zero. If it starts
865 * at zero it is effectively unlimited for the purposes of
868 if (--stopcount == 0)
872 * Iterate. If the vnode was ripped out from under us
873 * info.vp will already point to the next vnode, otherwise
874 * we have to obtain the next valid vnode ourselves.
877 info.vp = TAILQ_NEXT(vp, v_nmntvnodes);
880 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry);
881 lwkt_reltoken(&mp->mnt_token);
886 * Remove any vnodes in the vnode table belonging to mount point mp.
888 * If FORCECLOSE is not specified, there should not be any active ones,
889 * return error if any are found (nb: this is a user error, not a
890 * system error). If FORCECLOSE is specified, detach any active vnodes
893 * If WRITECLOSE is set, only flush out regular file vnodes open for
896 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
898 * `rootrefs' specifies the base reference count for the root vnode
899 * of this filesystem. The root vnode is considered busy if its
900 * v_refcnt exceeds this value. On a successful return, vflush()
901 * will call vrele() on the root vnode exactly rootrefs times.
902 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
906 static int busyprt = 0; /* print out busy vnodes */
907 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
910 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
919 vflush(struct mount *mp, int rootrefs, int flags)
921 struct thread *td = curthread; /* XXX */
922 struct vnode *rootvp = NULL;
924 struct vflush_info vflush_info;
927 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
928 ("vflush: bad args"));
930 * Get the filesystem root vnode. We can vput() it
931 * immediately, since with rootrefs > 0, it won't go away.
933 if ((error = VFS_ROOT(mp, &rootvp)) != 0) {
934 if ((flags & FORCECLOSE) == 0)
937 /* continue anyway */
943 vflush_info.busy = 0;
944 vflush_info.flags = flags;
946 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
948 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
950 * If just the root vnode is busy, and if its refcount
951 * is equal to `rootrefs', then go ahead and kill it.
953 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
954 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs"));
955 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) {
957 vgone_vxlocked(rootvp);
959 vflush_info.busy = 0;
962 if (vflush_info.busy)
964 for (; rootrefs > 0; rootrefs--)
970 * The scan callback is made with an VX locked vnode.
973 vflush_scan(struct mount *mp, struct vnode *vp, void *data)
975 struct vflush_info *info = data;
977 int flags = info->flags;
980 * Generally speaking try to deactivate on 0 refs (catch-all)
982 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
985 * Skip over a vnodes marked VSYSTEM.
987 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
992 * Do not force-close VCHR or VBLK vnodes
994 if (vp->v_type == VCHR || vp->v_type == VBLK)
995 flags &= ~(WRITECLOSE|FORCECLOSE);
998 * If WRITECLOSE is set, flush out unlinked but still open
999 * files (even if open only for reading) and regular file
1000 * vnodes open for writing.
1002 if ((flags & WRITECLOSE) &&
1003 (vp->v_type == VNON ||
1004 (VOP_GETATTR(vp, &vattr) == 0 &&
1005 vattr.va_nlink > 0)) &&
1006 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1011 * If we are the only holder (refcnt of 1) or the vnode is in
1012 * termination (refcnt < 0), we can vgone the vnode.
1014 if (VREFCNT(vp) <= 1) {
1020 * If FORCECLOSE is set, forcibly destroy the vnode and then move
1021 * it to a dummymount structure so vop_*() functions don't deref
1024 if (flags & FORCECLOSE) {
1027 if (vp->v_mount == NULL)
1028 insmntque(vp, &dummymount);
1032 if (vp->v_type == VCHR || vp->v_type == VBLK)
1033 kprintf("vflush: Warning, cannot destroy busy device vnode\n");
1036 vprint("vflush: busy vnode", vp);
1043 add_bio_ops(struct bio_ops *ops)
1045 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry);
1049 rem_bio_ops(struct bio_ops *ops)
1051 TAILQ_REMOVE(&bio_ops_list, ops, entry);
1055 * This calls the bio_ops io_sync function either for a mount point
1058 * WARNING: softdeps is weirdly coded and just isn't happy unless
1059 * io_sync is called with a NULL mount from the general syncing code.
1062 bio_ops_sync(struct mount *mp)
1064 struct bio_ops *ops;
1067 if ((ops = mp->mnt_bioops) != NULL)
1070 TAILQ_FOREACH(ops, &bio_ops_list, entry) {
1077 * Lookup a mount point by nch
1080 mount_get_by_nc(struct namecache *ncp)
1082 struct mount *mp = NULL;
1084 lwkt_gettoken(&mountlist_token);
1085 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1086 if (ncp == mp->mnt_ncmountpt.ncp)
1089 lwkt_reltoken(&mountlist_token);